Upload
lamhuong
View
215
Download
1
Embed Size (px)
Citation preview
WORKING PAPERS
ADVERTISING INTENSITY MARKET SHARE CONCENTRATION AND
DEGREE OF COOPERATION
William F Long
Working Paper No 65
June 1982
nc Bureau rl Ecooomics working papers are preJiminary materials circulated to stimulate discussioa and critical commenL AD data cootained in them are in the pabk domain Dis indodes informatioa obtaiaed by the Ccmmiioo which has become part of pablic record The IDd coodnsions set forth are those rl tile authors aod do ad middot reflect tflt 1im of other members rl the Bureau of Economics other Commissioo staff or the Commission itseJC Uponrequest agle copies rl the paper will be provided References in pobUcatioos to FfC Bureau of Ecooomks working papers by FfC ecoooomts (other than adalowiedgemeot by a mer that be b access to such aupublished materials) should be cleared with the author to protect tbe tentative dlaracter of these papers
BUREAU OF ECONOMICS FEDERAL TRADE COMJrflSSION
WASillNGTON DC 20580
ADVERTISING INTENSITY MARKET SHARE CONCENTRATION
AND DEGREE OF COOPERATION
William F Long
The represen ta t ions and conclus ions presented herein are those of the
author and have no t been adopted in whole or in part by the Federal Trade
Commission or its Bureau of Economics The Assis tant D irector o f the Bureau
of Economics for Financial Sta tistics has certif ied tha t he has
approved the discl osure avoidance proc edures used by the staff of the
Bus iness Program to ensure tha t the data included in this pap r do
indiv idual company line of bus iness data
reviewed and
Line of
no t1 id entify
ith
ADVERTI SING INTENSITY ARKET SHARE
CONCENTRATI6 a n d DEGREE of COO ERATION
Wi l lia m F L o n g
The purpo ses of this study are to assess the role that advertising
plays in several explicit models of industrial organizationl and to formushy
late procedures for testing hypotheses which that asse ssment ge nerates
In addition a technique for the explicit introduction of the degree of coshy
operation in such models is used and the impact of cooperation is explored
I bull THE GENERAL SETTING
I assume a static industry with N firms By industry I mean a group
of firms that pro duce goods which are perceived as substitutes The goods
may be perfect substitutes but nee d not be Other goods outside the group
in questio n are assumed to be irrelevant
thUsing pi q and a for price quantity and advertising for the 1--i i
firm respectively the price of the goo d is assumed to be a function of all
quantities and advertising outlays ie
(1)
where bars under letters indicate vectors I define y as the elasticity ofij
the j goo ds price with respect to the goods advertising i e
thLetting c bull c (q ) be total production cost for the i-- firm i i i
and ll its pro fit i
(2)
-De fining some additional variab les let si
v bull a s be adve rtising intensity Define S bull rs as industry sales i i i i
ith
Page 2
A bull tai as industry adverti sing and zi bull siS as the market share of the
firm
II TWO POLAR CASE OLIGOPOLY MODELS
Cournot Let each firm maximize its profit independently of all other
firms That is
(3)
- 1
si- -- - 1 - 0yii ai
This implies that
-(4) v bull yiii
Let r bull (yij) Further let Ad be a diagonal matrix with the same diagonal
as the matrix A Then the N equations in (4) may be written compactly as
(5) v bull rd l
where is a vector of ones Since rd is a diagonal matrix (5) may be reshy
written as
(6 )
where if is a vector
z on the diagonal7
Chamberlin
3
v bull
d is a diagonal matrix containing the elements of z
4 At the other extreme let each firm maximize total profit
for all firms Letting IT trribull
an
Page 3
( 7 )
- 0
This implies that
(8)
(9)
-
v bull bull
bull
bull
zl [ Y21 Y22 bull bull Y2N J
-1zl 0
-1z2 r z
0 1
Page 4
Writ ing (8) compactly gives
the matrix in the two equationsA comparison of (6) and (9) shows that
differs only in that all the elasticities are present in (9) but only the
awn-elasticities are present in ( 6 ) Since advertising expense mus t be I non-negat ive the first-order requirements are that v be the maximum of t
i
the amount shown in 6) or (9) and zero
III A SIMPLIFIED DEMAND STRUCTURE
So far the demand equation system is comple tely general I will now
move to a particular demand structure which is charac terized by an industry
elasticity of price with respect to advertising a general brand-swi tching
elas ticity and a dependence of the firm-specific elas ticity on its relative
size I make the expli cit assumpt ion that
n pound
)l
0 AA
Consider the identification of y as an elasticity of industry price
Page 5
(10)
where y is an
- (y- a)
industry demand elasticity a is a brand-swi tching
el astici ty and z1 is the market share In matrix no tation
(11) r bull ai + (y- a) t
wi th respe ct to advertising 0
le t ting S denote dS observe that
0(12) SS bull
-
-1 0s t s - -1 0S t d - 9- -
apl apl aplaa1 aa2 a 0 a
middot( a Pl
( 13)
-
-0 a bull 0
a
) 0 X s
-1 0dAA bull then d
a bull
J 0
(JA) middot and
Let the sc alar (SS)(AA) be denoted by y so ya zft
If
1
d
-
I
I
(11)
--
y
I -f
l J- a)
a) l zt
Substitutin g into (13) gives
i [ ai + (y(14) -I- bull S A
bull az + (y -
Pa ge6
I I II
bull
since l bull 1
Under these assumpt ons then y is the elasticity of industry sales
(and average industry p ice) with respect to industry advertisin g Note
that if y bull 0 so that there is no industry sales effect there will still
be positive advertising if there are brand-switching effects
Turning now to the brand-switching ef fect observe that
(15)
Noting that
-1 deg -1d z -z -
-
0 a
Page 7
(15) may be written as
0 (AA)
lx 2x 1x
ith
1y
i
First assume no
that if there are
will be no changes in market shares
- 0a bull
Page 8
Equation (17) is a general statement of the relation betbulleen changes in
advertising levels and changes in market shares which would acompany them It
may be used to make some general observations about brand-switching effects
industry effect eg y bull r t bull 0 and assume further
equal percentage changes in all advertising le =ls then there
Under these further assumptions
0l and dz -1 bull (iA) ltr assumption the second term on the right hand side of this expression is
- By
zero
necessary for rt bull 0Consequently for market shares to be unaffected it is
when y bull r l bull 0 This requirement is satisfied by equation (11)
Secondly consider effects on market shares if one firm inc ases its abull ar-
tising and others do not As a general assumption we require smiddot ch a change o
lead to a decrease in the shares of all the other firms hat is let
0 0aiai gt 0 and a a bull 0 for j + i For any matrix X with N rows and its trans-j j
pose X let X
row of X
bull middotmiddotmiddot where is a column vector which contains the
elements of the Under the assumptions about the iis given just
d-1 0 0above it follows that r becomes (aiai) iimiddot ehFor jzj lt 0 under these conditions then it follows that the j- element
N of (I- l ) lt 0 so middotij - il lt 0 or yij - yik lt 0 If this
kbulll N condition holds for all Y ij where j + i then it follows that yii - z
k Y gta
k- 1 ik since the th firms share must increase to offset the decreases in all other
shares
ei ith
1v ( ) h i
7urning
where is the
leads to z
to the r matrix given in (11) we note that 1 = cr + (y - cr) i i zi elementary vector with 1 in the position and 0 elsewhere This
bull cr + (y - cr) zi Since bull required is thatcr at- zY wYij (y - cr)zi - cr + (y - cr) zilt 0 or cr gt 0
These observations may be summarized by substituting (11) into (17) bull
giving
(18)
i s
d -1 zz -
-
-
+ (y - cr) tz
d -1 0cr(I - l ) a a bull
cr 1 z - (y - cr) 0 a
o
If ai increases and the other a s are const t cr must be positive to havej
a positive effect on zi A positive cr also bull plies a negative ffect of aJ
f all a s in-on zi other a s (including a ) fixed j
eluding a1 change by the same percentage zi remains unchanged Finally k i
if cr bull 0 there is no brand-switching effect
For a to have no effect on p requires yij bull 0 From (10) thisj i
amounts to y bull cr that is the industry demand elasticity and the brandshyswitching elasticity are equal If y gt cr gt 0 the industry elasticity yij doninates If y lt cr y lt 0 the brand-switching elasticity dominates ij
making
Page 10
Substituting ( 1 1) now in (6) the Cournot case and in (9) the
Chamberlin ease gives
-(19) Cournot vO a + (y - a) z
Chamberli n vl - y lmiddot th
The 1- element is then
-(20) Cournot vO a+ (y - a) zi i
-Chamberlin vl yi
In (19) and (20) the superscript 0 is used to denote total
independence of decision and the superscript l is used to denote
total interdependence of decision making
The relation between v and vf depends solely on t relation between
y and a f y gt a so that the industry effect dominat 3 vf v with
equality in the trirlal case of z bull 1 Cooperative irms would have ai
higher advertising intensity than non-cooperative fi
If there is equality between the industry elasticity and the
y Whe therbrand-switching elasticity or y bull there is cooperation among the firms has no effect on advertising inten-sity And if brand-switching is more important with y lt a e get
vi lt v cooperative firms will have a lower advertising intensity than
non-cooperative firms The three situations are depicted in Figure L
Given the results for the firm level variables calculation of the
corresponding re sults for the industry is s-traightforward Let
B bull t zf bull z z be the Herfindahl index of concen tration and notethat
z - bull l z bull 1 and that V bull A S bull a i 5 bull (siIS) (asi)i
bull middot Then for the general case from (6) and (9)
bull
Page lOa
y lt1 bull y y
10
f
V0 bull z [ a1 + (y - cr) J bull a + (y - a) H vl bull Y bull y
11Page
( 2 1) vo bull z d -1 r bull r z bull z- d d- i )11
l -1 N N v bull z d r z bull r z bull l y z - ij jibulll jbulll
and for the simplified case from (19)
(22
Since the forms of the equations in (22) are the same as in (20) all
of the analysis of (20) holdS for (22) except that His substituted for
middot z Advertising intensity will be higher for an industry with cooperativei
firms than for one with non-coope rative firms if the industry demand elas-
ticity dominates the brand-switchine lasticity If the ewo elasticities
are equal then so will be advertiSl intensity for the cooperative firm
and non-cooperative firm industries If brand-switching dominates the
cooperative firm industry will have lower advertising intensity
In the industry model context the possibility of a critical concen-
tration level may be easily introduced That is assume that at values
of R below R the firms are non-cooperative but that for alues equal to
or greater than H they are cooperative That is
(23) V bull a + (y - a) H if H lt H bull y if H gtmiddot H
This function is shown in Figure 2 for the three relamptions be eween y and a
IV THE D GRE OF COOEtRATION
The models depicted in Figure 2 are based on a very simple notion about
the relation between concentration and the cooperationnon-cooperation charac-
teristic They use what is essentilly a step function relating concentra-
Figure
I -
J H H
Page l la
2
v v v
a
I J I y a bull Y y ---i
I I I I I I II I
H 10 H H 10 H 1 0
bullbull
bull bull
Page 12
tion and the extent or degree of coo peration as shown in Figure 3 Using
o as a symbol for the degree of cooperation eq uation (23) can be rewritten
as
bull(24) v (1 - o) vo + ovl bull (1 o) [a+ - a) H ] + oy- (y
vhere
(25) o bull o(H) bull 0 if B lt B
Hbull 1 if B gt
The specific relation between concentration (H) and the deg ree of
cooperation ( o) shown in Figure 3 is more restrictive than necessary and
bas less appeal an one which shows a smooth increase of o from 0 to 1 as
B increases fro to 1 Such a smooth function is shown in Figure 4 he
degree of cooper ion is low until H gets close to H increases greatly as
B goes through R attaining a high level for an H larger than B but still
close to it and then goes to 1 as B goes to 1 More formally
(26) 0 bull o(H)
bull 0 if H bull 0
bull 1 if H bull 1
do gt (IdB
d2o- gt 0 if H B dBZ
0 if H H
lt 0 if R gt H-
If (26) is now substituted for (25) in (24) a corresponding smoothing
of the functions shown in Figure 2 is accomplished he smooth functions
Figure 3
I
10
H 10 H
Page 12a
Figure
Page 12b
4
10
H 10 H
l
(27)
Cl - o) o
[ (1 -
[ (1
t y ij z
j
are given in Figure 5
Page 13
The equation in (24) was constructed as a hybrid of two results derived
from first-order equations for the non-cooperative and cooperative models
It may also be constructed directly by using l bull (1 - o) + on as thei i
th 2objective function for the i-- firm Treating o as a constant the first
order equation is
0 an aa
i middot
n i
Using equations (3) and (7) and setting --- bull 0 givesaai
(28) - 0
After collecting terms and converting to matrix notation we get
(29) - Cl - o) + 0 z-1vi yii i
N
j bulll
+ 0
+ o v1
d-1 r z z
-(30) v Cl - o) rd
-
o) r +d
d r J _ -1- d z
and middot
d) r + 0 rjd
l r z
v - z --(31) zv
- (1 0- +d) t r z- d-
If we now impose the simplified demand structure by substituting (11)
we get
- (1 - o) a +( 32) + oy and
Figure
l
-y
v
Y
H
r
I I
v
a
y
H 10 H
v
a
bulla
HH 10
bullbull
(33)
i
Page 14
v - (1 - o) [ a l + (y - a) ] The corresponding equation for Vis given in (24)
If the relation between industry advertising intensity (V) and
concentration (H) is non - linear via the impact of concentration
on the degree of cooperation does the relation take the form of a
auadratic as some other investigators have proposed (Greer ( 1971 ) Martin (1979) Strickland amp Weiss (1976) ) That question can
be approached from two perspectives - conceptual and empirical In
the next section I will try to apply some statistical material to it
here I want to explore the question in the context of the models
developed above
Referring to equation (24) taking the deriviative with respect
to H and letting bull (H) gives
bull( 3 3a ) 3V (y - a) (1 - o) + (1 - H) 3H dH
Given that o and H each are bounded by 0 and 1 and that o is an in-
creasing function of H the term in the brackets in non-negative The
sign of the deriviative then is the same as the sign of (y - a) If the industry effect is dominant (y gt a) then advertising in-
tensity will increase with concentration If the two effects are equal
then advertising intensity will not vbullry with concentration and if the
brand-switching e ffect is larger advertising intensity will decrease
with concentration
For given advertising elasticity parameters then it is not pos-
sible to show an increase in A S up to some H and the n a decrease
he relation may be non - linear but it is monotonic over the permis-
sible range of H
Page 15
One possibility is to introduce a relation between (y - a) and H
If industries characterized as dominated by the brand-switching effect
also were more concentrated then a cross industry comparison would
show an inverted U shape relation between AS and H I know of no
support for such an association between ( y - a) and H
V ERROR SPECUCATION
So far I have implicitly assumed that there are no errors in the
model I now assume that the error
thvariable in the equation for the i- firm is additive that its mean is zero
that its variance is inversely proportionate to its sales and that all error
term covarlances are zero hat is
34) v bull (1 - o) [ a + (y - a) z ] + oy + u i i i
- - -
In vector notation
(35) bull (1 - o) [ at + lt - a) J +
E() bull 0 Cov (u) -
0
If we now ca lculate V bull bull we get
oy + U bull
(36) V bull (1 - o) la + (y - a)R J +
U bull E(U) bull 0 Var (U) bull
P age 16
The error term specifi c ation which follows Hall
wi th the ratio of profits
6 heir analysis seems
as wellbull particularly given
and Weiss was
applied by them to equa tions to equity or
asse ts as dependen t variables to carry the same
weight in the present context the substantial
evidence presented by marketing analysis on the tendency of company de-
cision makers t o use the advertising to sales ratio as the decision
7variable
The zero covariance assumption on the other hand1 is particularly
troublesome Within a probably not true since
many even ts outside the
single industry it is
industry would have
industries
similar effec ts on all the
8firms in the industry Between it is probably not true either
since many companies produce in more than one indus try and even ts in the
firm would tend to effect all of its activities The extension of the
model to allow for non-zero covariances is certainly called for
--- -
Page Ll
Athird characteristic of this specification also deserves some comment
only on equation is shown for the firm or industry Several haveinvestigators
included an advertising equation ina larger system of equations (Comanor and
Wilson (1974) (1979) and Weiss (1976) so in myMartin Strickland Idid
dissertation1 Ihave ewo defenses for presenting only single equation results
here One is the suggestion by Comanor Wilson that simultaneousand equation
verybias may not be important in this context he second is that Iam will-
ing to see the model extended to include a profitability equation and I intend
to move that direction in the near futurein
Onthe inclusion of aconcentration equation on the other hand I amskep-
tical he problem isthat in some fundamental sense equations in models of
industrial organization should be oriented to the decision-making contexts of firms
Ifeach firm in anindustry with Nfirms has only one decision variable then
there can be only Nindependent equations which are based on the first-order
optimization equations It not show large ofis difficult of course to a number
functional relations which follow from the first-order conditions point here he
isthat only Nof them can be independent
Consider for illustrative purposes only a situation in wich teh N
itsquantities say q are set exogeneously and each firm determines advertising1
of and level say ai where a is the true optimal value a a bulla
1 1 i i + i
Ifp a) then the Nq
s and the Na s will determine N p s say pii bullp
i(l i 1 i
Now given Eand 9 we can determine both the vector s andthe vector v since
- bull til
ands We can also determine S s z bull s-1 s-ii
v and we
industry concentration isHbull (alternatively
V and Hare functions of the same
pendent Furthermore if there are
VM) and (H1 H2 bull bull bull )contain
VI STATISTI CAL APPLICATIONS
Page 18
Given the observed values of might be tempted to conclude that we
have a two equation system in which v and z are jointly determined That would i i
be inappropriate however since the 2N random variables (v ) are determined by
onlt N basic random variables t The variance-covariance matrix for ( ) must
be singular bull
The extension of this observation to industry level variables is straightshy
forward Since observed industry advertising intensity is V bull l v and observed 4
C4 bull z ) it follows thati l i
set of random variables t and cannot be indeshy-
M industries then the variables (V1 v2 at most M independent random variables
Given the assumptions made above y a and o are constant within a
given industry That being the case (34) can be written as
(37)
s1 bull (1 - o) (y - a)
If and
So + +
(1 - o) a + oy
81 are esti mated for an industry using a suitable statisticalSo procedure giving and 81 an estimate of y is then available The sumSo
Intra-industry analysis
1974 of Business Re ort -
categories are identified
Page 19
of So and S1 is y so y bull So + S1 is the estimate of y
If o is assumed to be bounded by zero and one the sign of 61 is also
the sign of y - a The estimate of So is of cou se an estimate of
(1 - o ) a + oy Moreove if y gt a then a is less than 6o- (1 - B) a+ 0(
Cor esponding results would
hold if y bull a or y lt a
If 61 gt 0 then we would estimate that a lt 6o
bull
For purposes of estimating this model 32 manufacturing industry categories
were selected The data are fram the Line of Business forms filed with
the Federal Trade Commission they are discussed in detail in the Annual Line
which is available from the FTC The 32 industry
in Table 1
Simple regressions for equ ation (37) were run for each of the 32 industries
with data for each of the firms which filed in the industry constituting an ob-
servation The results are given in Table 2 lines lA 2A
1974
the intercept
For all the A
equations generalized least quares was used th both and market
share bein scaled bv the square root of sales
The 32 industries were taken from the larger set of all manufacturing industry
categories in the 1974 data files Two selection cr iteria were used thereLB
had to be at least ten reporting companies and the industry had to be a consumer
goods industry or a producer goods industry which is strongly associated with a
related consumer goods industry so that adve rtising by the firms in the producer
goods industry impacts on demand in the related industry (eg soft drink syrup
and soft drinks) There were 135 industries with at least ten companies - 103
producer goods and 32 consumer goods
273
274
TABLE 1 INDUSTRY CATEGORIES USED
FTC
COOE OESCRI P T I ON _
2001 MEAT PACKINGt SAUSAGE S ANO OTHER PREPARED
MEAT PROOUCTS
2004 DAIRY PRODUCTS EXC F LUID MILK
2005 CANNED SPECIALTIES
2001 FROZEN SPECIALTIES
OEHYORATFOt AND
I NC LU 0 ING PRE SERVE S
SOUP MIXESbull
SEASO lNGSt ANO
OTHER PET FOOJ
RAIN MILL PROOUCTSt
ANO PREPARED
RELATED PRODUCTS
PRODUCTS
Z008 CANNEOt ORIEOt PICKLED FRUITS
ANO VEGET ABLES JAMS JELLIES DEHYDRATED I G TA3t E
S40CES ANO SALAD ORESSINGS
2010 DOG CAT A O
2012 FlOUR t OTHER RICE
MILLING fiLENOE) FLOUR
2014 B EADt CAKE ANO
20113 CO FECTIONERY
2026 BOTTLED A O CANNED SOFT DRINKS 2021 FLAVORING EXTRACTS AND SYRUPS EC
2029 MISC FOOOS AND KINDRED PROJUCTS EXC
ROASTED COFFEE
2301 MENS AND BOYS SUITS AND COATS
2302 MENS ANO BOYS FURNISHINGS
230 3 kOMENS AND MISSES OUTER EAR
2551 HOUSEHOLD FURNITURE
2702 PERIODICALS
Pqe 19a 1
RELATED 1972
src cooe
2011 f 3
202tX2026
2032
203pound
2033t4t5
2047 2041t4t5
2051 2065 2086
2081
209XZ05
231
232
233
251 272
2703 BOOKS
2704 ISC PVeLISHING
2806 ORGANIC FlEERS
2807 DRUG$ ETHICAL
23234
PT 2 S 3
2808 ORUCSt PROPRIETARY
2809 PERFU ES t COSMETICS ANO OTHER TaiLET
PREPARATIONS
2810 SOAP AND OTHER CLEANING PREPARATIONS 2815
2901
PEST1CIDES AND AGR ICULTURAL
PE OLEUM REF I N ING CHEMICALS EC
3606 HOUSEHOLD COOKING EQUIPMENT
3612 HOUSEHOLD APPLIANC ES NEC INCLUJING ELECshy
rqlC HOUSE ARFS AND FANS ANO SE I G MACHINES
3617 RADIO AND T V RECE I VING SETS 3808 PHOTOG APHIC E UIP amp SUP LfESt EXC
3qo3
PHO TOCOPYING
SPOR TING A D E QU I P M ENT f SUP P L I E S ATHLETTC GOODS NEC
3904 DOLLS GAMESt TOYS AND CHLOqENtS VE ICLES
PT283
2844
234tX2844
287 9
291 3631
3634t6t9
36H
PT386l
3949
15
943
middot
l A6l f 2 R(GRESSION RESULTS
EQ I NO
NR CODE cu 121
lA 2001 A (
2A 2004 r (
3A 2005 a
-R
ClS amp31
16
17
1 z
I TER-CEPT bull t
ooo9t21At 00022311 oooeos uu
OOl49tCAI 0010 2 middot 000960CCI
o049tl1At o04t55C61
MARKET SH ARE bull 5 I
-010670Ctl
-017600181
-001050
-0211AO
-0077 0
AOVER-TIS INti
lbl
000()39 000072
000018 0 0024liAI
-000079
SHARf
4ampgtVf Q bull
111
000140
-000170
RSQ
COl
66 485 862
607 6 3 t ()2
blti6 633
flASTI-
CI T Y PH
-007271(1
000441
-u0288l
clASTSHU (t)Ef
bull 10
tto
-234
210
( OOlltOCCI -126520 000922 181 001790
4A 200 7 18 00590liAI -038670 4e16 -032110 lell u c
OOOS74e 000700 025940
00067018) 0021001AI -02l100CAI
5 26
920
5A 6 c
200 8 26 0 03Sl9CA I 0012 26181 o 024511(
-005540
-0 81420 0001961AI O()Ol90CAI ooll240
441 604 891
-0020 19 274
6A 2010 16 Oe04224CAI o359lOIAI 9oq ll4015dCAI 061 6 -middot 005435CAI 00011 8 1AI 917 c 00535ltAI -009850((1 OOOA51J CAI -001610161 971
7A 2012 001862 000010 351 00187) oo1 a oooq45 ono1o6 439 c 001109 -027050 Oe0067SIAI -OOll60C81 897
- middotmiddot- -
8A 201t ll 000757 Ol4140CCI 814 ott9oocet 095 8 oooqo6 C C J 000069(81 811 c ootl2lampBI -080910CCI Oe009lUBI -OOtlOOCCI 960I
I DIIJIT pound11 1V OTIUTVTfAll-1pound1t- A _ 00 R - qq r - Q
1jl)OQCD
0
-middot middot--
l
_ __ _
I
i
0 00141 c c
91)7
I
T ABLE 2 REG ESSIOH RESULTS CCONTI
EO NR
IND CODE
iR bull
cos
I TEq-
CEPT
t1ARKE T
SHA E
AOVER-
Tl SING
SHARpound-cent
AuVER
RSO ILAST1-
C I T Y ELAST
SH CPtf (lt Cl) bull 11 I 41 I 5I (61 171 I 8 I f9) c 10)
9A 2 0 1 8 19 002397181 -007640 367 -005 246 le46 6 000605 000550CAI 597
-000029 G14180 OOl654CAJ -024li01At
lOA 2026 12 0028901AI -OlZl90 -0114 0 1 1 1 5 759
Ll (
llA 2021
f3 c
12A lOZ9
B
(
13A 2301
B
c
14A 2302
B
c
154 201
n (
16A 2551
8
10
35
10
20
19
24
001897(()
00141 6CCI
Oll9lHCI
010604 middot middot
Oll083
OOJ11JCAt
002059141
001411CIH
-000412
000293(81
000187
001159 (3)
ooo6a4cct
o00792CCI
OOl40dCAt
000006
-000076
OOlltl6fA)
o o0 5 64CCt
-034570
-o1middot o3o
-1659 90
Ol51lO
-046190CAt
Ol5450CA)
-005l60
002940
-030060(()
-0 43l50CCt
0 33650
-019080
00004
bullJ00741CAI
-000069
000466
0001 51CAt
000947CAt
O OOI)70CAt OOl750CA J
000175CCI
000671CAI
OOll5) f A t 002207CBI
0004b7CA I
- 007 1 70
001760
-001410CAt
- 0 16200CAt
-004210
- 1 207 70
705
96 0
542
466 805
460
614
921
832
9 61
996
424
550
9 15
56 1 651
884
-006114
019 037
03504lCA)
oo409Q
-141644((
491 -0 17664
607
871
lell
084
101
072
103
108 11IllOQfl)-- l
-037880fCJ OOl6l4CAJ -040l70fCI 0()
c
I
I
bull 9
9t
middot middot middotbullmiddot bull lraquo
TABLE 2 REGRE SION RESUlTS ICONTt
bull
EO I NO tHl I filER- f4A_K T AOVER- sbulluRE ttSQ flA Jl- iLAS T NR CODE cos CEPT SHARE JISING AOVfR bull (JTY SHR CiJCF
110Cl I f amp Itt 15J tbt bull 1c u caa
llA 2702 ll o079 6 81 Bt -019110 S6s -OlltO 111
a 004679 ooooo1 t69
( oOl26o -015160 001560 -022690 ltJr()I
ldA 270 3 19 0065lt -OIt2l60 220 -Ol57a2 11 8 e o 0 71t6IA t ol08 75tAt ( OOl983CiU -060llOC() OOl7191A t -O lltdOC A) middot 89
lqA 2704 10 005082 -Oil060 22b -l)i1977 lelO 6 -000080 oo2772C4 t 910 c 000907 -011 200 Oe04105CA t -046960 995
20A 2d06 1 II
O OlltStiAt -001120 837 00014 -))th O01380amp At -oooo 11 821 ( OOl526CAt -026700181 0002 5UAt 000530 969
2lA 2807 28 OOt9lt9CAt -00891 0 516 -0 03962 225I8 002095CCt OOOl lliAt 690 ( o02965CA t -056650((1 OOOt50CAJ -003l70ttl 911I
I 22A 2808 15 ot3l06 119730 655 1130) 090
0 - 007526CC t Oe003961At 835- middot-
c I oo7523 -067210 001l 2tA t -0091t80CCt 951
2lA 2809 21 Oe2067lCAt -1218101At 71 2 -10095((1 121 ttlIaOQ8 00797418) OOOI8lltt 686 Ill( 0092lU A t -Olltl70Ct t 000605CAt -OOit8701At 951 _
---- -
32 009506CAt
0 bull 06 981 I A J 006901 cA bull
00 24A 2810 -010960 5t 5 -OOltSO 7 56
8 -000011 s11 c -5811701At oooaoatAJ oot520IAI 888
-- 37 --- middot - -- -1 ) l lbull t5JJ r-l bullhmiddot1 --- - --
l5A
26A
27A
c
2RA
c
2 9A
31A
llA
24
13
597
Ci)55
14
l 6h
RSQ ELASTI- (lASTCOE f-
TABLE 2
EQ I NO NR CODE IU
REGRESSION RESULTS CCONTa
NR INTER- MARKET
(10S CEPT SHARE AOVER- SHARfcent
TIS lNG AOVER
) CITY StiR
c It) bull 5 co a I 1) bull 0 bull 9 c l 0 bull( 2 1
2815 16 003019 -009990 275 -o069bmiddotJ
6 001446 000224 282 c 002178 -054630 o02221CCI -0075 00 646
2901 29 00021t2 CAI 0)0520 592 0007 6 1 068
0 00105C8 t 000012141 60)g( OOOlJ51AJ -00126018) ooou271At -000u80CA) 974
36oa 1 3 003296(4) -oca29o -oo-q94
001755 000203 lt91t 002499((1 -04l820CCI OOl374CAI 0068)0 946
3612 0033281CI 011520 195 01684) oso
00 30 42C OI 00012 2 424 oo579ltAt -2 7l7801At 002 100IAI -003220 819
8 3617 btel 011789 081002259 009530
001448 oOOl OitCCI 732 OOl7 22CAI -0 46160CAI 0004961AI -OOll40181 989c
30A 3808 14 0021t 3 6fAI -000110 628 o023291C) -005 6 middot- _ middot- 0bull 02 04 1 IC I ooooo9 637 ( 001584((1 -0161)1)0 00021 91AI -0 00060
390) 14 002790CCt 006590 h8 t 0 0938 4 070
B OOl89bCCJ 000144 74
0021961(1 -OItlOlO OOl323CAI -004800 915
B
middot--- - middot 4 - middot- - dfll3904 12 004563181 0)8300 879 0428651(1 089 OQIll004 277CA I ooozstcea 919
c 0041)3181 -0-11250 0011881(1 -006270 972 0Ill
- J - bull bull bull bullbullbull middot - - middot - middot gt middotbull L)ofbullflbull middot middotfJltl -tn- middotmiddotmiddotmiddoto +f middot t middot bull
c
is
positive
2
Page 0 determineThe consumer goodproducer good split is sometimes difficult to
Sollle support for the split used in this paper is a comparison of goodness - of - fit
distributions of a
industries
set
advertising are
that industry
a sales increase of
the
highest aD10U3 those
measures for the two groups of industries Table 3 contains
measures for the 32- consumer goods industries and the 103 producer goods
2 Median R is 5 7 for the consumer goods set and 40 for the producer goods
Estimates - of y the elasticity of demand with respect to
given in column of Table 2 The hypothesis that y bull 0 with
alternative was applied for each ittdustry As shown in the
could not be rtjected for 24 of the 32 industries Of the
and 10 are positive For the remalnifg eight three are
and five are significantly positive
Of the threertegative elasetcities the largest (in absolute value)
cosmetics (code 2809) With all elasticity of -1 01 a one
industry advertising would gwerate a decrease in industry sales
more than one percetu
At the other extrema is toys and games (code 39 04) For
a one percent ittcrease in industry advertising would lead to
four-tenths of a pampreentmiddot Though not significantly different from zero
elasticity estimatbull for pt-bpuetary drugs (code 2808)
which are
Both drug industries have
to sales rafios (211 pe1 cent for 28 08 and 13 4 per cent
other hand flaIJo ing ext1ac s (mainly soft-drink SYUlS) also
is for
percent increase in
of slightly
a middottwo - tailed
table that hypothesis
24 14 are negative
significantly negative
9
the codetics and p-ropriacuy very high advertising
for 28 09) bull On the
has a high ind UY
advenisiltg t() sales ratio ( 8 0 per cent) bull but its estimated elasticity is very
small and not signifieantly different uom zero
middot It doe$ not seem unusual to find many elasticities near zero with a few
negative and a few positive If aggregate consumption is insensitive to aggregate
advertising as seems likely ld then what one industry gains must be lost by
others
Page 21
elasticity of demand does
(10) we can
if t 0 then
Finally a determination concerning the industry
not fix the individual firms elasticity Going back to equation
write y bull (1 - ) a + z y Since a gt 0 it follows thatii i i
Y gt 0 If Ylt 0 however y s sign is indeterminateij ij
Turning next to the coefficient of market share 20 of them are negative and
12 are positive with three of each sign being significantly different from zero
The largest positive value which is significant is for pet food (code 2010) and
a close second is mens amp boys suits and coats (code 23 01) The largest signifi
cant negative value is for cosmetics (code 28 09)
In the model some relations among y a 0 and 61 are implied If middota is
negative (1- y) is greater than one Since 61Y bull (1- a) (1 - ) 61Y is
pCsitive but it will be less than one if 0 is large enough Seventeen of the A
ys are negative in each of the seventeen industries 61 is also negative and les
1 (algebraicly) than y so 61Y is greater than one his is not evidence that the
degree of cooperation is low or zero in any industry of course A high o togethegt
with a negative y which is small (in absolute terms) relative to a can give a
value of 61Y greater than one On the other hand if 61Y were less than one
o would have to be greater than zero
For y gt 0 two conditions may hold If y lt a it follows that 61 lt 0 since
(y - a ) lt 0 It also follows that 61Y lt 0 regardless of the value of o bull
In addition the relation is an if-and-only-if one ie if 61Y is negative
then y lt a Three industries have negative values for 61y dairy products
except milk (code 20 04) synthetic fibers (code 2806) and photographic equipment
(code 3808) In all three cases y is quite small so a does not have to be very
large to give a negative value to al
If y gt a the model shows that 61-- gt 0 whatever the value of 0 bull There are
12 cases where both y and 61 are positive with 61 ranging from near zero
(grain milling code 20 12) to 1 20 ( proprietary drugs code 2808) Several
coeffiei-en_rs
--
15
o f to zero
Tab le 3 - Consumer Goods
Indus tr ies
2Distributions of R and Producer Goods
Page 20a
Consumer Producer
Goods Goods
0 - 9 9 0 3
10 - 19 9 0 1 1
20 - 29 9 3 20
30 - 39 9 3 17
40 - 49 9 5 16
50 - 59 9 8
69 9 6 Ii 1060
--f 670 79 9 2
80 - 89 9 4 4
90 - 100 1 1
Totals 32 10 3
Notes
2is used R i s cal-
culated as F F + ((N - K- 1 ) K ] 1 b_ F is the F
statistic Lor the hypothesis that all o f the non - intercept terms are simul taneously equal
1
2
--
-- Since generalized least squ-area regress ion
Page 12
o f these cases have values o f 6 1 Y rea ter than 0 8 (pet food (20 10) bread amp cakes
(20 14 ) miscellaneous processed foods (20 29 ) men s and boys suit s and coats
(23 01 ) p rop riet ary drugs (28 08 ) elect ric housewares (36 1 2) radio and
IV sets 17 and games and toys (39 For be than(36 ) 04)) 6 1 Y to greater 0 8
both o and cry mus t b e less than 0 2 That is in these industries the results
shown in Tab le 2 would hold only with very strong dominance of marke t demand
effects over brand swit ching e ffects and a very low degree o f cooperation
Using the concentration rat ios given in Table 4 the average concentration
ratio for the eight indust ries is 38 1 For the remaining 24 industries it is
43 2 This result would lend some support t o the proposition that concentrat ion
and cooperation are related if the group of eight industries are identi fied
as having low degrees of cooperation
One f inal note on model predict ions S ince both cr and o are assumed to
be non-negative 60 should also be non-negat ive In only one o f the 32 cas es
is 60 negative and it is not signif icant bull
cor pJ
7 1
7 0 4
1 4 7 5 1
)
7 3 z a o a t s 1 1 3 3 5 ( i V Z 8 tJ1 2 1 - 1 1 0 7 3 5 c middot v 2 8 1 0 3 2 -0 0 1 4 3 0 C 1 V 9 8 3 2 8 1 5 1 6 - 0 0 6 t1 6 9 C _ V 1 11 0 2 9 0 1 29 0 00 7 6 1 C l k V 7 3 1 3 6 08 1 3 - o n 4 9 9 4 3 3 3 6 1 2 2 4 0 1 6 4 3
3 1 7 1 3 0 1 1 7 d 7 8 3 3 8 0 8 0 0 2 3 i 2 3 0 6
P 4
3 9 0 3 1 4
1 2
0 09 3 8 4 39 1)1 0 4 l 8 6 5
T A LE 4 e I NCU S T R Y C A T E G O R Y
F TC NOM e tOOE ass
C 1 I c 2 )
2 0 0 1 1 6 2 0 0r 1 7 zo o l Z 2 0 0 7 1 8
z o o 2 6
z o t a 1 6
2 1) 1 2 1 5 2 0 1 4 1 1 2 0 1 a 1 9 2 0 26 1 2 2 0 2 7 1 0 2 0 29 - 3 5
2 3 0 1 10
2 3 02 z o
2 3 03 1 9
2 5 5 1 2 4 Z 1 0 Z 1 1 2 0 3 1 9 2 7 4 1 0 z a o 1 3
2 A 0 7 z s
E LA S T I -C I T Y
c 3 )
- 0 09 7 2 7
0 00 449
- 0 0 2 8 9 3
- 0 3 2 7 70 - 0 OZ il I q
o 4 u t s a 0 0 1 3 7 3
o 1 4 o o-0 0 5 2 4o -0 1 94 0 1
-C l 6 l l 4 0 1 JJ 3 7 0 3 t 0 4 2 0 0 4 0 9
- 0 4 1 8 44 - O l 7 t ) 4 -ll 7 1 4 0 b -0 3 5 7 8 Z
- 0 1 9 7 7 0 1)0 3 3 4
-o n 9 2
IJ A T A
CONV
ON-C ONV
( 4 )
t 1NV
C 1t-V C NV c ) v C - JNV C rlNV c n v c a v C ON V ( i) l V
C CtV
C t1tV C IJNV
C 1 V
c 1 v
Pa ge 2La
HEQ F t Noe x
c 5 )
6 2q
6 89 1 48
6 1 4 ) q 3
1 9 7 1 9
q z t s 5 a e z
l l 9 1
4 o 7 1 2 0
c oqt 1 Z
4 4 )-q 1 z
5 5 9 1 0 64 1 3 3
n 3
Inter industry analysi
Page 23
Equation ( 3 6 ) may be used to fo rmulate a rough test o f the cooperat ion
vs e f ficiency controversy I f we rest ricted ourselves to some subset o f
indus tries where the relat ion be tween y and a i s the same we would be lookingbull
at a sample o f indus tries for which the dependence o f V on H and o is a
quadratic wi th no squared terms
I f we now sub s titute (26) into ( 36 ) we get
(38) v bull a + (y - a) R + (y - a) o (H) - (y - a) H o (H) + u
If it is only d if ferences in marke t shares that affect advertising inten-
s i ty levels i f concentration does not af fect cooperation and i f o bull 0
( 38 ) reduces t o
( 39 ) V bull a + (y - a) H + U
Let this b e the null hyp o thes is and note that V is a linear func-
tion o f H If y gt a and the null hyp othesis holds the expected relat ion
between V and H is as shown by the pos itively sloped straight line in Figure
6 b etween points A and C
In this context the alterna ive hypothesis the degree o f
cooperation ( o) is a positive function o f concentrat ion
is that
(H) I f the func-
tion is as described in (26) and Figure 4 the relation between V and H
would be as shown by the curved line connec t ing po ints A and C in Figure 6 One way to test the null hypothe s is against this al ternative is to test fo r
the hypothesized curvature in the function relating V and H
Fi gure
R
6
v
ID c
H 1 0
I
i I I
sfigned
clasJ s
in def ini g
effec t is the
I
Page middot2lf
One charac teristic of produc ts which may be useful
classes wi th respec t to significanc e of the brand-switching
conveniencenon-convenienc e distinc tion developed by Porter ( 1 97 )
Using his identif ication of IRS indus tries as a s tar t ing point
each of the 32 industries used in this s tudy to one of the two
The ass ignments are not ed in Tab le 4
Using the conveniencenon-convenienc e dis t inc tion the
divided into two sub-samples
sample was
Four regressions were then run bull ene for e ch
sub-sample one for the to tal sample and one for the total sample with
a dummy variable for convenience goods Each regression was run wi th LB s
and wi th indus tries as ob servations The results are given in Tab le 5
For the LB level equations ( 1A - 1D) equation(37) was used ome
rearranging of terms gives v bull y - ( 1-o ) (y - a) ( 1-z ) + ui i i
bull y - 6 1 ( 1 - z ) + ui When observations are pooled across indus trif s yi
is an industry variable In these regressions the values o f y es t imated
in the first stage were used Since industries have been grouped so as
to reduce differences in (y - a) that term is t reated as a cons tant
within each group Finally H the Herfindahl index has been subst ituted
for o the degree of cooperation The resultant equation is linear
y ( 1 - z ) and H x ( l - z ) This is the equation for whi ch results arei i
reported in Table 5 In all regress ions all variables were weighted by the
square roo t of sales
in
raquo
- 34
3 1
i l T 4 8LE S R E G E S S I ON R E SUL T S i X T f N S I ON S T O T Y PE Of
AND L E V E L OF C ONC E kT R A T I O
E O O tt S
tP bull T Y P E NR C ONV O B S tO N-CONV
E L A S T I shy
T I C J T Y
O E - t-41( T SliA R E
H( fl F
1 1 - K O x middot K O X E R F
- H S a 1 1 - H S t 1 8 1 1 9 1
R S O
1 1 0 1
q z
t - bull 9 f 2 1
Z t P D l 6 fC ONV - 1 9 1 s s c 2 1
e - l O I e bull u q a
2 3 1 0 1 1 Nl N- t O V 5 7 1 ( 1 2 19 1 6 1 - J l S 2 5 4
l l l l t z q o a 1 - J o v z I
l ( I NO l l A l l ol u l t o o c bull J u l l
d S
c 3 4 t t r a l
bull Z D 1 0 l l 4 l l 2 3 -gt 2 l l 2 0 - 4 lS 5 4 l C 4 1 o e 6 t l b b 5 1 1 0 1 1 1 c - 1 9 5 1 t d S 2 1 tV
OQ10 T-R A T I O S A E G I VE N l P A RE N T H E S E S l E V E L S f S I G 1 F l C ANC E A - 99 5 B - q 9 t - s NS J UC E NO I Nl E R C ( P T I S U (j f middot) I N f Hf t OUA TIOI S 6 E f 3R E A O J U S T H E N T F OR 1-t E T E R O SIC E O A S T f C I T V pIraquo
1
middot
I
R SO I S C A L C UL A T E D A S f I f bull 1 1 bull 1lt 1 K t t- E E N I S THf U A1 B E R O f fJe S E fi V A T I O S F t S HtE F - S T A T I S
I N DE P E ND E N T y II
T I C F O T H E U Y P O HIE S I S f A l l C JE F F I C H N T A R E QU A L HI l f RO A f40 K I S T H f UUM 8 E R OF Q ( A fL f S
I l l
1 A
1 6
t C
f 1 r R I l l
1 2 1
L Jt
L B
l q
L 1
on s T Y P E bull 2 )
P OUC T
X
H S I l l
l O l A I bull l 0 7 1
O O b O J
1 2 00 1
1 b 2 1 C I l l Z U
l l S b 1 -1 1 0 1
rlt F X l l - fi C f l
bull 1 I
l l
- ()4 2 3 l f H
1 - 2 7 9 1
K O X
1 1 - ti E R F I
l d l
S4 l i A I 2 l b
1 3 00
K O X -t f R F t S J
I 1 - HER F t
)
60 9
z o bull
0
t l l
3 8 8
1 7 6
5 64
64
N f bull
tl S bull
I l l
I 4 1
t ONV
NON-t ONV
A l l
A ll
c o v NO- ( ONV
1 4 1
1 5 1
- l S b
c - 1 1 2 1
5 8 2 C A I 1 6 7 0 1
middot I l 9 H
l 1 il
1 4 1 6 1
E L A C) T J shy
l t C I T Y
I S I
l b l
- 0 2 0
1 - 1 4 9 1
O Z 3 2 1 A I C b 6 b l
0 0 9 2 -J
1 1 b 4 1
o z t t t l 1 2 2 1 1
oru - Ht tt F
1 gt X bull b I
Page 25 bull
Mean val -as ior relevan t variables for the two sub-samples are
Conv Non-Conv
Adv Sales 4 7 2 3
iit Share 2 7 3 2
Ela cicity - 05 3 02 2
Herf indahl 7 3 1 860
That the dist nct ion between coft9enienee goods and non - convenience
goods mat ters is supported not only by these means but also by the regression
results Equat ions lA and lB are different in almost every respec t Only the
concentrat ionmarket share int eract ion term seems to mat ter for convenience
good s Jus t he opposite holds for non - convenience goods both elasticity1
and market sl e are highly significant but the concentrat ionmarket share L
int erac t ion t=rm is not
Given that one minus market share is used these equ3 tions the coeffici t
of market share is the negative of what is sh in col (6) of Table 5 For the
regress ion for non - convenience LB s reported in equation lB then LB s with larger
market shares have lower advertising to sales ratios
On the question of the impact of the degree of cooperation the story is mixed
For convenienc good s the coefficient is pos itive as equation (31) predicts when
there is a po sitive degree of cooperat ion Fo r the non-convenience goods LB s middot
concentration has no effec t
Aggregat ion t o the indus try level and then rer unning the equations $ives the
results shown in (2a-2d) of Tab le 5 As expected R2 goes up and levels of significanc e
fmiddotJr individual coefficients gJ down In add ition the coeffic ient of H ( l -H) inmiddot
quat ion 2A is no t significant though it is po s i tive Since 2A is the indus try level
I
--
all but one of them and that one is not significant For 18 of the 31
wo uld have on its adve rtising to sal es ratio
cognate of LA the variable H ( l-H) pl ays the same role in 2A as H ( l -z 1) p lays
in LA
Vconomies of scale in advert ising
The literature on advert ing contains a number of definitionbull o f economies
of scale and of findings eone rning its presence and magnitude Before I loo k
a t the data with the hope o f mmenting about scale effect s then I wan t to
define the term as I use it
By economies of scale I mean a decrease in the number o f units of adver-
tising which are needed to generate a unit of sales as advert ising is increased
Given the definition of the firm s own elast icity of demand with respect to
advertising (see the text foll wing equation (1) ) the technical definition
is that rii gt 1 Diseconomies of scale are def ined symmetrically i e
Yu lt 1
ith a cro s s - section oi observations on advertising intensity
(v bull a s1) and onmiddot advertising (a ) we may regress the former on the latter 1 i i
The results of such regress ions for the 32 industries are given in Table 2
lines lB ZB bull bull bull 32B For the 32 eases the regression coefficient is positive it
positive coefficients the coefficient i s signif icant
The que tion now is whether these results ar e evidence of pervasive dis-
economies of scale in advertislng I think no t Given my as sumption that the
data I observe are equilibrium results I may ri ghtly conclud e that almost every-
where high levels of advert ising are as sociated in equilibrium with high ratios
of advertising to sales I may no t conclud e anything however about the
impact that a move by some firm away from its equilibrium level o f advertising
middot- Page 27
Given the assumed relation in equation ( 37 ) there is a corre spondshy
ing equilibrium equation relating v and ai The equation is non - linear i
and I have not found a simp le way to characterize it It is po ssib le to
det ermi ne its deriviative with respect to a b y takin$ tne de riviative of ( 3 7 ) i
When that is done the result is
v - 1(40) i bull sl a a z ( 1-z )i i ira-l
Sinc e all the other terms in (40) are non-negative the sign dep ends only
on the sign of 6 1 which is in turn de pend ent only on the sign of (y - a)
That the signs of the coefficients of a in equations lB 2B bull bull bull 32B arei
not always equal to the signs of the coefficients of z in LA 2A bull bull bull 32A i
is still a mystery to me and something to be explored
bull
Summary
My purpose in writing this pa per wa s t o attempt to integrate explicit micro-
economic mode lling reasoned econome tric specification of error terms and high
quality data to exp lore some questions about adver tising by large firms
al quite important improvements could be made in the first two of those areas
and much data ma ssag ing is yet to be done in the third
Concerning the subst an ce of what I have done I think six things are im-
Page 2 8
Severshy
V and Conclusions
portant
1 Explicit modeling is worth the ef for t if for no other reas on than
that it provides a basis for choosing fr om a varie ty of funct ional
forms
2 The same is true for error specif ica t ion I did in fact look at scatshy
ter diagrams for all 32 industry categor ies and they vir tually all show
the het eroscedast icity which I as sumed and for which I correc ted
3 The predicted relation be tween advertis ing in tensity and market share
shows up clearly in on ly 25 of the case s examined I have ye t to
explore why that may b e the cas e
4 Some evidence concerning the presence and impact of coopera tion was
produced but it is not clear ly pervasive
5 The dis tinction between convenience goods and non-c onvenie nce goods is
unambiguously a good one
6 Virtually no evidence concerning econ omies of scale in adver tis ing can
be gleaned fr om this study given its assumpt ions
More work is called for on many of these issues my study seems to have
raised more questions than it has answer ed This is pr obab lly due to the richshy
ne ss of the data source since I had the oppor tunity to address several issue s
at once
- -- - Q --
Foo tnotes
Manager Line o f Business Program Federal Trade Commission The author depended heavily on several staf f members of the LB Pro gram for statistical and clerical support in particular Joe Cholka George Pascoe and Haro lene Jenkins Helpful comments were given b y Richard Caves Dennis Mueller Michael Porter and F M Scherer When the paper was in its early stages o f development both S teve Garber and Jon Rasmussen former economists in the LB Program provided eery useful cri tiques The views expressed here are my own o f course
1 The basic exp licit model of the role of advertising is Dorfman and S teiner ( 1954 ) and the models in this paper may be seen as ano ther extens ion of their work
2 For a review o f models of oligopoly which treat prices and quantities as relevant variables see Shubik ( 1 959) Extens ions to take advertis ing int o account are fairly s traightforward
3 Dorfman and S teiner dealt with a price-setting firm For a monopolis t inverting the demand relation to show Pbull P (Q A) _instead o f Q- (P A) has no effect on the results o f the analysis For oligopolists who face a sys tem of demand relations inversion does have some impac t on the results The dif ferences are not trivial nonetheless they will no t be explored in this study
4 Shubik ( 1 959) p
5 From a formal mathematical point o f view the function is similar to a1 i Legrangian funct ion of the form - r rr
L bull 11 + A ( j - (i i i+j
where the indicates some fixed level of the profits industry Following the same logic which resul ts in and income constraint as the marginal utility of the of the indus tr in the obj ective of firm i
6 Hall amp Weiss ( 1 967) p 323
7 See Schmallensee ( 1 97 2 ) Ch 2 pp 1 6-47
8 On this poin t see Imel amp Helmberger ( 1 9 7 1 )
9 Long ( 1 970) p 130 - 1 39
1 0 See Schmallensee ( 1 972) Ch 3 pp 48-8 7
rrj i+j ) of the rest of the
the identification profits of the rest
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980
ADVERTISING INTENSITY MARKET SHARE CONCENTRATION
AND DEGREE OF COOPERATION
William F Long
The represen ta t ions and conclus ions presented herein are those of the
author and have no t been adopted in whole or in part by the Federal Trade
Commission or its Bureau of Economics The Assis tant D irector o f the Bureau
of Economics for Financial Sta tistics has certif ied tha t he has
approved the discl osure avoidance proc edures used by the staff of the
Bus iness Program to ensure tha t the data included in this pap r do
indiv idual company line of bus iness data
reviewed and
Line of
no t1 id entify
ith
ADVERTI SING INTENSITY ARKET SHARE
CONCENTRATI6 a n d DEGREE of COO ERATION
Wi l lia m F L o n g
The purpo ses of this study are to assess the role that advertising
plays in several explicit models of industrial organizationl and to formushy
late procedures for testing hypotheses which that asse ssment ge nerates
In addition a technique for the explicit introduction of the degree of coshy
operation in such models is used and the impact of cooperation is explored
I bull THE GENERAL SETTING
I assume a static industry with N firms By industry I mean a group
of firms that pro duce goods which are perceived as substitutes The goods
may be perfect substitutes but nee d not be Other goods outside the group
in questio n are assumed to be irrelevant
thUsing pi q and a for price quantity and advertising for the 1--i i
firm respectively the price of the goo d is assumed to be a function of all
quantities and advertising outlays ie
(1)
where bars under letters indicate vectors I define y as the elasticity ofij
the j goo ds price with respect to the goods advertising i e
thLetting c bull c (q ) be total production cost for the i-- firm i i i
and ll its pro fit i
(2)
-De fining some additional variab les let si
v bull a s be adve rtising intensity Define S bull rs as industry sales i i i i
ith
Page 2
A bull tai as industry adverti sing and zi bull siS as the market share of the
firm
II TWO POLAR CASE OLIGOPOLY MODELS
Cournot Let each firm maximize its profit independently of all other
firms That is
(3)
- 1
si- -- - 1 - 0yii ai
This implies that
-(4) v bull yiii
Let r bull (yij) Further let Ad be a diagonal matrix with the same diagonal
as the matrix A Then the N equations in (4) may be written compactly as
(5) v bull rd l
where is a vector of ones Since rd is a diagonal matrix (5) may be reshy
written as
(6 )
where if is a vector
z on the diagonal7
Chamberlin
3
v bull
d is a diagonal matrix containing the elements of z
4 At the other extreme let each firm maximize total profit
for all firms Letting IT trribull
an
Page 3
( 7 )
- 0
This implies that
(8)
(9)
-
v bull bull
bull
bull
zl [ Y21 Y22 bull bull Y2N J
-1zl 0
-1z2 r z
0 1
Page 4
Writ ing (8) compactly gives
the matrix in the two equationsA comparison of (6) and (9) shows that
differs only in that all the elasticities are present in (9) but only the
awn-elasticities are present in ( 6 ) Since advertising expense mus t be I non-negat ive the first-order requirements are that v be the maximum of t
i
the amount shown in 6) or (9) and zero
III A SIMPLIFIED DEMAND STRUCTURE
So far the demand equation system is comple tely general I will now
move to a particular demand structure which is charac terized by an industry
elasticity of price with respect to advertising a general brand-swi tching
elas ticity and a dependence of the firm-specific elas ticity on its relative
size I make the expli cit assumpt ion that
n pound
)l
0 AA
Consider the identification of y as an elasticity of industry price
Page 5
(10)
where y is an
- (y- a)
industry demand elasticity a is a brand-swi tching
el astici ty and z1 is the market share In matrix no tation
(11) r bull ai + (y- a) t
wi th respe ct to advertising 0
le t ting S denote dS observe that
0(12) SS bull
-
-1 0s t s - -1 0S t d - 9- -
apl apl aplaa1 aa2 a 0 a
middot( a Pl
( 13)
-
-0 a bull 0
a
) 0 X s
-1 0dAA bull then d
a bull
J 0
(JA) middot and
Let the sc alar (SS)(AA) be denoted by y so ya zft
If
1
d
-
I
I
(11)
--
y
I -f
l J- a)
a) l zt
Substitutin g into (13) gives
i [ ai + (y(14) -I- bull S A
bull az + (y -
Pa ge6
I I II
bull
since l bull 1
Under these assumpt ons then y is the elasticity of industry sales
(and average industry p ice) with respect to industry advertisin g Note
that if y bull 0 so that there is no industry sales effect there will still
be positive advertising if there are brand-switching effects
Turning now to the brand-switching ef fect observe that
(15)
Noting that
-1 deg -1d z -z -
-
0 a
Page 7
(15) may be written as
0 (AA)
lx 2x 1x
ith
1y
i
First assume no
that if there are
will be no changes in market shares
- 0a bull
Page 8
Equation (17) is a general statement of the relation betbulleen changes in
advertising levels and changes in market shares which would acompany them It
may be used to make some general observations about brand-switching effects
industry effect eg y bull r t bull 0 and assume further
equal percentage changes in all advertising le =ls then there
Under these further assumptions
0l and dz -1 bull (iA) ltr assumption the second term on the right hand side of this expression is
- By
zero
necessary for rt bull 0Consequently for market shares to be unaffected it is
when y bull r l bull 0 This requirement is satisfied by equation (11)
Secondly consider effects on market shares if one firm inc ases its abull ar-
tising and others do not As a general assumption we require smiddot ch a change o
lead to a decrease in the shares of all the other firms hat is let
0 0aiai gt 0 and a a bull 0 for j + i For any matrix X with N rows and its trans-j j
pose X let X
row of X
bull middotmiddotmiddot where is a column vector which contains the
elements of the Under the assumptions about the iis given just
d-1 0 0above it follows that r becomes (aiai) iimiddot ehFor jzj lt 0 under these conditions then it follows that the j- element
N of (I- l ) lt 0 so middotij - il lt 0 or yij - yik lt 0 If this
kbulll N condition holds for all Y ij where j + i then it follows that yii - z
k Y gta
k- 1 ik since the th firms share must increase to offset the decreases in all other
shares
ei ith
1v ( ) h i
7urning
where is the
leads to z
to the r matrix given in (11) we note that 1 = cr + (y - cr) i i zi elementary vector with 1 in the position and 0 elsewhere This
bull cr + (y - cr) zi Since bull required is thatcr at- zY wYij (y - cr)zi - cr + (y - cr) zilt 0 or cr gt 0
These observations may be summarized by substituting (11) into (17) bull
giving
(18)
i s
d -1 zz -
-
-
+ (y - cr) tz
d -1 0cr(I - l ) a a bull
cr 1 z - (y - cr) 0 a
o
If ai increases and the other a s are const t cr must be positive to havej
a positive effect on zi A positive cr also bull plies a negative ffect of aJ
f all a s in-on zi other a s (including a ) fixed j
eluding a1 change by the same percentage zi remains unchanged Finally k i
if cr bull 0 there is no brand-switching effect
For a to have no effect on p requires yij bull 0 From (10) thisj i
amounts to y bull cr that is the industry demand elasticity and the brandshyswitching elasticity are equal If y gt cr gt 0 the industry elasticity yij doninates If y lt cr y lt 0 the brand-switching elasticity dominates ij
making
Page 10
Substituting ( 1 1) now in (6) the Cournot case and in (9) the
Chamberlin ease gives
-(19) Cournot vO a + (y - a) z
Chamberli n vl - y lmiddot th
The 1- element is then
-(20) Cournot vO a+ (y - a) zi i
-Chamberlin vl yi
In (19) and (20) the superscript 0 is used to denote total
independence of decision and the superscript l is used to denote
total interdependence of decision making
The relation between v and vf depends solely on t relation between
y and a f y gt a so that the industry effect dominat 3 vf v with
equality in the trirlal case of z bull 1 Cooperative irms would have ai
higher advertising intensity than non-cooperative fi
If there is equality between the industry elasticity and the
y Whe therbrand-switching elasticity or y bull there is cooperation among the firms has no effect on advertising inten-sity And if brand-switching is more important with y lt a e get
vi lt v cooperative firms will have a lower advertising intensity than
non-cooperative firms The three situations are depicted in Figure L
Given the results for the firm level variables calculation of the
corresponding re sults for the industry is s-traightforward Let
B bull t zf bull z z be the Herfindahl index of concen tration and notethat
z - bull l z bull 1 and that V bull A S bull a i 5 bull (siIS) (asi)i
bull middot Then for the general case from (6) and (9)
bull
Page lOa
y lt1 bull y y
10
f
V0 bull z [ a1 + (y - cr) J bull a + (y - a) H vl bull Y bull y
11Page
( 2 1) vo bull z d -1 r bull r z bull z- d d- i )11
l -1 N N v bull z d r z bull r z bull l y z - ij jibulll jbulll
and for the simplified case from (19)
(22
Since the forms of the equations in (22) are the same as in (20) all
of the analysis of (20) holdS for (22) except that His substituted for
middot z Advertising intensity will be higher for an industry with cooperativei
firms than for one with non-coope rative firms if the industry demand elas-
ticity dominates the brand-switchine lasticity If the ewo elasticities
are equal then so will be advertiSl intensity for the cooperative firm
and non-cooperative firm industries If brand-switching dominates the
cooperative firm industry will have lower advertising intensity
In the industry model context the possibility of a critical concen-
tration level may be easily introduced That is assume that at values
of R below R the firms are non-cooperative but that for alues equal to
or greater than H they are cooperative That is
(23) V bull a + (y - a) H if H lt H bull y if H gtmiddot H
This function is shown in Figure 2 for the three relamptions be eween y and a
IV THE D GRE OF COOEtRATION
The models depicted in Figure 2 are based on a very simple notion about
the relation between concentration and the cooperationnon-cooperation charac-
teristic They use what is essentilly a step function relating concentra-
Figure
I -
J H H
Page l la
2
v v v
a
I J I y a bull Y y ---i
I I I I I I II I
H 10 H H 10 H 1 0
bullbull
bull bull
Page 12
tion and the extent or degree of coo peration as shown in Figure 3 Using
o as a symbol for the degree of cooperation eq uation (23) can be rewritten
as
bull(24) v (1 - o) vo + ovl bull (1 o) [a+ - a) H ] + oy- (y
vhere
(25) o bull o(H) bull 0 if B lt B
Hbull 1 if B gt
The specific relation between concentration (H) and the deg ree of
cooperation ( o) shown in Figure 3 is more restrictive than necessary and
bas less appeal an one which shows a smooth increase of o from 0 to 1 as
B increases fro to 1 Such a smooth function is shown in Figure 4 he
degree of cooper ion is low until H gets close to H increases greatly as
B goes through R attaining a high level for an H larger than B but still
close to it and then goes to 1 as B goes to 1 More formally
(26) 0 bull o(H)
bull 0 if H bull 0
bull 1 if H bull 1
do gt (IdB
d2o- gt 0 if H B dBZ
0 if H H
lt 0 if R gt H-
If (26) is now substituted for (25) in (24) a corresponding smoothing
of the functions shown in Figure 2 is accomplished he smooth functions
Figure 3
I
10
H 10 H
Page 12a
Figure
Page 12b
4
10
H 10 H
l
(27)
Cl - o) o
[ (1 -
[ (1
t y ij z
j
are given in Figure 5
Page 13
The equation in (24) was constructed as a hybrid of two results derived
from first-order equations for the non-cooperative and cooperative models
It may also be constructed directly by using l bull (1 - o) + on as thei i
th 2objective function for the i-- firm Treating o as a constant the first
order equation is
0 an aa
i middot
n i
Using equations (3) and (7) and setting --- bull 0 givesaai
(28) - 0
After collecting terms and converting to matrix notation we get
(29) - Cl - o) + 0 z-1vi yii i
N
j bulll
+ 0
+ o v1
d-1 r z z
-(30) v Cl - o) rd
-
o) r +d
d r J _ -1- d z
and middot
d) r + 0 rjd
l r z
v - z --(31) zv
- (1 0- +d) t r z- d-
If we now impose the simplified demand structure by substituting (11)
we get
- (1 - o) a +( 32) + oy and
Figure
l
-y
v
Y
H
r
I I
v
a
y
H 10 H
v
a
bulla
HH 10
bullbull
(33)
i
Page 14
v - (1 - o) [ a l + (y - a) ] The corresponding equation for Vis given in (24)
If the relation between industry advertising intensity (V) and
concentration (H) is non - linear via the impact of concentration
on the degree of cooperation does the relation take the form of a
auadratic as some other investigators have proposed (Greer ( 1971 ) Martin (1979) Strickland amp Weiss (1976) ) That question can
be approached from two perspectives - conceptual and empirical In
the next section I will try to apply some statistical material to it
here I want to explore the question in the context of the models
developed above
Referring to equation (24) taking the deriviative with respect
to H and letting bull (H) gives
bull( 3 3a ) 3V (y - a) (1 - o) + (1 - H) 3H dH
Given that o and H each are bounded by 0 and 1 and that o is an in-
creasing function of H the term in the brackets in non-negative The
sign of the deriviative then is the same as the sign of (y - a) If the industry effect is dominant (y gt a) then advertising in-
tensity will increase with concentration If the two effects are equal
then advertising intensity will not vbullry with concentration and if the
brand-switching e ffect is larger advertising intensity will decrease
with concentration
For given advertising elasticity parameters then it is not pos-
sible to show an increase in A S up to some H and the n a decrease
he relation may be non - linear but it is monotonic over the permis-
sible range of H
Page 15
One possibility is to introduce a relation between (y - a) and H
If industries characterized as dominated by the brand-switching effect
also were more concentrated then a cross industry comparison would
show an inverted U shape relation between AS and H I know of no
support for such an association between ( y - a) and H
V ERROR SPECUCATION
So far I have implicitly assumed that there are no errors in the
model I now assume that the error
thvariable in the equation for the i- firm is additive that its mean is zero
that its variance is inversely proportionate to its sales and that all error
term covarlances are zero hat is
34) v bull (1 - o) [ a + (y - a) z ] + oy + u i i i
- - -
In vector notation
(35) bull (1 - o) [ at + lt - a) J +
E() bull 0 Cov (u) -
0
If we now ca lculate V bull bull we get
oy + U bull
(36) V bull (1 - o) la + (y - a)R J +
U bull E(U) bull 0 Var (U) bull
P age 16
The error term specifi c ation which follows Hall
wi th the ratio of profits
6 heir analysis seems
as wellbull particularly given
and Weiss was
applied by them to equa tions to equity or
asse ts as dependen t variables to carry the same
weight in the present context the substantial
evidence presented by marketing analysis on the tendency of company de-
cision makers t o use the advertising to sales ratio as the decision
7variable
The zero covariance assumption on the other hand1 is particularly
troublesome Within a probably not true since
many even ts outside the
single industry it is
industry would have
industries
similar effec ts on all the
8firms in the industry Between it is probably not true either
since many companies produce in more than one indus try and even ts in the
firm would tend to effect all of its activities The extension of the
model to allow for non-zero covariances is certainly called for
--- -
Page Ll
Athird characteristic of this specification also deserves some comment
only on equation is shown for the firm or industry Several haveinvestigators
included an advertising equation ina larger system of equations (Comanor and
Wilson (1974) (1979) and Weiss (1976) so in myMartin Strickland Idid
dissertation1 Ihave ewo defenses for presenting only single equation results
here One is the suggestion by Comanor Wilson that simultaneousand equation
verybias may not be important in this context he second is that Iam will-
ing to see the model extended to include a profitability equation and I intend
to move that direction in the near futurein
Onthe inclusion of aconcentration equation on the other hand I amskep-
tical he problem isthat in some fundamental sense equations in models of
industrial organization should be oriented to the decision-making contexts of firms
Ifeach firm in anindustry with Nfirms has only one decision variable then
there can be only Nindependent equations which are based on the first-order
optimization equations It not show large ofis difficult of course to a number
functional relations which follow from the first-order conditions point here he
isthat only Nof them can be independent
Consider for illustrative purposes only a situation in wich teh N
itsquantities say q are set exogeneously and each firm determines advertising1
of and level say ai where a is the true optimal value a a bulla
1 1 i i + i
Ifp a) then the Nq
s and the Na s will determine N p s say pii bullp
i(l i 1 i
Now given Eand 9 we can determine both the vector s andthe vector v since
- bull til
ands We can also determine S s z bull s-1 s-ii
v and we
industry concentration isHbull (alternatively
V and Hare functions of the same
pendent Furthermore if there are
VM) and (H1 H2 bull bull bull )contain
VI STATISTI CAL APPLICATIONS
Page 18
Given the observed values of might be tempted to conclude that we
have a two equation system in which v and z are jointly determined That would i i
be inappropriate however since the 2N random variables (v ) are determined by
onlt N basic random variables t The variance-covariance matrix for ( ) must
be singular bull
The extension of this observation to industry level variables is straightshy
forward Since observed industry advertising intensity is V bull l v and observed 4
C4 bull z ) it follows thati l i
set of random variables t and cannot be indeshy-
M industries then the variables (V1 v2 at most M independent random variables
Given the assumptions made above y a and o are constant within a
given industry That being the case (34) can be written as
(37)
s1 bull (1 - o) (y - a)
If and
So + +
(1 - o) a + oy
81 are esti mated for an industry using a suitable statisticalSo procedure giving and 81 an estimate of y is then available The sumSo
Intra-industry analysis
1974 of Business Re ort -
categories are identified
Page 19
of So and S1 is y so y bull So + S1 is the estimate of y
If o is assumed to be bounded by zero and one the sign of 61 is also
the sign of y - a The estimate of So is of cou se an estimate of
(1 - o ) a + oy Moreove if y gt a then a is less than 6o- (1 - B) a+ 0(
Cor esponding results would
hold if y bull a or y lt a
If 61 gt 0 then we would estimate that a lt 6o
bull
For purposes of estimating this model 32 manufacturing industry categories
were selected The data are fram the Line of Business forms filed with
the Federal Trade Commission they are discussed in detail in the Annual Line
which is available from the FTC The 32 industry
in Table 1
Simple regressions for equ ation (37) were run for each of the 32 industries
with data for each of the firms which filed in the industry constituting an ob-
servation The results are given in Table 2 lines lA 2A
1974
the intercept
For all the A
equations generalized least quares was used th both and market
share bein scaled bv the square root of sales
The 32 industries were taken from the larger set of all manufacturing industry
categories in the 1974 data files Two selection cr iteria were used thereLB
had to be at least ten reporting companies and the industry had to be a consumer
goods industry or a producer goods industry which is strongly associated with a
related consumer goods industry so that adve rtising by the firms in the producer
goods industry impacts on demand in the related industry (eg soft drink syrup
and soft drinks) There were 135 industries with at least ten companies - 103
producer goods and 32 consumer goods
273
274
TABLE 1 INDUSTRY CATEGORIES USED
FTC
COOE OESCRI P T I ON _
2001 MEAT PACKINGt SAUSAGE S ANO OTHER PREPARED
MEAT PROOUCTS
2004 DAIRY PRODUCTS EXC F LUID MILK
2005 CANNED SPECIALTIES
2001 FROZEN SPECIALTIES
OEHYORATFOt AND
I NC LU 0 ING PRE SERVE S
SOUP MIXESbull
SEASO lNGSt ANO
OTHER PET FOOJ
RAIN MILL PROOUCTSt
ANO PREPARED
RELATED PRODUCTS
PRODUCTS
Z008 CANNEOt ORIEOt PICKLED FRUITS
ANO VEGET ABLES JAMS JELLIES DEHYDRATED I G TA3t E
S40CES ANO SALAD ORESSINGS
2010 DOG CAT A O
2012 FlOUR t OTHER RICE
MILLING fiLENOE) FLOUR
2014 B EADt CAKE ANO
20113 CO FECTIONERY
2026 BOTTLED A O CANNED SOFT DRINKS 2021 FLAVORING EXTRACTS AND SYRUPS EC
2029 MISC FOOOS AND KINDRED PROJUCTS EXC
ROASTED COFFEE
2301 MENS AND BOYS SUITS AND COATS
2302 MENS ANO BOYS FURNISHINGS
230 3 kOMENS AND MISSES OUTER EAR
2551 HOUSEHOLD FURNITURE
2702 PERIODICALS
Pqe 19a 1
RELATED 1972
src cooe
2011 f 3
202tX2026
2032
203pound
2033t4t5
2047 2041t4t5
2051 2065 2086
2081
209XZ05
231
232
233
251 272
2703 BOOKS
2704 ISC PVeLISHING
2806 ORGANIC FlEERS
2807 DRUG$ ETHICAL
23234
PT 2 S 3
2808 ORUCSt PROPRIETARY
2809 PERFU ES t COSMETICS ANO OTHER TaiLET
PREPARATIONS
2810 SOAP AND OTHER CLEANING PREPARATIONS 2815
2901
PEST1CIDES AND AGR ICULTURAL
PE OLEUM REF I N ING CHEMICALS EC
3606 HOUSEHOLD COOKING EQUIPMENT
3612 HOUSEHOLD APPLIANC ES NEC INCLUJING ELECshy
rqlC HOUSE ARFS AND FANS ANO SE I G MACHINES
3617 RADIO AND T V RECE I VING SETS 3808 PHOTOG APHIC E UIP amp SUP LfESt EXC
3qo3
PHO TOCOPYING
SPOR TING A D E QU I P M ENT f SUP P L I E S ATHLETTC GOODS NEC
3904 DOLLS GAMESt TOYS AND CHLOqENtS VE ICLES
PT283
2844
234tX2844
287 9
291 3631
3634t6t9
36H
PT386l
3949
15
943
middot
l A6l f 2 R(GRESSION RESULTS
EQ I NO
NR CODE cu 121
lA 2001 A (
2A 2004 r (
3A 2005 a
-R
ClS amp31
16
17
1 z
I TER-CEPT bull t
ooo9t21At 00022311 oooeos uu
OOl49tCAI 0010 2 middot 000960CCI
o049tl1At o04t55C61
MARKET SH ARE bull 5 I
-010670Ctl
-017600181
-001050
-0211AO
-0077 0
AOVER-TIS INti
lbl
000()39 000072
000018 0 0024liAI
-000079
SHARf
4ampgtVf Q bull
111
000140
-000170
RSQ
COl
66 485 862
607 6 3 t ()2
blti6 633
flASTI-
CI T Y PH
-007271(1
000441
-u0288l
clASTSHU (t)Ef
bull 10
tto
-234
210
( OOlltOCCI -126520 000922 181 001790
4A 200 7 18 00590liAI -038670 4e16 -032110 lell u c
OOOS74e 000700 025940
00067018) 0021001AI -02l100CAI
5 26
920
5A 6 c
200 8 26 0 03Sl9CA I 0012 26181 o 024511(
-005540
-0 81420 0001961AI O()Ol90CAI ooll240
441 604 891
-0020 19 274
6A 2010 16 Oe04224CAI o359lOIAI 9oq ll4015dCAI 061 6 -middot 005435CAI 00011 8 1AI 917 c 00535ltAI -009850((1 OOOA51J CAI -001610161 971
7A 2012 001862 000010 351 00187) oo1 a oooq45 ono1o6 439 c 001109 -027050 Oe0067SIAI -OOll60C81 897
- middotmiddot- -
8A 201t ll 000757 Ol4140CCI 814 ott9oocet 095 8 oooqo6 C C J 000069(81 811 c ootl2lampBI -080910CCI Oe009lUBI -OOtlOOCCI 960I
I DIIJIT pound11 1V OTIUTVTfAll-1pound1t- A _ 00 R - qq r - Q
1jl)OQCD
0
-middot middot--
l
_ __ _
I
i
0 00141 c c
91)7
I
T ABLE 2 REG ESSIOH RESULTS CCONTI
EO NR
IND CODE
iR bull
cos
I TEq-
CEPT
t1ARKE T
SHA E
AOVER-
Tl SING
SHARpound-cent
AuVER
RSO ILAST1-
C I T Y ELAST
SH CPtf (lt Cl) bull 11 I 41 I 5I (61 171 I 8 I f9) c 10)
9A 2 0 1 8 19 002397181 -007640 367 -005 246 le46 6 000605 000550CAI 597
-000029 G14180 OOl654CAJ -024li01At
lOA 2026 12 0028901AI -OlZl90 -0114 0 1 1 1 5 759
Ll (
llA 2021
f3 c
12A lOZ9
B
(
13A 2301
B
c
14A 2302
B
c
154 201
n (
16A 2551
8
10
35
10
20
19
24
001897(()
00141 6CCI
Oll9lHCI
010604 middot middot
Oll083
OOJ11JCAt
002059141
001411CIH
-000412
000293(81
000187
001159 (3)
ooo6a4cct
o00792CCI
OOl40dCAt
000006
-000076
OOlltl6fA)
o o0 5 64CCt
-034570
-o1middot o3o
-1659 90
Ol51lO
-046190CAt
Ol5450CA)
-005l60
002940
-030060(()
-0 43l50CCt
0 33650
-019080
00004
bullJ00741CAI
-000069
000466
0001 51CAt
000947CAt
O OOI)70CAt OOl750CA J
000175CCI
000671CAI
OOll5) f A t 002207CBI
0004b7CA I
- 007 1 70
001760
-001410CAt
- 0 16200CAt
-004210
- 1 207 70
705
96 0
542
466 805
460
614
921
832
9 61
996
424
550
9 15
56 1 651
884
-006114
019 037
03504lCA)
oo409Q
-141644((
491 -0 17664
607
871
lell
084
101
072
103
108 11IllOQfl)-- l
-037880fCJ OOl6l4CAJ -040l70fCI 0()
c
I
I
bull 9
9t
middot middot middotbullmiddot bull lraquo
TABLE 2 REGRE SION RESUlTS ICONTt
bull
EO I NO tHl I filER- f4A_K T AOVER- sbulluRE ttSQ flA Jl- iLAS T NR CODE cos CEPT SHARE JISING AOVfR bull (JTY SHR CiJCF
110Cl I f amp Itt 15J tbt bull 1c u caa
llA 2702 ll o079 6 81 Bt -019110 S6s -OlltO 111
a 004679 ooooo1 t69
( oOl26o -015160 001560 -022690 ltJr()I
ldA 270 3 19 0065lt -OIt2l60 220 -Ol57a2 11 8 e o 0 71t6IA t ol08 75tAt ( OOl983CiU -060llOC() OOl7191A t -O lltdOC A) middot 89
lqA 2704 10 005082 -Oil060 22b -l)i1977 lelO 6 -000080 oo2772C4 t 910 c 000907 -011 200 Oe04105CA t -046960 995
20A 2d06 1 II
O OlltStiAt -001120 837 00014 -))th O01380amp At -oooo 11 821 ( OOl526CAt -026700181 0002 5UAt 000530 969
2lA 2807 28 OOt9lt9CAt -00891 0 516 -0 03962 225I8 002095CCt OOOl lliAt 690 ( o02965CA t -056650((1 OOOt50CAJ -003l70ttl 911I
I 22A 2808 15 ot3l06 119730 655 1130) 090
0 - 007526CC t Oe003961At 835- middot-
c I oo7523 -067210 001l 2tA t -0091t80CCt 951
2lA 2809 21 Oe2067lCAt -1218101At 71 2 -10095((1 121 ttlIaOQ8 00797418) OOOI8lltt 686 Ill( 0092lU A t -Olltl70Ct t 000605CAt -OOit8701At 951 _
---- -
32 009506CAt
0 bull 06 981 I A J 006901 cA bull
00 24A 2810 -010960 5t 5 -OOltSO 7 56
8 -000011 s11 c -5811701At oooaoatAJ oot520IAI 888
-- 37 --- middot - -- -1 ) l lbull t5JJ r-l bullhmiddot1 --- - --
l5A
26A
27A
c
2RA
c
2 9A
31A
llA
24
13
597
Ci)55
14
l 6h
RSQ ELASTI- (lASTCOE f-
TABLE 2
EQ I NO NR CODE IU
REGRESSION RESULTS CCONTa
NR INTER- MARKET
(10S CEPT SHARE AOVER- SHARfcent
TIS lNG AOVER
) CITY StiR
c It) bull 5 co a I 1) bull 0 bull 9 c l 0 bull( 2 1
2815 16 003019 -009990 275 -o069bmiddotJ
6 001446 000224 282 c 002178 -054630 o02221CCI -0075 00 646
2901 29 00021t2 CAI 0)0520 592 0007 6 1 068
0 00105C8 t 000012141 60)g( OOOlJ51AJ -00126018) ooou271At -000u80CA) 974
36oa 1 3 003296(4) -oca29o -oo-q94
001755 000203 lt91t 002499((1 -04l820CCI OOl374CAI 0068)0 946
3612 0033281CI 011520 195 01684) oso
00 30 42C OI 00012 2 424 oo579ltAt -2 7l7801At 002 100IAI -003220 819
8 3617 btel 011789 081002259 009530
001448 oOOl OitCCI 732 OOl7 22CAI -0 46160CAI 0004961AI -OOll40181 989c
30A 3808 14 0021t 3 6fAI -000110 628 o023291C) -005 6 middot- _ middot- 0bull 02 04 1 IC I ooooo9 637 ( 001584((1 -0161)1)0 00021 91AI -0 00060
390) 14 002790CCt 006590 h8 t 0 0938 4 070
B OOl89bCCJ 000144 74
0021961(1 -OItlOlO OOl323CAI -004800 915
B
middot--- - middot 4 - middot- - dfll3904 12 004563181 0)8300 879 0428651(1 089 OQIll004 277CA I ooozstcea 919
c 0041)3181 -0-11250 0011881(1 -006270 972 0Ill
- J - bull bull bull bullbullbull middot - - middot - middot gt middotbull L)ofbullflbull middot middotfJltl -tn- middotmiddotmiddotmiddoto +f middot t middot bull
c
is
positive
2
Page 0 determineThe consumer goodproducer good split is sometimes difficult to
Sollle support for the split used in this paper is a comparison of goodness - of - fit
distributions of a
industries
set
advertising are
that industry
a sales increase of
the
highest aD10U3 those
measures for the two groups of industries Table 3 contains
measures for the 32- consumer goods industries and the 103 producer goods
2 Median R is 5 7 for the consumer goods set and 40 for the producer goods
Estimates - of y the elasticity of demand with respect to
given in column of Table 2 The hypothesis that y bull 0 with
alternative was applied for each ittdustry As shown in the
could not be rtjected for 24 of the 32 industries Of the
and 10 are positive For the remalnifg eight three are
and five are significantly positive
Of the threertegative elasetcities the largest (in absolute value)
cosmetics (code 2809) With all elasticity of -1 01 a one
industry advertising would gwerate a decrease in industry sales
more than one percetu
At the other extrema is toys and games (code 39 04) For
a one percent ittcrease in industry advertising would lead to
four-tenths of a pampreentmiddot Though not significantly different from zero
elasticity estimatbull for pt-bpuetary drugs (code 2808)
which are
Both drug industries have
to sales rafios (211 pe1 cent for 28 08 and 13 4 per cent
other hand flaIJo ing ext1ac s (mainly soft-drink SYUlS) also
is for
percent increase in
of slightly
a middottwo - tailed
table that hypothesis
24 14 are negative
significantly negative
9
the codetics and p-ropriacuy very high advertising
for 28 09) bull On the
has a high ind UY
advenisiltg t() sales ratio ( 8 0 per cent) bull but its estimated elasticity is very
small and not signifieantly different uom zero
middot It doe$ not seem unusual to find many elasticities near zero with a few
negative and a few positive If aggregate consumption is insensitive to aggregate
advertising as seems likely ld then what one industry gains must be lost by
others
Page 21
elasticity of demand does
(10) we can
if t 0 then
Finally a determination concerning the industry
not fix the individual firms elasticity Going back to equation
write y bull (1 - ) a + z y Since a gt 0 it follows thatii i i
Y gt 0 If Ylt 0 however y s sign is indeterminateij ij
Turning next to the coefficient of market share 20 of them are negative and
12 are positive with three of each sign being significantly different from zero
The largest positive value which is significant is for pet food (code 2010) and
a close second is mens amp boys suits and coats (code 23 01) The largest signifi
cant negative value is for cosmetics (code 28 09)
In the model some relations among y a 0 and 61 are implied If middota is
negative (1- y) is greater than one Since 61Y bull (1- a) (1 - ) 61Y is
pCsitive but it will be less than one if 0 is large enough Seventeen of the A
ys are negative in each of the seventeen industries 61 is also negative and les
1 (algebraicly) than y so 61Y is greater than one his is not evidence that the
degree of cooperation is low or zero in any industry of course A high o togethegt
with a negative y which is small (in absolute terms) relative to a can give a
value of 61Y greater than one On the other hand if 61Y were less than one
o would have to be greater than zero
For y gt 0 two conditions may hold If y lt a it follows that 61 lt 0 since
(y - a ) lt 0 It also follows that 61Y lt 0 regardless of the value of o bull
In addition the relation is an if-and-only-if one ie if 61Y is negative
then y lt a Three industries have negative values for 61y dairy products
except milk (code 20 04) synthetic fibers (code 2806) and photographic equipment
(code 3808) In all three cases y is quite small so a does not have to be very
large to give a negative value to al
If y gt a the model shows that 61-- gt 0 whatever the value of 0 bull There are
12 cases where both y and 61 are positive with 61 ranging from near zero
(grain milling code 20 12) to 1 20 ( proprietary drugs code 2808) Several
coeffiei-en_rs
--
15
o f to zero
Tab le 3 - Consumer Goods
Indus tr ies
2Distributions of R and Producer Goods
Page 20a
Consumer Producer
Goods Goods
0 - 9 9 0 3
10 - 19 9 0 1 1
20 - 29 9 3 20
30 - 39 9 3 17
40 - 49 9 5 16
50 - 59 9 8
69 9 6 Ii 1060
--f 670 79 9 2
80 - 89 9 4 4
90 - 100 1 1
Totals 32 10 3
Notes
2is used R i s cal-
culated as F F + ((N - K- 1 ) K ] 1 b_ F is the F
statistic Lor the hypothesis that all o f the non - intercept terms are simul taneously equal
1
2
--
-- Since generalized least squ-area regress ion
Page 12
o f these cases have values o f 6 1 Y rea ter than 0 8 (pet food (20 10) bread amp cakes
(20 14 ) miscellaneous processed foods (20 29 ) men s and boys suit s and coats
(23 01 ) p rop riet ary drugs (28 08 ) elect ric housewares (36 1 2) radio and
IV sets 17 and games and toys (39 For be than(36 ) 04)) 6 1 Y to greater 0 8
both o and cry mus t b e less than 0 2 That is in these industries the results
shown in Tab le 2 would hold only with very strong dominance of marke t demand
effects over brand swit ching e ffects and a very low degree o f cooperation
Using the concentration rat ios given in Table 4 the average concentration
ratio for the eight indust ries is 38 1 For the remaining 24 industries it is
43 2 This result would lend some support t o the proposition that concentrat ion
and cooperation are related if the group of eight industries are identi fied
as having low degrees of cooperation
One f inal note on model predict ions S ince both cr and o are assumed to
be non-negative 60 should also be non-negat ive In only one o f the 32 cas es
is 60 negative and it is not signif icant bull
cor pJ
7 1
7 0 4
1 4 7 5 1
)
7 3 z a o a t s 1 1 3 3 5 ( i V Z 8 tJ1 2 1 - 1 1 0 7 3 5 c middot v 2 8 1 0 3 2 -0 0 1 4 3 0 C 1 V 9 8 3 2 8 1 5 1 6 - 0 0 6 t1 6 9 C _ V 1 11 0 2 9 0 1 29 0 00 7 6 1 C l k V 7 3 1 3 6 08 1 3 - o n 4 9 9 4 3 3 3 6 1 2 2 4 0 1 6 4 3
3 1 7 1 3 0 1 1 7 d 7 8 3 3 8 0 8 0 0 2 3 i 2 3 0 6
P 4
3 9 0 3 1 4
1 2
0 09 3 8 4 39 1)1 0 4 l 8 6 5
T A LE 4 e I NCU S T R Y C A T E G O R Y
F TC NOM e tOOE ass
C 1 I c 2 )
2 0 0 1 1 6 2 0 0r 1 7 zo o l Z 2 0 0 7 1 8
z o o 2 6
z o t a 1 6
2 1) 1 2 1 5 2 0 1 4 1 1 2 0 1 a 1 9 2 0 26 1 2 2 0 2 7 1 0 2 0 29 - 3 5
2 3 0 1 10
2 3 02 z o
2 3 03 1 9
2 5 5 1 2 4 Z 1 0 Z 1 1 2 0 3 1 9 2 7 4 1 0 z a o 1 3
2 A 0 7 z s
E LA S T I -C I T Y
c 3 )
- 0 09 7 2 7
0 00 449
- 0 0 2 8 9 3
- 0 3 2 7 70 - 0 OZ il I q
o 4 u t s a 0 0 1 3 7 3
o 1 4 o o-0 0 5 2 4o -0 1 94 0 1
-C l 6 l l 4 0 1 JJ 3 7 0 3 t 0 4 2 0 0 4 0 9
- 0 4 1 8 44 - O l 7 t ) 4 -ll 7 1 4 0 b -0 3 5 7 8 Z
- 0 1 9 7 7 0 1)0 3 3 4
-o n 9 2
IJ A T A
CONV
ON-C ONV
( 4 )
t 1NV
C 1t-V C NV c ) v C - JNV C rlNV c n v c a v C ON V ( i) l V
C CtV
C t1tV C IJNV
C 1 V
c 1 v
Pa ge 2La
HEQ F t Noe x
c 5 )
6 2q
6 89 1 48
6 1 4 ) q 3
1 9 7 1 9
q z t s 5 a e z
l l 9 1
4 o 7 1 2 0
c oqt 1 Z
4 4 )-q 1 z
5 5 9 1 0 64 1 3 3
n 3
Inter industry analysi
Page 23
Equation ( 3 6 ) may be used to fo rmulate a rough test o f the cooperat ion
vs e f ficiency controversy I f we rest ricted ourselves to some subset o f
indus tries where the relat ion be tween y and a i s the same we would be lookingbull
at a sample o f indus tries for which the dependence o f V on H and o is a
quadratic wi th no squared terms
I f we now sub s titute (26) into ( 36 ) we get
(38) v bull a + (y - a) R + (y - a) o (H) - (y - a) H o (H) + u
If it is only d if ferences in marke t shares that affect advertising inten-
s i ty levels i f concentration does not af fect cooperation and i f o bull 0
( 38 ) reduces t o
( 39 ) V bull a + (y - a) H + U
Let this b e the null hyp o thes is and note that V is a linear func-
tion o f H If y gt a and the null hyp othesis holds the expected relat ion
between V and H is as shown by the pos itively sloped straight line in Figure
6 b etween points A and C
In this context the alterna ive hypothesis the degree o f
cooperation ( o) is a positive function o f concentrat ion
is that
(H) I f the func-
tion is as described in (26) and Figure 4 the relation between V and H
would be as shown by the curved line connec t ing po ints A and C in Figure 6 One way to test the null hypothe s is against this al ternative is to test fo r
the hypothesized curvature in the function relating V and H
Fi gure
R
6
v
ID c
H 1 0
I
i I I
sfigned
clasJ s
in def ini g
effec t is the
I
Page middot2lf
One charac teristic of produc ts which may be useful
classes wi th respec t to significanc e of the brand-switching
conveniencenon-convenienc e distinc tion developed by Porter ( 1 97 )
Using his identif ication of IRS indus tries as a s tar t ing point
each of the 32 industries used in this s tudy to one of the two
The ass ignments are not ed in Tab le 4
Using the conveniencenon-convenienc e dis t inc tion the
divided into two sub-samples
sample was
Four regressions were then run bull ene for e ch
sub-sample one for the to tal sample and one for the total sample with
a dummy variable for convenience goods Each regression was run wi th LB s
and wi th indus tries as ob servations The results are given in Tab le 5
For the LB level equations ( 1A - 1D) equation(37) was used ome
rearranging of terms gives v bull y - ( 1-o ) (y - a) ( 1-z ) + ui i i
bull y - 6 1 ( 1 - z ) + ui When observations are pooled across indus trif s yi
is an industry variable In these regressions the values o f y es t imated
in the first stage were used Since industries have been grouped so as
to reduce differences in (y - a) that term is t reated as a cons tant
within each group Finally H the Herfindahl index has been subst ituted
for o the degree of cooperation The resultant equation is linear
y ( 1 - z ) and H x ( l - z ) This is the equation for whi ch results arei i
reported in Table 5 In all regress ions all variables were weighted by the
square roo t of sales
in
raquo
- 34
3 1
i l T 4 8LE S R E G E S S I ON R E SUL T S i X T f N S I ON S T O T Y PE Of
AND L E V E L OF C ONC E kT R A T I O
E O O tt S
tP bull T Y P E NR C ONV O B S tO N-CONV
E L A S T I shy
T I C J T Y
O E - t-41( T SliA R E
H( fl F
1 1 - K O x middot K O X E R F
- H S a 1 1 - H S t 1 8 1 1 9 1
R S O
1 1 0 1
q z
t - bull 9 f 2 1
Z t P D l 6 fC ONV - 1 9 1 s s c 2 1
e - l O I e bull u q a
2 3 1 0 1 1 Nl N- t O V 5 7 1 ( 1 2 19 1 6 1 - J l S 2 5 4
l l l l t z q o a 1 - J o v z I
l ( I NO l l A l l ol u l t o o c bull J u l l
d S
c 3 4 t t r a l
bull Z D 1 0 l l 4 l l 2 3 -gt 2 l l 2 0 - 4 lS 5 4 l C 4 1 o e 6 t l b b 5 1 1 0 1 1 1 c - 1 9 5 1 t d S 2 1 tV
OQ10 T-R A T I O S A E G I VE N l P A RE N T H E S E S l E V E L S f S I G 1 F l C ANC E A - 99 5 B - q 9 t - s NS J UC E NO I Nl E R C ( P T I S U (j f middot) I N f Hf t OUA TIOI S 6 E f 3R E A O J U S T H E N T F OR 1-t E T E R O SIC E O A S T f C I T V pIraquo
1
middot
I
R SO I S C A L C UL A T E D A S f I f bull 1 1 bull 1lt 1 K t t- E E N I S THf U A1 B E R O f fJe S E fi V A T I O S F t S HtE F - S T A T I S
I N DE P E ND E N T y II
T I C F O T H E U Y P O HIE S I S f A l l C JE F F I C H N T A R E QU A L HI l f RO A f40 K I S T H f UUM 8 E R OF Q ( A fL f S
I l l
1 A
1 6
t C
f 1 r R I l l
1 2 1
L Jt
L B
l q
L 1
on s T Y P E bull 2 )
P OUC T
X
H S I l l
l O l A I bull l 0 7 1
O O b O J
1 2 00 1
1 b 2 1 C I l l Z U
l l S b 1 -1 1 0 1
rlt F X l l - fi C f l
bull 1 I
l l
- ()4 2 3 l f H
1 - 2 7 9 1
K O X
1 1 - ti E R F I
l d l
S4 l i A I 2 l b
1 3 00
K O X -t f R F t S J
I 1 - HER F t
)
60 9
z o bull
0
t l l
3 8 8
1 7 6
5 64
64
N f bull
tl S bull
I l l
I 4 1
t ONV
NON-t ONV
A l l
A ll
c o v NO- ( ONV
1 4 1
1 5 1
- l S b
c - 1 1 2 1
5 8 2 C A I 1 6 7 0 1
middot I l 9 H
l 1 il
1 4 1 6 1
E L A C) T J shy
l t C I T Y
I S I
l b l
- 0 2 0
1 - 1 4 9 1
O Z 3 2 1 A I C b 6 b l
0 0 9 2 -J
1 1 b 4 1
o z t t t l 1 2 2 1 1
oru - Ht tt F
1 gt X bull b I
Page 25 bull
Mean val -as ior relevan t variables for the two sub-samples are
Conv Non-Conv
Adv Sales 4 7 2 3
iit Share 2 7 3 2
Ela cicity - 05 3 02 2
Herf indahl 7 3 1 860
That the dist nct ion between coft9enienee goods and non - convenience
goods mat ters is supported not only by these means but also by the regression
results Equat ions lA and lB are different in almost every respec t Only the
concentrat ionmarket share int eract ion term seems to mat ter for convenience
good s Jus t he opposite holds for non - convenience goods both elasticity1
and market sl e are highly significant but the concentrat ionmarket share L
int erac t ion t=rm is not
Given that one minus market share is used these equ3 tions the coeffici t
of market share is the negative of what is sh in col (6) of Table 5 For the
regress ion for non - convenience LB s reported in equation lB then LB s with larger
market shares have lower advertising to sales ratios
On the question of the impact of the degree of cooperation the story is mixed
For convenienc good s the coefficient is pos itive as equation (31) predicts when
there is a po sitive degree of cooperat ion Fo r the non-convenience goods LB s middot
concentration has no effec t
Aggregat ion t o the indus try level and then rer unning the equations $ives the
results shown in (2a-2d) of Tab le 5 As expected R2 goes up and levels of significanc e
fmiddotJr individual coefficients gJ down In add ition the coeffic ient of H ( l -H) inmiddot
quat ion 2A is no t significant though it is po s i tive Since 2A is the indus try level
I
--
all but one of them and that one is not significant For 18 of the 31
wo uld have on its adve rtising to sal es ratio
cognate of LA the variable H ( l-H) pl ays the same role in 2A as H ( l -z 1) p lays
in LA
Vconomies of scale in advert ising
The literature on advert ing contains a number of definitionbull o f economies
of scale and of findings eone rning its presence and magnitude Before I loo k
a t the data with the hope o f mmenting about scale effect s then I wan t to
define the term as I use it
By economies of scale I mean a decrease in the number o f units of adver-
tising which are needed to generate a unit of sales as advert ising is increased
Given the definition of the firm s own elast icity of demand with respect to
advertising (see the text foll wing equation (1) ) the technical definition
is that rii gt 1 Diseconomies of scale are def ined symmetrically i e
Yu lt 1
ith a cro s s - section oi observations on advertising intensity
(v bull a s1) and onmiddot advertising (a ) we may regress the former on the latter 1 i i
The results of such regress ions for the 32 industries are given in Table 2
lines lB ZB bull bull bull 32B For the 32 eases the regression coefficient is positive it
positive coefficients the coefficient i s signif icant
The que tion now is whether these results ar e evidence of pervasive dis-
economies of scale in advertislng I think no t Given my as sumption that the
data I observe are equilibrium results I may ri ghtly conclud e that almost every-
where high levels of advert ising are as sociated in equilibrium with high ratios
of advertising to sales I may no t conclud e anything however about the
impact that a move by some firm away from its equilibrium level o f advertising
middot- Page 27
Given the assumed relation in equation ( 37 ) there is a corre spondshy
ing equilibrium equation relating v and ai The equation is non - linear i
and I have not found a simp le way to characterize it It is po ssib le to
det ermi ne its deriviative with respect to a b y takin$ tne de riviative of ( 3 7 ) i
When that is done the result is
v - 1(40) i bull sl a a z ( 1-z )i i ira-l
Sinc e all the other terms in (40) are non-negative the sign dep ends only
on the sign of 6 1 which is in turn de pend ent only on the sign of (y - a)
That the signs of the coefficients of a in equations lB 2B bull bull bull 32B arei
not always equal to the signs of the coefficients of z in LA 2A bull bull bull 32A i
is still a mystery to me and something to be explored
bull
Summary
My purpose in writing this pa per wa s t o attempt to integrate explicit micro-
economic mode lling reasoned econome tric specification of error terms and high
quality data to exp lore some questions about adver tising by large firms
al quite important improvements could be made in the first two of those areas
and much data ma ssag ing is yet to be done in the third
Concerning the subst an ce of what I have done I think six things are im-
Page 2 8
Severshy
V and Conclusions
portant
1 Explicit modeling is worth the ef for t if for no other reas on than
that it provides a basis for choosing fr om a varie ty of funct ional
forms
2 The same is true for error specif ica t ion I did in fact look at scatshy
ter diagrams for all 32 industry categor ies and they vir tually all show
the het eroscedast icity which I as sumed and for which I correc ted
3 The predicted relation be tween advertis ing in tensity and market share
shows up clearly in on ly 25 of the case s examined I have ye t to
explore why that may b e the cas e
4 Some evidence concerning the presence and impact of coopera tion was
produced but it is not clear ly pervasive
5 The dis tinction between convenience goods and non-c onvenie nce goods is
unambiguously a good one
6 Virtually no evidence concerning econ omies of scale in adver tis ing can
be gleaned fr om this study given its assumpt ions
More work is called for on many of these issues my study seems to have
raised more questions than it has answer ed This is pr obab lly due to the richshy
ne ss of the data source since I had the oppor tunity to address several issue s
at once
- -- - Q --
Foo tnotes
Manager Line o f Business Program Federal Trade Commission The author depended heavily on several staf f members of the LB Pro gram for statistical and clerical support in particular Joe Cholka George Pascoe and Haro lene Jenkins Helpful comments were given b y Richard Caves Dennis Mueller Michael Porter and F M Scherer When the paper was in its early stages o f development both S teve Garber and Jon Rasmussen former economists in the LB Program provided eery useful cri tiques The views expressed here are my own o f course
1 The basic exp licit model of the role of advertising is Dorfman and S teiner ( 1954 ) and the models in this paper may be seen as ano ther extens ion of their work
2 For a review o f models of oligopoly which treat prices and quantities as relevant variables see Shubik ( 1 959) Extens ions to take advertis ing int o account are fairly s traightforward
3 Dorfman and S teiner dealt with a price-setting firm For a monopolis t inverting the demand relation to show Pbull P (Q A) _instead o f Q- (P A) has no effect on the results o f the analysis For oligopolists who face a sys tem of demand relations inversion does have some impac t on the results The dif ferences are not trivial nonetheless they will no t be explored in this study
4 Shubik ( 1 959) p
5 From a formal mathematical point o f view the function is similar to a1 i Legrangian funct ion of the form - r rr
L bull 11 + A ( j - (i i i+j
where the indicates some fixed level of the profits industry Following the same logic which resul ts in and income constraint as the marginal utility of the of the indus tr in the obj ective of firm i
6 Hall amp Weiss ( 1 967) p 323
7 See Schmallensee ( 1 97 2 ) Ch 2 pp 1 6-47
8 On this poin t see Imel amp Helmberger ( 1 9 7 1 )
9 Long ( 1 970) p 130 - 1 39
1 0 See Schmallensee ( 1 972) Ch 3 pp 48-8 7
rrj i+j ) of the rest of the
the identification profits of the rest
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980
ith
ADVERTI SING INTENSITY ARKET SHARE
CONCENTRATI6 a n d DEGREE of COO ERATION
Wi l lia m F L o n g
The purpo ses of this study are to assess the role that advertising
plays in several explicit models of industrial organizationl and to formushy
late procedures for testing hypotheses which that asse ssment ge nerates
In addition a technique for the explicit introduction of the degree of coshy
operation in such models is used and the impact of cooperation is explored
I bull THE GENERAL SETTING
I assume a static industry with N firms By industry I mean a group
of firms that pro duce goods which are perceived as substitutes The goods
may be perfect substitutes but nee d not be Other goods outside the group
in questio n are assumed to be irrelevant
thUsing pi q and a for price quantity and advertising for the 1--i i
firm respectively the price of the goo d is assumed to be a function of all
quantities and advertising outlays ie
(1)
where bars under letters indicate vectors I define y as the elasticity ofij
the j goo ds price with respect to the goods advertising i e
thLetting c bull c (q ) be total production cost for the i-- firm i i i
and ll its pro fit i
(2)
-De fining some additional variab les let si
v bull a s be adve rtising intensity Define S bull rs as industry sales i i i i
ith
Page 2
A bull tai as industry adverti sing and zi bull siS as the market share of the
firm
II TWO POLAR CASE OLIGOPOLY MODELS
Cournot Let each firm maximize its profit independently of all other
firms That is
(3)
- 1
si- -- - 1 - 0yii ai
This implies that
-(4) v bull yiii
Let r bull (yij) Further let Ad be a diagonal matrix with the same diagonal
as the matrix A Then the N equations in (4) may be written compactly as
(5) v bull rd l
where is a vector of ones Since rd is a diagonal matrix (5) may be reshy
written as
(6 )
where if is a vector
z on the diagonal7
Chamberlin
3
v bull
d is a diagonal matrix containing the elements of z
4 At the other extreme let each firm maximize total profit
for all firms Letting IT trribull
an
Page 3
( 7 )
- 0
This implies that
(8)
(9)
-
v bull bull
bull
bull
zl [ Y21 Y22 bull bull Y2N J
-1zl 0
-1z2 r z
0 1
Page 4
Writ ing (8) compactly gives
the matrix in the two equationsA comparison of (6) and (9) shows that
differs only in that all the elasticities are present in (9) but only the
awn-elasticities are present in ( 6 ) Since advertising expense mus t be I non-negat ive the first-order requirements are that v be the maximum of t
i
the amount shown in 6) or (9) and zero
III A SIMPLIFIED DEMAND STRUCTURE
So far the demand equation system is comple tely general I will now
move to a particular demand structure which is charac terized by an industry
elasticity of price with respect to advertising a general brand-swi tching
elas ticity and a dependence of the firm-specific elas ticity on its relative
size I make the expli cit assumpt ion that
n pound
)l
0 AA
Consider the identification of y as an elasticity of industry price
Page 5
(10)
where y is an
- (y- a)
industry demand elasticity a is a brand-swi tching
el astici ty and z1 is the market share In matrix no tation
(11) r bull ai + (y- a) t
wi th respe ct to advertising 0
le t ting S denote dS observe that
0(12) SS bull
-
-1 0s t s - -1 0S t d - 9- -
apl apl aplaa1 aa2 a 0 a
middot( a Pl
( 13)
-
-0 a bull 0
a
) 0 X s
-1 0dAA bull then d
a bull
J 0
(JA) middot and
Let the sc alar (SS)(AA) be denoted by y so ya zft
If
1
d
-
I
I
(11)
--
y
I -f
l J- a)
a) l zt
Substitutin g into (13) gives
i [ ai + (y(14) -I- bull S A
bull az + (y -
Pa ge6
I I II
bull
since l bull 1
Under these assumpt ons then y is the elasticity of industry sales
(and average industry p ice) with respect to industry advertisin g Note
that if y bull 0 so that there is no industry sales effect there will still
be positive advertising if there are brand-switching effects
Turning now to the brand-switching ef fect observe that
(15)
Noting that
-1 deg -1d z -z -
-
0 a
Page 7
(15) may be written as
0 (AA)
lx 2x 1x
ith
1y
i
First assume no
that if there are
will be no changes in market shares
- 0a bull
Page 8
Equation (17) is a general statement of the relation betbulleen changes in
advertising levels and changes in market shares which would acompany them It
may be used to make some general observations about brand-switching effects
industry effect eg y bull r t bull 0 and assume further
equal percentage changes in all advertising le =ls then there
Under these further assumptions
0l and dz -1 bull (iA) ltr assumption the second term on the right hand side of this expression is
- By
zero
necessary for rt bull 0Consequently for market shares to be unaffected it is
when y bull r l bull 0 This requirement is satisfied by equation (11)
Secondly consider effects on market shares if one firm inc ases its abull ar-
tising and others do not As a general assumption we require smiddot ch a change o
lead to a decrease in the shares of all the other firms hat is let
0 0aiai gt 0 and a a bull 0 for j + i For any matrix X with N rows and its trans-j j
pose X let X
row of X
bull middotmiddotmiddot where is a column vector which contains the
elements of the Under the assumptions about the iis given just
d-1 0 0above it follows that r becomes (aiai) iimiddot ehFor jzj lt 0 under these conditions then it follows that the j- element
N of (I- l ) lt 0 so middotij - il lt 0 or yij - yik lt 0 If this
kbulll N condition holds for all Y ij where j + i then it follows that yii - z
k Y gta
k- 1 ik since the th firms share must increase to offset the decreases in all other
shares
ei ith
1v ( ) h i
7urning
where is the
leads to z
to the r matrix given in (11) we note that 1 = cr + (y - cr) i i zi elementary vector with 1 in the position and 0 elsewhere This
bull cr + (y - cr) zi Since bull required is thatcr at- zY wYij (y - cr)zi - cr + (y - cr) zilt 0 or cr gt 0
These observations may be summarized by substituting (11) into (17) bull
giving
(18)
i s
d -1 zz -
-
-
+ (y - cr) tz
d -1 0cr(I - l ) a a bull
cr 1 z - (y - cr) 0 a
o
If ai increases and the other a s are const t cr must be positive to havej
a positive effect on zi A positive cr also bull plies a negative ffect of aJ
f all a s in-on zi other a s (including a ) fixed j
eluding a1 change by the same percentage zi remains unchanged Finally k i
if cr bull 0 there is no brand-switching effect
For a to have no effect on p requires yij bull 0 From (10) thisj i
amounts to y bull cr that is the industry demand elasticity and the brandshyswitching elasticity are equal If y gt cr gt 0 the industry elasticity yij doninates If y lt cr y lt 0 the brand-switching elasticity dominates ij
making
Page 10
Substituting ( 1 1) now in (6) the Cournot case and in (9) the
Chamberlin ease gives
-(19) Cournot vO a + (y - a) z
Chamberli n vl - y lmiddot th
The 1- element is then
-(20) Cournot vO a+ (y - a) zi i
-Chamberlin vl yi
In (19) and (20) the superscript 0 is used to denote total
independence of decision and the superscript l is used to denote
total interdependence of decision making
The relation between v and vf depends solely on t relation between
y and a f y gt a so that the industry effect dominat 3 vf v with
equality in the trirlal case of z bull 1 Cooperative irms would have ai
higher advertising intensity than non-cooperative fi
If there is equality between the industry elasticity and the
y Whe therbrand-switching elasticity or y bull there is cooperation among the firms has no effect on advertising inten-sity And if brand-switching is more important with y lt a e get
vi lt v cooperative firms will have a lower advertising intensity than
non-cooperative firms The three situations are depicted in Figure L
Given the results for the firm level variables calculation of the
corresponding re sults for the industry is s-traightforward Let
B bull t zf bull z z be the Herfindahl index of concen tration and notethat
z - bull l z bull 1 and that V bull A S bull a i 5 bull (siIS) (asi)i
bull middot Then for the general case from (6) and (9)
bull
Page lOa
y lt1 bull y y
10
f
V0 bull z [ a1 + (y - cr) J bull a + (y - a) H vl bull Y bull y
11Page
( 2 1) vo bull z d -1 r bull r z bull z- d d- i )11
l -1 N N v bull z d r z bull r z bull l y z - ij jibulll jbulll
and for the simplified case from (19)
(22
Since the forms of the equations in (22) are the same as in (20) all
of the analysis of (20) holdS for (22) except that His substituted for
middot z Advertising intensity will be higher for an industry with cooperativei
firms than for one with non-coope rative firms if the industry demand elas-
ticity dominates the brand-switchine lasticity If the ewo elasticities
are equal then so will be advertiSl intensity for the cooperative firm
and non-cooperative firm industries If brand-switching dominates the
cooperative firm industry will have lower advertising intensity
In the industry model context the possibility of a critical concen-
tration level may be easily introduced That is assume that at values
of R below R the firms are non-cooperative but that for alues equal to
or greater than H they are cooperative That is
(23) V bull a + (y - a) H if H lt H bull y if H gtmiddot H
This function is shown in Figure 2 for the three relamptions be eween y and a
IV THE D GRE OF COOEtRATION
The models depicted in Figure 2 are based on a very simple notion about
the relation between concentration and the cooperationnon-cooperation charac-
teristic They use what is essentilly a step function relating concentra-
Figure
I -
J H H
Page l la
2
v v v
a
I J I y a bull Y y ---i
I I I I I I II I
H 10 H H 10 H 1 0
bullbull
bull bull
Page 12
tion and the extent or degree of coo peration as shown in Figure 3 Using
o as a symbol for the degree of cooperation eq uation (23) can be rewritten
as
bull(24) v (1 - o) vo + ovl bull (1 o) [a+ - a) H ] + oy- (y
vhere
(25) o bull o(H) bull 0 if B lt B
Hbull 1 if B gt
The specific relation between concentration (H) and the deg ree of
cooperation ( o) shown in Figure 3 is more restrictive than necessary and
bas less appeal an one which shows a smooth increase of o from 0 to 1 as
B increases fro to 1 Such a smooth function is shown in Figure 4 he
degree of cooper ion is low until H gets close to H increases greatly as
B goes through R attaining a high level for an H larger than B but still
close to it and then goes to 1 as B goes to 1 More formally
(26) 0 bull o(H)
bull 0 if H bull 0
bull 1 if H bull 1
do gt (IdB
d2o- gt 0 if H B dBZ
0 if H H
lt 0 if R gt H-
If (26) is now substituted for (25) in (24) a corresponding smoothing
of the functions shown in Figure 2 is accomplished he smooth functions
Figure 3
I
10
H 10 H
Page 12a
Figure
Page 12b
4
10
H 10 H
l
(27)
Cl - o) o
[ (1 -
[ (1
t y ij z
j
are given in Figure 5
Page 13
The equation in (24) was constructed as a hybrid of two results derived
from first-order equations for the non-cooperative and cooperative models
It may also be constructed directly by using l bull (1 - o) + on as thei i
th 2objective function for the i-- firm Treating o as a constant the first
order equation is
0 an aa
i middot
n i
Using equations (3) and (7) and setting --- bull 0 givesaai
(28) - 0
After collecting terms and converting to matrix notation we get
(29) - Cl - o) + 0 z-1vi yii i
N
j bulll
+ 0
+ o v1
d-1 r z z
-(30) v Cl - o) rd
-
o) r +d
d r J _ -1- d z
and middot
d) r + 0 rjd
l r z
v - z --(31) zv
- (1 0- +d) t r z- d-
If we now impose the simplified demand structure by substituting (11)
we get
- (1 - o) a +( 32) + oy and
Figure
l
-y
v
Y
H
r
I I
v
a
y
H 10 H
v
a
bulla
HH 10
bullbull
(33)
i
Page 14
v - (1 - o) [ a l + (y - a) ] The corresponding equation for Vis given in (24)
If the relation between industry advertising intensity (V) and
concentration (H) is non - linear via the impact of concentration
on the degree of cooperation does the relation take the form of a
auadratic as some other investigators have proposed (Greer ( 1971 ) Martin (1979) Strickland amp Weiss (1976) ) That question can
be approached from two perspectives - conceptual and empirical In
the next section I will try to apply some statistical material to it
here I want to explore the question in the context of the models
developed above
Referring to equation (24) taking the deriviative with respect
to H and letting bull (H) gives
bull( 3 3a ) 3V (y - a) (1 - o) + (1 - H) 3H dH
Given that o and H each are bounded by 0 and 1 and that o is an in-
creasing function of H the term in the brackets in non-negative The
sign of the deriviative then is the same as the sign of (y - a) If the industry effect is dominant (y gt a) then advertising in-
tensity will increase with concentration If the two effects are equal
then advertising intensity will not vbullry with concentration and if the
brand-switching e ffect is larger advertising intensity will decrease
with concentration
For given advertising elasticity parameters then it is not pos-
sible to show an increase in A S up to some H and the n a decrease
he relation may be non - linear but it is monotonic over the permis-
sible range of H
Page 15
One possibility is to introduce a relation between (y - a) and H
If industries characterized as dominated by the brand-switching effect
also were more concentrated then a cross industry comparison would
show an inverted U shape relation between AS and H I know of no
support for such an association between ( y - a) and H
V ERROR SPECUCATION
So far I have implicitly assumed that there are no errors in the
model I now assume that the error
thvariable in the equation for the i- firm is additive that its mean is zero
that its variance is inversely proportionate to its sales and that all error
term covarlances are zero hat is
34) v bull (1 - o) [ a + (y - a) z ] + oy + u i i i
- - -
In vector notation
(35) bull (1 - o) [ at + lt - a) J +
E() bull 0 Cov (u) -
0
If we now ca lculate V bull bull we get
oy + U bull
(36) V bull (1 - o) la + (y - a)R J +
U bull E(U) bull 0 Var (U) bull
P age 16
The error term specifi c ation which follows Hall
wi th the ratio of profits
6 heir analysis seems
as wellbull particularly given
and Weiss was
applied by them to equa tions to equity or
asse ts as dependen t variables to carry the same
weight in the present context the substantial
evidence presented by marketing analysis on the tendency of company de-
cision makers t o use the advertising to sales ratio as the decision
7variable
The zero covariance assumption on the other hand1 is particularly
troublesome Within a probably not true since
many even ts outside the
single industry it is
industry would have
industries
similar effec ts on all the
8firms in the industry Between it is probably not true either
since many companies produce in more than one indus try and even ts in the
firm would tend to effect all of its activities The extension of the
model to allow for non-zero covariances is certainly called for
--- -
Page Ll
Athird characteristic of this specification also deserves some comment
only on equation is shown for the firm or industry Several haveinvestigators
included an advertising equation ina larger system of equations (Comanor and
Wilson (1974) (1979) and Weiss (1976) so in myMartin Strickland Idid
dissertation1 Ihave ewo defenses for presenting only single equation results
here One is the suggestion by Comanor Wilson that simultaneousand equation
verybias may not be important in this context he second is that Iam will-
ing to see the model extended to include a profitability equation and I intend
to move that direction in the near futurein
Onthe inclusion of aconcentration equation on the other hand I amskep-
tical he problem isthat in some fundamental sense equations in models of
industrial organization should be oriented to the decision-making contexts of firms
Ifeach firm in anindustry with Nfirms has only one decision variable then
there can be only Nindependent equations which are based on the first-order
optimization equations It not show large ofis difficult of course to a number
functional relations which follow from the first-order conditions point here he
isthat only Nof them can be independent
Consider for illustrative purposes only a situation in wich teh N
itsquantities say q are set exogeneously and each firm determines advertising1
of and level say ai where a is the true optimal value a a bulla
1 1 i i + i
Ifp a) then the Nq
s and the Na s will determine N p s say pii bullp
i(l i 1 i
Now given Eand 9 we can determine both the vector s andthe vector v since
- bull til
ands We can also determine S s z bull s-1 s-ii
v and we
industry concentration isHbull (alternatively
V and Hare functions of the same
pendent Furthermore if there are
VM) and (H1 H2 bull bull bull )contain
VI STATISTI CAL APPLICATIONS
Page 18
Given the observed values of might be tempted to conclude that we
have a two equation system in which v and z are jointly determined That would i i
be inappropriate however since the 2N random variables (v ) are determined by
onlt N basic random variables t The variance-covariance matrix for ( ) must
be singular bull
The extension of this observation to industry level variables is straightshy
forward Since observed industry advertising intensity is V bull l v and observed 4
C4 bull z ) it follows thati l i
set of random variables t and cannot be indeshy-
M industries then the variables (V1 v2 at most M independent random variables
Given the assumptions made above y a and o are constant within a
given industry That being the case (34) can be written as
(37)
s1 bull (1 - o) (y - a)
If and
So + +
(1 - o) a + oy
81 are esti mated for an industry using a suitable statisticalSo procedure giving and 81 an estimate of y is then available The sumSo
Intra-industry analysis
1974 of Business Re ort -
categories are identified
Page 19
of So and S1 is y so y bull So + S1 is the estimate of y
If o is assumed to be bounded by zero and one the sign of 61 is also
the sign of y - a The estimate of So is of cou se an estimate of
(1 - o ) a + oy Moreove if y gt a then a is less than 6o- (1 - B) a+ 0(
Cor esponding results would
hold if y bull a or y lt a
If 61 gt 0 then we would estimate that a lt 6o
bull
For purposes of estimating this model 32 manufacturing industry categories
were selected The data are fram the Line of Business forms filed with
the Federal Trade Commission they are discussed in detail in the Annual Line
which is available from the FTC The 32 industry
in Table 1
Simple regressions for equ ation (37) were run for each of the 32 industries
with data for each of the firms which filed in the industry constituting an ob-
servation The results are given in Table 2 lines lA 2A
1974
the intercept
For all the A
equations generalized least quares was used th both and market
share bein scaled bv the square root of sales
The 32 industries were taken from the larger set of all manufacturing industry
categories in the 1974 data files Two selection cr iteria were used thereLB
had to be at least ten reporting companies and the industry had to be a consumer
goods industry or a producer goods industry which is strongly associated with a
related consumer goods industry so that adve rtising by the firms in the producer
goods industry impacts on demand in the related industry (eg soft drink syrup
and soft drinks) There were 135 industries with at least ten companies - 103
producer goods and 32 consumer goods
273
274
TABLE 1 INDUSTRY CATEGORIES USED
FTC
COOE OESCRI P T I ON _
2001 MEAT PACKINGt SAUSAGE S ANO OTHER PREPARED
MEAT PROOUCTS
2004 DAIRY PRODUCTS EXC F LUID MILK
2005 CANNED SPECIALTIES
2001 FROZEN SPECIALTIES
OEHYORATFOt AND
I NC LU 0 ING PRE SERVE S
SOUP MIXESbull
SEASO lNGSt ANO
OTHER PET FOOJ
RAIN MILL PROOUCTSt
ANO PREPARED
RELATED PRODUCTS
PRODUCTS
Z008 CANNEOt ORIEOt PICKLED FRUITS
ANO VEGET ABLES JAMS JELLIES DEHYDRATED I G TA3t E
S40CES ANO SALAD ORESSINGS
2010 DOG CAT A O
2012 FlOUR t OTHER RICE
MILLING fiLENOE) FLOUR
2014 B EADt CAKE ANO
20113 CO FECTIONERY
2026 BOTTLED A O CANNED SOFT DRINKS 2021 FLAVORING EXTRACTS AND SYRUPS EC
2029 MISC FOOOS AND KINDRED PROJUCTS EXC
ROASTED COFFEE
2301 MENS AND BOYS SUITS AND COATS
2302 MENS ANO BOYS FURNISHINGS
230 3 kOMENS AND MISSES OUTER EAR
2551 HOUSEHOLD FURNITURE
2702 PERIODICALS
Pqe 19a 1
RELATED 1972
src cooe
2011 f 3
202tX2026
2032
203pound
2033t4t5
2047 2041t4t5
2051 2065 2086
2081
209XZ05
231
232
233
251 272
2703 BOOKS
2704 ISC PVeLISHING
2806 ORGANIC FlEERS
2807 DRUG$ ETHICAL
23234
PT 2 S 3
2808 ORUCSt PROPRIETARY
2809 PERFU ES t COSMETICS ANO OTHER TaiLET
PREPARATIONS
2810 SOAP AND OTHER CLEANING PREPARATIONS 2815
2901
PEST1CIDES AND AGR ICULTURAL
PE OLEUM REF I N ING CHEMICALS EC
3606 HOUSEHOLD COOKING EQUIPMENT
3612 HOUSEHOLD APPLIANC ES NEC INCLUJING ELECshy
rqlC HOUSE ARFS AND FANS ANO SE I G MACHINES
3617 RADIO AND T V RECE I VING SETS 3808 PHOTOG APHIC E UIP amp SUP LfESt EXC
3qo3
PHO TOCOPYING
SPOR TING A D E QU I P M ENT f SUP P L I E S ATHLETTC GOODS NEC
3904 DOLLS GAMESt TOYS AND CHLOqENtS VE ICLES
PT283
2844
234tX2844
287 9
291 3631
3634t6t9
36H
PT386l
3949
15
943
middot
l A6l f 2 R(GRESSION RESULTS
EQ I NO
NR CODE cu 121
lA 2001 A (
2A 2004 r (
3A 2005 a
-R
ClS amp31
16
17
1 z
I TER-CEPT bull t
ooo9t21At 00022311 oooeos uu
OOl49tCAI 0010 2 middot 000960CCI
o049tl1At o04t55C61
MARKET SH ARE bull 5 I
-010670Ctl
-017600181
-001050
-0211AO
-0077 0
AOVER-TIS INti
lbl
000()39 000072
000018 0 0024liAI
-000079
SHARf
4ampgtVf Q bull
111
000140
-000170
RSQ
COl
66 485 862
607 6 3 t ()2
blti6 633
flASTI-
CI T Y PH
-007271(1
000441
-u0288l
clASTSHU (t)Ef
bull 10
tto
-234
210
( OOlltOCCI -126520 000922 181 001790
4A 200 7 18 00590liAI -038670 4e16 -032110 lell u c
OOOS74e 000700 025940
00067018) 0021001AI -02l100CAI
5 26
920
5A 6 c
200 8 26 0 03Sl9CA I 0012 26181 o 024511(
-005540
-0 81420 0001961AI O()Ol90CAI ooll240
441 604 891
-0020 19 274
6A 2010 16 Oe04224CAI o359lOIAI 9oq ll4015dCAI 061 6 -middot 005435CAI 00011 8 1AI 917 c 00535ltAI -009850((1 OOOA51J CAI -001610161 971
7A 2012 001862 000010 351 00187) oo1 a oooq45 ono1o6 439 c 001109 -027050 Oe0067SIAI -OOll60C81 897
- middotmiddot- -
8A 201t ll 000757 Ol4140CCI 814 ott9oocet 095 8 oooqo6 C C J 000069(81 811 c ootl2lampBI -080910CCI Oe009lUBI -OOtlOOCCI 960I
I DIIJIT pound11 1V OTIUTVTfAll-1pound1t- A _ 00 R - qq r - Q
1jl)OQCD
0
-middot middot--
l
_ __ _
I
i
0 00141 c c
91)7
I
T ABLE 2 REG ESSIOH RESULTS CCONTI
EO NR
IND CODE
iR bull
cos
I TEq-
CEPT
t1ARKE T
SHA E
AOVER-
Tl SING
SHARpound-cent
AuVER
RSO ILAST1-
C I T Y ELAST
SH CPtf (lt Cl) bull 11 I 41 I 5I (61 171 I 8 I f9) c 10)
9A 2 0 1 8 19 002397181 -007640 367 -005 246 le46 6 000605 000550CAI 597
-000029 G14180 OOl654CAJ -024li01At
lOA 2026 12 0028901AI -OlZl90 -0114 0 1 1 1 5 759
Ll (
llA 2021
f3 c
12A lOZ9
B
(
13A 2301
B
c
14A 2302
B
c
154 201
n (
16A 2551
8
10
35
10
20
19
24
001897(()
00141 6CCI
Oll9lHCI
010604 middot middot
Oll083
OOJ11JCAt
002059141
001411CIH
-000412
000293(81
000187
001159 (3)
ooo6a4cct
o00792CCI
OOl40dCAt
000006
-000076
OOlltl6fA)
o o0 5 64CCt
-034570
-o1middot o3o
-1659 90
Ol51lO
-046190CAt
Ol5450CA)
-005l60
002940
-030060(()
-0 43l50CCt
0 33650
-019080
00004
bullJ00741CAI
-000069
000466
0001 51CAt
000947CAt
O OOI)70CAt OOl750CA J
000175CCI
000671CAI
OOll5) f A t 002207CBI
0004b7CA I
- 007 1 70
001760
-001410CAt
- 0 16200CAt
-004210
- 1 207 70
705
96 0
542
466 805
460
614
921
832
9 61
996
424
550
9 15
56 1 651
884
-006114
019 037
03504lCA)
oo409Q
-141644((
491 -0 17664
607
871
lell
084
101
072
103
108 11IllOQfl)-- l
-037880fCJ OOl6l4CAJ -040l70fCI 0()
c
I
I
bull 9
9t
middot middot middotbullmiddot bull lraquo
TABLE 2 REGRE SION RESUlTS ICONTt
bull
EO I NO tHl I filER- f4A_K T AOVER- sbulluRE ttSQ flA Jl- iLAS T NR CODE cos CEPT SHARE JISING AOVfR bull (JTY SHR CiJCF
110Cl I f amp Itt 15J tbt bull 1c u caa
llA 2702 ll o079 6 81 Bt -019110 S6s -OlltO 111
a 004679 ooooo1 t69
( oOl26o -015160 001560 -022690 ltJr()I
ldA 270 3 19 0065lt -OIt2l60 220 -Ol57a2 11 8 e o 0 71t6IA t ol08 75tAt ( OOl983CiU -060llOC() OOl7191A t -O lltdOC A) middot 89
lqA 2704 10 005082 -Oil060 22b -l)i1977 lelO 6 -000080 oo2772C4 t 910 c 000907 -011 200 Oe04105CA t -046960 995
20A 2d06 1 II
O OlltStiAt -001120 837 00014 -))th O01380amp At -oooo 11 821 ( OOl526CAt -026700181 0002 5UAt 000530 969
2lA 2807 28 OOt9lt9CAt -00891 0 516 -0 03962 225I8 002095CCt OOOl lliAt 690 ( o02965CA t -056650((1 OOOt50CAJ -003l70ttl 911I
I 22A 2808 15 ot3l06 119730 655 1130) 090
0 - 007526CC t Oe003961At 835- middot-
c I oo7523 -067210 001l 2tA t -0091t80CCt 951
2lA 2809 21 Oe2067lCAt -1218101At 71 2 -10095((1 121 ttlIaOQ8 00797418) OOOI8lltt 686 Ill( 0092lU A t -Olltl70Ct t 000605CAt -OOit8701At 951 _
---- -
32 009506CAt
0 bull 06 981 I A J 006901 cA bull
00 24A 2810 -010960 5t 5 -OOltSO 7 56
8 -000011 s11 c -5811701At oooaoatAJ oot520IAI 888
-- 37 --- middot - -- -1 ) l lbull t5JJ r-l bullhmiddot1 --- - --
l5A
26A
27A
c
2RA
c
2 9A
31A
llA
24
13
597
Ci)55
14
l 6h
RSQ ELASTI- (lASTCOE f-
TABLE 2
EQ I NO NR CODE IU
REGRESSION RESULTS CCONTa
NR INTER- MARKET
(10S CEPT SHARE AOVER- SHARfcent
TIS lNG AOVER
) CITY StiR
c It) bull 5 co a I 1) bull 0 bull 9 c l 0 bull( 2 1
2815 16 003019 -009990 275 -o069bmiddotJ
6 001446 000224 282 c 002178 -054630 o02221CCI -0075 00 646
2901 29 00021t2 CAI 0)0520 592 0007 6 1 068
0 00105C8 t 000012141 60)g( OOOlJ51AJ -00126018) ooou271At -000u80CA) 974
36oa 1 3 003296(4) -oca29o -oo-q94
001755 000203 lt91t 002499((1 -04l820CCI OOl374CAI 0068)0 946
3612 0033281CI 011520 195 01684) oso
00 30 42C OI 00012 2 424 oo579ltAt -2 7l7801At 002 100IAI -003220 819
8 3617 btel 011789 081002259 009530
001448 oOOl OitCCI 732 OOl7 22CAI -0 46160CAI 0004961AI -OOll40181 989c
30A 3808 14 0021t 3 6fAI -000110 628 o023291C) -005 6 middot- _ middot- 0bull 02 04 1 IC I ooooo9 637 ( 001584((1 -0161)1)0 00021 91AI -0 00060
390) 14 002790CCt 006590 h8 t 0 0938 4 070
B OOl89bCCJ 000144 74
0021961(1 -OItlOlO OOl323CAI -004800 915
B
middot--- - middot 4 - middot- - dfll3904 12 004563181 0)8300 879 0428651(1 089 OQIll004 277CA I ooozstcea 919
c 0041)3181 -0-11250 0011881(1 -006270 972 0Ill
- J - bull bull bull bullbullbull middot - - middot - middot gt middotbull L)ofbullflbull middot middotfJltl -tn- middotmiddotmiddotmiddoto +f middot t middot bull
c
is
positive
2
Page 0 determineThe consumer goodproducer good split is sometimes difficult to
Sollle support for the split used in this paper is a comparison of goodness - of - fit
distributions of a
industries
set
advertising are
that industry
a sales increase of
the
highest aD10U3 those
measures for the two groups of industries Table 3 contains
measures for the 32- consumer goods industries and the 103 producer goods
2 Median R is 5 7 for the consumer goods set and 40 for the producer goods
Estimates - of y the elasticity of demand with respect to
given in column of Table 2 The hypothesis that y bull 0 with
alternative was applied for each ittdustry As shown in the
could not be rtjected for 24 of the 32 industries Of the
and 10 are positive For the remalnifg eight three are
and five are significantly positive
Of the threertegative elasetcities the largest (in absolute value)
cosmetics (code 2809) With all elasticity of -1 01 a one
industry advertising would gwerate a decrease in industry sales
more than one percetu
At the other extrema is toys and games (code 39 04) For
a one percent ittcrease in industry advertising would lead to
four-tenths of a pampreentmiddot Though not significantly different from zero
elasticity estimatbull for pt-bpuetary drugs (code 2808)
which are
Both drug industries have
to sales rafios (211 pe1 cent for 28 08 and 13 4 per cent
other hand flaIJo ing ext1ac s (mainly soft-drink SYUlS) also
is for
percent increase in
of slightly
a middottwo - tailed
table that hypothesis
24 14 are negative
significantly negative
9
the codetics and p-ropriacuy very high advertising
for 28 09) bull On the
has a high ind UY
advenisiltg t() sales ratio ( 8 0 per cent) bull but its estimated elasticity is very
small and not signifieantly different uom zero
middot It doe$ not seem unusual to find many elasticities near zero with a few
negative and a few positive If aggregate consumption is insensitive to aggregate
advertising as seems likely ld then what one industry gains must be lost by
others
Page 21
elasticity of demand does
(10) we can
if t 0 then
Finally a determination concerning the industry
not fix the individual firms elasticity Going back to equation
write y bull (1 - ) a + z y Since a gt 0 it follows thatii i i
Y gt 0 If Ylt 0 however y s sign is indeterminateij ij
Turning next to the coefficient of market share 20 of them are negative and
12 are positive with three of each sign being significantly different from zero
The largest positive value which is significant is for pet food (code 2010) and
a close second is mens amp boys suits and coats (code 23 01) The largest signifi
cant negative value is for cosmetics (code 28 09)
In the model some relations among y a 0 and 61 are implied If middota is
negative (1- y) is greater than one Since 61Y bull (1- a) (1 - ) 61Y is
pCsitive but it will be less than one if 0 is large enough Seventeen of the A
ys are negative in each of the seventeen industries 61 is also negative and les
1 (algebraicly) than y so 61Y is greater than one his is not evidence that the
degree of cooperation is low or zero in any industry of course A high o togethegt
with a negative y which is small (in absolute terms) relative to a can give a
value of 61Y greater than one On the other hand if 61Y were less than one
o would have to be greater than zero
For y gt 0 two conditions may hold If y lt a it follows that 61 lt 0 since
(y - a ) lt 0 It also follows that 61Y lt 0 regardless of the value of o bull
In addition the relation is an if-and-only-if one ie if 61Y is negative
then y lt a Three industries have negative values for 61y dairy products
except milk (code 20 04) synthetic fibers (code 2806) and photographic equipment
(code 3808) In all three cases y is quite small so a does not have to be very
large to give a negative value to al
If y gt a the model shows that 61-- gt 0 whatever the value of 0 bull There are
12 cases where both y and 61 are positive with 61 ranging from near zero
(grain milling code 20 12) to 1 20 ( proprietary drugs code 2808) Several
coeffiei-en_rs
--
15
o f to zero
Tab le 3 - Consumer Goods
Indus tr ies
2Distributions of R and Producer Goods
Page 20a
Consumer Producer
Goods Goods
0 - 9 9 0 3
10 - 19 9 0 1 1
20 - 29 9 3 20
30 - 39 9 3 17
40 - 49 9 5 16
50 - 59 9 8
69 9 6 Ii 1060
--f 670 79 9 2
80 - 89 9 4 4
90 - 100 1 1
Totals 32 10 3
Notes
2is used R i s cal-
culated as F F + ((N - K- 1 ) K ] 1 b_ F is the F
statistic Lor the hypothesis that all o f the non - intercept terms are simul taneously equal
1
2
--
-- Since generalized least squ-area regress ion
Page 12
o f these cases have values o f 6 1 Y rea ter than 0 8 (pet food (20 10) bread amp cakes
(20 14 ) miscellaneous processed foods (20 29 ) men s and boys suit s and coats
(23 01 ) p rop riet ary drugs (28 08 ) elect ric housewares (36 1 2) radio and
IV sets 17 and games and toys (39 For be than(36 ) 04)) 6 1 Y to greater 0 8
both o and cry mus t b e less than 0 2 That is in these industries the results
shown in Tab le 2 would hold only with very strong dominance of marke t demand
effects over brand swit ching e ffects and a very low degree o f cooperation
Using the concentration rat ios given in Table 4 the average concentration
ratio for the eight indust ries is 38 1 For the remaining 24 industries it is
43 2 This result would lend some support t o the proposition that concentrat ion
and cooperation are related if the group of eight industries are identi fied
as having low degrees of cooperation
One f inal note on model predict ions S ince both cr and o are assumed to
be non-negative 60 should also be non-negat ive In only one o f the 32 cas es
is 60 negative and it is not signif icant bull
cor pJ
7 1
7 0 4
1 4 7 5 1
)
7 3 z a o a t s 1 1 3 3 5 ( i V Z 8 tJ1 2 1 - 1 1 0 7 3 5 c middot v 2 8 1 0 3 2 -0 0 1 4 3 0 C 1 V 9 8 3 2 8 1 5 1 6 - 0 0 6 t1 6 9 C _ V 1 11 0 2 9 0 1 29 0 00 7 6 1 C l k V 7 3 1 3 6 08 1 3 - o n 4 9 9 4 3 3 3 6 1 2 2 4 0 1 6 4 3
3 1 7 1 3 0 1 1 7 d 7 8 3 3 8 0 8 0 0 2 3 i 2 3 0 6
P 4
3 9 0 3 1 4
1 2
0 09 3 8 4 39 1)1 0 4 l 8 6 5
T A LE 4 e I NCU S T R Y C A T E G O R Y
F TC NOM e tOOE ass
C 1 I c 2 )
2 0 0 1 1 6 2 0 0r 1 7 zo o l Z 2 0 0 7 1 8
z o o 2 6
z o t a 1 6
2 1) 1 2 1 5 2 0 1 4 1 1 2 0 1 a 1 9 2 0 26 1 2 2 0 2 7 1 0 2 0 29 - 3 5
2 3 0 1 10
2 3 02 z o
2 3 03 1 9
2 5 5 1 2 4 Z 1 0 Z 1 1 2 0 3 1 9 2 7 4 1 0 z a o 1 3
2 A 0 7 z s
E LA S T I -C I T Y
c 3 )
- 0 09 7 2 7
0 00 449
- 0 0 2 8 9 3
- 0 3 2 7 70 - 0 OZ il I q
o 4 u t s a 0 0 1 3 7 3
o 1 4 o o-0 0 5 2 4o -0 1 94 0 1
-C l 6 l l 4 0 1 JJ 3 7 0 3 t 0 4 2 0 0 4 0 9
- 0 4 1 8 44 - O l 7 t ) 4 -ll 7 1 4 0 b -0 3 5 7 8 Z
- 0 1 9 7 7 0 1)0 3 3 4
-o n 9 2
IJ A T A
CONV
ON-C ONV
( 4 )
t 1NV
C 1t-V C NV c ) v C - JNV C rlNV c n v c a v C ON V ( i) l V
C CtV
C t1tV C IJNV
C 1 V
c 1 v
Pa ge 2La
HEQ F t Noe x
c 5 )
6 2q
6 89 1 48
6 1 4 ) q 3
1 9 7 1 9
q z t s 5 a e z
l l 9 1
4 o 7 1 2 0
c oqt 1 Z
4 4 )-q 1 z
5 5 9 1 0 64 1 3 3
n 3
Inter industry analysi
Page 23
Equation ( 3 6 ) may be used to fo rmulate a rough test o f the cooperat ion
vs e f ficiency controversy I f we rest ricted ourselves to some subset o f
indus tries where the relat ion be tween y and a i s the same we would be lookingbull
at a sample o f indus tries for which the dependence o f V on H and o is a
quadratic wi th no squared terms
I f we now sub s titute (26) into ( 36 ) we get
(38) v bull a + (y - a) R + (y - a) o (H) - (y - a) H o (H) + u
If it is only d if ferences in marke t shares that affect advertising inten-
s i ty levels i f concentration does not af fect cooperation and i f o bull 0
( 38 ) reduces t o
( 39 ) V bull a + (y - a) H + U
Let this b e the null hyp o thes is and note that V is a linear func-
tion o f H If y gt a and the null hyp othesis holds the expected relat ion
between V and H is as shown by the pos itively sloped straight line in Figure
6 b etween points A and C
In this context the alterna ive hypothesis the degree o f
cooperation ( o) is a positive function o f concentrat ion
is that
(H) I f the func-
tion is as described in (26) and Figure 4 the relation between V and H
would be as shown by the curved line connec t ing po ints A and C in Figure 6 One way to test the null hypothe s is against this al ternative is to test fo r
the hypothesized curvature in the function relating V and H
Fi gure
R
6
v
ID c
H 1 0
I
i I I
sfigned
clasJ s
in def ini g
effec t is the
I
Page middot2lf
One charac teristic of produc ts which may be useful
classes wi th respec t to significanc e of the brand-switching
conveniencenon-convenienc e distinc tion developed by Porter ( 1 97 )
Using his identif ication of IRS indus tries as a s tar t ing point
each of the 32 industries used in this s tudy to one of the two
The ass ignments are not ed in Tab le 4
Using the conveniencenon-convenienc e dis t inc tion the
divided into two sub-samples
sample was
Four regressions were then run bull ene for e ch
sub-sample one for the to tal sample and one for the total sample with
a dummy variable for convenience goods Each regression was run wi th LB s
and wi th indus tries as ob servations The results are given in Tab le 5
For the LB level equations ( 1A - 1D) equation(37) was used ome
rearranging of terms gives v bull y - ( 1-o ) (y - a) ( 1-z ) + ui i i
bull y - 6 1 ( 1 - z ) + ui When observations are pooled across indus trif s yi
is an industry variable In these regressions the values o f y es t imated
in the first stage were used Since industries have been grouped so as
to reduce differences in (y - a) that term is t reated as a cons tant
within each group Finally H the Herfindahl index has been subst ituted
for o the degree of cooperation The resultant equation is linear
y ( 1 - z ) and H x ( l - z ) This is the equation for whi ch results arei i
reported in Table 5 In all regress ions all variables were weighted by the
square roo t of sales
in
raquo
- 34
3 1
i l T 4 8LE S R E G E S S I ON R E SUL T S i X T f N S I ON S T O T Y PE Of
AND L E V E L OF C ONC E kT R A T I O
E O O tt S
tP bull T Y P E NR C ONV O B S tO N-CONV
E L A S T I shy
T I C J T Y
O E - t-41( T SliA R E
H( fl F
1 1 - K O x middot K O X E R F
- H S a 1 1 - H S t 1 8 1 1 9 1
R S O
1 1 0 1
q z
t - bull 9 f 2 1
Z t P D l 6 fC ONV - 1 9 1 s s c 2 1
e - l O I e bull u q a
2 3 1 0 1 1 Nl N- t O V 5 7 1 ( 1 2 19 1 6 1 - J l S 2 5 4
l l l l t z q o a 1 - J o v z I
l ( I NO l l A l l ol u l t o o c bull J u l l
d S
c 3 4 t t r a l
bull Z D 1 0 l l 4 l l 2 3 -gt 2 l l 2 0 - 4 lS 5 4 l C 4 1 o e 6 t l b b 5 1 1 0 1 1 1 c - 1 9 5 1 t d S 2 1 tV
OQ10 T-R A T I O S A E G I VE N l P A RE N T H E S E S l E V E L S f S I G 1 F l C ANC E A - 99 5 B - q 9 t - s NS J UC E NO I Nl E R C ( P T I S U (j f middot) I N f Hf t OUA TIOI S 6 E f 3R E A O J U S T H E N T F OR 1-t E T E R O SIC E O A S T f C I T V pIraquo
1
middot
I
R SO I S C A L C UL A T E D A S f I f bull 1 1 bull 1lt 1 K t t- E E N I S THf U A1 B E R O f fJe S E fi V A T I O S F t S HtE F - S T A T I S
I N DE P E ND E N T y II
T I C F O T H E U Y P O HIE S I S f A l l C JE F F I C H N T A R E QU A L HI l f RO A f40 K I S T H f UUM 8 E R OF Q ( A fL f S
I l l
1 A
1 6
t C
f 1 r R I l l
1 2 1
L Jt
L B
l q
L 1
on s T Y P E bull 2 )
P OUC T
X
H S I l l
l O l A I bull l 0 7 1
O O b O J
1 2 00 1
1 b 2 1 C I l l Z U
l l S b 1 -1 1 0 1
rlt F X l l - fi C f l
bull 1 I
l l
- ()4 2 3 l f H
1 - 2 7 9 1
K O X
1 1 - ti E R F I
l d l
S4 l i A I 2 l b
1 3 00
K O X -t f R F t S J
I 1 - HER F t
)
60 9
z o bull
0
t l l
3 8 8
1 7 6
5 64
64
N f bull
tl S bull
I l l
I 4 1
t ONV
NON-t ONV
A l l
A ll
c o v NO- ( ONV
1 4 1
1 5 1
- l S b
c - 1 1 2 1
5 8 2 C A I 1 6 7 0 1
middot I l 9 H
l 1 il
1 4 1 6 1
E L A C) T J shy
l t C I T Y
I S I
l b l
- 0 2 0
1 - 1 4 9 1
O Z 3 2 1 A I C b 6 b l
0 0 9 2 -J
1 1 b 4 1
o z t t t l 1 2 2 1 1
oru - Ht tt F
1 gt X bull b I
Page 25 bull
Mean val -as ior relevan t variables for the two sub-samples are
Conv Non-Conv
Adv Sales 4 7 2 3
iit Share 2 7 3 2
Ela cicity - 05 3 02 2
Herf indahl 7 3 1 860
That the dist nct ion between coft9enienee goods and non - convenience
goods mat ters is supported not only by these means but also by the regression
results Equat ions lA and lB are different in almost every respec t Only the
concentrat ionmarket share int eract ion term seems to mat ter for convenience
good s Jus t he opposite holds for non - convenience goods both elasticity1
and market sl e are highly significant but the concentrat ionmarket share L
int erac t ion t=rm is not
Given that one minus market share is used these equ3 tions the coeffici t
of market share is the negative of what is sh in col (6) of Table 5 For the
regress ion for non - convenience LB s reported in equation lB then LB s with larger
market shares have lower advertising to sales ratios
On the question of the impact of the degree of cooperation the story is mixed
For convenienc good s the coefficient is pos itive as equation (31) predicts when
there is a po sitive degree of cooperat ion Fo r the non-convenience goods LB s middot
concentration has no effec t
Aggregat ion t o the indus try level and then rer unning the equations $ives the
results shown in (2a-2d) of Tab le 5 As expected R2 goes up and levels of significanc e
fmiddotJr individual coefficients gJ down In add ition the coeffic ient of H ( l -H) inmiddot
quat ion 2A is no t significant though it is po s i tive Since 2A is the indus try level
I
--
all but one of them and that one is not significant For 18 of the 31
wo uld have on its adve rtising to sal es ratio
cognate of LA the variable H ( l-H) pl ays the same role in 2A as H ( l -z 1) p lays
in LA
Vconomies of scale in advert ising
The literature on advert ing contains a number of definitionbull o f economies
of scale and of findings eone rning its presence and magnitude Before I loo k
a t the data with the hope o f mmenting about scale effect s then I wan t to
define the term as I use it
By economies of scale I mean a decrease in the number o f units of adver-
tising which are needed to generate a unit of sales as advert ising is increased
Given the definition of the firm s own elast icity of demand with respect to
advertising (see the text foll wing equation (1) ) the technical definition
is that rii gt 1 Diseconomies of scale are def ined symmetrically i e
Yu lt 1
ith a cro s s - section oi observations on advertising intensity
(v bull a s1) and onmiddot advertising (a ) we may regress the former on the latter 1 i i
The results of such regress ions for the 32 industries are given in Table 2
lines lB ZB bull bull bull 32B For the 32 eases the regression coefficient is positive it
positive coefficients the coefficient i s signif icant
The que tion now is whether these results ar e evidence of pervasive dis-
economies of scale in advertislng I think no t Given my as sumption that the
data I observe are equilibrium results I may ri ghtly conclud e that almost every-
where high levels of advert ising are as sociated in equilibrium with high ratios
of advertising to sales I may no t conclud e anything however about the
impact that a move by some firm away from its equilibrium level o f advertising
middot- Page 27
Given the assumed relation in equation ( 37 ) there is a corre spondshy
ing equilibrium equation relating v and ai The equation is non - linear i
and I have not found a simp le way to characterize it It is po ssib le to
det ermi ne its deriviative with respect to a b y takin$ tne de riviative of ( 3 7 ) i
When that is done the result is
v - 1(40) i bull sl a a z ( 1-z )i i ira-l
Sinc e all the other terms in (40) are non-negative the sign dep ends only
on the sign of 6 1 which is in turn de pend ent only on the sign of (y - a)
That the signs of the coefficients of a in equations lB 2B bull bull bull 32B arei
not always equal to the signs of the coefficients of z in LA 2A bull bull bull 32A i
is still a mystery to me and something to be explored
bull
Summary
My purpose in writing this pa per wa s t o attempt to integrate explicit micro-
economic mode lling reasoned econome tric specification of error terms and high
quality data to exp lore some questions about adver tising by large firms
al quite important improvements could be made in the first two of those areas
and much data ma ssag ing is yet to be done in the third
Concerning the subst an ce of what I have done I think six things are im-
Page 2 8
Severshy
V and Conclusions
portant
1 Explicit modeling is worth the ef for t if for no other reas on than
that it provides a basis for choosing fr om a varie ty of funct ional
forms
2 The same is true for error specif ica t ion I did in fact look at scatshy
ter diagrams for all 32 industry categor ies and they vir tually all show
the het eroscedast icity which I as sumed and for which I correc ted
3 The predicted relation be tween advertis ing in tensity and market share
shows up clearly in on ly 25 of the case s examined I have ye t to
explore why that may b e the cas e
4 Some evidence concerning the presence and impact of coopera tion was
produced but it is not clear ly pervasive
5 The dis tinction between convenience goods and non-c onvenie nce goods is
unambiguously a good one
6 Virtually no evidence concerning econ omies of scale in adver tis ing can
be gleaned fr om this study given its assumpt ions
More work is called for on many of these issues my study seems to have
raised more questions than it has answer ed This is pr obab lly due to the richshy
ne ss of the data source since I had the oppor tunity to address several issue s
at once
- -- - Q --
Foo tnotes
Manager Line o f Business Program Federal Trade Commission The author depended heavily on several staf f members of the LB Pro gram for statistical and clerical support in particular Joe Cholka George Pascoe and Haro lene Jenkins Helpful comments were given b y Richard Caves Dennis Mueller Michael Porter and F M Scherer When the paper was in its early stages o f development both S teve Garber and Jon Rasmussen former economists in the LB Program provided eery useful cri tiques The views expressed here are my own o f course
1 The basic exp licit model of the role of advertising is Dorfman and S teiner ( 1954 ) and the models in this paper may be seen as ano ther extens ion of their work
2 For a review o f models of oligopoly which treat prices and quantities as relevant variables see Shubik ( 1 959) Extens ions to take advertis ing int o account are fairly s traightforward
3 Dorfman and S teiner dealt with a price-setting firm For a monopolis t inverting the demand relation to show Pbull P (Q A) _instead o f Q- (P A) has no effect on the results o f the analysis For oligopolists who face a sys tem of demand relations inversion does have some impac t on the results The dif ferences are not trivial nonetheless they will no t be explored in this study
4 Shubik ( 1 959) p
5 From a formal mathematical point o f view the function is similar to a1 i Legrangian funct ion of the form - r rr
L bull 11 + A ( j - (i i i+j
where the indicates some fixed level of the profits industry Following the same logic which resul ts in and income constraint as the marginal utility of the of the indus tr in the obj ective of firm i
6 Hall amp Weiss ( 1 967) p 323
7 See Schmallensee ( 1 97 2 ) Ch 2 pp 1 6-47
8 On this poin t see Imel amp Helmberger ( 1 9 7 1 )
9 Long ( 1 970) p 130 - 1 39
1 0 See Schmallensee ( 1 972) Ch 3 pp 48-8 7
rrj i+j ) of the rest of the
the identification profits of the rest
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980
ith
Page 2
A bull tai as industry adverti sing and zi bull siS as the market share of the
firm
II TWO POLAR CASE OLIGOPOLY MODELS
Cournot Let each firm maximize its profit independently of all other
firms That is
(3)
- 1
si- -- - 1 - 0yii ai
This implies that
-(4) v bull yiii
Let r bull (yij) Further let Ad be a diagonal matrix with the same diagonal
as the matrix A Then the N equations in (4) may be written compactly as
(5) v bull rd l
where is a vector of ones Since rd is a diagonal matrix (5) may be reshy
written as
(6 )
where if is a vector
z on the diagonal7
Chamberlin
3
v bull
d is a diagonal matrix containing the elements of z
4 At the other extreme let each firm maximize total profit
for all firms Letting IT trribull
an
Page 3
( 7 )
- 0
This implies that
(8)
(9)
-
v bull bull
bull
bull
zl [ Y21 Y22 bull bull Y2N J
-1zl 0
-1z2 r z
0 1
Page 4
Writ ing (8) compactly gives
the matrix in the two equationsA comparison of (6) and (9) shows that
differs only in that all the elasticities are present in (9) but only the
awn-elasticities are present in ( 6 ) Since advertising expense mus t be I non-negat ive the first-order requirements are that v be the maximum of t
i
the amount shown in 6) or (9) and zero
III A SIMPLIFIED DEMAND STRUCTURE
So far the demand equation system is comple tely general I will now
move to a particular demand structure which is charac terized by an industry
elasticity of price with respect to advertising a general brand-swi tching
elas ticity and a dependence of the firm-specific elas ticity on its relative
size I make the expli cit assumpt ion that
n pound
)l
0 AA
Consider the identification of y as an elasticity of industry price
Page 5
(10)
where y is an
- (y- a)
industry demand elasticity a is a brand-swi tching
el astici ty and z1 is the market share In matrix no tation
(11) r bull ai + (y- a) t
wi th respe ct to advertising 0
le t ting S denote dS observe that
0(12) SS bull
-
-1 0s t s - -1 0S t d - 9- -
apl apl aplaa1 aa2 a 0 a
middot( a Pl
( 13)
-
-0 a bull 0
a
) 0 X s
-1 0dAA bull then d
a bull
J 0
(JA) middot and
Let the sc alar (SS)(AA) be denoted by y so ya zft
If
1
d
-
I
I
(11)
--
y
I -f
l J- a)
a) l zt
Substitutin g into (13) gives
i [ ai + (y(14) -I- bull S A
bull az + (y -
Pa ge6
I I II
bull
since l bull 1
Under these assumpt ons then y is the elasticity of industry sales
(and average industry p ice) with respect to industry advertisin g Note
that if y bull 0 so that there is no industry sales effect there will still
be positive advertising if there are brand-switching effects
Turning now to the brand-switching ef fect observe that
(15)
Noting that
-1 deg -1d z -z -
-
0 a
Page 7
(15) may be written as
0 (AA)
lx 2x 1x
ith
1y
i
First assume no
that if there are
will be no changes in market shares
- 0a bull
Page 8
Equation (17) is a general statement of the relation betbulleen changes in
advertising levels and changes in market shares which would acompany them It
may be used to make some general observations about brand-switching effects
industry effect eg y bull r t bull 0 and assume further
equal percentage changes in all advertising le =ls then there
Under these further assumptions
0l and dz -1 bull (iA) ltr assumption the second term on the right hand side of this expression is
- By
zero
necessary for rt bull 0Consequently for market shares to be unaffected it is
when y bull r l bull 0 This requirement is satisfied by equation (11)
Secondly consider effects on market shares if one firm inc ases its abull ar-
tising and others do not As a general assumption we require smiddot ch a change o
lead to a decrease in the shares of all the other firms hat is let
0 0aiai gt 0 and a a bull 0 for j + i For any matrix X with N rows and its trans-j j
pose X let X
row of X
bull middotmiddotmiddot where is a column vector which contains the
elements of the Under the assumptions about the iis given just
d-1 0 0above it follows that r becomes (aiai) iimiddot ehFor jzj lt 0 under these conditions then it follows that the j- element
N of (I- l ) lt 0 so middotij - il lt 0 or yij - yik lt 0 If this
kbulll N condition holds for all Y ij where j + i then it follows that yii - z
k Y gta
k- 1 ik since the th firms share must increase to offset the decreases in all other
shares
ei ith
1v ( ) h i
7urning
where is the
leads to z
to the r matrix given in (11) we note that 1 = cr + (y - cr) i i zi elementary vector with 1 in the position and 0 elsewhere This
bull cr + (y - cr) zi Since bull required is thatcr at- zY wYij (y - cr)zi - cr + (y - cr) zilt 0 or cr gt 0
These observations may be summarized by substituting (11) into (17) bull
giving
(18)
i s
d -1 zz -
-
-
+ (y - cr) tz
d -1 0cr(I - l ) a a bull
cr 1 z - (y - cr) 0 a
o
If ai increases and the other a s are const t cr must be positive to havej
a positive effect on zi A positive cr also bull plies a negative ffect of aJ
f all a s in-on zi other a s (including a ) fixed j
eluding a1 change by the same percentage zi remains unchanged Finally k i
if cr bull 0 there is no brand-switching effect
For a to have no effect on p requires yij bull 0 From (10) thisj i
amounts to y bull cr that is the industry demand elasticity and the brandshyswitching elasticity are equal If y gt cr gt 0 the industry elasticity yij doninates If y lt cr y lt 0 the brand-switching elasticity dominates ij
making
Page 10
Substituting ( 1 1) now in (6) the Cournot case and in (9) the
Chamberlin ease gives
-(19) Cournot vO a + (y - a) z
Chamberli n vl - y lmiddot th
The 1- element is then
-(20) Cournot vO a+ (y - a) zi i
-Chamberlin vl yi
In (19) and (20) the superscript 0 is used to denote total
independence of decision and the superscript l is used to denote
total interdependence of decision making
The relation between v and vf depends solely on t relation between
y and a f y gt a so that the industry effect dominat 3 vf v with
equality in the trirlal case of z bull 1 Cooperative irms would have ai
higher advertising intensity than non-cooperative fi
If there is equality between the industry elasticity and the
y Whe therbrand-switching elasticity or y bull there is cooperation among the firms has no effect on advertising inten-sity And if brand-switching is more important with y lt a e get
vi lt v cooperative firms will have a lower advertising intensity than
non-cooperative firms The three situations are depicted in Figure L
Given the results for the firm level variables calculation of the
corresponding re sults for the industry is s-traightforward Let
B bull t zf bull z z be the Herfindahl index of concen tration and notethat
z - bull l z bull 1 and that V bull A S bull a i 5 bull (siIS) (asi)i
bull middot Then for the general case from (6) and (9)
bull
Page lOa
y lt1 bull y y
10
f
V0 bull z [ a1 + (y - cr) J bull a + (y - a) H vl bull Y bull y
11Page
( 2 1) vo bull z d -1 r bull r z bull z- d d- i )11
l -1 N N v bull z d r z bull r z bull l y z - ij jibulll jbulll
and for the simplified case from (19)
(22
Since the forms of the equations in (22) are the same as in (20) all
of the analysis of (20) holdS for (22) except that His substituted for
middot z Advertising intensity will be higher for an industry with cooperativei
firms than for one with non-coope rative firms if the industry demand elas-
ticity dominates the brand-switchine lasticity If the ewo elasticities
are equal then so will be advertiSl intensity for the cooperative firm
and non-cooperative firm industries If brand-switching dominates the
cooperative firm industry will have lower advertising intensity
In the industry model context the possibility of a critical concen-
tration level may be easily introduced That is assume that at values
of R below R the firms are non-cooperative but that for alues equal to
or greater than H they are cooperative That is
(23) V bull a + (y - a) H if H lt H bull y if H gtmiddot H
This function is shown in Figure 2 for the three relamptions be eween y and a
IV THE D GRE OF COOEtRATION
The models depicted in Figure 2 are based on a very simple notion about
the relation between concentration and the cooperationnon-cooperation charac-
teristic They use what is essentilly a step function relating concentra-
Figure
I -
J H H
Page l la
2
v v v
a
I J I y a bull Y y ---i
I I I I I I II I
H 10 H H 10 H 1 0
bullbull
bull bull
Page 12
tion and the extent or degree of coo peration as shown in Figure 3 Using
o as a symbol for the degree of cooperation eq uation (23) can be rewritten
as
bull(24) v (1 - o) vo + ovl bull (1 o) [a+ - a) H ] + oy- (y
vhere
(25) o bull o(H) bull 0 if B lt B
Hbull 1 if B gt
The specific relation between concentration (H) and the deg ree of
cooperation ( o) shown in Figure 3 is more restrictive than necessary and
bas less appeal an one which shows a smooth increase of o from 0 to 1 as
B increases fro to 1 Such a smooth function is shown in Figure 4 he
degree of cooper ion is low until H gets close to H increases greatly as
B goes through R attaining a high level for an H larger than B but still
close to it and then goes to 1 as B goes to 1 More formally
(26) 0 bull o(H)
bull 0 if H bull 0
bull 1 if H bull 1
do gt (IdB
d2o- gt 0 if H B dBZ
0 if H H
lt 0 if R gt H-
If (26) is now substituted for (25) in (24) a corresponding smoothing
of the functions shown in Figure 2 is accomplished he smooth functions
Figure 3
I
10
H 10 H
Page 12a
Figure
Page 12b
4
10
H 10 H
l
(27)
Cl - o) o
[ (1 -
[ (1
t y ij z
j
are given in Figure 5
Page 13
The equation in (24) was constructed as a hybrid of two results derived
from first-order equations for the non-cooperative and cooperative models
It may also be constructed directly by using l bull (1 - o) + on as thei i
th 2objective function for the i-- firm Treating o as a constant the first
order equation is
0 an aa
i middot
n i
Using equations (3) and (7) and setting --- bull 0 givesaai
(28) - 0
After collecting terms and converting to matrix notation we get
(29) - Cl - o) + 0 z-1vi yii i
N
j bulll
+ 0
+ o v1
d-1 r z z
-(30) v Cl - o) rd
-
o) r +d
d r J _ -1- d z
and middot
d) r + 0 rjd
l r z
v - z --(31) zv
- (1 0- +d) t r z- d-
If we now impose the simplified demand structure by substituting (11)
we get
- (1 - o) a +( 32) + oy and
Figure
l
-y
v
Y
H
r
I I
v
a
y
H 10 H
v
a
bulla
HH 10
bullbull
(33)
i
Page 14
v - (1 - o) [ a l + (y - a) ] The corresponding equation for Vis given in (24)
If the relation between industry advertising intensity (V) and
concentration (H) is non - linear via the impact of concentration
on the degree of cooperation does the relation take the form of a
auadratic as some other investigators have proposed (Greer ( 1971 ) Martin (1979) Strickland amp Weiss (1976) ) That question can
be approached from two perspectives - conceptual and empirical In
the next section I will try to apply some statistical material to it
here I want to explore the question in the context of the models
developed above
Referring to equation (24) taking the deriviative with respect
to H and letting bull (H) gives
bull( 3 3a ) 3V (y - a) (1 - o) + (1 - H) 3H dH
Given that o and H each are bounded by 0 and 1 and that o is an in-
creasing function of H the term in the brackets in non-negative The
sign of the deriviative then is the same as the sign of (y - a) If the industry effect is dominant (y gt a) then advertising in-
tensity will increase with concentration If the two effects are equal
then advertising intensity will not vbullry with concentration and if the
brand-switching e ffect is larger advertising intensity will decrease
with concentration
For given advertising elasticity parameters then it is not pos-
sible to show an increase in A S up to some H and the n a decrease
he relation may be non - linear but it is monotonic over the permis-
sible range of H
Page 15
One possibility is to introduce a relation between (y - a) and H
If industries characterized as dominated by the brand-switching effect
also were more concentrated then a cross industry comparison would
show an inverted U shape relation between AS and H I know of no
support for such an association between ( y - a) and H
V ERROR SPECUCATION
So far I have implicitly assumed that there are no errors in the
model I now assume that the error
thvariable in the equation for the i- firm is additive that its mean is zero
that its variance is inversely proportionate to its sales and that all error
term covarlances are zero hat is
34) v bull (1 - o) [ a + (y - a) z ] + oy + u i i i
- - -
In vector notation
(35) bull (1 - o) [ at + lt - a) J +
E() bull 0 Cov (u) -
0
If we now ca lculate V bull bull we get
oy + U bull
(36) V bull (1 - o) la + (y - a)R J +
U bull E(U) bull 0 Var (U) bull
P age 16
The error term specifi c ation which follows Hall
wi th the ratio of profits
6 heir analysis seems
as wellbull particularly given
and Weiss was
applied by them to equa tions to equity or
asse ts as dependen t variables to carry the same
weight in the present context the substantial
evidence presented by marketing analysis on the tendency of company de-
cision makers t o use the advertising to sales ratio as the decision
7variable
The zero covariance assumption on the other hand1 is particularly
troublesome Within a probably not true since
many even ts outside the
single industry it is
industry would have
industries
similar effec ts on all the
8firms in the industry Between it is probably not true either
since many companies produce in more than one indus try and even ts in the
firm would tend to effect all of its activities The extension of the
model to allow for non-zero covariances is certainly called for
--- -
Page Ll
Athird characteristic of this specification also deserves some comment
only on equation is shown for the firm or industry Several haveinvestigators
included an advertising equation ina larger system of equations (Comanor and
Wilson (1974) (1979) and Weiss (1976) so in myMartin Strickland Idid
dissertation1 Ihave ewo defenses for presenting only single equation results
here One is the suggestion by Comanor Wilson that simultaneousand equation
verybias may not be important in this context he second is that Iam will-
ing to see the model extended to include a profitability equation and I intend
to move that direction in the near futurein
Onthe inclusion of aconcentration equation on the other hand I amskep-
tical he problem isthat in some fundamental sense equations in models of
industrial organization should be oriented to the decision-making contexts of firms
Ifeach firm in anindustry with Nfirms has only one decision variable then
there can be only Nindependent equations which are based on the first-order
optimization equations It not show large ofis difficult of course to a number
functional relations which follow from the first-order conditions point here he
isthat only Nof them can be independent
Consider for illustrative purposes only a situation in wich teh N
itsquantities say q are set exogeneously and each firm determines advertising1
of and level say ai where a is the true optimal value a a bulla
1 1 i i + i
Ifp a) then the Nq
s and the Na s will determine N p s say pii bullp
i(l i 1 i
Now given Eand 9 we can determine both the vector s andthe vector v since
- bull til
ands We can also determine S s z bull s-1 s-ii
v and we
industry concentration isHbull (alternatively
V and Hare functions of the same
pendent Furthermore if there are
VM) and (H1 H2 bull bull bull )contain
VI STATISTI CAL APPLICATIONS
Page 18
Given the observed values of might be tempted to conclude that we
have a two equation system in which v and z are jointly determined That would i i
be inappropriate however since the 2N random variables (v ) are determined by
onlt N basic random variables t The variance-covariance matrix for ( ) must
be singular bull
The extension of this observation to industry level variables is straightshy
forward Since observed industry advertising intensity is V bull l v and observed 4
C4 bull z ) it follows thati l i
set of random variables t and cannot be indeshy-
M industries then the variables (V1 v2 at most M independent random variables
Given the assumptions made above y a and o are constant within a
given industry That being the case (34) can be written as
(37)
s1 bull (1 - o) (y - a)
If and
So + +
(1 - o) a + oy
81 are esti mated for an industry using a suitable statisticalSo procedure giving and 81 an estimate of y is then available The sumSo
Intra-industry analysis
1974 of Business Re ort -
categories are identified
Page 19
of So and S1 is y so y bull So + S1 is the estimate of y
If o is assumed to be bounded by zero and one the sign of 61 is also
the sign of y - a The estimate of So is of cou se an estimate of
(1 - o ) a + oy Moreove if y gt a then a is less than 6o- (1 - B) a+ 0(
Cor esponding results would
hold if y bull a or y lt a
If 61 gt 0 then we would estimate that a lt 6o
bull
For purposes of estimating this model 32 manufacturing industry categories
were selected The data are fram the Line of Business forms filed with
the Federal Trade Commission they are discussed in detail in the Annual Line
which is available from the FTC The 32 industry
in Table 1
Simple regressions for equ ation (37) were run for each of the 32 industries
with data for each of the firms which filed in the industry constituting an ob-
servation The results are given in Table 2 lines lA 2A
1974
the intercept
For all the A
equations generalized least quares was used th both and market
share bein scaled bv the square root of sales
The 32 industries were taken from the larger set of all manufacturing industry
categories in the 1974 data files Two selection cr iteria were used thereLB
had to be at least ten reporting companies and the industry had to be a consumer
goods industry or a producer goods industry which is strongly associated with a
related consumer goods industry so that adve rtising by the firms in the producer
goods industry impacts on demand in the related industry (eg soft drink syrup
and soft drinks) There were 135 industries with at least ten companies - 103
producer goods and 32 consumer goods
273
274
TABLE 1 INDUSTRY CATEGORIES USED
FTC
COOE OESCRI P T I ON _
2001 MEAT PACKINGt SAUSAGE S ANO OTHER PREPARED
MEAT PROOUCTS
2004 DAIRY PRODUCTS EXC F LUID MILK
2005 CANNED SPECIALTIES
2001 FROZEN SPECIALTIES
OEHYORATFOt AND
I NC LU 0 ING PRE SERVE S
SOUP MIXESbull
SEASO lNGSt ANO
OTHER PET FOOJ
RAIN MILL PROOUCTSt
ANO PREPARED
RELATED PRODUCTS
PRODUCTS
Z008 CANNEOt ORIEOt PICKLED FRUITS
ANO VEGET ABLES JAMS JELLIES DEHYDRATED I G TA3t E
S40CES ANO SALAD ORESSINGS
2010 DOG CAT A O
2012 FlOUR t OTHER RICE
MILLING fiLENOE) FLOUR
2014 B EADt CAKE ANO
20113 CO FECTIONERY
2026 BOTTLED A O CANNED SOFT DRINKS 2021 FLAVORING EXTRACTS AND SYRUPS EC
2029 MISC FOOOS AND KINDRED PROJUCTS EXC
ROASTED COFFEE
2301 MENS AND BOYS SUITS AND COATS
2302 MENS ANO BOYS FURNISHINGS
230 3 kOMENS AND MISSES OUTER EAR
2551 HOUSEHOLD FURNITURE
2702 PERIODICALS
Pqe 19a 1
RELATED 1972
src cooe
2011 f 3
202tX2026
2032
203pound
2033t4t5
2047 2041t4t5
2051 2065 2086
2081
209XZ05
231
232
233
251 272
2703 BOOKS
2704 ISC PVeLISHING
2806 ORGANIC FlEERS
2807 DRUG$ ETHICAL
23234
PT 2 S 3
2808 ORUCSt PROPRIETARY
2809 PERFU ES t COSMETICS ANO OTHER TaiLET
PREPARATIONS
2810 SOAP AND OTHER CLEANING PREPARATIONS 2815
2901
PEST1CIDES AND AGR ICULTURAL
PE OLEUM REF I N ING CHEMICALS EC
3606 HOUSEHOLD COOKING EQUIPMENT
3612 HOUSEHOLD APPLIANC ES NEC INCLUJING ELECshy
rqlC HOUSE ARFS AND FANS ANO SE I G MACHINES
3617 RADIO AND T V RECE I VING SETS 3808 PHOTOG APHIC E UIP amp SUP LfESt EXC
3qo3
PHO TOCOPYING
SPOR TING A D E QU I P M ENT f SUP P L I E S ATHLETTC GOODS NEC
3904 DOLLS GAMESt TOYS AND CHLOqENtS VE ICLES
PT283
2844
234tX2844
287 9
291 3631
3634t6t9
36H
PT386l
3949
15
943
middot
l A6l f 2 R(GRESSION RESULTS
EQ I NO
NR CODE cu 121
lA 2001 A (
2A 2004 r (
3A 2005 a
-R
ClS amp31
16
17
1 z
I TER-CEPT bull t
ooo9t21At 00022311 oooeos uu
OOl49tCAI 0010 2 middot 000960CCI
o049tl1At o04t55C61
MARKET SH ARE bull 5 I
-010670Ctl
-017600181
-001050
-0211AO
-0077 0
AOVER-TIS INti
lbl
000()39 000072
000018 0 0024liAI
-000079
SHARf
4ampgtVf Q bull
111
000140
-000170
RSQ
COl
66 485 862
607 6 3 t ()2
blti6 633
flASTI-
CI T Y PH
-007271(1
000441
-u0288l
clASTSHU (t)Ef
bull 10
tto
-234
210
( OOlltOCCI -126520 000922 181 001790
4A 200 7 18 00590liAI -038670 4e16 -032110 lell u c
OOOS74e 000700 025940
00067018) 0021001AI -02l100CAI
5 26
920
5A 6 c
200 8 26 0 03Sl9CA I 0012 26181 o 024511(
-005540
-0 81420 0001961AI O()Ol90CAI ooll240
441 604 891
-0020 19 274
6A 2010 16 Oe04224CAI o359lOIAI 9oq ll4015dCAI 061 6 -middot 005435CAI 00011 8 1AI 917 c 00535ltAI -009850((1 OOOA51J CAI -001610161 971
7A 2012 001862 000010 351 00187) oo1 a oooq45 ono1o6 439 c 001109 -027050 Oe0067SIAI -OOll60C81 897
- middotmiddot- -
8A 201t ll 000757 Ol4140CCI 814 ott9oocet 095 8 oooqo6 C C J 000069(81 811 c ootl2lampBI -080910CCI Oe009lUBI -OOtlOOCCI 960I
I DIIJIT pound11 1V OTIUTVTfAll-1pound1t- A _ 00 R - qq r - Q
1jl)OQCD
0
-middot middot--
l
_ __ _
I
i
0 00141 c c
91)7
I
T ABLE 2 REG ESSIOH RESULTS CCONTI
EO NR
IND CODE
iR bull
cos
I TEq-
CEPT
t1ARKE T
SHA E
AOVER-
Tl SING
SHARpound-cent
AuVER
RSO ILAST1-
C I T Y ELAST
SH CPtf (lt Cl) bull 11 I 41 I 5I (61 171 I 8 I f9) c 10)
9A 2 0 1 8 19 002397181 -007640 367 -005 246 le46 6 000605 000550CAI 597
-000029 G14180 OOl654CAJ -024li01At
lOA 2026 12 0028901AI -OlZl90 -0114 0 1 1 1 5 759
Ll (
llA 2021
f3 c
12A lOZ9
B
(
13A 2301
B
c
14A 2302
B
c
154 201
n (
16A 2551
8
10
35
10
20
19
24
001897(()
00141 6CCI
Oll9lHCI
010604 middot middot
Oll083
OOJ11JCAt
002059141
001411CIH
-000412
000293(81
000187
001159 (3)
ooo6a4cct
o00792CCI
OOl40dCAt
000006
-000076
OOlltl6fA)
o o0 5 64CCt
-034570
-o1middot o3o
-1659 90
Ol51lO
-046190CAt
Ol5450CA)
-005l60
002940
-030060(()
-0 43l50CCt
0 33650
-019080
00004
bullJ00741CAI
-000069
000466
0001 51CAt
000947CAt
O OOI)70CAt OOl750CA J
000175CCI
000671CAI
OOll5) f A t 002207CBI
0004b7CA I
- 007 1 70
001760
-001410CAt
- 0 16200CAt
-004210
- 1 207 70
705
96 0
542
466 805
460
614
921
832
9 61
996
424
550
9 15
56 1 651
884
-006114
019 037
03504lCA)
oo409Q
-141644((
491 -0 17664
607
871
lell
084
101
072
103
108 11IllOQfl)-- l
-037880fCJ OOl6l4CAJ -040l70fCI 0()
c
I
I
bull 9
9t
middot middot middotbullmiddot bull lraquo
TABLE 2 REGRE SION RESUlTS ICONTt
bull
EO I NO tHl I filER- f4A_K T AOVER- sbulluRE ttSQ flA Jl- iLAS T NR CODE cos CEPT SHARE JISING AOVfR bull (JTY SHR CiJCF
110Cl I f amp Itt 15J tbt bull 1c u caa
llA 2702 ll o079 6 81 Bt -019110 S6s -OlltO 111
a 004679 ooooo1 t69
( oOl26o -015160 001560 -022690 ltJr()I
ldA 270 3 19 0065lt -OIt2l60 220 -Ol57a2 11 8 e o 0 71t6IA t ol08 75tAt ( OOl983CiU -060llOC() OOl7191A t -O lltdOC A) middot 89
lqA 2704 10 005082 -Oil060 22b -l)i1977 lelO 6 -000080 oo2772C4 t 910 c 000907 -011 200 Oe04105CA t -046960 995
20A 2d06 1 II
O OlltStiAt -001120 837 00014 -))th O01380amp At -oooo 11 821 ( OOl526CAt -026700181 0002 5UAt 000530 969
2lA 2807 28 OOt9lt9CAt -00891 0 516 -0 03962 225I8 002095CCt OOOl lliAt 690 ( o02965CA t -056650((1 OOOt50CAJ -003l70ttl 911I
I 22A 2808 15 ot3l06 119730 655 1130) 090
0 - 007526CC t Oe003961At 835- middot-
c I oo7523 -067210 001l 2tA t -0091t80CCt 951
2lA 2809 21 Oe2067lCAt -1218101At 71 2 -10095((1 121 ttlIaOQ8 00797418) OOOI8lltt 686 Ill( 0092lU A t -Olltl70Ct t 000605CAt -OOit8701At 951 _
---- -
32 009506CAt
0 bull 06 981 I A J 006901 cA bull
00 24A 2810 -010960 5t 5 -OOltSO 7 56
8 -000011 s11 c -5811701At oooaoatAJ oot520IAI 888
-- 37 --- middot - -- -1 ) l lbull t5JJ r-l bullhmiddot1 --- - --
l5A
26A
27A
c
2RA
c
2 9A
31A
llA
24
13
597
Ci)55
14
l 6h
RSQ ELASTI- (lASTCOE f-
TABLE 2
EQ I NO NR CODE IU
REGRESSION RESULTS CCONTa
NR INTER- MARKET
(10S CEPT SHARE AOVER- SHARfcent
TIS lNG AOVER
) CITY StiR
c It) bull 5 co a I 1) bull 0 bull 9 c l 0 bull( 2 1
2815 16 003019 -009990 275 -o069bmiddotJ
6 001446 000224 282 c 002178 -054630 o02221CCI -0075 00 646
2901 29 00021t2 CAI 0)0520 592 0007 6 1 068
0 00105C8 t 000012141 60)g( OOOlJ51AJ -00126018) ooou271At -000u80CA) 974
36oa 1 3 003296(4) -oca29o -oo-q94
001755 000203 lt91t 002499((1 -04l820CCI OOl374CAI 0068)0 946
3612 0033281CI 011520 195 01684) oso
00 30 42C OI 00012 2 424 oo579ltAt -2 7l7801At 002 100IAI -003220 819
8 3617 btel 011789 081002259 009530
001448 oOOl OitCCI 732 OOl7 22CAI -0 46160CAI 0004961AI -OOll40181 989c
30A 3808 14 0021t 3 6fAI -000110 628 o023291C) -005 6 middot- _ middot- 0bull 02 04 1 IC I ooooo9 637 ( 001584((1 -0161)1)0 00021 91AI -0 00060
390) 14 002790CCt 006590 h8 t 0 0938 4 070
B OOl89bCCJ 000144 74
0021961(1 -OItlOlO OOl323CAI -004800 915
B
middot--- - middot 4 - middot- - dfll3904 12 004563181 0)8300 879 0428651(1 089 OQIll004 277CA I ooozstcea 919
c 0041)3181 -0-11250 0011881(1 -006270 972 0Ill
- J - bull bull bull bullbullbull middot - - middot - middot gt middotbull L)ofbullflbull middot middotfJltl -tn- middotmiddotmiddotmiddoto +f middot t middot bull
c
is
positive
2
Page 0 determineThe consumer goodproducer good split is sometimes difficult to
Sollle support for the split used in this paper is a comparison of goodness - of - fit
distributions of a
industries
set
advertising are
that industry
a sales increase of
the
highest aD10U3 those
measures for the two groups of industries Table 3 contains
measures for the 32- consumer goods industries and the 103 producer goods
2 Median R is 5 7 for the consumer goods set and 40 for the producer goods
Estimates - of y the elasticity of demand with respect to
given in column of Table 2 The hypothesis that y bull 0 with
alternative was applied for each ittdustry As shown in the
could not be rtjected for 24 of the 32 industries Of the
and 10 are positive For the remalnifg eight three are
and five are significantly positive
Of the threertegative elasetcities the largest (in absolute value)
cosmetics (code 2809) With all elasticity of -1 01 a one
industry advertising would gwerate a decrease in industry sales
more than one percetu
At the other extrema is toys and games (code 39 04) For
a one percent ittcrease in industry advertising would lead to
four-tenths of a pampreentmiddot Though not significantly different from zero
elasticity estimatbull for pt-bpuetary drugs (code 2808)
which are
Both drug industries have
to sales rafios (211 pe1 cent for 28 08 and 13 4 per cent
other hand flaIJo ing ext1ac s (mainly soft-drink SYUlS) also
is for
percent increase in
of slightly
a middottwo - tailed
table that hypothesis
24 14 are negative
significantly negative
9
the codetics and p-ropriacuy very high advertising
for 28 09) bull On the
has a high ind UY
advenisiltg t() sales ratio ( 8 0 per cent) bull but its estimated elasticity is very
small and not signifieantly different uom zero
middot It doe$ not seem unusual to find many elasticities near zero with a few
negative and a few positive If aggregate consumption is insensitive to aggregate
advertising as seems likely ld then what one industry gains must be lost by
others
Page 21
elasticity of demand does
(10) we can
if t 0 then
Finally a determination concerning the industry
not fix the individual firms elasticity Going back to equation
write y bull (1 - ) a + z y Since a gt 0 it follows thatii i i
Y gt 0 If Ylt 0 however y s sign is indeterminateij ij
Turning next to the coefficient of market share 20 of them are negative and
12 are positive with three of each sign being significantly different from zero
The largest positive value which is significant is for pet food (code 2010) and
a close second is mens amp boys suits and coats (code 23 01) The largest signifi
cant negative value is for cosmetics (code 28 09)
In the model some relations among y a 0 and 61 are implied If middota is
negative (1- y) is greater than one Since 61Y bull (1- a) (1 - ) 61Y is
pCsitive but it will be less than one if 0 is large enough Seventeen of the A
ys are negative in each of the seventeen industries 61 is also negative and les
1 (algebraicly) than y so 61Y is greater than one his is not evidence that the
degree of cooperation is low or zero in any industry of course A high o togethegt
with a negative y which is small (in absolute terms) relative to a can give a
value of 61Y greater than one On the other hand if 61Y were less than one
o would have to be greater than zero
For y gt 0 two conditions may hold If y lt a it follows that 61 lt 0 since
(y - a ) lt 0 It also follows that 61Y lt 0 regardless of the value of o bull
In addition the relation is an if-and-only-if one ie if 61Y is negative
then y lt a Three industries have negative values for 61y dairy products
except milk (code 20 04) synthetic fibers (code 2806) and photographic equipment
(code 3808) In all three cases y is quite small so a does not have to be very
large to give a negative value to al
If y gt a the model shows that 61-- gt 0 whatever the value of 0 bull There are
12 cases where both y and 61 are positive with 61 ranging from near zero
(grain milling code 20 12) to 1 20 ( proprietary drugs code 2808) Several
coeffiei-en_rs
--
15
o f to zero
Tab le 3 - Consumer Goods
Indus tr ies
2Distributions of R and Producer Goods
Page 20a
Consumer Producer
Goods Goods
0 - 9 9 0 3
10 - 19 9 0 1 1
20 - 29 9 3 20
30 - 39 9 3 17
40 - 49 9 5 16
50 - 59 9 8
69 9 6 Ii 1060
--f 670 79 9 2
80 - 89 9 4 4
90 - 100 1 1
Totals 32 10 3
Notes
2is used R i s cal-
culated as F F + ((N - K- 1 ) K ] 1 b_ F is the F
statistic Lor the hypothesis that all o f the non - intercept terms are simul taneously equal
1
2
--
-- Since generalized least squ-area regress ion
Page 12
o f these cases have values o f 6 1 Y rea ter than 0 8 (pet food (20 10) bread amp cakes
(20 14 ) miscellaneous processed foods (20 29 ) men s and boys suit s and coats
(23 01 ) p rop riet ary drugs (28 08 ) elect ric housewares (36 1 2) radio and
IV sets 17 and games and toys (39 For be than(36 ) 04)) 6 1 Y to greater 0 8
both o and cry mus t b e less than 0 2 That is in these industries the results
shown in Tab le 2 would hold only with very strong dominance of marke t demand
effects over brand swit ching e ffects and a very low degree o f cooperation
Using the concentration rat ios given in Table 4 the average concentration
ratio for the eight indust ries is 38 1 For the remaining 24 industries it is
43 2 This result would lend some support t o the proposition that concentrat ion
and cooperation are related if the group of eight industries are identi fied
as having low degrees of cooperation
One f inal note on model predict ions S ince both cr and o are assumed to
be non-negative 60 should also be non-negat ive In only one o f the 32 cas es
is 60 negative and it is not signif icant bull
cor pJ
7 1
7 0 4
1 4 7 5 1
)
7 3 z a o a t s 1 1 3 3 5 ( i V Z 8 tJ1 2 1 - 1 1 0 7 3 5 c middot v 2 8 1 0 3 2 -0 0 1 4 3 0 C 1 V 9 8 3 2 8 1 5 1 6 - 0 0 6 t1 6 9 C _ V 1 11 0 2 9 0 1 29 0 00 7 6 1 C l k V 7 3 1 3 6 08 1 3 - o n 4 9 9 4 3 3 3 6 1 2 2 4 0 1 6 4 3
3 1 7 1 3 0 1 1 7 d 7 8 3 3 8 0 8 0 0 2 3 i 2 3 0 6
P 4
3 9 0 3 1 4
1 2
0 09 3 8 4 39 1)1 0 4 l 8 6 5
T A LE 4 e I NCU S T R Y C A T E G O R Y
F TC NOM e tOOE ass
C 1 I c 2 )
2 0 0 1 1 6 2 0 0r 1 7 zo o l Z 2 0 0 7 1 8
z o o 2 6
z o t a 1 6
2 1) 1 2 1 5 2 0 1 4 1 1 2 0 1 a 1 9 2 0 26 1 2 2 0 2 7 1 0 2 0 29 - 3 5
2 3 0 1 10
2 3 02 z o
2 3 03 1 9
2 5 5 1 2 4 Z 1 0 Z 1 1 2 0 3 1 9 2 7 4 1 0 z a o 1 3
2 A 0 7 z s
E LA S T I -C I T Y
c 3 )
- 0 09 7 2 7
0 00 449
- 0 0 2 8 9 3
- 0 3 2 7 70 - 0 OZ il I q
o 4 u t s a 0 0 1 3 7 3
o 1 4 o o-0 0 5 2 4o -0 1 94 0 1
-C l 6 l l 4 0 1 JJ 3 7 0 3 t 0 4 2 0 0 4 0 9
- 0 4 1 8 44 - O l 7 t ) 4 -ll 7 1 4 0 b -0 3 5 7 8 Z
- 0 1 9 7 7 0 1)0 3 3 4
-o n 9 2
IJ A T A
CONV
ON-C ONV
( 4 )
t 1NV
C 1t-V C NV c ) v C - JNV C rlNV c n v c a v C ON V ( i) l V
C CtV
C t1tV C IJNV
C 1 V
c 1 v
Pa ge 2La
HEQ F t Noe x
c 5 )
6 2q
6 89 1 48
6 1 4 ) q 3
1 9 7 1 9
q z t s 5 a e z
l l 9 1
4 o 7 1 2 0
c oqt 1 Z
4 4 )-q 1 z
5 5 9 1 0 64 1 3 3
n 3
Inter industry analysi
Page 23
Equation ( 3 6 ) may be used to fo rmulate a rough test o f the cooperat ion
vs e f ficiency controversy I f we rest ricted ourselves to some subset o f
indus tries where the relat ion be tween y and a i s the same we would be lookingbull
at a sample o f indus tries for which the dependence o f V on H and o is a
quadratic wi th no squared terms
I f we now sub s titute (26) into ( 36 ) we get
(38) v bull a + (y - a) R + (y - a) o (H) - (y - a) H o (H) + u
If it is only d if ferences in marke t shares that affect advertising inten-
s i ty levels i f concentration does not af fect cooperation and i f o bull 0
( 38 ) reduces t o
( 39 ) V bull a + (y - a) H + U
Let this b e the null hyp o thes is and note that V is a linear func-
tion o f H If y gt a and the null hyp othesis holds the expected relat ion
between V and H is as shown by the pos itively sloped straight line in Figure
6 b etween points A and C
In this context the alterna ive hypothesis the degree o f
cooperation ( o) is a positive function o f concentrat ion
is that
(H) I f the func-
tion is as described in (26) and Figure 4 the relation between V and H
would be as shown by the curved line connec t ing po ints A and C in Figure 6 One way to test the null hypothe s is against this al ternative is to test fo r
the hypothesized curvature in the function relating V and H
Fi gure
R
6
v
ID c
H 1 0
I
i I I
sfigned
clasJ s
in def ini g
effec t is the
I
Page middot2lf
One charac teristic of produc ts which may be useful
classes wi th respec t to significanc e of the brand-switching
conveniencenon-convenienc e distinc tion developed by Porter ( 1 97 )
Using his identif ication of IRS indus tries as a s tar t ing point
each of the 32 industries used in this s tudy to one of the two
The ass ignments are not ed in Tab le 4
Using the conveniencenon-convenienc e dis t inc tion the
divided into two sub-samples
sample was
Four regressions were then run bull ene for e ch
sub-sample one for the to tal sample and one for the total sample with
a dummy variable for convenience goods Each regression was run wi th LB s
and wi th indus tries as ob servations The results are given in Tab le 5
For the LB level equations ( 1A - 1D) equation(37) was used ome
rearranging of terms gives v bull y - ( 1-o ) (y - a) ( 1-z ) + ui i i
bull y - 6 1 ( 1 - z ) + ui When observations are pooled across indus trif s yi
is an industry variable In these regressions the values o f y es t imated
in the first stage were used Since industries have been grouped so as
to reduce differences in (y - a) that term is t reated as a cons tant
within each group Finally H the Herfindahl index has been subst ituted
for o the degree of cooperation The resultant equation is linear
y ( 1 - z ) and H x ( l - z ) This is the equation for whi ch results arei i
reported in Table 5 In all regress ions all variables were weighted by the
square roo t of sales
in
raquo
- 34
3 1
i l T 4 8LE S R E G E S S I ON R E SUL T S i X T f N S I ON S T O T Y PE Of
AND L E V E L OF C ONC E kT R A T I O
E O O tt S
tP bull T Y P E NR C ONV O B S tO N-CONV
E L A S T I shy
T I C J T Y
O E - t-41( T SliA R E
H( fl F
1 1 - K O x middot K O X E R F
- H S a 1 1 - H S t 1 8 1 1 9 1
R S O
1 1 0 1
q z
t - bull 9 f 2 1
Z t P D l 6 fC ONV - 1 9 1 s s c 2 1
e - l O I e bull u q a
2 3 1 0 1 1 Nl N- t O V 5 7 1 ( 1 2 19 1 6 1 - J l S 2 5 4
l l l l t z q o a 1 - J o v z I
l ( I NO l l A l l ol u l t o o c bull J u l l
d S
c 3 4 t t r a l
bull Z D 1 0 l l 4 l l 2 3 -gt 2 l l 2 0 - 4 lS 5 4 l C 4 1 o e 6 t l b b 5 1 1 0 1 1 1 c - 1 9 5 1 t d S 2 1 tV
OQ10 T-R A T I O S A E G I VE N l P A RE N T H E S E S l E V E L S f S I G 1 F l C ANC E A - 99 5 B - q 9 t - s NS J UC E NO I Nl E R C ( P T I S U (j f middot) I N f Hf t OUA TIOI S 6 E f 3R E A O J U S T H E N T F OR 1-t E T E R O SIC E O A S T f C I T V pIraquo
1
middot
I
R SO I S C A L C UL A T E D A S f I f bull 1 1 bull 1lt 1 K t t- E E N I S THf U A1 B E R O f fJe S E fi V A T I O S F t S HtE F - S T A T I S
I N DE P E ND E N T y II
T I C F O T H E U Y P O HIE S I S f A l l C JE F F I C H N T A R E QU A L HI l f RO A f40 K I S T H f UUM 8 E R OF Q ( A fL f S
I l l
1 A
1 6
t C
f 1 r R I l l
1 2 1
L Jt
L B
l q
L 1
on s T Y P E bull 2 )
P OUC T
X
H S I l l
l O l A I bull l 0 7 1
O O b O J
1 2 00 1
1 b 2 1 C I l l Z U
l l S b 1 -1 1 0 1
rlt F X l l - fi C f l
bull 1 I
l l
- ()4 2 3 l f H
1 - 2 7 9 1
K O X
1 1 - ti E R F I
l d l
S4 l i A I 2 l b
1 3 00
K O X -t f R F t S J
I 1 - HER F t
)
60 9
z o bull
0
t l l
3 8 8
1 7 6
5 64
64
N f bull
tl S bull
I l l
I 4 1
t ONV
NON-t ONV
A l l
A ll
c o v NO- ( ONV
1 4 1
1 5 1
- l S b
c - 1 1 2 1
5 8 2 C A I 1 6 7 0 1
middot I l 9 H
l 1 il
1 4 1 6 1
E L A C) T J shy
l t C I T Y
I S I
l b l
- 0 2 0
1 - 1 4 9 1
O Z 3 2 1 A I C b 6 b l
0 0 9 2 -J
1 1 b 4 1
o z t t t l 1 2 2 1 1
oru - Ht tt F
1 gt X bull b I
Page 25 bull
Mean val -as ior relevan t variables for the two sub-samples are
Conv Non-Conv
Adv Sales 4 7 2 3
iit Share 2 7 3 2
Ela cicity - 05 3 02 2
Herf indahl 7 3 1 860
That the dist nct ion between coft9enienee goods and non - convenience
goods mat ters is supported not only by these means but also by the regression
results Equat ions lA and lB are different in almost every respec t Only the
concentrat ionmarket share int eract ion term seems to mat ter for convenience
good s Jus t he opposite holds for non - convenience goods both elasticity1
and market sl e are highly significant but the concentrat ionmarket share L
int erac t ion t=rm is not
Given that one minus market share is used these equ3 tions the coeffici t
of market share is the negative of what is sh in col (6) of Table 5 For the
regress ion for non - convenience LB s reported in equation lB then LB s with larger
market shares have lower advertising to sales ratios
On the question of the impact of the degree of cooperation the story is mixed
For convenienc good s the coefficient is pos itive as equation (31) predicts when
there is a po sitive degree of cooperat ion Fo r the non-convenience goods LB s middot
concentration has no effec t
Aggregat ion t o the indus try level and then rer unning the equations $ives the
results shown in (2a-2d) of Tab le 5 As expected R2 goes up and levels of significanc e
fmiddotJr individual coefficients gJ down In add ition the coeffic ient of H ( l -H) inmiddot
quat ion 2A is no t significant though it is po s i tive Since 2A is the indus try level
I
--
all but one of them and that one is not significant For 18 of the 31
wo uld have on its adve rtising to sal es ratio
cognate of LA the variable H ( l-H) pl ays the same role in 2A as H ( l -z 1) p lays
in LA
Vconomies of scale in advert ising
The literature on advert ing contains a number of definitionbull o f economies
of scale and of findings eone rning its presence and magnitude Before I loo k
a t the data with the hope o f mmenting about scale effect s then I wan t to
define the term as I use it
By economies of scale I mean a decrease in the number o f units of adver-
tising which are needed to generate a unit of sales as advert ising is increased
Given the definition of the firm s own elast icity of demand with respect to
advertising (see the text foll wing equation (1) ) the technical definition
is that rii gt 1 Diseconomies of scale are def ined symmetrically i e
Yu lt 1
ith a cro s s - section oi observations on advertising intensity
(v bull a s1) and onmiddot advertising (a ) we may regress the former on the latter 1 i i
The results of such regress ions for the 32 industries are given in Table 2
lines lB ZB bull bull bull 32B For the 32 eases the regression coefficient is positive it
positive coefficients the coefficient i s signif icant
The que tion now is whether these results ar e evidence of pervasive dis-
economies of scale in advertislng I think no t Given my as sumption that the
data I observe are equilibrium results I may ri ghtly conclud e that almost every-
where high levels of advert ising are as sociated in equilibrium with high ratios
of advertising to sales I may no t conclud e anything however about the
impact that a move by some firm away from its equilibrium level o f advertising
middot- Page 27
Given the assumed relation in equation ( 37 ) there is a corre spondshy
ing equilibrium equation relating v and ai The equation is non - linear i
and I have not found a simp le way to characterize it It is po ssib le to
det ermi ne its deriviative with respect to a b y takin$ tne de riviative of ( 3 7 ) i
When that is done the result is
v - 1(40) i bull sl a a z ( 1-z )i i ira-l
Sinc e all the other terms in (40) are non-negative the sign dep ends only
on the sign of 6 1 which is in turn de pend ent only on the sign of (y - a)
That the signs of the coefficients of a in equations lB 2B bull bull bull 32B arei
not always equal to the signs of the coefficients of z in LA 2A bull bull bull 32A i
is still a mystery to me and something to be explored
bull
Summary
My purpose in writing this pa per wa s t o attempt to integrate explicit micro-
economic mode lling reasoned econome tric specification of error terms and high
quality data to exp lore some questions about adver tising by large firms
al quite important improvements could be made in the first two of those areas
and much data ma ssag ing is yet to be done in the third
Concerning the subst an ce of what I have done I think six things are im-
Page 2 8
Severshy
V and Conclusions
portant
1 Explicit modeling is worth the ef for t if for no other reas on than
that it provides a basis for choosing fr om a varie ty of funct ional
forms
2 The same is true for error specif ica t ion I did in fact look at scatshy
ter diagrams for all 32 industry categor ies and they vir tually all show
the het eroscedast icity which I as sumed and for which I correc ted
3 The predicted relation be tween advertis ing in tensity and market share
shows up clearly in on ly 25 of the case s examined I have ye t to
explore why that may b e the cas e
4 Some evidence concerning the presence and impact of coopera tion was
produced but it is not clear ly pervasive
5 The dis tinction between convenience goods and non-c onvenie nce goods is
unambiguously a good one
6 Virtually no evidence concerning econ omies of scale in adver tis ing can
be gleaned fr om this study given its assumpt ions
More work is called for on many of these issues my study seems to have
raised more questions than it has answer ed This is pr obab lly due to the richshy
ne ss of the data source since I had the oppor tunity to address several issue s
at once
- -- - Q --
Foo tnotes
Manager Line o f Business Program Federal Trade Commission The author depended heavily on several staf f members of the LB Pro gram for statistical and clerical support in particular Joe Cholka George Pascoe and Haro lene Jenkins Helpful comments were given b y Richard Caves Dennis Mueller Michael Porter and F M Scherer When the paper was in its early stages o f development both S teve Garber and Jon Rasmussen former economists in the LB Program provided eery useful cri tiques The views expressed here are my own o f course
1 The basic exp licit model of the role of advertising is Dorfman and S teiner ( 1954 ) and the models in this paper may be seen as ano ther extens ion of their work
2 For a review o f models of oligopoly which treat prices and quantities as relevant variables see Shubik ( 1 959) Extens ions to take advertis ing int o account are fairly s traightforward
3 Dorfman and S teiner dealt with a price-setting firm For a monopolis t inverting the demand relation to show Pbull P (Q A) _instead o f Q- (P A) has no effect on the results o f the analysis For oligopolists who face a sys tem of demand relations inversion does have some impac t on the results The dif ferences are not trivial nonetheless they will no t be explored in this study
4 Shubik ( 1 959) p
5 From a formal mathematical point o f view the function is similar to a1 i Legrangian funct ion of the form - r rr
L bull 11 + A ( j - (i i i+j
where the indicates some fixed level of the profits industry Following the same logic which resul ts in and income constraint as the marginal utility of the of the indus tr in the obj ective of firm i
6 Hall amp Weiss ( 1 967) p 323
7 See Schmallensee ( 1 97 2 ) Ch 2 pp 1 6-47
8 On this poin t see Imel amp Helmberger ( 1 9 7 1 )
9 Long ( 1 970) p 130 - 1 39
1 0 See Schmallensee ( 1 972) Ch 3 pp 48-8 7
rrj i+j ) of the rest of the
the identification profits of the rest
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980
an
Page 3
( 7 )
- 0
This implies that
(8)
(9)
-
v bull bull
bull
bull
zl [ Y21 Y22 bull bull Y2N J
-1zl 0
-1z2 r z
0 1
Page 4
Writ ing (8) compactly gives
the matrix in the two equationsA comparison of (6) and (9) shows that
differs only in that all the elasticities are present in (9) but only the
awn-elasticities are present in ( 6 ) Since advertising expense mus t be I non-negat ive the first-order requirements are that v be the maximum of t
i
the amount shown in 6) or (9) and zero
III A SIMPLIFIED DEMAND STRUCTURE
So far the demand equation system is comple tely general I will now
move to a particular demand structure which is charac terized by an industry
elasticity of price with respect to advertising a general brand-swi tching
elas ticity and a dependence of the firm-specific elas ticity on its relative
size I make the expli cit assumpt ion that
n pound
)l
0 AA
Consider the identification of y as an elasticity of industry price
Page 5
(10)
where y is an
- (y- a)
industry demand elasticity a is a brand-swi tching
el astici ty and z1 is the market share In matrix no tation
(11) r bull ai + (y- a) t
wi th respe ct to advertising 0
le t ting S denote dS observe that
0(12) SS bull
-
-1 0s t s - -1 0S t d - 9- -
apl apl aplaa1 aa2 a 0 a
middot( a Pl
( 13)
-
-0 a bull 0
a
) 0 X s
-1 0dAA bull then d
a bull
J 0
(JA) middot and
Let the sc alar (SS)(AA) be denoted by y so ya zft
If
1
d
-
I
I
(11)
--
y
I -f
l J- a)
a) l zt
Substitutin g into (13) gives
i [ ai + (y(14) -I- bull S A
bull az + (y -
Pa ge6
I I II
bull
since l bull 1
Under these assumpt ons then y is the elasticity of industry sales
(and average industry p ice) with respect to industry advertisin g Note
that if y bull 0 so that there is no industry sales effect there will still
be positive advertising if there are brand-switching effects
Turning now to the brand-switching ef fect observe that
(15)
Noting that
-1 deg -1d z -z -
-
0 a
Page 7
(15) may be written as
0 (AA)
lx 2x 1x
ith
1y
i
First assume no
that if there are
will be no changes in market shares
- 0a bull
Page 8
Equation (17) is a general statement of the relation betbulleen changes in
advertising levels and changes in market shares which would acompany them It
may be used to make some general observations about brand-switching effects
industry effect eg y bull r t bull 0 and assume further
equal percentage changes in all advertising le =ls then there
Under these further assumptions
0l and dz -1 bull (iA) ltr assumption the second term on the right hand side of this expression is
- By
zero
necessary for rt bull 0Consequently for market shares to be unaffected it is
when y bull r l bull 0 This requirement is satisfied by equation (11)
Secondly consider effects on market shares if one firm inc ases its abull ar-
tising and others do not As a general assumption we require smiddot ch a change o
lead to a decrease in the shares of all the other firms hat is let
0 0aiai gt 0 and a a bull 0 for j + i For any matrix X with N rows and its trans-j j
pose X let X
row of X
bull middotmiddotmiddot where is a column vector which contains the
elements of the Under the assumptions about the iis given just
d-1 0 0above it follows that r becomes (aiai) iimiddot ehFor jzj lt 0 under these conditions then it follows that the j- element
N of (I- l ) lt 0 so middotij - il lt 0 or yij - yik lt 0 If this
kbulll N condition holds for all Y ij where j + i then it follows that yii - z
k Y gta
k- 1 ik since the th firms share must increase to offset the decreases in all other
shares
ei ith
1v ( ) h i
7urning
where is the
leads to z
to the r matrix given in (11) we note that 1 = cr + (y - cr) i i zi elementary vector with 1 in the position and 0 elsewhere This
bull cr + (y - cr) zi Since bull required is thatcr at- zY wYij (y - cr)zi - cr + (y - cr) zilt 0 or cr gt 0
These observations may be summarized by substituting (11) into (17) bull
giving
(18)
i s
d -1 zz -
-
-
+ (y - cr) tz
d -1 0cr(I - l ) a a bull
cr 1 z - (y - cr) 0 a
o
If ai increases and the other a s are const t cr must be positive to havej
a positive effect on zi A positive cr also bull plies a negative ffect of aJ
f all a s in-on zi other a s (including a ) fixed j
eluding a1 change by the same percentage zi remains unchanged Finally k i
if cr bull 0 there is no brand-switching effect
For a to have no effect on p requires yij bull 0 From (10) thisj i
amounts to y bull cr that is the industry demand elasticity and the brandshyswitching elasticity are equal If y gt cr gt 0 the industry elasticity yij doninates If y lt cr y lt 0 the brand-switching elasticity dominates ij
making
Page 10
Substituting ( 1 1) now in (6) the Cournot case and in (9) the
Chamberlin ease gives
-(19) Cournot vO a + (y - a) z
Chamberli n vl - y lmiddot th
The 1- element is then
-(20) Cournot vO a+ (y - a) zi i
-Chamberlin vl yi
In (19) and (20) the superscript 0 is used to denote total
independence of decision and the superscript l is used to denote
total interdependence of decision making
The relation between v and vf depends solely on t relation between
y and a f y gt a so that the industry effect dominat 3 vf v with
equality in the trirlal case of z bull 1 Cooperative irms would have ai
higher advertising intensity than non-cooperative fi
If there is equality between the industry elasticity and the
y Whe therbrand-switching elasticity or y bull there is cooperation among the firms has no effect on advertising inten-sity And if brand-switching is more important with y lt a e get
vi lt v cooperative firms will have a lower advertising intensity than
non-cooperative firms The three situations are depicted in Figure L
Given the results for the firm level variables calculation of the
corresponding re sults for the industry is s-traightforward Let
B bull t zf bull z z be the Herfindahl index of concen tration and notethat
z - bull l z bull 1 and that V bull A S bull a i 5 bull (siIS) (asi)i
bull middot Then for the general case from (6) and (9)
bull
Page lOa
y lt1 bull y y
10
f
V0 bull z [ a1 + (y - cr) J bull a + (y - a) H vl bull Y bull y
11Page
( 2 1) vo bull z d -1 r bull r z bull z- d d- i )11
l -1 N N v bull z d r z bull r z bull l y z - ij jibulll jbulll
and for the simplified case from (19)
(22
Since the forms of the equations in (22) are the same as in (20) all
of the analysis of (20) holdS for (22) except that His substituted for
middot z Advertising intensity will be higher for an industry with cooperativei
firms than for one with non-coope rative firms if the industry demand elas-
ticity dominates the brand-switchine lasticity If the ewo elasticities
are equal then so will be advertiSl intensity for the cooperative firm
and non-cooperative firm industries If brand-switching dominates the
cooperative firm industry will have lower advertising intensity
In the industry model context the possibility of a critical concen-
tration level may be easily introduced That is assume that at values
of R below R the firms are non-cooperative but that for alues equal to
or greater than H they are cooperative That is
(23) V bull a + (y - a) H if H lt H bull y if H gtmiddot H
This function is shown in Figure 2 for the three relamptions be eween y and a
IV THE D GRE OF COOEtRATION
The models depicted in Figure 2 are based on a very simple notion about
the relation between concentration and the cooperationnon-cooperation charac-
teristic They use what is essentilly a step function relating concentra-
Figure
I -
J H H
Page l la
2
v v v
a
I J I y a bull Y y ---i
I I I I I I II I
H 10 H H 10 H 1 0
bullbull
bull bull
Page 12
tion and the extent or degree of coo peration as shown in Figure 3 Using
o as a symbol for the degree of cooperation eq uation (23) can be rewritten
as
bull(24) v (1 - o) vo + ovl bull (1 o) [a+ - a) H ] + oy- (y
vhere
(25) o bull o(H) bull 0 if B lt B
Hbull 1 if B gt
The specific relation between concentration (H) and the deg ree of
cooperation ( o) shown in Figure 3 is more restrictive than necessary and
bas less appeal an one which shows a smooth increase of o from 0 to 1 as
B increases fro to 1 Such a smooth function is shown in Figure 4 he
degree of cooper ion is low until H gets close to H increases greatly as
B goes through R attaining a high level for an H larger than B but still
close to it and then goes to 1 as B goes to 1 More formally
(26) 0 bull o(H)
bull 0 if H bull 0
bull 1 if H bull 1
do gt (IdB
d2o- gt 0 if H B dBZ
0 if H H
lt 0 if R gt H-
If (26) is now substituted for (25) in (24) a corresponding smoothing
of the functions shown in Figure 2 is accomplished he smooth functions
Figure 3
I
10
H 10 H
Page 12a
Figure
Page 12b
4
10
H 10 H
l
(27)
Cl - o) o
[ (1 -
[ (1
t y ij z
j
are given in Figure 5
Page 13
The equation in (24) was constructed as a hybrid of two results derived
from first-order equations for the non-cooperative and cooperative models
It may also be constructed directly by using l bull (1 - o) + on as thei i
th 2objective function for the i-- firm Treating o as a constant the first
order equation is
0 an aa
i middot
n i
Using equations (3) and (7) and setting --- bull 0 givesaai
(28) - 0
After collecting terms and converting to matrix notation we get
(29) - Cl - o) + 0 z-1vi yii i
N
j bulll
+ 0
+ o v1
d-1 r z z
-(30) v Cl - o) rd
-
o) r +d
d r J _ -1- d z
and middot
d) r + 0 rjd
l r z
v - z --(31) zv
- (1 0- +d) t r z- d-
If we now impose the simplified demand structure by substituting (11)
we get
- (1 - o) a +( 32) + oy and
Figure
l
-y
v
Y
H
r
I I
v
a
y
H 10 H
v
a
bulla
HH 10
bullbull
(33)
i
Page 14
v - (1 - o) [ a l + (y - a) ] The corresponding equation for Vis given in (24)
If the relation between industry advertising intensity (V) and
concentration (H) is non - linear via the impact of concentration
on the degree of cooperation does the relation take the form of a
auadratic as some other investigators have proposed (Greer ( 1971 ) Martin (1979) Strickland amp Weiss (1976) ) That question can
be approached from two perspectives - conceptual and empirical In
the next section I will try to apply some statistical material to it
here I want to explore the question in the context of the models
developed above
Referring to equation (24) taking the deriviative with respect
to H and letting bull (H) gives
bull( 3 3a ) 3V (y - a) (1 - o) + (1 - H) 3H dH
Given that o and H each are bounded by 0 and 1 and that o is an in-
creasing function of H the term in the brackets in non-negative The
sign of the deriviative then is the same as the sign of (y - a) If the industry effect is dominant (y gt a) then advertising in-
tensity will increase with concentration If the two effects are equal
then advertising intensity will not vbullry with concentration and if the
brand-switching e ffect is larger advertising intensity will decrease
with concentration
For given advertising elasticity parameters then it is not pos-
sible to show an increase in A S up to some H and the n a decrease
he relation may be non - linear but it is monotonic over the permis-
sible range of H
Page 15
One possibility is to introduce a relation between (y - a) and H
If industries characterized as dominated by the brand-switching effect
also were more concentrated then a cross industry comparison would
show an inverted U shape relation between AS and H I know of no
support for such an association between ( y - a) and H
V ERROR SPECUCATION
So far I have implicitly assumed that there are no errors in the
model I now assume that the error
thvariable in the equation for the i- firm is additive that its mean is zero
that its variance is inversely proportionate to its sales and that all error
term covarlances are zero hat is
34) v bull (1 - o) [ a + (y - a) z ] + oy + u i i i
- - -
In vector notation
(35) bull (1 - o) [ at + lt - a) J +
E() bull 0 Cov (u) -
0
If we now ca lculate V bull bull we get
oy + U bull
(36) V bull (1 - o) la + (y - a)R J +
U bull E(U) bull 0 Var (U) bull
P age 16
The error term specifi c ation which follows Hall
wi th the ratio of profits
6 heir analysis seems
as wellbull particularly given
and Weiss was
applied by them to equa tions to equity or
asse ts as dependen t variables to carry the same
weight in the present context the substantial
evidence presented by marketing analysis on the tendency of company de-
cision makers t o use the advertising to sales ratio as the decision
7variable
The zero covariance assumption on the other hand1 is particularly
troublesome Within a probably not true since
many even ts outside the
single industry it is
industry would have
industries
similar effec ts on all the
8firms in the industry Between it is probably not true either
since many companies produce in more than one indus try and even ts in the
firm would tend to effect all of its activities The extension of the
model to allow for non-zero covariances is certainly called for
--- -
Page Ll
Athird characteristic of this specification also deserves some comment
only on equation is shown for the firm or industry Several haveinvestigators
included an advertising equation ina larger system of equations (Comanor and
Wilson (1974) (1979) and Weiss (1976) so in myMartin Strickland Idid
dissertation1 Ihave ewo defenses for presenting only single equation results
here One is the suggestion by Comanor Wilson that simultaneousand equation
verybias may not be important in this context he second is that Iam will-
ing to see the model extended to include a profitability equation and I intend
to move that direction in the near futurein
Onthe inclusion of aconcentration equation on the other hand I amskep-
tical he problem isthat in some fundamental sense equations in models of
industrial organization should be oriented to the decision-making contexts of firms
Ifeach firm in anindustry with Nfirms has only one decision variable then
there can be only Nindependent equations which are based on the first-order
optimization equations It not show large ofis difficult of course to a number
functional relations which follow from the first-order conditions point here he
isthat only Nof them can be independent
Consider for illustrative purposes only a situation in wich teh N
itsquantities say q are set exogeneously and each firm determines advertising1
of and level say ai where a is the true optimal value a a bulla
1 1 i i + i
Ifp a) then the Nq
s and the Na s will determine N p s say pii bullp
i(l i 1 i
Now given Eand 9 we can determine both the vector s andthe vector v since
- bull til
ands We can also determine S s z bull s-1 s-ii
v and we
industry concentration isHbull (alternatively
V and Hare functions of the same
pendent Furthermore if there are
VM) and (H1 H2 bull bull bull )contain
VI STATISTI CAL APPLICATIONS
Page 18
Given the observed values of might be tempted to conclude that we
have a two equation system in which v and z are jointly determined That would i i
be inappropriate however since the 2N random variables (v ) are determined by
onlt N basic random variables t The variance-covariance matrix for ( ) must
be singular bull
The extension of this observation to industry level variables is straightshy
forward Since observed industry advertising intensity is V bull l v and observed 4
C4 bull z ) it follows thati l i
set of random variables t and cannot be indeshy-
M industries then the variables (V1 v2 at most M independent random variables
Given the assumptions made above y a and o are constant within a
given industry That being the case (34) can be written as
(37)
s1 bull (1 - o) (y - a)
If and
So + +
(1 - o) a + oy
81 are esti mated for an industry using a suitable statisticalSo procedure giving and 81 an estimate of y is then available The sumSo
Intra-industry analysis
1974 of Business Re ort -
categories are identified
Page 19
of So and S1 is y so y bull So + S1 is the estimate of y
If o is assumed to be bounded by zero and one the sign of 61 is also
the sign of y - a The estimate of So is of cou se an estimate of
(1 - o ) a + oy Moreove if y gt a then a is less than 6o- (1 - B) a+ 0(
Cor esponding results would
hold if y bull a or y lt a
If 61 gt 0 then we would estimate that a lt 6o
bull
For purposes of estimating this model 32 manufacturing industry categories
were selected The data are fram the Line of Business forms filed with
the Federal Trade Commission they are discussed in detail in the Annual Line
which is available from the FTC The 32 industry
in Table 1
Simple regressions for equ ation (37) were run for each of the 32 industries
with data for each of the firms which filed in the industry constituting an ob-
servation The results are given in Table 2 lines lA 2A
1974
the intercept
For all the A
equations generalized least quares was used th both and market
share bein scaled bv the square root of sales
The 32 industries were taken from the larger set of all manufacturing industry
categories in the 1974 data files Two selection cr iteria were used thereLB
had to be at least ten reporting companies and the industry had to be a consumer
goods industry or a producer goods industry which is strongly associated with a
related consumer goods industry so that adve rtising by the firms in the producer
goods industry impacts on demand in the related industry (eg soft drink syrup
and soft drinks) There were 135 industries with at least ten companies - 103
producer goods and 32 consumer goods
273
274
TABLE 1 INDUSTRY CATEGORIES USED
FTC
COOE OESCRI P T I ON _
2001 MEAT PACKINGt SAUSAGE S ANO OTHER PREPARED
MEAT PROOUCTS
2004 DAIRY PRODUCTS EXC F LUID MILK
2005 CANNED SPECIALTIES
2001 FROZEN SPECIALTIES
OEHYORATFOt AND
I NC LU 0 ING PRE SERVE S
SOUP MIXESbull
SEASO lNGSt ANO
OTHER PET FOOJ
RAIN MILL PROOUCTSt
ANO PREPARED
RELATED PRODUCTS
PRODUCTS
Z008 CANNEOt ORIEOt PICKLED FRUITS
ANO VEGET ABLES JAMS JELLIES DEHYDRATED I G TA3t E
S40CES ANO SALAD ORESSINGS
2010 DOG CAT A O
2012 FlOUR t OTHER RICE
MILLING fiLENOE) FLOUR
2014 B EADt CAKE ANO
20113 CO FECTIONERY
2026 BOTTLED A O CANNED SOFT DRINKS 2021 FLAVORING EXTRACTS AND SYRUPS EC
2029 MISC FOOOS AND KINDRED PROJUCTS EXC
ROASTED COFFEE
2301 MENS AND BOYS SUITS AND COATS
2302 MENS ANO BOYS FURNISHINGS
230 3 kOMENS AND MISSES OUTER EAR
2551 HOUSEHOLD FURNITURE
2702 PERIODICALS
Pqe 19a 1
RELATED 1972
src cooe
2011 f 3
202tX2026
2032
203pound
2033t4t5
2047 2041t4t5
2051 2065 2086
2081
209XZ05
231
232
233
251 272
2703 BOOKS
2704 ISC PVeLISHING
2806 ORGANIC FlEERS
2807 DRUG$ ETHICAL
23234
PT 2 S 3
2808 ORUCSt PROPRIETARY
2809 PERFU ES t COSMETICS ANO OTHER TaiLET
PREPARATIONS
2810 SOAP AND OTHER CLEANING PREPARATIONS 2815
2901
PEST1CIDES AND AGR ICULTURAL
PE OLEUM REF I N ING CHEMICALS EC
3606 HOUSEHOLD COOKING EQUIPMENT
3612 HOUSEHOLD APPLIANC ES NEC INCLUJING ELECshy
rqlC HOUSE ARFS AND FANS ANO SE I G MACHINES
3617 RADIO AND T V RECE I VING SETS 3808 PHOTOG APHIC E UIP amp SUP LfESt EXC
3qo3
PHO TOCOPYING
SPOR TING A D E QU I P M ENT f SUP P L I E S ATHLETTC GOODS NEC
3904 DOLLS GAMESt TOYS AND CHLOqENtS VE ICLES
PT283
2844
234tX2844
287 9
291 3631
3634t6t9
36H
PT386l
3949
15
943
middot
l A6l f 2 R(GRESSION RESULTS
EQ I NO
NR CODE cu 121
lA 2001 A (
2A 2004 r (
3A 2005 a
-R
ClS amp31
16
17
1 z
I TER-CEPT bull t
ooo9t21At 00022311 oooeos uu
OOl49tCAI 0010 2 middot 000960CCI
o049tl1At o04t55C61
MARKET SH ARE bull 5 I
-010670Ctl
-017600181
-001050
-0211AO
-0077 0
AOVER-TIS INti
lbl
000()39 000072
000018 0 0024liAI
-000079
SHARf
4ampgtVf Q bull
111
000140
-000170
RSQ
COl
66 485 862
607 6 3 t ()2
blti6 633
flASTI-
CI T Y PH
-007271(1
000441
-u0288l
clASTSHU (t)Ef
bull 10
tto
-234
210
( OOlltOCCI -126520 000922 181 001790
4A 200 7 18 00590liAI -038670 4e16 -032110 lell u c
OOOS74e 000700 025940
00067018) 0021001AI -02l100CAI
5 26
920
5A 6 c
200 8 26 0 03Sl9CA I 0012 26181 o 024511(
-005540
-0 81420 0001961AI O()Ol90CAI ooll240
441 604 891
-0020 19 274
6A 2010 16 Oe04224CAI o359lOIAI 9oq ll4015dCAI 061 6 -middot 005435CAI 00011 8 1AI 917 c 00535ltAI -009850((1 OOOA51J CAI -001610161 971
7A 2012 001862 000010 351 00187) oo1 a oooq45 ono1o6 439 c 001109 -027050 Oe0067SIAI -OOll60C81 897
- middotmiddot- -
8A 201t ll 000757 Ol4140CCI 814 ott9oocet 095 8 oooqo6 C C J 000069(81 811 c ootl2lampBI -080910CCI Oe009lUBI -OOtlOOCCI 960I
I DIIJIT pound11 1V OTIUTVTfAll-1pound1t- A _ 00 R - qq r - Q
1jl)OQCD
0
-middot middot--
l
_ __ _
I
i
0 00141 c c
91)7
I
T ABLE 2 REG ESSIOH RESULTS CCONTI
EO NR
IND CODE
iR bull
cos
I TEq-
CEPT
t1ARKE T
SHA E
AOVER-
Tl SING
SHARpound-cent
AuVER
RSO ILAST1-
C I T Y ELAST
SH CPtf (lt Cl) bull 11 I 41 I 5I (61 171 I 8 I f9) c 10)
9A 2 0 1 8 19 002397181 -007640 367 -005 246 le46 6 000605 000550CAI 597
-000029 G14180 OOl654CAJ -024li01At
lOA 2026 12 0028901AI -OlZl90 -0114 0 1 1 1 5 759
Ll (
llA 2021
f3 c
12A lOZ9
B
(
13A 2301
B
c
14A 2302
B
c
154 201
n (
16A 2551
8
10
35
10
20
19
24
001897(()
00141 6CCI
Oll9lHCI
010604 middot middot
Oll083
OOJ11JCAt
002059141
001411CIH
-000412
000293(81
000187
001159 (3)
ooo6a4cct
o00792CCI
OOl40dCAt
000006
-000076
OOlltl6fA)
o o0 5 64CCt
-034570
-o1middot o3o
-1659 90
Ol51lO
-046190CAt
Ol5450CA)
-005l60
002940
-030060(()
-0 43l50CCt
0 33650
-019080
00004
bullJ00741CAI
-000069
000466
0001 51CAt
000947CAt
O OOI)70CAt OOl750CA J
000175CCI
000671CAI
OOll5) f A t 002207CBI
0004b7CA I
- 007 1 70
001760
-001410CAt
- 0 16200CAt
-004210
- 1 207 70
705
96 0
542
466 805
460
614
921
832
9 61
996
424
550
9 15
56 1 651
884
-006114
019 037
03504lCA)
oo409Q
-141644((
491 -0 17664
607
871
lell
084
101
072
103
108 11IllOQfl)-- l
-037880fCJ OOl6l4CAJ -040l70fCI 0()
c
I
I
bull 9
9t
middot middot middotbullmiddot bull lraquo
TABLE 2 REGRE SION RESUlTS ICONTt
bull
EO I NO tHl I filER- f4A_K T AOVER- sbulluRE ttSQ flA Jl- iLAS T NR CODE cos CEPT SHARE JISING AOVfR bull (JTY SHR CiJCF
110Cl I f amp Itt 15J tbt bull 1c u caa
llA 2702 ll o079 6 81 Bt -019110 S6s -OlltO 111
a 004679 ooooo1 t69
( oOl26o -015160 001560 -022690 ltJr()I
ldA 270 3 19 0065lt -OIt2l60 220 -Ol57a2 11 8 e o 0 71t6IA t ol08 75tAt ( OOl983CiU -060llOC() OOl7191A t -O lltdOC A) middot 89
lqA 2704 10 005082 -Oil060 22b -l)i1977 lelO 6 -000080 oo2772C4 t 910 c 000907 -011 200 Oe04105CA t -046960 995
20A 2d06 1 II
O OlltStiAt -001120 837 00014 -))th O01380amp At -oooo 11 821 ( OOl526CAt -026700181 0002 5UAt 000530 969
2lA 2807 28 OOt9lt9CAt -00891 0 516 -0 03962 225I8 002095CCt OOOl lliAt 690 ( o02965CA t -056650((1 OOOt50CAJ -003l70ttl 911I
I 22A 2808 15 ot3l06 119730 655 1130) 090
0 - 007526CC t Oe003961At 835- middot-
c I oo7523 -067210 001l 2tA t -0091t80CCt 951
2lA 2809 21 Oe2067lCAt -1218101At 71 2 -10095((1 121 ttlIaOQ8 00797418) OOOI8lltt 686 Ill( 0092lU A t -Olltl70Ct t 000605CAt -OOit8701At 951 _
---- -
32 009506CAt
0 bull 06 981 I A J 006901 cA bull
00 24A 2810 -010960 5t 5 -OOltSO 7 56
8 -000011 s11 c -5811701At oooaoatAJ oot520IAI 888
-- 37 --- middot - -- -1 ) l lbull t5JJ r-l bullhmiddot1 --- - --
l5A
26A
27A
c
2RA
c
2 9A
31A
llA
24
13
597
Ci)55
14
l 6h
RSQ ELASTI- (lASTCOE f-
TABLE 2
EQ I NO NR CODE IU
REGRESSION RESULTS CCONTa
NR INTER- MARKET
(10S CEPT SHARE AOVER- SHARfcent
TIS lNG AOVER
) CITY StiR
c It) bull 5 co a I 1) bull 0 bull 9 c l 0 bull( 2 1
2815 16 003019 -009990 275 -o069bmiddotJ
6 001446 000224 282 c 002178 -054630 o02221CCI -0075 00 646
2901 29 00021t2 CAI 0)0520 592 0007 6 1 068
0 00105C8 t 000012141 60)g( OOOlJ51AJ -00126018) ooou271At -000u80CA) 974
36oa 1 3 003296(4) -oca29o -oo-q94
001755 000203 lt91t 002499((1 -04l820CCI OOl374CAI 0068)0 946
3612 0033281CI 011520 195 01684) oso
00 30 42C OI 00012 2 424 oo579ltAt -2 7l7801At 002 100IAI -003220 819
8 3617 btel 011789 081002259 009530
001448 oOOl OitCCI 732 OOl7 22CAI -0 46160CAI 0004961AI -OOll40181 989c
30A 3808 14 0021t 3 6fAI -000110 628 o023291C) -005 6 middot- _ middot- 0bull 02 04 1 IC I ooooo9 637 ( 001584((1 -0161)1)0 00021 91AI -0 00060
390) 14 002790CCt 006590 h8 t 0 0938 4 070
B OOl89bCCJ 000144 74
0021961(1 -OItlOlO OOl323CAI -004800 915
B
middot--- - middot 4 - middot- - dfll3904 12 004563181 0)8300 879 0428651(1 089 OQIll004 277CA I ooozstcea 919
c 0041)3181 -0-11250 0011881(1 -006270 972 0Ill
- J - bull bull bull bullbullbull middot - - middot - middot gt middotbull L)ofbullflbull middot middotfJltl -tn- middotmiddotmiddotmiddoto +f middot t middot bull
c
is
positive
2
Page 0 determineThe consumer goodproducer good split is sometimes difficult to
Sollle support for the split used in this paper is a comparison of goodness - of - fit
distributions of a
industries
set
advertising are
that industry
a sales increase of
the
highest aD10U3 those
measures for the two groups of industries Table 3 contains
measures for the 32- consumer goods industries and the 103 producer goods
2 Median R is 5 7 for the consumer goods set and 40 for the producer goods
Estimates - of y the elasticity of demand with respect to
given in column of Table 2 The hypothesis that y bull 0 with
alternative was applied for each ittdustry As shown in the
could not be rtjected for 24 of the 32 industries Of the
and 10 are positive For the remalnifg eight three are
and five are significantly positive
Of the threertegative elasetcities the largest (in absolute value)
cosmetics (code 2809) With all elasticity of -1 01 a one
industry advertising would gwerate a decrease in industry sales
more than one percetu
At the other extrema is toys and games (code 39 04) For
a one percent ittcrease in industry advertising would lead to
four-tenths of a pampreentmiddot Though not significantly different from zero
elasticity estimatbull for pt-bpuetary drugs (code 2808)
which are
Both drug industries have
to sales rafios (211 pe1 cent for 28 08 and 13 4 per cent
other hand flaIJo ing ext1ac s (mainly soft-drink SYUlS) also
is for
percent increase in
of slightly
a middottwo - tailed
table that hypothesis
24 14 are negative
significantly negative
9
the codetics and p-ropriacuy very high advertising
for 28 09) bull On the
has a high ind UY
advenisiltg t() sales ratio ( 8 0 per cent) bull but its estimated elasticity is very
small and not signifieantly different uom zero
middot It doe$ not seem unusual to find many elasticities near zero with a few
negative and a few positive If aggregate consumption is insensitive to aggregate
advertising as seems likely ld then what one industry gains must be lost by
others
Page 21
elasticity of demand does
(10) we can
if t 0 then
Finally a determination concerning the industry
not fix the individual firms elasticity Going back to equation
write y bull (1 - ) a + z y Since a gt 0 it follows thatii i i
Y gt 0 If Ylt 0 however y s sign is indeterminateij ij
Turning next to the coefficient of market share 20 of them are negative and
12 are positive with three of each sign being significantly different from zero
The largest positive value which is significant is for pet food (code 2010) and
a close second is mens amp boys suits and coats (code 23 01) The largest signifi
cant negative value is for cosmetics (code 28 09)
In the model some relations among y a 0 and 61 are implied If middota is
negative (1- y) is greater than one Since 61Y bull (1- a) (1 - ) 61Y is
pCsitive but it will be less than one if 0 is large enough Seventeen of the A
ys are negative in each of the seventeen industries 61 is also negative and les
1 (algebraicly) than y so 61Y is greater than one his is not evidence that the
degree of cooperation is low or zero in any industry of course A high o togethegt
with a negative y which is small (in absolute terms) relative to a can give a
value of 61Y greater than one On the other hand if 61Y were less than one
o would have to be greater than zero
For y gt 0 two conditions may hold If y lt a it follows that 61 lt 0 since
(y - a ) lt 0 It also follows that 61Y lt 0 regardless of the value of o bull
In addition the relation is an if-and-only-if one ie if 61Y is negative
then y lt a Three industries have negative values for 61y dairy products
except milk (code 20 04) synthetic fibers (code 2806) and photographic equipment
(code 3808) In all three cases y is quite small so a does not have to be very
large to give a negative value to al
If y gt a the model shows that 61-- gt 0 whatever the value of 0 bull There are
12 cases where both y and 61 are positive with 61 ranging from near zero
(grain milling code 20 12) to 1 20 ( proprietary drugs code 2808) Several
coeffiei-en_rs
--
15
o f to zero
Tab le 3 - Consumer Goods
Indus tr ies
2Distributions of R and Producer Goods
Page 20a
Consumer Producer
Goods Goods
0 - 9 9 0 3
10 - 19 9 0 1 1
20 - 29 9 3 20
30 - 39 9 3 17
40 - 49 9 5 16
50 - 59 9 8
69 9 6 Ii 1060
--f 670 79 9 2
80 - 89 9 4 4
90 - 100 1 1
Totals 32 10 3
Notes
2is used R i s cal-
culated as F F + ((N - K- 1 ) K ] 1 b_ F is the F
statistic Lor the hypothesis that all o f the non - intercept terms are simul taneously equal
1
2
--
-- Since generalized least squ-area regress ion
Page 12
o f these cases have values o f 6 1 Y rea ter than 0 8 (pet food (20 10) bread amp cakes
(20 14 ) miscellaneous processed foods (20 29 ) men s and boys suit s and coats
(23 01 ) p rop riet ary drugs (28 08 ) elect ric housewares (36 1 2) radio and
IV sets 17 and games and toys (39 For be than(36 ) 04)) 6 1 Y to greater 0 8
both o and cry mus t b e less than 0 2 That is in these industries the results
shown in Tab le 2 would hold only with very strong dominance of marke t demand
effects over brand swit ching e ffects and a very low degree o f cooperation
Using the concentration rat ios given in Table 4 the average concentration
ratio for the eight indust ries is 38 1 For the remaining 24 industries it is
43 2 This result would lend some support t o the proposition that concentrat ion
and cooperation are related if the group of eight industries are identi fied
as having low degrees of cooperation
One f inal note on model predict ions S ince both cr and o are assumed to
be non-negative 60 should also be non-negat ive In only one o f the 32 cas es
is 60 negative and it is not signif icant bull
cor pJ
7 1
7 0 4
1 4 7 5 1
)
7 3 z a o a t s 1 1 3 3 5 ( i V Z 8 tJ1 2 1 - 1 1 0 7 3 5 c middot v 2 8 1 0 3 2 -0 0 1 4 3 0 C 1 V 9 8 3 2 8 1 5 1 6 - 0 0 6 t1 6 9 C _ V 1 11 0 2 9 0 1 29 0 00 7 6 1 C l k V 7 3 1 3 6 08 1 3 - o n 4 9 9 4 3 3 3 6 1 2 2 4 0 1 6 4 3
3 1 7 1 3 0 1 1 7 d 7 8 3 3 8 0 8 0 0 2 3 i 2 3 0 6
P 4
3 9 0 3 1 4
1 2
0 09 3 8 4 39 1)1 0 4 l 8 6 5
T A LE 4 e I NCU S T R Y C A T E G O R Y
F TC NOM e tOOE ass
C 1 I c 2 )
2 0 0 1 1 6 2 0 0r 1 7 zo o l Z 2 0 0 7 1 8
z o o 2 6
z o t a 1 6
2 1) 1 2 1 5 2 0 1 4 1 1 2 0 1 a 1 9 2 0 26 1 2 2 0 2 7 1 0 2 0 29 - 3 5
2 3 0 1 10
2 3 02 z o
2 3 03 1 9
2 5 5 1 2 4 Z 1 0 Z 1 1 2 0 3 1 9 2 7 4 1 0 z a o 1 3
2 A 0 7 z s
E LA S T I -C I T Y
c 3 )
- 0 09 7 2 7
0 00 449
- 0 0 2 8 9 3
- 0 3 2 7 70 - 0 OZ il I q
o 4 u t s a 0 0 1 3 7 3
o 1 4 o o-0 0 5 2 4o -0 1 94 0 1
-C l 6 l l 4 0 1 JJ 3 7 0 3 t 0 4 2 0 0 4 0 9
- 0 4 1 8 44 - O l 7 t ) 4 -ll 7 1 4 0 b -0 3 5 7 8 Z
- 0 1 9 7 7 0 1)0 3 3 4
-o n 9 2
IJ A T A
CONV
ON-C ONV
( 4 )
t 1NV
C 1t-V C NV c ) v C - JNV C rlNV c n v c a v C ON V ( i) l V
C CtV
C t1tV C IJNV
C 1 V
c 1 v
Pa ge 2La
HEQ F t Noe x
c 5 )
6 2q
6 89 1 48
6 1 4 ) q 3
1 9 7 1 9
q z t s 5 a e z
l l 9 1
4 o 7 1 2 0
c oqt 1 Z
4 4 )-q 1 z
5 5 9 1 0 64 1 3 3
n 3
Inter industry analysi
Page 23
Equation ( 3 6 ) may be used to fo rmulate a rough test o f the cooperat ion
vs e f ficiency controversy I f we rest ricted ourselves to some subset o f
indus tries where the relat ion be tween y and a i s the same we would be lookingbull
at a sample o f indus tries for which the dependence o f V on H and o is a
quadratic wi th no squared terms
I f we now sub s titute (26) into ( 36 ) we get
(38) v bull a + (y - a) R + (y - a) o (H) - (y - a) H o (H) + u
If it is only d if ferences in marke t shares that affect advertising inten-
s i ty levels i f concentration does not af fect cooperation and i f o bull 0
( 38 ) reduces t o
( 39 ) V bull a + (y - a) H + U
Let this b e the null hyp o thes is and note that V is a linear func-
tion o f H If y gt a and the null hyp othesis holds the expected relat ion
between V and H is as shown by the pos itively sloped straight line in Figure
6 b etween points A and C
In this context the alterna ive hypothesis the degree o f
cooperation ( o) is a positive function o f concentrat ion
is that
(H) I f the func-
tion is as described in (26) and Figure 4 the relation between V and H
would be as shown by the curved line connec t ing po ints A and C in Figure 6 One way to test the null hypothe s is against this al ternative is to test fo r
the hypothesized curvature in the function relating V and H
Fi gure
R
6
v
ID c
H 1 0
I
i I I
sfigned
clasJ s
in def ini g
effec t is the
I
Page middot2lf
One charac teristic of produc ts which may be useful
classes wi th respec t to significanc e of the brand-switching
conveniencenon-convenienc e distinc tion developed by Porter ( 1 97 )
Using his identif ication of IRS indus tries as a s tar t ing point
each of the 32 industries used in this s tudy to one of the two
The ass ignments are not ed in Tab le 4
Using the conveniencenon-convenienc e dis t inc tion the
divided into two sub-samples
sample was
Four regressions were then run bull ene for e ch
sub-sample one for the to tal sample and one for the total sample with
a dummy variable for convenience goods Each regression was run wi th LB s
and wi th indus tries as ob servations The results are given in Tab le 5
For the LB level equations ( 1A - 1D) equation(37) was used ome
rearranging of terms gives v bull y - ( 1-o ) (y - a) ( 1-z ) + ui i i
bull y - 6 1 ( 1 - z ) + ui When observations are pooled across indus trif s yi
is an industry variable In these regressions the values o f y es t imated
in the first stage were used Since industries have been grouped so as
to reduce differences in (y - a) that term is t reated as a cons tant
within each group Finally H the Herfindahl index has been subst ituted
for o the degree of cooperation The resultant equation is linear
y ( 1 - z ) and H x ( l - z ) This is the equation for whi ch results arei i
reported in Table 5 In all regress ions all variables were weighted by the
square roo t of sales
in
raquo
- 34
3 1
i l T 4 8LE S R E G E S S I ON R E SUL T S i X T f N S I ON S T O T Y PE Of
AND L E V E L OF C ONC E kT R A T I O
E O O tt S
tP bull T Y P E NR C ONV O B S tO N-CONV
E L A S T I shy
T I C J T Y
O E - t-41( T SliA R E
H( fl F
1 1 - K O x middot K O X E R F
- H S a 1 1 - H S t 1 8 1 1 9 1
R S O
1 1 0 1
q z
t - bull 9 f 2 1
Z t P D l 6 fC ONV - 1 9 1 s s c 2 1
e - l O I e bull u q a
2 3 1 0 1 1 Nl N- t O V 5 7 1 ( 1 2 19 1 6 1 - J l S 2 5 4
l l l l t z q o a 1 - J o v z I
l ( I NO l l A l l ol u l t o o c bull J u l l
d S
c 3 4 t t r a l
bull Z D 1 0 l l 4 l l 2 3 -gt 2 l l 2 0 - 4 lS 5 4 l C 4 1 o e 6 t l b b 5 1 1 0 1 1 1 c - 1 9 5 1 t d S 2 1 tV
OQ10 T-R A T I O S A E G I VE N l P A RE N T H E S E S l E V E L S f S I G 1 F l C ANC E A - 99 5 B - q 9 t - s NS J UC E NO I Nl E R C ( P T I S U (j f middot) I N f Hf t OUA TIOI S 6 E f 3R E A O J U S T H E N T F OR 1-t E T E R O SIC E O A S T f C I T V pIraquo
1
middot
I
R SO I S C A L C UL A T E D A S f I f bull 1 1 bull 1lt 1 K t t- E E N I S THf U A1 B E R O f fJe S E fi V A T I O S F t S HtE F - S T A T I S
I N DE P E ND E N T y II
T I C F O T H E U Y P O HIE S I S f A l l C JE F F I C H N T A R E QU A L HI l f RO A f40 K I S T H f UUM 8 E R OF Q ( A fL f S
I l l
1 A
1 6
t C
f 1 r R I l l
1 2 1
L Jt
L B
l q
L 1
on s T Y P E bull 2 )
P OUC T
X
H S I l l
l O l A I bull l 0 7 1
O O b O J
1 2 00 1
1 b 2 1 C I l l Z U
l l S b 1 -1 1 0 1
rlt F X l l - fi C f l
bull 1 I
l l
- ()4 2 3 l f H
1 - 2 7 9 1
K O X
1 1 - ti E R F I
l d l
S4 l i A I 2 l b
1 3 00
K O X -t f R F t S J
I 1 - HER F t
)
60 9
z o bull
0
t l l
3 8 8
1 7 6
5 64
64
N f bull
tl S bull
I l l
I 4 1
t ONV
NON-t ONV
A l l
A ll
c o v NO- ( ONV
1 4 1
1 5 1
- l S b
c - 1 1 2 1
5 8 2 C A I 1 6 7 0 1
middot I l 9 H
l 1 il
1 4 1 6 1
E L A C) T J shy
l t C I T Y
I S I
l b l
- 0 2 0
1 - 1 4 9 1
O Z 3 2 1 A I C b 6 b l
0 0 9 2 -J
1 1 b 4 1
o z t t t l 1 2 2 1 1
oru - Ht tt F
1 gt X bull b I
Page 25 bull
Mean val -as ior relevan t variables for the two sub-samples are
Conv Non-Conv
Adv Sales 4 7 2 3
iit Share 2 7 3 2
Ela cicity - 05 3 02 2
Herf indahl 7 3 1 860
That the dist nct ion between coft9enienee goods and non - convenience
goods mat ters is supported not only by these means but also by the regression
results Equat ions lA and lB are different in almost every respec t Only the
concentrat ionmarket share int eract ion term seems to mat ter for convenience
good s Jus t he opposite holds for non - convenience goods both elasticity1
and market sl e are highly significant but the concentrat ionmarket share L
int erac t ion t=rm is not
Given that one minus market share is used these equ3 tions the coeffici t
of market share is the negative of what is sh in col (6) of Table 5 For the
regress ion for non - convenience LB s reported in equation lB then LB s with larger
market shares have lower advertising to sales ratios
On the question of the impact of the degree of cooperation the story is mixed
For convenienc good s the coefficient is pos itive as equation (31) predicts when
there is a po sitive degree of cooperat ion Fo r the non-convenience goods LB s middot
concentration has no effec t
Aggregat ion t o the indus try level and then rer unning the equations $ives the
results shown in (2a-2d) of Tab le 5 As expected R2 goes up and levels of significanc e
fmiddotJr individual coefficients gJ down In add ition the coeffic ient of H ( l -H) inmiddot
quat ion 2A is no t significant though it is po s i tive Since 2A is the indus try level
I
--
all but one of them and that one is not significant For 18 of the 31
wo uld have on its adve rtising to sal es ratio
cognate of LA the variable H ( l-H) pl ays the same role in 2A as H ( l -z 1) p lays
in LA
Vconomies of scale in advert ising
The literature on advert ing contains a number of definitionbull o f economies
of scale and of findings eone rning its presence and magnitude Before I loo k
a t the data with the hope o f mmenting about scale effect s then I wan t to
define the term as I use it
By economies of scale I mean a decrease in the number o f units of adver-
tising which are needed to generate a unit of sales as advert ising is increased
Given the definition of the firm s own elast icity of demand with respect to
advertising (see the text foll wing equation (1) ) the technical definition
is that rii gt 1 Diseconomies of scale are def ined symmetrically i e
Yu lt 1
ith a cro s s - section oi observations on advertising intensity
(v bull a s1) and onmiddot advertising (a ) we may regress the former on the latter 1 i i
The results of such regress ions for the 32 industries are given in Table 2
lines lB ZB bull bull bull 32B For the 32 eases the regression coefficient is positive it
positive coefficients the coefficient i s signif icant
The que tion now is whether these results ar e evidence of pervasive dis-
economies of scale in advertislng I think no t Given my as sumption that the
data I observe are equilibrium results I may ri ghtly conclud e that almost every-
where high levels of advert ising are as sociated in equilibrium with high ratios
of advertising to sales I may no t conclud e anything however about the
impact that a move by some firm away from its equilibrium level o f advertising
middot- Page 27
Given the assumed relation in equation ( 37 ) there is a corre spondshy
ing equilibrium equation relating v and ai The equation is non - linear i
and I have not found a simp le way to characterize it It is po ssib le to
det ermi ne its deriviative with respect to a b y takin$ tne de riviative of ( 3 7 ) i
When that is done the result is
v - 1(40) i bull sl a a z ( 1-z )i i ira-l
Sinc e all the other terms in (40) are non-negative the sign dep ends only
on the sign of 6 1 which is in turn de pend ent only on the sign of (y - a)
That the signs of the coefficients of a in equations lB 2B bull bull bull 32B arei
not always equal to the signs of the coefficients of z in LA 2A bull bull bull 32A i
is still a mystery to me and something to be explored
bull
Summary
My purpose in writing this pa per wa s t o attempt to integrate explicit micro-
economic mode lling reasoned econome tric specification of error terms and high
quality data to exp lore some questions about adver tising by large firms
al quite important improvements could be made in the first two of those areas
and much data ma ssag ing is yet to be done in the third
Concerning the subst an ce of what I have done I think six things are im-
Page 2 8
Severshy
V and Conclusions
portant
1 Explicit modeling is worth the ef for t if for no other reas on than
that it provides a basis for choosing fr om a varie ty of funct ional
forms
2 The same is true for error specif ica t ion I did in fact look at scatshy
ter diagrams for all 32 industry categor ies and they vir tually all show
the het eroscedast icity which I as sumed and for which I correc ted
3 The predicted relation be tween advertis ing in tensity and market share
shows up clearly in on ly 25 of the case s examined I have ye t to
explore why that may b e the cas e
4 Some evidence concerning the presence and impact of coopera tion was
produced but it is not clear ly pervasive
5 The dis tinction between convenience goods and non-c onvenie nce goods is
unambiguously a good one
6 Virtually no evidence concerning econ omies of scale in adver tis ing can
be gleaned fr om this study given its assumpt ions
More work is called for on many of these issues my study seems to have
raised more questions than it has answer ed This is pr obab lly due to the richshy
ne ss of the data source since I had the oppor tunity to address several issue s
at once
- -- - Q --
Foo tnotes
Manager Line o f Business Program Federal Trade Commission The author depended heavily on several staf f members of the LB Pro gram for statistical and clerical support in particular Joe Cholka George Pascoe and Haro lene Jenkins Helpful comments were given b y Richard Caves Dennis Mueller Michael Porter and F M Scherer When the paper was in its early stages o f development both S teve Garber and Jon Rasmussen former economists in the LB Program provided eery useful cri tiques The views expressed here are my own o f course
1 The basic exp licit model of the role of advertising is Dorfman and S teiner ( 1954 ) and the models in this paper may be seen as ano ther extens ion of their work
2 For a review o f models of oligopoly which treat prices and quantities as relevant variables see Shubik ( 1 959) Extens ions to take advertis ing int o account are fairly s traightforward
3 Dorfman and S teiner dealt with a price-setting firm For a monopolis t inverting the demand relation to show Pbull P (Q A) _instead o f Q- (P A) has no effect on the results o f the analysis For oligopolists who face a sys tem of demand relations inversion does have some impac t on the results The dif ferences are not trivial nonetheless they will no t be explored in this study
4 Shubik ( 1 959) p
5 From a formal mathematical point o f view the function is similar to a1 i Legrangian funct ion of the form - r rr
L bull 11 + A ( j - (i i i+j
where the indicates some fixed level of the profits industry Following the same logic which resul ts in and income constraint as the marginal utility of the of the indus tr in the obj ective of firm i
6 Hall amp Weiss ( 1 967) p 323
7 See Schmallensee ( 1 97 2 ) Ch 2 pp 1 6-47
8 On this poin t see Imel amp Helmberger ( 1 9 7 1 )
9 Long ( 1 970) p 130 - 1 39
1 0 See Schmallensee ( 1 972) Ch 3 pp 48-8 7
rrj i+j ) of the rest of the
the identification profits of the rest
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980
(9)
-
v bull bull
bull
bull
zl [ Y21 Y22 bull bull Y2N J
-1zl 0
-1z2 r z
0 1
Page 4
Writ ing (8) compactly gives
the matrix in the two equationsA comparison of (6) and (9) shows that
differs only in that all the elasticities are present in (9) but only the
awn-elasticities are present in ( 6 ) Since advertising expense mus t be I non-negat ive the first-order requirements are that v be the maximum of t
i
the amount shown in 6) or (9) and zero
III A SIMPLIFIED DEMAND STRUCTURE
So far the demand equation system is comple tely general I will now
move to a particular demand structure which is charac terized by an industry
elasticity of price with respect to advertising a general brand-swi tching
elas ticity and a dependence of the firm-specific elas ticity on its relative
size I make the expli cit assumpt ion that
n pound
)l
0 AA
Consider the identification of y as an elasticity of industry price
Page 5
(10)
where y is an
- (y- a)
industry demand elasticity a is a brand-swi tching
el astici ty and z1 is the market share In matrix no tation
(11) r bull ai + (y- a) t
wi th respe ct to advertising 0
le t ting S denote dS observe that
0(12) SS bull
-
-1 0s t s - -1 0S t d - 9- -
apl apl aplaa1 aa2 a 0 a
middot( a Pl
( 13)
-
-0 a bull 0
a
) 0 X s
-1 0dAA bull then d
a bull
J 0
(JA) middot and
Let the sc alar (SS)(AA) be denoted by y so ya zft
If
1
d
-
I
I
(11)
--
y
I -f
l J- a)
a) l zt
Substitutin g into (13) gives
i [ ai + (y(14) -I- bull S A
bull az + (y -
Pa ge6
I I II
bull
since l bull 1
Under these assumpt ons then y is the elasticity of industry sales
(and average industry p ice) with respect to industry advertisin g Note
that if y bull 0 so that there is no industry sales effect there will still
be positive advertising if there are brand-switching effects
Turning now to the brand-switching ef fect observe that
(15)
Noting that
-1 deg -1d z -z -
-
0 a
Page 7
(15) may be written as
0 (AA)
lx 2x 1x
ith
1y
i
First assume no
that if there are
will be no changes in market shares
- 0a bull
Page 8
Equation (17) is a general statement of the relation betbulleen changes in
advertising levels and changes in market shares which would acompany them It
may be used to make some general observations about brand-switching effects
industry effect eg y bull r t bull 0 and assume further
equal percentage changes in all advertising le =ls then there
Under these further assumptions
0l and dz -1 bull (iA) ltr assumption the second term on the right hand side of this expression is
- By
zero
necessary for rt bull 0Consequently for market shares to be unaffected it is
when y bull r l bull 0 This requirement is satisfied by equation (11)
Secondly consider effects on market shares if one firm inc ases its abull ar-
tising and others do not As a general assumption we require smiddot ch a change o
lead to a decrease in the shares of all the other firms hat is let
0 0aiai gt 0 and a a bull 0 for j + i For any matrix X with N rows and its trans-j j
pose X let X
row of X
bull middotmiddotmiddot where is a column vector which contains the
elements of the Under the assumptions about the iis given just
d-1 0 0above it follows that r becomes (aiai) iimiddot ehFor jzj lt 0 under these conditions then it follows that the j- element
N of (I- l ) lt 0 so middotij - il lt 0 or yij - yik lt 0 If this
kbulll N condition holds for all Y ij where j + i then it follows that yii - z
k Y gta
k- 1 ik since the th firms share must increase to offset the decreases in all other
shares
ei ith
1v ( ) h i
7urning
where is the
leads to z
to the r matrix given in (11) we note that 1 = cr + (y - cr) i i zi elementary vector with 1 in the position and 0 elsewhere This
bull cr + (y - cr) zi Since bull required is thatcr at- zY wYij (y - cr)zi - cr + (y - cr) zilt 0 or cr gt 0
These observations may be summarized by substituting (11) into (17) bull
giving
(18)
i s
d -1 zz -
-
-
+ (y - cr) tz
d -1 0cr(I - l ) a a bull
cr 1 z - (y - cr) 0 a
o
If ai increases and the other a s are const t cr must be positive to havej
a positive effect on zi A positive cr also bull plies a negative ffect of aJ
f all a s in-on zi other a s (including a ) fixed j
eluding a1 change by the same percentage zi remains unchanged Finally k i
if cr bull 0 there is no brand-switching effect
For a to have no effect on p requires yij bull 0 From (10) thisj i
amounts to y bull cr that is the industry demand elasticity and the brandshyswitching elasticity are equal If y gt cr gt 0 the industry elasticity yij doninates If y lt cr y lt 0 the brand-switching elasticity dominates ij
making
Page 10
Substituting ( 1 1) now in (6) the Cournot case and in (9) the
Chamberlin ease gives
-(19) Cournot vO a + (y - a) z
Chamberli n vl - y lmiddot th
The 1- element is then
-(20) Cournot vO a+ (y - a) zi i
-Chamberlin vl yi
In (19) and (20) the superscript 0 is used to denote total
independence of decision and the superscript l is used to denote
total interdependence of decision making
The relation between v and vf depends solely on t relation between
y and a f y gt a so that the industry effect dominat 3 vf v with
equality in the trirlal case of z bull 1 Cooperative irms would have ai
higher advertising intensity than non-cooperative fi
If there is equality between the industry elasticity and the
y Whe therbrand-switching elasticity or y bull there is cooperation among the firms has no effect on advertising inten-sity And if brand-switching is more important with y lt a e get
vi lt v cooperative firms will have a lower advertising intensity than
non-cooperative firms The three situations are depicted in Figure L
Given the results for the firm level variables calculation of the
corresponding re sults for the industry is s-traightforward Let
B bull t zf bull z z be the Herfindahl index of concen tration and notethat
z - bull l z bull 1 and that V bull A S bull a i 5 bull (siIS) (asi)i
bull middot Then for the general case from (6) and (9)
bull
Page lOa
y lt1 bull y y
10
f
V0 bull z [ a1 + (y - cr) J bull a + (y - a) H vl bull Y bull y
11Page
( 2 1) vo bull z d -1 r bull r z bull z- d d- i )11
l -1 N N v bull z d r z bull r z bull l y z - ij jibulll jbulll
and for the simplified case from (19)
(22
Since the forms of the equations in (22) are the same as in (20) all
of the analysis of (20) holdS for (22) except that His substituted for
middot z Advertising intensity will be higher for an industry with cooperativei
firms than for one with non-coope rative firms if the industry demand elas-
ticity dominates the brand-switchine lasticity If the ewo elasticities
are equal then so will be advertiSl intensity for the cooperative firm
and non-cooperative firm industries If brand-switching dominates the
cooperative firm industry will have lower advertising intensity
In the industry model context the possibility of a critical concen-
tration level may be easily introduced That is assume that at values
of R below R the firms are non-cooperative but that for alues equal to
or greater than H they are cooperative That is
(23) V bull a + (y - a) H if H lt H bull y if H gtmiddot H
This function is shown in Figure 2 for the three relamptions be eween y and a
IV THE D GRE OF COOEtRATION
The models depicted in Figure 2 are based on a very simple notion about
the relation between concentration and the cooperationnon-cooperation charac-
teristic They use what is essentilly a step function relating concentra-
Figure
I -
J H H
Page l la
2
v v v
a
I J I y a bull Y y ---i
I I I I I I II I
H 10 H H 10 H 1 0
bullbull
bull bull
Page 12
tion and the extent or degree of coo peration as shown in Figure 3 Using
o as a symbol for the degree of cooperation eq uation (23) can be rewritten
as
bull(24) v (1 - o) vo + ovl bull (1 o) [a+ - a) H ] + oy- (y
vhere
(25) o bull o(H) bull 0 if B lt B
Hbull 1 if B gt
The specific relation between concentration (H) and the deg ree of
cooperation ( o) shown in Figure 3 is more restrictive than necessary and
bas less appeal an one which shows a smooth increase of o from 0 to 1 as
B increases fro to 1 Such a smooth function is shown in Figure 4 he
degree of cooper ion is low until H gets close to H increases greatly as
B goes through R attaining a high level for an H larger than B but still
close to it and then goes to 1 as B goes to 1 More formally
(26) 0 bull o(H)
bull 0 if H bull 0
bull 1 if H bull 1
do gt (IdB
d2o- gt 0 if H B dBZ
0 if H H
lt 0 if R gt H-
If (26) is now substituted for (25) in (24) a corresponding smoothing
of the functions shown in Figure 2 is accomplished he smooth functions
Figure 3
I
10
H 10 H
Page 12a
Figure
Page 12b
4
10
H 10 H
l
(27)
Cl - o) o
[ (1 -
[ (1
t y ij z
j
are given in Figure 5
Page 13
The equation in (24) was constructed as a hybrid of two results derived
from first-order equations for the non-cooperative and cooperative models
It may also be constructed directly by using l bull (1 - o) + on as thei i
th 2objective function for the i-- firm Treating o as a constant the first
order equation is
0 an aa
i middot
n i
Using equations (3) and (7) and setting --- bull 0 givesaai
(28) - 0
After collecting terms and converting to matrix notation we get
(29) - Cl - o) + 0 z-1vi yii i
N
j bulll
+ 0
+ o v1
d-1 r z z
-(30) v Cl - o) rd
-
o) r +d
d r J _ -1- d z
and middot
d) r + 0 rjd
l r z
v - z --(31) zv
- (1 0- +d) t r z- d-
If we now impose the simplified demand structure by substituting (11)
we get
- (1 - o) a +( 32) + oy and
Figure
l
-y
v
Y
H
r
I I
v
a
y
H 10 H
v
a
bulla
HH 10
bullbull
(33)
i
Page 14
v - (1 - o) [ a l + (y - a) ] The corresponding equation for Vis given in (24)
If the relation between industry advertising intensity (V) and
concentration (H) is non - linear via the impact of concentration
on the degree of cooperation does the relation take the form of a
auadratic as some other investigators have proposed (Greer ( 1971 ) Martin (1979) Strickland amp Weiss (1976) ) That question can
be approached from two perspectives - conceptual and empirical In
the next section I will try to apply some statistical material to it
here I want to explore the question in the context of the models
developed above
Referring to equation (24) taking the deriviative with respect
to H and letting bull (H) gives
bull( 3 3a ) 3V (y - a) (1 - o) + (1 - H) 3H dH
Given that o and H each are bounded by 0 and 1 and that o is an in-
creasing function of H the term in the brackets in non-negative The
sign of the deriviative then is the same as the sign of (y - a) If the industry effect is dominant (y gt a) then advertising in-
tensity will increase with concentration If the two effects are equal
then advertising intensity will not vbullry with concentration and if the
brand-switching e ffect is larger advertising intensity will decrease
with concentration
For given advertising elasticity parameters then it is not pos-
sible to show an increase in A S up to some H and the n a decrease
he relation may be non - linear but it is monotonic over the permis-
sible range of H
Page 15
One possibility is to introduce a relation between (y - a) and H
If industries characterized as dominated by the brand-switching effect
also were more concentrated then a cross industry comparison would
show an inverted U shape relation between AS and H I know of no
support for such an association between ( y - a) and H
V ERROR SPECUCATION
So far I have implicitly assumed that there are no errors in the
model I now assume that the error
thvariable in the equation for the i- firm is additive that its mean is zero
that its variance is inversely proportionate to its sales and that all error
term covarlances are zero hat is
34) v bull (1 - o) [ a + (y - a) z ] + oy + u i i i
- - -
In vector notation
(35) bull (1 - o) [ at + lt - a) J +
E() bull 0 Cov (u) -
0
If we now ca lculate V bull bull we get
oy + U bull
(36) V bull (1 - o) la + (y - a)R J +
U bull E(U) bull 0 Var (U) bull
P age 16
The error term specifi c ation which follows Hall
wi th the ratio of profits
6 heir analysis seems
as wellbull particularly given
and Weiss was
applied by them to equa tions to equity or
asse ts as dependen t variables to carry the same
weight in the present context the substantial
evidence presented by marketing analysis on the tendency of company de-
cision makers t o use the advertising to sales ratio as the decision
7variable
The zero covariance assumption on the other hand1 is particularly
troublesome Within a probably not true since
many even ts outside the
single industry it is
industry would have
industries
similar effec ts on all the
8firms in the industry Between it is probably not true either
since many companies produce in more than one indus try and even ts in the
firm would tend to effect all of its activities The extension of the
model to allow for non-zero covariances is certainly called for
--- -
Page Ll
Athird characteristic of this specification also deserves some comment
only on equation is shown for the firm or industry Several haveinvestigators
included an advertising equation ina larger system of equations (Comanor and
Wilson (1974) (1979) and Weiss (1976) so in myMartin Strickland Idid
dissertation1 Ihave ewo defenses for presenting only single equation results
here One is the suggestion by Comanor Wilson that simultaneousand equation
verybias may not be important in this context he second is that Iam will-
ing to see the model extended to include a profitability equation and I intend
to move that direction in the near futurein
Onthe inclusion of aconcentration equation on the other hand I amskep-
tical he problem isthat in some fundamental sense equations in models of
industrial organization should be oriented to the decision-making contexts of firms
Ifeach firm in anindustry with Nfirms has only one decision variable then
there can be only Nindependent equations which are based on the first-order
optimization equations It not show large ofis difficult of course to a number
functional relations which follow from the first-order conditions point here he
isthat only Nof them can be independent
Consider for illustrative purposes only a situation in wich teh N
itsquantities say q are set exogeneously and each firm determines advertising1
of and level say ai where a is the true optimal value a a bulla
1 1 i i + i
Ifp a) then the Nq
s and the Na s will determine N p s say pii bullp
i(l i 1 i
Now given Eand 9 we can determine both the vector s andthe vector v since
- bull til
ands We can also determine S s z bull s-1 s-ii
v and we
industry concentration isHbull (alternatively
V and Hare functions of the same
pendent Furthermore if there are
VM) and (H1 H2 bull bull bull )contain
VI STATISTI CAL APPLICATIONS
Page 18
Given the observed values of might be tempted to conclude that we
have a two equation system in which v and z are jointly determined That would i i
be inappropriate however since the 2N random variables (v ) are determined by
onlt N basic random variables t The variance-covariance matrix for ( ) must
be singular bull
The extension of this observation to industry level variables is straightshy
forward Since observed industry advertising intensity is V bull l v and observed 4
C4 bull z ) it follows thati l i
set of random variables t and cannot be indeshy-
M industries then the variables (V1 v2 at most M independent random variables
Given the assumptions made above y a and o are constant within a
given industry That being the case (34) can be written as
(37)
s1 bull (1 - o) (y - a)
If and
So + +
(1 - o) a + oy
81 are esti mated for an industry using a suitable statisticalSo procedure giving and 81 an estimate of y is then available The sumSo
Intra-industry analysis
1974 of Business Re ort -
categories are identified
Page 19
of So and S1 is y so y bull So + S1 is the estimate of y
If o is assumed to be bounded by zero and one the sign of 61 is also
the sign of y - a The estimate of So is of cou se an estimate of
(1 - o ) a + oy Moreove if y gt a then a is less than 6o- (1 - B) a+ 0(
Cor esponding results would
hold if y bull a or y lt a
If 61 gt 0 then we would estimate that a lt 6o
bull
For purposes of estimating this model 32 manufacturing industry categories
were selected The data are fram the Line of Business forms filed with
the Federal Trade Commission they are discussed in detail in the Annual Line
which is available from the FTC The 32 industry
in Table 1
Simple regressions for equ ation (37) were run for each of the 32 industries
with data for each of the firms which filed in the industry constituting an ob-
servation The results are given in Table 2 lines lA 2A
1974
the intercept
For all the A
equations generalized least quares was used th both and market
share bein scaled bv the square root of sales
The 32 industries were taken from the larger set of all manufacturing industry
categories in the 1974 data files Two selection cr iteria were used thereLB
had to be at least ten reporting companies and the industry had to be a consumer
goods industry or a producer goods industry which is strongly associated with a
related consumer goods industry so that adve rtising by the firms in the producer
goods industry impacts on demand in the related industry (eg soft drink syrup
and soft drinks) There were 135 industries with at least ten companies - 103
producer goods and 32 consumer goods
273
274
TABLE 1 INDUSTRY CATEGORIES USED
FTC
COOE OESCRI P T I ON _
2001 MEAT PACKINGt SAUSAGE S ANO OTHER PREPARED
MEAT PROOUCTS
2004 DAIRY PRODUCTS EXC F LUID MILK
2005 CANNED SPECIALTIES
2001 FROZEN SPECIALTIES
OEHYORATFOt AND
I NC LU 0 ING PRE SERVE S
SOUP MIXESbull
SEASO lNGSt ANO
OTHER PET FOOJ
RAIN MILL PROOUCTSt
ANO PREPARED
RELATED PRODUCTS
PRODUCTS
Z008 CANNEOt ORIEOt PICKLED FRUITS
ANO VEGET ABLES JAMS JELLIES DEHYDRATED I G TA3t E
S40CES ANO SALAD ORESSINGS
2010 DOG CAT A O
2012 FlOUR t OTHER RICE
MILLING fiLENOE) FLOUR
2014 B EADt CAKE ANO
20113 CO FECTIONERY
2026 BOTTLED A O CANNED SOFT DRINKS 2021 FLAVORING EXTRACTS AND SYRUPS EC
2029 MISC FOOOS AND KINDRED PROJUCTS EXC
ROASTED COFFEE
2301 MENS AND BOYS SUITS AND COATS
2302 MENS ANO BOYS FURNISHINGS
230 3 kOMENS AND MISSES OUTER EAR
2551 HOUSEHOLD FURNITURE
2702 PERIODICALS
Pqe 19a 1
RELATED 1972
src cooe
2011 f 3
202tX2026
2032
203pound
2033t4t5
2047 2041t4t5
2051 2065 2086
2081
209XZ05
231
232
233
251 272
2703 BOOKS
2704 ISC PVeLISHING
2806 ORGANIC FlEERS
2807 DRUG$ ETHICAL
23234
PT 2 S 3
2808 ORUCSt PROPRIETARY
2809 PERFU ES t COSMETICS ANO OTHER TaiLET
PREPARATIONS
2810 SOAP AND OTHER CLEANING PREPARATIONS 2815
2901
PEST1CIDES AND AGR ICULTURAL
PE OLEUM REF I N ING CHEMICALS EC
3606 HOUSEHOLD COOKING EQUIPMENT
3612 HOUSEHOLD APPLIANC ES NEC INCLUJING ELECshy
rqlC HOUSE ARFS AND FANS ANO SE I G MACHINES
3617 RADIO AND T V RECE I VING SETS 3808 PHOTOG APHIC E UIP amp SUP LfESt EXC
3qo3
PHO TOCOPYING
SPOR TING A D E QU I P M ENT f SUP P L I E S ATHLETTC GOODS NEC
3904 DOLLS GAMESt TOYS AND CHLOqENtS VE ICLES
PT283
2844
234tX2844
287 9
291 3631
3634t6t9
36H
PT386l
3949
15
943
middot
l A6l f 2 R(GRESSION RESULTS
EQ I NO
NR CODE cu 121
lA 2001 A (
2A 2004 r (
3A 2005 a
-R
ClS amp31
16
17
1 z
I TER-CEPT bull t
ooo9t21At 00022311 oooeos uu
OOl49tCAI 0010 2 middot 000960CCI
o049tl1At o04t55C61
MARKET SH ARE bull 5 I
-010670Ctl
-017600181
-001050
-0211AO
-0077 0
AOVER-TIS INti
lbl
000()39 000072
000018 0 0024liAI
-000079
SHARf
4ampgtVf Q bull
111
000140
-000170
RSQ
COl
66 485 862
607 6 3 t ()2
blti6 633
flASTI-
CI T Y PH
-007271(1
000441
-u0288l
clASTSHU (t)Ef
bull 10
tto
-234
210
( OOlltOCCI -126520 000922 181 001790
4A 200 7 18 00590liAI -038670 4e16 -032110 lell u c
OOOS74e 000700 025940
00067018) 0021001AI -02l100CAI
5 26
920
5A 6 c
200 8 26 0 03Sl9CA I 0012 26181 o 024511(
-005540
-0 81420 0001961AI O()Ol90CAI ooll240
441 604 891
-0020 19 274
6A 2010 16 Oe04224CAI o359lOIAI 9oq ll4015dCAI 061 6 -middot 005435CAI 00011 8 1AI 917 c 00535ltAI -009850((1 OOOA51J CAI -001610161 971
7A 2012 001862 000010 351 00187) oo1 a oooq45 ono1o6 439 c 001109 -027050 Oe0067SIAI -OOll60C81 897
- middotmiddot- -
8A 201t ll 000757 Ol4140CCI 814 ott9oocet 095 8 oooqo6 C C J 000069(81 811 c ootl2lampBI -080910CCI Oe009lUBI -OOtlOOCCI 960I
I DIIJIT pound11 1V OTIUTVTfAll-1pound1t- A _ 00 R - qq r - Q
1jl)OQCD
0
-middot middot--
l
_ __ _
I
i
0 00141 c c
91)7
I
T ABLE 2 REG ESSIOH RESULTS CCONTI
EO NR
IND CODE
iR bull
cos
I TEq-
CEPT
t1ARKE T
SHA E
AOVER-
Tl SING
SHARpound-cent
AuVER
RSO ILAST1-
C I T Y ELAST
SH CPtf (lt Cl) bull 11 I 41 I 5I (61 171 I 8 I f9) c 10)
9A 2 0 1 8 19 002397181 -007640 367 -005 246 le46 6 000605 000550CAI 597
-000029 G14180 OOl654CAJ -024li01At
lOA 2026 12 0028901AI -OlZl90 -0114 0 1 1 1 5 759
Ll (
llA 2021
f3 c
12A lOZ9
B
(
13A 2301
B
c
14A 2302
B
c
154 201
n (
16A 2551
8
10
35
10
20
19
24
001897(()
00141 6CCI
Oll9lHCI
010604 middot middot
Oll083
OOJ11JCAt
002059141
001411CIH
-000412
000293(81
000187
001159 (3)
ooo6a4cct
o00792CCI
OOl40dCAt
000006
-000076
OOlltl6fA)
o o0 5 64CCt
-034570
-o1middot o3o
-1659 90
Ol51lO
-046190CAt
Ol5450CA)
-005l60
002940
-030060(()
-0 43l50CCt
0 33650
-019080
00004
bullJ00741CAI
-000069
000466
0001 51CAt
000947CAt
O OOI)70CAt OOl750CA J
000175CCI
000671CAI
OOll5) f A t 002207CBI
0004b7CA I
- 007 1 70
001760
-001410CAt
- 0 16200CAt
-004210
- 1 207 70
705
96 0
542
466 805
460
614
921
832
9 61
996
424
550
9 15
56 1 651
884
-006114
019 037
03504lCA)
oo409Q
-141644((
491 -0 17664
607
871
lell
084
101
072
103
108 11IllOQfl)-- l
-037880fCJ OOl6l4CAJ -040l70fCI 0()
c
I
I
bull 9
9t
middot middot middotbullmiddot bull lraquo
TABLE 2 REGRE SION RESUlTS ICONTt
bull
EO I NO tHl I filER- f4A_K T AOVER- sbulluRE ttSQ flA Jl- iLAS T NR CODE cos CEPT SHARE JISING AOVfR bull (JTY SHR CiJCF
110Cl I f amp Itt 15J tbt bull 1c u caa
llA 2702 ll o079 6 81 Bt -019110 S6s -OlltO 111
a 004679 ooooo1 t69
( oOl26o -015160 001560 -022690 ltJr()I
ldA 270 3 19 0065lt -OIt2l60 220 -Ol57a2 11 8 e o 0 71t6IA t ol08 75tAt ( OOl983CiU -060llOC() OOl7191A t -O lltdOC A) middot 89
lqA 2704 10 005082 -Oil060 22b -l)i1977 lelO 6 -000080 oo2772C4 t 910 c 000907 -011 200 Oe04105CA t -046960 995
20A 2d06 1 II
O OlltStiAt -001120 837 00014 -))th O01380amp At -oooo 11 821 ( OOl526CAt -026700181 0002 5UAt 000530 969
2lA 2807 28 OOt9lt9CAt -00891 0 516 -0 03962 225I8 002095CCt OOOl lliAt 690 ( o02965CA t -056650((1 OOOt50CAJ -003l70ttl 911I
I 22A 2808 15 ot3l06 119730 655 1130) 090
0 - 007526CC t Oe003961At 835- middot-
c I oo7523 -067210 001l 2tA t -0091t80CCt 951
2lA 2809 21 Oe2067lCAt -1218101At 71 2 -10095((1 121 ttlIaOQ8 00797418) OOOI8lltt 686 Ill( 0092lU A t -Olltl70Ct t 000605CAt -OOit8701At 951 _
---- -
32 009506CAt
0 bull 06 981 I A J 006901 cA bull
00 24A 2810 -010960 5t 5 -OOltSO 7 56
8 -000011 s11 c -5811701At oooaoatAJ oot520IAI 888
-- 37 --- middot - -- -1 ) l lbull t5JJ r-l bullhmiddot1 --- - --
l5A
26A
27A
c
2RA
c
2 9A
31A
llA
24
13
597
Ci)55
14
l 6h
RSQ ELASTI- (lASTCOE f-
TABLE 2
EQ I NO NR CODE IU
REGRESSION RESULTS CCONTa
NR INTER- MARKET
(10S CEPT SHARE AOVER- SHARfcent
TIS lNG AOVER
) CITY StiR
c It) bull 5 co a I 1) bull 0 bull 9 c l 0 bull( 2 1
2815 16 003019 -009990 275 -o069bmiddotJ
6 001446 000224 282 c 002178 -054630 o02221CCI -0075 00 646
2901 29 00021t2 CAI 0)0520 592 0007 6 1 068
0 00105C8 t 000012141 60)g( OOOlJ51AJ -00126018) ooou271At -000u80CA) 974
36oa 1 3 003296(4) -oca29o -oo-q94
001755 000203 lt91t 002499((1 -04l820CCI OOl374CAI 0068)0 946
3612 0033281CI 011520 195 01684) oso
00 30 42C OI 00012 2 424 oo579ltAt -2 7l7801At 002 100IAI -003220 819
8 3617 btel 011789 081002259 009530
001448 oOOl OitCCI 732 OOl7 22CAI -0 46160CAI 0004961AI -OOll40181 989c
30A 3808 14 0021t 3 6fAI -000110 628 o023291C) -005 6 middot- _ middot- 0bull 02 04 1 IC I ooooo9 637 ( 001584((1 -0161)1)0 00021 91AI -0 00060
390) 14 002790CCt 006590 h8 t 0 0938 4 070
B OOl89bCCJ 000144 74
0021961(1 -OItlOlO OOl323CAI -004800 915
B
middot--- - middot 4 - middot- - dfll3904 12 004563181 0)8300 879 0428651(1 089 OQIll004 277CA I ooozstcea 919
c 0041)3181 -0-11250 0011881(1 -006270 972 0Ill
- J - bull bull bull bullbullbull middot - - middot - middot gt middotbull L)ofbullflbull middot middotfJltl -tn- middotmiddotmiddotmiddoto +f middot t middot bull
c
is
positive
2
Page 0 determineThe consumer goodproducer good split is sometimes difficult to
Sollle support for the split used in this paper is a comparison of goodness - of - fit
distributions of a
industries
set
advertising are
that industry
a sales increase of
the
highest aD10U3 those
measures for the two groups of industries Table 3 contains
measures for the 32- consumer goods industries and the 103 producer goods
2 Median R is 5 7 for the consumer goods set and 40 for the producer goods
Estimates - of y the elasticity of demand with respect to
given in column of Table 2 The hypothesis that y bull 0 with
alternative was applied for each ittdustry As shown in the
could not be rtjected for 24 of the 32 industries Of the
and 10 are positive For the remalnifg eight three are
and five are significantly positive
Of the threertegative elasetcities the largest (in absolute value)
cosmetics (code 2809) With all elasticity of -1 01 a one
industry advertising would gwerate a decrease in industry sales
more than one percetu
At the other extrema is toys and games (code 39 04) For
a one percent ittcrease in industry advertising would lead to
four-tenths of a pampreentmiddot Though not significantly different from zero
elasticity estimatbull for pt-bpuetary drugs (code 2808)
which are
Both drug industries have
to sales rafios (211 pe1 cent for 28 08 and 13 4 per cent
other hand flaIJo ing ext1ac s (mainly soft-drink SYUlS) also
is for
percent increase in
of slightly
a middottwo - tailed
table that hypothesis
24 14 are negative
significantly negative
9
the codetics and p-ropriacuy very high advertising
for 28 09) bull On the
has a high ind UY
advenisiltg t() sales ratio ( 8 0 per cent) bull but its estimated elasticity is very
small and not signifieantly different uom zero
middot It doe$ not seem unusual to find many elasticities near zero with a few
negative and a few positive If aggregate consumption is insensitive to aggregate
advertising as seems likely ld then what one industry gains must be lost by
others
Page 21
elasticity of demand does
(10) we can
if t 0 then
Finally a determination concerning the industry
not fix the individual firms elasticity Going back to equation
write y bull (1 - ) a + z y Since a gt 0 it follows thatii i i
Y gt 0 If Ylt 0 however y s sign is indeterminateij ij
Turning next to the coefficient of market share 20 of them are negative and
12 are positive with three of each sign being significantly different from zero
The largest positive value which is significant is for pet food (code 2010) and
a close second is mens amp boys suits and coats (code 23 01) The largest signifi
cant negative value is for cosmetics (code 28 09)
In the model some relations among y a 0 and 61 are implied If middota is
negative (1- y) is greater than one Since 61Y bull (1- a) (1 - ) 61Y is
pCsitive but it will be less than one if 0 is large enough Seventeen of the A
ys are negative in each of the seventeen industries 61 is also negative and les
1 (algebraicly) than y so 61Y is greater than one his is not evidence that the
degree of cooperation is low or zero in any industry of course A high o togethegt
with a negative y which is small (in absolute terms) relative to a can give a
value of 61Y greater than one On the other hand if 61Y were less than one
o would have to be greater than zero
For y gt 0 two conditions may hold If y lt a it follows that 61 lt 0 since
(y - a ) lt 0 It also follows that 61Y lt 0 regardless of the value of o bull
In addition the relation is an if-and-only-if one ie if 61Y is negative
then y lt a Three industries have negative values for 61y dairy products
except milk (code 20 04) synthetic fibers (code 2806) and photographic equipment
(code 3808) In all three cases y is quite small so a does not have to be very
large to give a negative value to al
If y gt a the model shows that 61-- gt 0 whatever the value of 0 bull There are
12 cases where both y and 61 are positive with 61 ranging from near zero
(grain milling code 20 12) to 1 20 ( proprietary drugs code 2808) Several
coeffiei-en_rs
--
15
o f to zero
Tab le 3 - Consumer Goods
Indus tr ies
2Distributions of R and Producer Goods
Page 20a
Consumer Producer
Goods Goods
0 - 9 9 0 3
10 - 19 9 0 1 1
20 - 29 9 3 20
30 - 39 9 3 17
40 - 49 9 5 16
50 - 59 9 8
69 9 6 Ii 1060
--f 670 79 9 2
80 - 89 9 4 4
90 - 100 1 1
Totals 32 10 3
Notes
2is used R i s cal-
culated as F F + ((N - K- 1 ) K ] 1 b_ F is the F
statistic Lor the hypothesis that all o f the non - intercept terms are simul taneously equal
1
2
--
-- Since generalized least squ-area regress ion
Page 12
o f these cases have values o f 6 1 Y rea ter than 0 8 (pet food (20 10) bread amp cakes
(20 14 ) miscellaneous processed foods (20 29 ) men s and boys suit s and coats
(23 01 ) p rop riet ary drugs (28 08 ) elect ric housewares (36 1 2) radio and
IV sets 17 and games and toys (39 For be than(36 ) 04)) 6 1 Y to greater 0 8
both o and cry mus t b e less than 0 2 That is in these industries the results
shown in Tab le 2 would hold only with very strong dominance of marke t demand
effects over brand swit ching e ffects and a very low degree o f cooperation
Using the concentration rat ios given in Table 4 the average concentration
ratio for the eight indust ries is 38 1 For the remaining 24 industries it is
43 2 This result would lend some support t o the proposition that concentrat ion
and cooperation are related if the group of eight industries are identi fied
as having low degrees of cooperation
One f inal note on model predict ions S ince both cr and o are assumed to
be non-negative 60 should also be non-negat ive In only one o f the 32 cas es
is 60 negative and it is not signif icant bull
cor pJ
7 1
7 0 4
1 4 7 5 1
)
7 3 z a o a t s 1 1 3 3 5 ( i V Z 8 tJ1 2 1 - 1 1 0 7 3 5 c middot v 2 8 1 0 3 2 -0 0 1 4 3 0 C 1 V 9 8 3 2 8 1 5 1 6 - 0 0 6 t1 6 9 C _ V 1 11 0 2 9 0 1 29 0 00 7 6 1 C l k V 7 3 1 3 6 08 1 3 - o n 4 9 9 4 3 3 3 6 1 2 2 4 0 1 6 4 3
3 1 7 1 3 0 1 1 7 d 7 8 3 3 8 0 8 0 0 2 3 i 2 3 0 6
P 4
3 9 0 3 1 4
1 2
0 09 3 8 4 39 1)1 0 4 l 8 6 5
T A LE 4 e I NCU S T R Y C A T E G O R Y
F TC NOM e tOOE ass
C 1 I c 2 )
2 0 0 1 1 6 2 0 0r 1 7 zo o l Z 2 0 0 7 1 8
z o o 2 6
z o t a 1 6
2 1) 1 2 1 5 2 0 1 4 1 1 2 0 1 a 1 9 2 0 26 1 2 2 0 2 7 1 0 2 0 29 - 3 5
2 3 0 1 10
2 3 02 z o
2 3 03 1 9
2 5 5 1 2 4 Z 1 0 Z 1 1 2 0 3 1 9 2 7 4 1 0 z a o 1 3
2 A 0 7 z s
E LA S T I -C I T Y
c 3 )
- 0 09 7 2 7
0 00 449
- 0 0 2 8 9 3
- 0 3 2 7 70 - 0 OZ il I q
o 4 u t s a 0 0 1 3 7 3
o 1 4 o o-0 0 5 2 4o -0 1 94 0 1
-C l 6 l l 4 0 1 JJ 3 7 0 3 t 0 4 2 0 0 4 0 9
- 0 4 1 8 44 - O l 7 t ) 4 -ll 7 1 4 0 b -0 3 5 7 8 Z
- 0 1 9 7 7 0 1)0 3 3 4
-o n 9 2
IJ A T A
CONV
ON-C ONV
( 4 )
t 1NV
C 1t-V C NV c ) v C - JNV C rlNV c n v c a v C ON V ( i) l V
C CtV
C t1tV C IJNV
C 1 V
c 1 v
Pa ge 2La
HEQ F t Noe x
c 5 )
6 2q
6 89 1 48
6 1 4 ) q 3
1 9 7 1 9
q z t s 5 a e z
l l 9 1
4 o 7 1 2 0
c oqt 1 Z
4 4 )-q 1 z
5 5 9 1 0 64 1 3 3
n 3
Inter industry analysi
Page 23
Equation ( 3 6 ) may be used to fo rmulate a rough test o f the cooperat ion
vs e f ficiency controversy I f we rest ricted ourselves to some subset o f
indus tries where the relat ion be tween y and a i s the same we would be lookingbull
at a sample o f indus tries for which the dependence o f V on H and o is a
quadratic wi th no squared terms
I f we now sub s titute (26) into ( 36 ) we get
(38) v bull a + (y - a) R + (y - a) o (H) - (y - a) H o (H) + u
If it is only d if ferences in marke t shares that affect advertising inten-
s i ty levels i f concentration does not af fect cooperation and i f o bull 0
( 38 ) reduces t o
( 39 ) V bull a + (y - a) H + U
Let this b e the null hyp o thes is and note that V is a linear func-
tion o f H If y gt a and the null hyp othesis holds the expected relat ion
between V and H is as shown by the pos itively sloped straight line in Figure
6 b etween points A and C
In this context the alterna ive hypothesis the degree o f
cooperation ( o) is a positive function o f concentrat ion
is that
(H) I f the func-
tion is as described in (26) and Figure 4 the relation between V and H
would be as shown by the curved line connec t ing po ints A and C in Figure 6 One way to test the null hypothe s is against this al ternative is to test fo r
the hypothesized curvature in the function relating V and H
Fi gure
R
6
v
ID c
H 1 0
I
i I I
sfigned
clasJ s
in def ini g
effec t is the
I
Page middot2lf
One charac teristic of produc ts which may be useful
classes wi th respec t to significanc e of the brand-switching
conveniencenon-convenienc e distinc tion developed by Porter ( 1 97 )
Using his identif ication of IRS indus tries as a s tar t ing point
each of the 32 industries used in this s tudy to one of the two
The ass ignments are not ed in Tab le 4
Using the conveniencenon-convenienc e dis t inc tion the
divided into two sub-samples
sample was
Four regressions were then run bull ene for e ch
sub-sample one for the to tal sample and one for the total sample with
a dummy variable for convenience goods Each regression was run wi th LB s
and wi th indus tries as ob servations The results are given in Tab le 5
For the LB level equations ( 1A - 1D) equation(37) was used ome
rearranging of terms gives v bull y - ( 1-o ) (y - a) ( 1-z ) + ui i i
bull y - 6 1 ( 1 - z ) + ui When observations are pooled across indus trif s yi
is an industry variable In these regressions the values o f y es t imated
in the first stage were used Since industries have been grouped so as
to reduce differences in (y - a) that term is t reated as a cons tant
within each group Finally H the Herfindahl index has been subst ituted
for o the degree of cooperation The resultant equation is linear
y ( 1 - z ) and H x ( l - z ) This is the equation for whi ch results arei i
reported in Table 5 In all regress ions all variables were weighted by the
square roo t of sales
in
raquo
- 34
3 1
i l T 4 8LE S R E G E S S I ON R E SUL T S i X T f N S I ON S T O T Y PE Of
AND L E V E L OF C ONC E kT R A T I O
E O O tt S
tP bull T Y P E NR C ONV O B S tO N-CONV
E L A S T I shy
T I C J T Y
O E - t-41( T SliA R E
H( fl F
1 1 - K O x middot K O X E R F
- H S a 1 1 - H S t 1 8 1 1 9 1
R S O
1 1 0 1
q z
t - bull 9 f 2 1
Z t P D l 6 fC ONV - 1 9 1 s s c 2 1
e - l O I e bull u q a
2 3 1 0 1 1 Nl N- t O V 5 7 1 ( 1 2 19 1 6 1 - J l S 2 5 4
l l l l t z q o a 1 - J o v z I
l ( I NO l l A l l ol u l t o o c bull J u l l
d S
c 3 4 t t r a l
bull Z D 1 0 l l 4 l l 2 3 -gt 2 l l 2 0 - 4 lS 5 4 l C 4 1 o e 6 t l b b 5 1 1 0 1 1 1 c - 1 9 5 1 t d S 2 1 tV
OQ10 T-R A T I O S A E G I VE N l P A RE N T H E S E S l E V E L S f S I G 1 F l C ANC E A - 99 5 B - q 9 t - s NS J UC E NO I Nl E R C ( P T I S U (j f middot) I N f Hf t OUA TIOI S 6 E f 3R E A O J U S T H E N T F OR 1-t E T E R O SIC E O A S T f C I T V pIraquo
1
middot
I
R SO I S C A L C UL A T E D A S f I f bull 1 1 bull 1lt 1 K t t- E E N I S THf U A1 B E R O f fJe S E fi V A T I O S F t S HtE F - S T A T I S
I N DE P E ND E N T y II
T I C F O T H E U Y P O HIE S I S f A l l C JE F F I C H N T A R E QU A L HI l f RO A f40 K I S T H f UUM 8 E R OF Q ( A fL f S
I l l
1 A
1 6
t C
f 1 r R I l l
1 2 1
L Jt
L B
l q
L 1
on s T Y P E bull 2 )
P OUC T
X
H S I l l
l O l A I bull l 0 7 1
O O b O J
1 2 00 1
1 b 2 1 C I l l Z U
l l S b 1 -1 1 0 1
rlt F X l l - fi C f l
bull 1 I
l l
- ()4 2 3 l f H
1 - 2 7 9 1
K O X
1 1 - ti E R F I
l d l
S4 l i A I 2 l b
1 3 00
K O X -t f R F t S J
I 1 - HER F t
)
60 9
z o bull
0
t l l
3 8 8
1 7 6
5 64
64
N f bull
tl S bull
I l l
I 4 1
t ONV
NON-t ONV
A l l
A ll
c o v NO- ( ONV
1 4 1
1 5 1
- l S b
c - 1 1 2 1
5 8 2 C A I 1 6 7 0 1
middot I l 9 H
l 1 il
1 4 1 6 1
E L A C) T J shy
l t C I T Y
I S I
l b l
- 0 2 0
1 - 1 4 9 1
O Z 3 2 1 A I C b 6 b l
0 0 9 2 -J
1 1 b 4 1
o z t t t l 1 2 2 1 1
oru - Ht tt F
1 gt X bull b I
Page 25 bull
Mean val -as ior relevan t variables for the two sub-samples are
Conv Non-Conv
Adv Sales 4 7 2 3
iit Share 2 7 3 2
Ela cicity - 05 3 02 2
Herf indahl 7 3 1 860
That the dist nct ion between coft9enienee goods and non - convenience
goods mat ters is supported not only by these means but also by the regression
results Equat ions lA and lB are different in almost every respec t Only the
concentrat ionmarket share int eract ion term seems to mat ter for convenience
good s Jus t he opposite holds for non - convenience goods both elasticity1
and market sl e are highly significant but the concentrat ionmarket share L
int erac t ion t=rm is not
Given that one minus market share is used these equ3 tions the coeffici t
of market share is the negative of what is sh in col (6) of Table 5 For the
regress ion for non - convenience LB s reported in equation lB then LB s with larger
market shares have lower advertising to sales ratios
On the question of the impact of the degree of cooperation the story is mixed
For convenienc good s the coefficient is pos itive as equation (31) predicts when
there is a po sitive degree of cooperat ion Fo r the non-convenience goods LB s middot
concentration has no effec t
Aggregat ion t o the indus try level and then rer unning the equations $ives the
results shown in (2a-2d) of Tab le 5 As expected R2 goes up and levels of significanc e
fmiddotJr individual coefficients gJ down In add ition the coeffic ient of H ( l -H) inmiddot
quat ion 2A is no t significant though it is po s i tive Since 2A is the indus try level
I
--
all but one of them and that one is not significant For 18 of the 31
wo uld have on its adve rtising to sal es ratio
cognate of LA the variable H ( l-H) pl ays the same role in 2A as H ( l -z 1) p lays
in LA
Vconomies of scale in advert ising
The literature on advert ing contains a number of definitionbull o f economies
of scale and of findings eone rning its presence and magnitude Before I loo k
a t the data with the hope o f mmenting about scale effect s then I wan t to
define the term as I use it
By economies of scale I mean a decrease in the number o f units of adver-
tising which are needed to generate a unit of sales as advert ising is increased
Given the definition of the firm s own elast icity of demand with respect to
advertising (see the text foll wing equation (1) ) the technical definition
is that rii gt 1 Diseconomies of scale are def ined symmetrically i e
Yu lt 1
ith a cro s s - section oi observations on advertising intensity
(v bull a s1) and onmiddot advertising (a ) we may regress the former on the latter 1 i i
The results of such regress ions for the 32 industries are given in Table 2
lines lB ZB bull bull bull 32B For the 32 eases the regression coefficient is positive it
positive coefficients the coefficient i s signif icant
The que tion now is whether these results ar e evidence of pervasive dis-
economies of scale in advertislng I think no t Given my as sumption that the
data I observe are equilibrium results I may ri ghtly conclud e that almost every-
where high levels of advert ising are as sociated in equilibrium with high ratios
of advertising to sales I may no t conclud e anything however about the
impact that a move by some firm away from its equilibrium level o f advertising
middot- Page 27
Given the assumed relation in equation ( 37 ) there is a corre spondshy
ing equilibrium equation relating v and ai The equation is non - linear i
and I have not found a simp le way to characterize it It is po ssib le to
det ermi ne its deriviative with respect to a b y takin$ tne de riviative of ( 3 7 ) i
When that is done the result is
v - 1(40) i bull sl a a z ( 1-z )i i ira-l
Sinc e all the other terms in (40) are non-negative the sign dep ends only
on the sign of 6 1 which is in turn de pend ent only on the sign of (y - a)
That the signs of the coefficients of a in equations lB 2B bull bull bull 32B arei
not always equal to the signs of the coefficients of z in LA 2A bull bull bull 32A i
is still a mystery to me and something to be explored
bull
Summary
My purpose in writing this pa per wa s t o attempt to integrate explicit micro-
economic mode lling reasoned econome tric specification of error terms and high
quality data to exp lore some questions about adver tising by large firms
al quite important improvements could be made in the first two of those areas
and much data ma ssag ing is yet to be done in the third
Concerning the subst an ce of what I have done I think six things are im-
Page 2 8
Severshy
V and Conclusions
portant
1 Explicit modeling is worth the ef for t if for no other reas on than
that it provides a basis for choosing fr om a varie ty of funct ional
forms
2 The same is true for error specif ica t ion I did in fact look at scatshy
ter diagrams for all 32 industry categor ies and they vir tually all show
the het eroscedast icity which I as sumed and for which I correc ted
3 The predicted relation be tween advertis ing in tensity and market share
shows up clearly in on ly 25 of the case s examined I have ye t to
explore why that may b e the cas e
4 Some evidence concerning the presence and impact of coopera tion was
produced but it is not clear ly pervasive
5 The dis tinction between convenience goods and non-c onvenie nce goods is
unambiguously a good one
6 Virtually no evidence concerning econ omies of scale in adver tis ing can
be gleaned fr om this study given its assumpt ions
More work is called for on many of these issues my study seems to have
raised more questions than it has answer ed This is pr obab lly due to the richshy
ne ss of the data source since I had the oppor tunity to address several issue s
at once
- -- - Q --
Foo tnotes
Manager Line o f Business Program Federal Trade Commission The author depended heavily on several staf f members of the LB Pro gram for statistical and clerical support in particular Joe Cholka George Pascoe and Haro lene Jenkins Helpful comments were given b y Richard Caves Dennis Mueller Michael Porter and F M Scherer When the paper was in its early stages o f development both S teve Garber and Jon Rasmussen former economists in the LB Program provided eery useful cri tiques The views expressed here are my own o f course
1 The basic exp licit model of the role of advertising is Dorfman and S teiner ( 1954 ) and the models in this paper may be seen as ano ther extens ion of their work
2 For a review o f models of oligopoly which treat prices and quantities as relevant variables see Shubik ( 1 959) Extens ions to take advertis ing int o account are fairly s traightforward
3 Dorfman and S teiner dealt with a price-setting firm For a monopolis t inverting the demand relation to show Pbull P (Q A) _instead o f Q- (P A) has no effect on the results o f the analysis For oligopolists who face a sys tem of demand relations inversion does have some impac t on the results The dif ferences are not trivial nonetheless they will no t be explored in this study
4 Shubik ( 1 959) p
5 From a formal mathematical point o f view the function is similar to a1 i Legrangian funct ion of the form - r rr
L bull 11 + A ( j - (i i i+j
where the indicates some fixed level of the profits industry Following the same logic which resul ts in and income constraint as the marginal utility of the of the indus tr in the obj ective of firm i
6 Hall amp Weiss ( 1 967) p 323
7 See Schmallensee ( 1 97 2 ) Ch 2 pp 1 6-47
8 On this poin t see Imel amp Helmberger ( 1 9 7 1 )
9 Long ( 1 970) p 130 - 1 39
1 0 See Schmallensee ( 1 972) Ch 3 pp 48-8 7
rrj i+j ) of the rest of the
the identification profits of the rest
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980
n pound
)l
0 AA
Consider the identification of y as an elasticity of industry price
Page 5
(10)
where y is an
- (y- a)
industry demand elasticity a is a brand-swi tching
el astici ty and z1 is the market share In matrix no tation
(11) r bull ai + (y- a) t
wi th respe ct to advertising 0
le t ting S denote dS observe that
0(12) SS bull
-
-1 0s t s - -1 0S t d - 9- -
apl apl aplaa1 aa2 a 0 a
middot( a Pl
( 13)
-
-0 a bull 0
a
) 0 X s
-1 0dAA bull then d
a bull
J 0
(JA) middot and
Let the sc alar (SS)(AA) be denoted by y so ya zft
If
1
d
-
I
I
(11)
--
y
I -f
l J- a)
a) l zt
Substitutin g into (13) gives
i [ ai + (y(14) -I- bull S A
bull az + (y -
Pa ge6
I I II
bull
since l bull 1
Under these assumpt ons then y is the elasticity of industry sales
(and average industry p ice) with respect to industry advertisin g Note
that if y bull 0 so that there is no industry sales effect there will still
be positive advertising if there are brand-switching effects
Turning now to the brand-switching ef fect observe that
(15)
Noting that
-1 deg -1d z -z -
-
0 a
Page 7
(15) may be written as
0 (AA)
lx 2x 1x
ith
1y
i
First assume no
that if there are
will be no changes in market shares
- 0a bull
Page 8
Equation (17) is a general statement of the relation betbulleen changes in
advertising levels and changes in market shares which would acompany them It
may be used to make some general observations about brand-switching effects
industry effect eg y bull r t bull 0 and assume further
equal percentage changes in all advertising le =ls then there
Under these further assumptions
0l and dz -1 bull (iA) ltr assumption the second term on the right hand side of this expression is
- By
zero
necessary for rt bull 0Consequently for market shares to be unaffected it is
when y bull r l bull 0 This requirement is satisfied by equation (11)
Secondly consider effects on market shares if one firm inc ases its abull ar-
tising and others do not As a general assumption we require smiddot ch a change o
lead to a decrease in the shares of all the other firms hat is let
0 0aiai gt 0 and a a bull 0 for j + i For any matrix X with N rows and its trans-j j
pose X let X
row of X
bull middotmiddotmiddot where is a column vector which contains the
elements of the Under the assumptions about the iis given just
d-1 0 0above it follows that r becomes (aiai) iimiddot ehFor jzj lt 0 under these conditions then it follows that the j- element
N of (I- l ) lt 0 so middotij - il lt 0 or yij - yik lt 0 If this
kbulll N condition holds for all Y ij where j + i then it follows that yii - z
k Y gta
k- 1 ik since the th firms share must increase to offset the decreases in all other
shares
ei ith
1v ( ) h i
7urning
where is the
leads to z
to the r matrix given in (11) we note that 1 = cr + (y - cr) i i zi elementary vector with 1 in the position and 0 elsewhere This
bull cr + (y - cr) zi Since bull required is thatcr at- zY wYij (y - cr)zi - cr + (y - cr) zilt 0 or cr gt 0
These observations may be summarized by substituting (11) into (17) bull
giving
(18)
i s
d -1 zz -
-
-
+ (y - cr) tz
d -1 0cr(I - l ) a a bull
cr 1 z - (y - cr) 0 a
o
If ai increases and the other a s are const t cr must be positive to havej
a positive effect on zi A positive cr also bull plies a negative ffect of aJ
f all a s in-on zi other a s (including a ) fixed j
eluding a1 change by the same percentage zi remains unchanged Finally k i
if cr bull 0 there is no brand-switching effect
For a to have no effect on p requires yij bull 0 From (10) thisj i
amounts to y bull cr that is the industry demand elasticity and the brandshyswitching elasticity are equal If y gt cr gt 0 the industry elasticity yij doninates If y lt cr y lt 0 the brand-switching elasticity dominates ij
making
Page 10
Substituting ( 1 1) now in (6) the Cournot case and in (9) the
Chamberlin ease gives
-(19) Cournot vO a + (y - a) z
Chamberli n vl - y lmiddot th
The 1- element is then
-(20) Cournot vO a+ (y - a) zi i
-Chamberlin vl yi
In (19) and (20) the superscript 0 is used to denote total
independence of decision and the superscript l is used to denote
total interdependence of decision making
The relation between v and vf depends solely on t relation between
y and a f y gt a so that the industry effect dominat 3 vf v with
equality in the trirlal case of z bull 1 Cooperative irms would have ai
higher advertising intensity than non-cooperative fi
If there is equality between the industry elasticity and the
y Whe therbrand-switching elasticity or y bull there is cooperation among the firms has no effect on advertising inten-sity And if brand-switching is more important with y lt a e get
vi lt v cooperative firms will have a lower advertising intensity than
non-cooperative firms The three situations are depicted in Figure L
Given the results for the firm level variables calculation of the
corresponding re sults for the industry is s-traightforward Let
B bull t zf bull z z be the Herfindahl index of concen tration and notethat
z - bull l z bull 1 and that V bull A S bull a i 5 bull (siIS) (asi)i
bull middot Then for the general case from (6) and (9)
bull
Page lOa
y lt1 bull y y
10
f
V0 bull z [ a1 + (y - cr) J bull a + (y - a) H vl bull Y bull y
11Page
( 2 1) vo bull z d -1 r bull r z bull z- d d- i )11
l -1 N N v bull z d r z bull r z bull l y z - ij jibulll jbulll
and for the simplified case from (19)
(22
Since the forms of the equations in (22) are the same as in (20) all
of the analysis of (20) holdS for (22) except that His substituted for
middot z Advertising intensity will be higher for an industry with cooperativei
firms than for one with non-coope rative firms if the industry demand elas-
ticity dominates the brand-switchine lasticity If the ewo elasticities
are equal then so will be advertiSl intensity for the cooperative firm
and non-cooperative firm industries If brand-switching dominates the
cooperative firm industry will have lower advertising intensity
In the industry model context the possibility of a critical concen-
tration level may be easily introduced That is assume that at values
of R below R the firms are non-cooperative but that for alues equal to
or greater than H they are cooperative That is
(23) V bull a + (y - a) H if H lt H bull y if H gtmiddot H
This function is shown in Figure 2 for the three relamptions be eween y and a
IV THE D GRE OF COOEtRATION
The models depicted in Figure 2 are based on a very simple notion about
the relation between concentration and the cooperationnon-cooperation charac-
teristic They use what is essentilly a step function relating concentra-
Figure
I -
J H H
Page l la
2
v v v
a
I J I y a bull Y y ---i
I I I I I I II I
H 10 H H 10 H 1 0
bullbull
bull bull
Page 12
tion and the extent or degree of coo peration as shown in Figure 3 Using
o as a symbol for the degree of cooperation eq uation (23) can be rewritten
as
bull(24) v (1 - o) vo + ovl bull (1 o) [a+ - a) H ] + oy- (y
vhere
(25) o bull o(H) bull 0 if B lt B
Hbull 1 if B gt
The specific relation between concentration (H) and the deg ree of
cooperation ( o) shown in Figure 3 is more restrictive than necessary and
bas less appeal an one which shows a smooth increase of o from 0 to 1 as
B increases fro to 1 Such a smooth function is shown in Figure 4 he
degree of cooper ion is low until H gets close to H increases greatly as
B goes through R attaining a high level for an H larger than B but still
close to it and then goes to 1 as B goes to 1 More formally
(26) 0 bull o(H)
bull 0 if H bull 0
bull 1 if H bull 1
do gt (IdB
d2o- gt 0 if H B dBZ
0 if H H
lt 0 if R gt H-
If (26) is now substituted for (25) in (24) a corresponding smoothing
of the functions shown in Figure 2 is accomplished he smooth functions
Figure 3
I
10
H 10 H
Page 12a
Figure
Page 12b
4
10
H 10 H
l
(27)
Cl - o) o
[ (1 -
[ (1
t y ij z
j
are given in Figure 5
Page 13
The equation in (24) was constructed as a hybrid of two results derived
from first-order equations for the non-cooperative and cooperative models
It may also be constructed directly by using l bull (1 - o) + on as thei i
th 2objective function for the i-- firm Treating o as a constant the first
order equation is
0 an aa
i middot
n i
Using equations (3) and (7) and setting --- bull 0 givesaai
(28) - 0
After collecting terms and converting to matrix notation we get
(29) - Cl - o) + 0 z-1vi yii i
N
j bulll
+ 0
+ o v1
d-1 r z z
-(30) v Cl - o) rd
-
o) r +d
d r J _ -1- d z
and middot
d) r + 0 rjd
l r z
v - z --(31) zv
- (1 0- +d) t r z- d-
If we now impose the simplified demand structure by substituting (11)
we get
- (1 - o) a +( 32) + oy and
Figure
l
-y
v
Y
H
r
I I
v
a
y
H 10 H
v
a
bulla
HH 10
bullbull
(33)
i
Page 14
v - (1 - o) [ a l + (y - a) ] The corresponding equation for Vis given in (24)
If the relation between industry advertising intensity (V) and
concentration (H) is non - linear via the impact of concentration
on the degree of cooperation does the relation take the form of a
auadratic as some other investigators have proposed (Greer ( 1971 ) Martin (1979) Strickland amp Weiss (1976) ) That question can
be approached from two perspectives - conceptual and empirical In
the next section I will try to apply some statistical material to it
here I want to explore the question in the context of the models
developed above
Referring to equation (24) taking the deriviative with respect
to H and letting bull (H) gives
bull( 3 3a ) 3V (y - a) (1 - o) + (1 - H) 3H dH
Given that o and H each are bounded by 0 and 1 and that o is an in-
creasing function of H the term in the brackets in non-negative The
sign of the deriviative then is the same as the sign of (y - a) If the industry effect is dominant (y gt a) then advertising in-
tensity will increase with concentration If the two effects are equal
then advertising intensity will not vbullry with concentration and if the
brand-switching e ffect is larger advertising intensity will decrease
with concentration
For given advertising elasticity parameters then it is not pos-
sible to show an increase in A S up to some H and the n a decrease
he relation may be non - linear but it is monotonic over the permis-
sible range of H
Page 15
One possibility is to introduce a relation between (y - a) and H
If industries characterized as dominated by the brand-switching effect
also were more concentrated then a cross industry comparison would
show an inverted U shape relation between AS and H I know of no
support for such an association between ( y - a) and H
V ERROR SPECUCATION
So far I have implicitly assumed that there are no errors in the
model I now assume that the error
thvariable in the equation for the i- firm is additive that its mean is zero
that its variance is inversely proportionate to its sales and that all error
term covarlances are zero hat is
34) v bull (1 - o) [ a + (y - a) z ] + oy + u i i i
- - -
In vector notation
(35) bull (1 - o) [ at + lt - a) J +
E() bull 0 Cov (u) -
0
If we now ca lculate V bull bull we get
oy + U bull
(36) V bull (1 - o) la + (y - a)R J +
U bull E(U) bull 0 Var (U) bull
P age 16
The error term specifi c ation which follows Hall
wi th the ratio of profits
6 heir analysis seems
as wellbull particularly given
and Weiss was
applied by them to equa tions to equity or
asse ts as dependen t variables to carry the same
weight in the present context the substantial
evidence presented by marketing analysis on the tendency of company de-
cision makers t o use the advertising to sales ratio as the decision
7variable
The zero covariance assumption on the other hand1 is particularly
troublesome Within a probably not true since
many even ts outside the
single industry it is
industry would have
industries
similar effec ts on all the
8firms in the industry Between it is probably not true either
since many companies produce in more than one indus try and even ts in the
firm would tend to effect all of its activities The extension of the
model to allow for non-zero covariances is certainly called for
--- -
Page Ll
Athird characteristic of this specification also deserves some comment
only on equation is shown for the firm or industry Several haveinvestigators
included an advertising equation ina larger system of equations (Comanor and
Wilson (1974) (1979) and Weiss (1976) so in myMartin Strickland Idid
dissertation1 Ihave ewo defenses for presenting only single equation results
here One is the suggestion by Comanor Wilson that simultaneousand equation
verybias may not be important in this context he second is that Iam will-
ing to see the model extended to include a profitability equation and I intend
to move that direction in the near futurein
Onthe inclusion of aconcentration equation on the other hand I amskep-
tical he problem isthat in some fundamental sense equations in models of
industrial organization should be oriented to the decision-making contexts of firms
Ifeach firm in anindustry with Nfirms has only one decision variable then
there can be only Nindependent equations which are based on the first-order
optimization equations It not show large ofis difficult of course to a number
functional relations which follow from the first-order conditions point here he
isthat only Nof them can be independent
Consider for illustrative purposes only a situation in wich teh N
itsquantities say q are set exogeneously and each firm determines advertising1
of and level say ai where a is the true optimal value a a bulla
1 1 i i + i
Ifp a) then the Nq
s and the Na s will determine N p s say pii bullp
i(l i 1 i
Now given Eand 9 we can determine both the vector s andthe vector v since
- bull til
ands We can also determine S s z bull s-1 s-ii
v and we
industry concentration isHbull (alternatively
V and Hare functions of the same
pendent Furthermore if there are
VM) and (H1 H2 bull bull bull )contain
VI STATISTI CAL APPLICATIONS
Page 18
Given the observed values of might be tempted to conclude that we
have a two equation system in which v and z are jointly determined That would i i
be inappropriate however since the 2N random variables (v ) are determined by
onlt N basic random variables t The variance-covariance matrix for ( ) must
be singular bull
The extension of this observation to industry level variables is straightshy
forward Since observed industry advertising intensity is V bull l v and observed 4
C4 bull z ) it follows thati l i
set of random variables t and cannot be indeshy-
M industries then the variables (V1 v2 at most M independent random variables
Given the assumptions made above y a and o are constant within a
given industry That being the case (34) can be written as
(37)
s1 bull (1 - o) (y - a)
If and
So + +
(1 - o) a + oy
81 are esti mated for an industry using a suitable statisticalSo procedure giving and 81 an estimate of y is then available The sumSo
Intra-industry analysis
1974 of Business Re ort -
categories are identified
Page 19
of So and S1 is y so y bull So + S1 is the estimate of y
If o is assumed to be bounded by zero and one the sign of 61 is also
the sign of y - a The estimate of So is of cou se an estimate of
(1 - o ) a + oy Moreove if y gt a then a is less than 6o- (1 - B) a+ 0(
Cor esponding results would
hold if y bull a or y lt a
If 61 gt 0 then we would estimate that a lt 6o
bull
For purposes of estimating this model 32 manufacturing industry categories
were selected The data are fram the Line of Business forms filed with
the Federal Trade Commission they are discussed in detail in the Annual Line
which is available from the FTC The 32 industry
in Table 1
Simple regressions for equ ation (37) were run for each of the 32 industries
with data for each of the firms which filed in the industry constituting an ob-
servation The results are given in Table 2 lines lA 2A
1974
the intercept
For all the A
equations generalized least quares was used th both and market
share bein scaled bv the square root of sales
The 32 industries were taken from the larger set of all manufacturing industry
categories in the 1974 data files Two selection cr iteria were used thereLB
had to be at least ten reporting companies and the industry had to be a consumer
goods industry or a producer goods industry which is strongly associated with a
related consumer goods industry so that adve rtising by the firms in the producer
goods industry impacts on demand in the related industry (eg soft drink syrup
and soft drinks) There were 135 industries with at least ten companies - 103
producer goods and 32 consumer goods
273
274
TABLE 1 INDUSTRY CATEGORIES USED
FTC
COOE OESCRI P T I ON _
2001 MEAT PACKINGt SAUSAGE S ANO OTHER PREPARED
MEAT PROOUCTS
2004 DAIRY PRODUCTS EXC F LUID MILK
2005 CANNED SPECIALTIES
2001 FROZEN SPECIALTIES
OEHYORATFOt AND
I NC LU 0 ING PRE SERVE S
SOUP MIXESbull
SEASO lNGSt ANO
OTHER PET FOOJ
RAIN MILL PROOUCTSt
ANO PREPARED
RELATED PRODUCTS
PRODUCTS
Z008 CANNEOt ORIEOt PICKLED FRUITS
ANO VEGET ABLES JAMS JELLIES DEHYDRATED I G TA3t E
S40CES ANO SALAD ORESSINGS
2010 DOG CAT A O
2012 FlOUR t OTHER RICE
MILLING fiLENOE) FLOUR
2014 B EADt CAKE ANO
20113 CO FECTIONERY
2026 BOTTLED A O CANNED SOFT DRINKS 2021 FLAVORING EXTRACTS AND SYRUPS EC
2029 MISC FOOOS AND KINDRED PROJUCTS EXC
ROASTED COFFEE
2301 MENS AND BOYS SUITS AND COATS
2302 MENS ANO BOYS FURNISHINGS
230 3 kOMENS AND MISSES OUTER EAR
2551 HOUSEHOLD FURNITURE
2702 PERIODICALS
Pqe 19a 1
RELATED 1972
src cooe
2011 f 3
202tX2026
2032
203pound
2033t4t5
2047 2041t4t5
2051 2065 2086
2081
209XZ05
231
232
233
251 272
2703 BOOKS
2704 ISC PVeLISHING
2806 ORGANIC FlEERS
2807 DRUG$ ETHICAL
23234
PT 2 S 3
2808 ORUCSt PROPRIETARY
2809 PERFU ES t COSMETICS ANO OTHER TaiLET
PREPARATIONS
2810 SOAP AND OTHER CLEANING PREPARATIONS 2815
2901
PEST1CIDES AND AGR ICULTURAL
PE OLEUM REF I N ING CHEMICALS EC
3606 HOUSEHOLD COOKING EQUIPMENT
3612 HOUSEHOLD APPLIANC ES NEC INCLUJING ELECshy
rqlC HOUSE ARFS AND FANS ANO SE I G MACHINES
3617 RADIO AND T V RECE I VING SETS 3808 PHOTOG APHIC E UIP amp SUP LfESt EXC
3qo3
PHO TOCOPYING
SPOR TING A D E QU I P M ENT f SUP P L I E S ATHLETTC GOODS NEC
3904 DOLLS GAMESt TOYS AND CHLOqENtS VE ICLES
PT283
2844
234tX2844
287 9
291 3631
3634t6t9
36H
PT386l
3949
15
943
middot
l A6l f 2 R(GRESSION RESULTS
EQ I NO
NR CODE cu 121
lA 2001 A (
2A 2004 r (
3A 2005 a
-R
ClS amp31
16
17
1 z
I TER-CEPT bull t
ooo9t21At 00022311 oooeos uu
OOl49tCAI 0010 2 middot 000960CCI
o049tl1At o04t55C61
MARKET SH ARE bull 5 I
-010670Ctl
-017600181
-001050
-0211AO
-0077 0
AOVER-TIS INti
lbl
000()39 000072
000018 0 0024liAI
-000079
SHARf
4ampgtVf Q bull
111
000140
-000170
RSQ
COl
66 485 862
607 6 3 t ()2
blti6 633
flASTI-
CI T Y PH
-007271(1
000441
-u0288l
clASTSHU (t)Ef
bull 10
tto
-234
210
( OOlltOCCI -126520 000922 181 001790
4A 200 7 18 00590liAI -038670 4e16 -032110 lell u c
OOOS74e 000700 025940
00067018) 0021001AI -02l100CAI
5 26
920
5A 6 c
200 8 26 0 03Sl9CA I 0012 26181 o 024511(
-005540
-0 81420 0001961AI O()Ol90CAI ooll240
441 604 891
-0020 19 274
6A 2010 16 Oe04224CAI o359lOIAI 9oq ll4015dCAI 061 6 -middot 005435CAI 00011 8 1AI 917 c 00535ltAI -009850((1 OOOA51J CAI -001610161 971
7A 2012 001862 000010 351 00187) oo1 a oooq45 ono1o6 439 c 001109 -027050 Oe0067SIAI -OOll60C81 897
- middotmiddot- -
8A 201t ll 000757 Ol4140CCI 814 ott9oocet 095 8 oooqo6 C C J 000069(81 811 c ootl2lampBI -080910CCI Oe009lUBI -OOtlOOCCI 960I
I DIIJIT pound11 1V OTIUTVTfAll-1pound1t- A _ 00 R - qq r - Q
1jl)OQCD
0
-middot middot--
l
_ __ _
I
i
0 00141 c c
91)7
I
T ABLE 2 REG ESSIOH RESULTS CCONTI
EO NR
IND CODE
iR bull
cos
I TEq-
CEPT
t1ARKE T
SHA E
AOVER-
Tl SING
SHARpound-cent
AuVER
RSO ILAST1-
C I T Y ELAST
SH CPtf (lt Cl) bull 11 I 41 I 5I (61 171 I 8 I f9) c 10)
9A 2 0 1 8 19 002397181 -007640 367 -005 246 le46 6 000605 000550CAI 597
-000029 G14180 OOl654CAJ -024li01At
lOA 2026 12 0028901AI -OlZl90 -0114 0 1 1 1 5 759
Ll (
llA 2021
f3 c
12A lOZ9
B
(
13A 2301
B
c
14A 2302
B
c
154 201
n (
16A 2551
8
10
35
10
20
19
24
001897(()
00141 6CCI
Oll9lHCI
010604 middot middot
Oll083
OOJ11JCAt
002059141
001411CIH
-000412
000293(81
000187
001159 (3)
ooo6a4cct
o00792CCI
OOl40dCAt
000006
-000076
OOlltl6fA)
o o0 5 64CCt
-034570
-o1middot o3o
-1659 90
Ol51lO
-046190CAt
Ol5450CA)
-005l60
002940
-030060(()
-0 43l50CCt
0 33650
-019080
00004
bullJ00741CAI
-000069
000466
0001 51CAt
000947CAt
O OOI)70CAt OOl750CA J
000175CCI
000671CAI
OOll5) f A t 002207CBI
0004b7CA I
- 007 1 70
001760
-001410CAt
- 0 16200CAt
-004210
- 1 207 70
705
96 0
542
466 805
460
614
921
832
9 61
996
424
550
9 15
56 1 651
884
-006114
019 037
03504lCA)
oo409Q
-141644((
491 -0 17664
607
871
lell
084
101
072
103
108 11IllOQfl)-- l
-037880fCJ OOl6l4CAJ -040l70fCI 0()
c
I
I
bull 9
9t
middot middot middotbullmiddot bull lraquo
TABLE 2 REGRE SION RESUlTS ICONTt
bull
EO I NO tHl I filER- f4A_K T AOVER- sbulluRE ttSQ flA Jl- iLAS T NR CODE cos CEPT SHARE JISING AOVfR bull (JTY SHR CiJCF
110Cl I f amp Itt 15J tbt bull 1c u caa
llA 2702 ll o079 6 81 Bt -019110 S6s -OlltO 111
a 004679 ooooo1 t69
( oOl26o -015160 001560 -022690 ltJr()I
ldA 270 3 19 0065lt -OIt2l60 220 -Ol57a2 11 8 e o 0 71t6IA t ol08 75tAt ( OOl983CiU -060llOC() OOl7191A t -O lltdOC A) middot 89
lqA 2704 10 005082 -Oil060 22b -l)i1977 lelO 6 -000080 oo2772C4 t 910 c 000907 -011 200 Oe04105CA t -046960 995
20A 2d06 1 II
O OlltStiAt -001120 837 00014 -))th O01380amp At -oooo 11 821 ( OOl526CAt -026700181 0002 5UAt 000530 969
2lA 2807 28 OOt9lt9CAt -00891 0 516 -0 03962 225I8 002095CCt OOOl lliAt 690 ( o02965CA t -056650((1 OOOt50CAJ -003l70ttl 911I
I 22A 2808 15 ot3l06 119730 655 1130) 090
0 - 007526CC t Oe003961At 835- middot-
c I oo7523 -067210 001l 2tA t -0091t80CCt 951
2lA 2809 21 Oe2067lCAt -1218101At 71 2 -10095((1 121 ttlIaOQ8 00797418) OOOI8lltt 686 Ill( 0092lU A t -Olltl70Ct t 000605CAt -OOit8701At 951 _
---- -
32 009506CAt
0 bull 06 981 I A J 006901 cA bull
00 24A 2810 -010960 5t 5 -OOltSO 7 56
8 -000011 s11 c -5811701At oooaoatAJ oot520IAI 888
-- 37 --- middot - -- -1 ) l lbull t5JJ r-l bullhmiddot1 --- - --
l5A
26A
27A
c
2RA
c
2 9A
31A
llA
24
13
597
Ci)55
14
l 6h
RSQ ELASTI- (lASTCOE f-
TABLE 2
EQ I NO NR CODE IU
REGRESSION RESULTS CCONTa
NR INTER- MARKET
(10S CEPT SHARE AOVER- SHARfcent
TIS lNG AOVER
) CITY StiR
c It) bull 5 co a I 1) bull 0 bull 9 c l 0 bull( 2 1
2815 16 003019 -009990 275 -o069bmiddotJ
6 001446 000224 282 c 002178 -054630 o02221CCI -0075 00 646
2901 29 00021t2 CAI 0)0520 592 0007 6 1 068
0 00105C8 t 000012141 60)g( OOOlJ51AJ -00126018) ooou271At -000u80CA) 974
36oa 1 3 003296(4) -oca29o -oo-q94
001755 000203 lt91t 002499((1 -04l820CCI OOl374CAI 0068)0 946
3612 0033281CI 011520 195 01684) oso
00 30 42C OI 00012 2 424 oo579ltAt -2 7l7801At 002 100IAI -003220 819
8 3617 btel 011789 081002259 009530
001448 oOOl OitCCI 732 OOl7 22CAI -0 46160CAI 0004961AI -OOll40181 989c
30A 3808 14 0021t 3 6fAI -000110 628 o023291C) -005 6 middot- _ middot- 0bull 02 04 1 IC I ooooo9 637 ( 001584((1 -0161)1)0 00021 91AI -0 00060
390) 14 002790CCt 006590 h8 t 0 0938 4 070
B OOl89bCCJ 000144 74
0021961(1 -OItlOlO OOl323CAI -004800 915
B
middot--- - middot 4 - middot- - dfll3904 12 004563181 0)8300 879 0428651(1 089 OQIll004 277CA I ooozstcea 919
c 0041)3181 -0-11250 0011881(1 -006270 972 0Ill
- J - bull bull bull bullbullbull middot - - middot - middot gt middotbull L)ofbullflbull middot middotfJltl -tn- middotmiddotmiddotmiddoto +f middot t middot bull
c
is
positive
2
Page 0 determineThe consumer goodproducer good split is sometimes difficult to
Sollle support for the split used in this paper is a comparison of goodness - of - fit
distributions of a
industries
set
advertising are
that industry
a sales increase of
the
highest aD10U3 those
measures for the two groups of industries Table 3 contains
measures for the 32- consumer goods industries and the 103 producer goods
2 Median R is 5 7 for the consumer goods set and 40 for the producer goods
Estimates - of y the elasticity of demand with respect to
given in column of Table 2 The hypothesis that y bull 0 with
alternative was applied for each ittdustry As shown in the
could not be rtjected for 24 of the 32 industries Of the
and 10 are positive For the remalnifg eight three are
and five are significantly positive
Of the threertegative elasetcities the largest (in absolute value)
cosmetics (code 2809) With all elasticity of -1 01 a one
industry advertising would gwerate a decrease in industry sales
more than one percetu
At the other extrema is toys and games (code 39 04) For
a one percent ittcrease in industry advertising would lead to
four-tenths of a pampreentmiddot Though not significantly different from zero
elasticity estimatbull for pt-bpuetary drugs (code 2808)
which are
Both drug industries have
to sales rafios (211 pe1 cent for 28 08 and 13 4 per cent
other hand flaIJo ing ext1ac s (mainly soft-drink SYUlS) also
is for
percent increase in
of slightly
a middottwo - tailed
table that hypothesis
24 14 are negative
significantly negative
9
the codetics and p-ropriacuy very high advertising
for 28 09) bull On the
has a high ind UY
advenisiltg t() sales ratio ( 8 0 per cent) bull but its estimated elasticity is very
small and not signifieantly different uom zero
middot It doe$ not seem unusual to find many elasticities near zero with a few
negative and a few positive If aggregate consumption is insensitive to aggregate
advertising as seems likely ld then what one industry gains must be lost by
others
Page 21
elasticity of demand does
(10) we can
if t 0 then
Finally a determination concerning the industry
not fix the individual firms elasticity Going back to equation
write y bull (1 - ) a + z y Since a gt 0 it follows thatii i i
Y gt 0 If Ylt 0 however y s sign is indeterminateij ij
Turning next to the coefficient of market share 20 of them are negative and
12 are positive with three of each sign being significantly different from zero
The largest positive value which is significant is for pet food (code 2010) and
a close second is mens amp boys suits and coats (code 23 01) The largest signifi
cant negative value is for cosmetics (code 28 09)
In the model some relations among y a 0 and 61 are implied If middota is
negative (1- y) is greater than one Since 61Y bull (1- a) (1 - ) 61Y is
pCsitive but it will be less than one if 0 is large enough Seventeen of the A
ys are negative in each of the seventeen industries 61 is also negative and les
1 (algebraicly) than y so 61Y is greater than one his is not evidence that the
degree of cooperation is low or zero in any industry of course A high o togethegt
with a negative y which is small (in absolute terms) relative to a can give a
value of 61Y greater than one On the other hand if 61Y were less than one
o would have to be greater than zero
For y gt 0 two conditions may hold If y lt a it follows that 61 lt 0 since
(y - a ) lt 0 It also follows that 61Y lt 0 regardless of the value of o bull
In addition the relation is an if-and-only-if one ie if 61Y is negative
then y lt a Three industries have negative values for 61y dairy products
except milk (code 20 04) synthetic fibers (code 2806) and photographic equipment
(code 3808) In all three cases y is quite small so a does not have to be very
large to give a negative value to al
If y gt a the model shows that 61-- gt 0 whatever the value of 0 bull There are
12 cases where both y and 61 are positive with 61 ranging from near zero
(grain milling code 20 12) to 1 20 ( proprietary drugs code 2808) Several
coeffiei-en_rs
--
15
o f to zero
Tab le 3 - Consumer Goods
Indus tr ies
2Distributions of R and Producer Goods
Page 20a
Consumer Producer
Goods Goods
0 - 9 9 0 3
10 - 19 9 0 1 1
20 - 29 9 3 20
30 - 39 9 3 17
40 - 49 9 5 16
50 - 59 9 8
69 9 6 Ii 1060
--f 670 79 9 2
80 - 89 9 4 4
90 - 100 1 1
Totals 32 10 3
Notes
2is used R i s cal-
culated as F F + ((N - K- 1 ) K ] 1 b_ F is the F
statistic Lor the hypothesis that all o f the non - intercept terms are simul taneously equal
1
2
--
-- Since generalized least squ-area regress ion
Page 12
o f these cases have values o f 6 1 Y rea ter than 0 8 (pet food (20 10) bread amp cakes
(20 14 ) miscellaneous processed foods (20 29 ) men s and boys suit s and coats
(23 01 ) p rop riet ary drugs (28 08 ) elect ric housewares (36 1 2) radio and
IV sets 17 and games and toys (39 For be than(36 ) 04)) 6 1 Y to greater 0 8
both o and cry mus t b e less than 0 2 That is in these industries the results
shown in Tab le 2 would hold only with very strong dominance of marke t demand
effects over brand swit ching e ffects and a very low degree o f cooperation
Using the concentration rat ios given in Table 4 the average concentration
ratio for the eight indust ries is 38 1 For the remaining 24 industries it is
43 2 This result would lend some support t o the proposition that concentrat ion
and cooperation are related if the group of eight industries are identi fied
as having low degrees of cooperation
One f inal note on model predict ions S ince both cr and o are assumed to
be non-negative 60 should also be non-negat ive In only one o f the 32 cas es
is 60 negative and it is not signif icant bull
cor pJ
7 1
7 0 4
1 4 7 5 1
)
7 3 z a o a t s 1 1 3 3 5 ( i V Z 8 tJ1 2 1 - 1 1 0 7 3 5 c middot v 2 8 1 0 3 2 -0 0 1 4 3 0 C 1 V 9 8 3 2 8 1 5 1 6 - 0 0 6 t1 6 9 C _ V 1 11 0 2 9 0 1 29 0 00 7 6 1 C l k V 7 3 1 3 6 08 1 3 - o n 4 9 9 4 3 3 3 6 1 2 2 4 0 1 6 4 3
3 1 7 1 3 0 1 1 7 d 7 8 3 3 8 0 8 0 0 2 3 i 2 3 0 6
P 4
3 9 0 3 1 4
1 2
0 09 3 8 4 39 1)1 0 4 l 8 6 5
T A LE 4 e I NCU S T R Y C A T E G O R Y
F TC NOM e tOOE ass
C 1 I c 2 )
2 0 0 1 1 6 2 0 0r 1 7 zo o l Z 2 0 0 7 1 8
z o o 2 6
z o t a 1 6
2 1) 1 2 1 5 2 0 1 4 1 1 2 0 1 a 1 9 2 0 26 1 2 2 0 2 7 1 0 2 0 29 - 3 5
2 3 0 1 10
2 3 02 z o
2 3 03 1 9
2 5 5 1 2 4 Z 1 0 Z 1 1 2 0 3 1 9 2 7 4 1 0 z a o 1 3
2 A 0 7 z s
E LA S T I -C I T Y
c 3 )
- 0 09 7 2 7
0 00 449
- 0 0 2 8 9 3
- 0 3 2 7 70 - 0 OZ il I q
o 4 u t s a 0 0 1 3 7 3
o 1 4 o o-0 0 5 2 4o -0 1 94 0 1
-C l 6 l l 4 0 1 JJ 3 7 0 3 t 0 4 2 0 0 4 0 9
- 0 4 1 8 44 - O l 7 t ) 4 -ll 7 1 4 0 b -0 3 5 7 8 Z
- 0 1 9 7 7 0 1)0 3 3 4
-o n 9 2
IJ A T A
CONV
ON-C ONV
( 4 )
t 1NV
C 1t-V C NV c ) v C - JNV C rlNV c n v c a v C ON V ( i) l V
C CtV
C t1tV C IJNV
C 1 V
c 1 v
Pa ge 2La
HEQ F t Noe x
c 5 )
6 2q
6 89 1 48
6 1 4 ) q 3
1 9 7 1 9
q z t s 5 a e z
l l 9 1
4 o 7 1 2 0
c oqt 1 Z
4 4 )-q 1 z
5 5 9 1 0 64 1 3 3
n 3
Inter industry analysi
Page 23
Equation ( 3 6 ) may be used to fo rmulate a rough test o f the cooperat ion
vs e f ficiency controversy I f we rest ricted ourselves to some subset o f
indus tries where the relat ion be tween y and a i s the same we would be lookingbull
at a sample o f indus tries for which the dependence o f V on H and o is a
quadratic wi th no squared terms
I f we now sub s titute (26) into ( 36 ) we get
(38) v bull a + (y - a) R + (y - a) o (H) - (y - a) H o (H) + u
If it is only d if ferences in marke t shares that affect advertising inten-
s i ty levels i f concentration does not af fect cooperation and i f o bull 0
( 38 ) reduces t o
( 39 ) V bull a + (y - a) H + U
Let this b e the null hyp o thes is and note that V is a linear func-
tion o f H If y gt a and the null hyp othesis holds the expected relat ion
between V and H is as shown by the pos itively sloped straight line in Figure
6 b etween points A and C
In this context the alterna ive hypothesis the degree o f
cooperation ( o) is a positive function o f concentrat ion
is that
(H) I f the func-
tion is as described in (26) and Figure 4 the relation between V and H
would be as shown by the curved line connec t ing po ints A and C in Figure 6 One way to test the null hypothe s is against this al ternative is to test fo r
the hypothesized curvature in the function relating V and H
Fi gure
R
6
v
ID c
H 1 0
I
i I I
sfigned
clasJ s
in def ini g
effec t is the
I
Page middot2lf
One charac teristic of produc ts which may be useful
classes wi th respec t to significanc e of the brand-switching
conveniencenon-convenienc e distinc tion developed by Porter ( 1 97 )
Using his identif ication of IRS indus tries as a s tar t ing point
each of the 32 industries used in this s tudy to one of the two
The ass ignments are not ed in Tab le 4
Using the conveniencenon-convenienc e dis t inc tion the
divided into two sub-samples
sample was
Four regressions were then run bull ene for e ch
sub-sample one for the to tal sample and one for the total sample with
a dummy variable for convenience goods Each regression was run wi th LB s
and wi th indus tries as ob servations The results are given in Tab le 5
For the LB level equations ( 1A - 1D) equation(37) was used ome
rearranging of terms gives v bull y - ( 1-o ) (y - a) ( 1-z ) + ui i i
bull y - 6 1 ( 1 - z ) + ui When observations are pooled across indus trif s yi
is an industry variable In these regressions the values o f y es t imated
in the first stage were used Since industries have been grouped so as
to reduce differences in (y - a) that term is t reated as a cons tant
within each group Finally H the Herfindahl index has been subst ituted
for o the degree of cooperation The resultant equation is linear
y ( 1 - z ) and H x ( l - z ) This is the equation for whi ch results arei i
reported in Table 5 In all regress ions all variables were weighted by the
square roo t of sales
in
raquo
- 34
3 1
i l T 4 8LE S R E G E S S I ON R E SUL T S i X T f N S I ON S T O T Y PE Of
AND L E V E L OF C ONC E kT R A T I O
E O O tt S
tP bull T Y P E NR C ONV O B S tO N-CONV
E L A S T I shy
T I C J T Y
O E - t-41( T SliA R E
H( fl F
1 1 - K O x middot K O X E R F
- H S a 1 1 - H S t 1 8 1 1 9 1
R S O
1 1 0 1
q z
t - bull 9 f 2 1
Z t P D l 6 fC ONV - 1 9 1 s s c 2 1
e - l O I e bull u q a
2 3 1 0 1 1 Nl N- t O V 5 7 1 ( 1 2 19 1 6 1 - J l S 2 5 4
l l l l t z q o a 1 - J o v z I
l ( I NO l l A l l ol u l t o o c bull J u l l
d S
c 3 4 t t r a l
bull Z D 1 0 l l 4 l l 2 3 -gt 2 l l 2 0 - 4 lS 5 4 l C 4 1 o e 6 t l b b 5 1 1 0 1 1 1 c - 1 9 5 1 t d S 2 1 tV
OQ10 T-R A T I O S A E G I VE N l P A RE N T H E S E S l E V E L S f S I G 1 F l C ANC E A - 99 5 B - q 9 t - s NS J UC E NO I Nl E R C ( P T I S U (j f middot) I N f Hf t OUA TIOI S 6 E f 3R E A O J U S T H E N T F OR 1-t E T E R O SIC E O A S T f C I T V pIraquo
1
middot
I
R SO I S C A L C UL A T E D A S f I f bull 1 1 bull 1lt 1 K t t- E E N I S THf U A1 B E R O f fJe S E fi V A T I O S F t S HtE F - S T A T I S
I N DE P E ND E N T y II
T I C F O T H E U Y P O HIE S I S f A l l C JE F F I C H N T A R E QU A L HI l f RO A f40 K I S T H f UUM 8 E R OF Q ( A fL f S
I l l
1 A
1 6
t C
f 1 r R I l l
1 2 1
L Jt
L B
l q
L 1
on s T Y P E bull 2 )
P OUC T
X
H S I l l
l O l A I bull l 0 7 1
O O b O J
1 2 00 1
1 b 2 1 C I l l Z U
l l S b 1 -1 1 0 1
rlt F X l l - fi C f l
bull 1 I
l l
- ()4 2 3 l f H
1 - 2 7 9 1
K O X
1 1 - ti E R F I
l d l
S4 l i A I 2 l b
1 3 00
K O X -t f R F t S J
I 1 - HER F t
)
60 9
z o bull
0
t l l
3 8 8
1 7 6
5 64
64
N f bull
tl S bull
I l l
I 4 1
t ONV
NON-t ONV
A l l
A ll
c o v NO- ( ONV
1 4 1
1 5 1
- l S b
c - 1 1 2 1
5 8 2 C A I 1 6 7 0 1
middot I l 9 H
l 1 il
1 4 1 6 1
E L A C) T J shy
l t C I T Y
I S I
l b l
- 0 2 0
1 - 1 4 9 1
O Z 3 2 1 A I C b 6 b l
0 0 9 2 -J
1 1 b 4 1
o z t t t l 1 2 2 1 1
oru - Ht tt F
1 gt X bull b I
Page 25 bull
Mean val -as ior relevan t variables for the two sub-samples are
Conv Non-Conv
Adv Sales 4 7 2 3
iit Share 2 7 3 2
Ela cicity - 05 3 02 2
Herf indahl 7 3 1 860
That the dist nct ion between coft9enienee goods and non - convenience
goods mat ters is supported not only by these means but also by the regression
results Equat ions lA and lB are different in almost every respec t Only the
concentrat ionmarket share int eract ion term seems to mat ter for convenience
good s Jus t he opposite holds for non - convenience goods both elasticity1
and market sl e are highly significant but the concentrat ionmarket share L
int erac t ion t=rm is not
Given that one minus market share is used these equ3 tions the coeffici t
of market share is the negative of what is sh in col (6) of Table 5 For the
regress ion for non - convenience LB s reported in equation lB then LB s with larger
market shares have lower advertising to sales ratios
On the question of the impact of the degree of cooperation the story is mixed
For convenienc good s the coefficient is pos itive as equation (31) predicts when
there is a po sitive degree of cooperat ion Fo r the non-convenience goods LB s middot
concentration has no effec t
Aggregat ion t o the indus try level and then rer unning the equations $ives the
results shown in (2a-2d) of Tab le 5 As expected R2 goes up and levels of significanc e
fmiddotJr individual coefficients gJ down In add ition the coeffic ient of H ( l -H) inmiddot
quat ion 2A is no t significant though it is po s i tive Since 2A is the indus try level
I
--
all but one of them and that one is not significant For 18 of the 31
wo uld have on its adve rtising to sal es ratio
cognate of LA the variable H ( l-H) pl ays the same role in 2A as H ( l -z 1) p lays
in LA
Vconomies of scale in advert ising
The literature on advert ing contains a number of definitionbull o f economies
of scale and of findings eone rning its presence and magnitude Before I loo k
a t the data with the hope o f mmenting about scale effect s then I wan t to
define the term as I use it
By economies of scale I mean a decrease in the number o f units of adver-
tising which are needed to generate a unit of sales as advert ising is increased
Given the definition of the firm s own elast icity of demand with respect to
advertising (see the text foll wing equation (1) ) the technical definition
is that rii gt 1 Diseconomies of scale are def ined symmetrically i e
Yu lt 1
ith a cro s s - section oi observations on advertising intensity
(v bull a s1) and onmiddot advertising (a ) we may regress the former on the latter 1 i i
The results of such regress ions for the 32 industries are given in Table 2
lines lB ZB bull bull bull 32B For the 32 eases the regression coefficient is positive it
positive coefficients the coefficient i s signif icant
The que tion now is whether these results ar e evidence of pervasive dis-
economies of scale in advertislng I think no t Given my as sumption that the
data I observe are equilibrium results I may ri ghtly conclud e that almost every-
where high levels of advert ising are as sociated in equilibrium with high ratios
of advertising to sales I may no t conclud e anything however about the
impact that a move by some firm away from its equilibrium level o f advertising
middot- Page 27
Given the assumed relation in equation ( 37 ) there is a corre spondshy
ing equilibrium equation relating v and ai The equation is non - linear i
and I have not found a simp le way to characterize it It is po ssib le to
det ermi ne its deriviative with respect to a b y takin$ tne de riviative of ( 3 7 ) i
When that is done the result is
v - 1(40) i bull sl a a z ( 1-z )i i ira-l
Sinc e all the other terms in (40) are non-negative the sign dep ends only
on the sign of 6 1 which is in turn de pend ent only on the sign of (y - a)
That the signs of the coefficients of a in equations lB 2B bull bull bull 32B arei
not always equal to the signs of the coefficients of z in LA 2A bull bull bull 32A i
is still a mystery to me and something to be explored
bull
Summary
My purpose in writing this pa per wa s t o attempt to integrate explicit micro-
economic mode lling reasoned econome tric specification of error terms and high
quality data to exp lore some questions about adver tising by large firms
al quite important improvements could be made in the first two of those areas
and much data ma ssag ing is yet to be done in the third
Concerning the subst an ce of what I have done I think six things are im-
Page 2 8
Severshy
V and Conclusions
portant
1 Explicit modeling is worth the ef for t if for no other reas on than
that it provides a basis for choosing fr om a varie ty of funct ional
forms
2 The same is true for error specif ica t ion I did in fact look at scatshy
ter diagrams for all 32 industry categor ies and they vir tually all show
the het eroscedast icity which I as sumed and for which I correc ted
3 The predicted relation be tween advertis ing in tensity and market share
shows up clearly in on ly 25 of the case s examined I have ye t to
explore why that may b e the cas e
4 Some evidence concerning the presence and impact of coopera tion was
produced but it is not clear ly pervasive
5 The dis tinction between convenience goods and non-c onvenie nce goods is
unambiguously a good one
6 Virtually no evidence concerning econ omies of scale in adver tis ing can
be gleaned fr om this study given its assumpt ions
More work is called for on many of these issues my study seems to have
raised more questions than it has answer ed This is pr obab lly due to the richshy
ne ss of the data source since I had the oppor tunity to address several issue s
at once
- -- - Q --
Foo tnotes
Manager Line o f Business Program Federal Trade Commission The author depended heavily on several staf f members of the LB Pro gram for statistical and clerical support in particular Joe Cholka George Pascoe and Haro lene Jenkins Helpful comments were given b y Richard Caves Dennis Mueller Michael Porter and F M Scherer When the paper was in its early stages o f development both S teve Garber and Jon Rasmussen former economists in the LB Program provided eery useful cri tiques The views expressed here are my own o f course
1 The basic exp licit model of the role of advertising is Dorfman and S teiner ( 1954 ) and the models in this paper may be seen as ano ther extens ion of their work
2 For a review o f models of oligopoly which treat prices and quantities as relevant variables see Shubik ( 1 959) Extens ions to take advertis ing int o account are fairly s traightforward
3 Dorfman and S teiner dealt with a price-setting firm For a monopolis t inverting the demand relation to show Pbull P (Q A) _instead o f Q- (P A) has no effect on the results o f the analysis For oligopolists who face a sys tem of demand relations inversion does have some impac t on the results The dif ferences are not trivial nonetheless they will no t be explored in this study
4 Shubik ( 1 959) p
5 From a formal mathematical point o f view the function is similar to a1 i Legrangian funct ion of the form - r rr
L bull 11 + A ( j - (i i i+j
where the indicates some fixed level of the profits industry Following the same logic which resul ts in and income constraint as the marginal utility of the of the indus tr in the obj ective of firm i
6 Hall amp Weiss ( 1 967) p 323
7 See Schmallensee ( 1 97 2 ) Ch 2 pp 1 6-47
8 On this poin t see Imel amp Helmberger ( 1 9 7 1 )
9 Long ( 1 970) p 130 - 1 39
1 0 See Schmallensee ( 1 972) Ch 3 pp 48-8 7
rrj i+j ) of the rest of the
the identification profits of the rest
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980
1
d
-
I
I
(11)
--
y
I -f
l J- a)
a) l zt
Substitutin g into (13) gives
i [ ai + (y(14) -I- bull S A
bull az + (y -
Pa ge6
I I II
bull
since l bull 1
Under these assumpt ons then y is the elasticity of industry sales
(and average industry p ice) with respect to industry advertisin g Note
that if y bull 0 so that there is no industry sales effect there will still
be positive advertising if there are brand-switching effects
Turning now to the brand-switching ef fect observe that
(15)
Noting that
-1 deg -1d z -z -
-
0 a
Page 7
(15) may be written as
0 (AA)
lx 2x 1x
ith
1y
i
First assume no
that if there are
will be no changes in market shares
- 0a bull
Page 8
Equation (17) is a general statement of the relation betbulleen changes in
advertising levels and changes in market shares which would acompany them It
may be used to make some general observations about brand-switching effects
industry effect eg y bull r t bull 0 and assume further
equal percentage changes in all advertising le =ls then there
Under these further assumptions
0l and dz -1 bull (iA) ltr assumption the second term on the right hand side of this expression is
- By
zero
necessary for rt bull 0Consequently for market shares to be unaffected it is
when y bull r l bull 0 This requirement is satisfied by equation (11)
Secondly consider effects on market shares if one firm inc ases its abull ar-
tising and others do not As a general assumption we require smiddot ch a change o
lead to a decrease in the shares of all the other firms hat is let
0 0aiai gt 0 and a a bull 0 for j + i For any matrix X with N rows and its trans-j j
pose X let X
row of X
bull middotmiddotmiddot where is a column vector which contains the
elements of the Under the assumptions about the iis given just
d-1 0 0above it follows that r becomes (aiai) iimiddot ehFor jzj lt 0 under these conditions then it follows that the j- element
N of (I- l ) lt 0 so middotij - il lt 0 or yij - yik lt 0 If this
kbulll N condition holds for all Y ij where j + i then it follows that yii - z
k Y gta
k- 1 ik since the th firms share must increase to offset the decreases in all other
shares
ei ith
1v ( ) h i
7urning
where is the
leads to z
to the r matrix given in (11) we note that 1 = cr + (y - cr) i i zi elementary vector with 1 in the position and 0 elsewhere This
bull cr + (y - cr) zi Since bull required is thatcr at- zY wYij (y - cr)zi - cr + (y - cr) zilt 0 or cr gt 0
These observations may be summarized by substituting (11) into (17) bull
giving
(18)
i s
d -1 zz -
-
-
+ (y - cr) tz
d -1 0cr(I - l ) a a bull
cr 1 z - (y - cr) 0 a
o
If ai increases and the other a s are const t cr must be positive to havej
a positive effect on zi A positive cr also bull plies a negative ffect of aJ
f all a s in-on zi other a s (including a ) fixed j
eluding a1 change by the same percentage zi remains unchanged Finally k i
if cr bull 0 there is no brand-switching effect
For a to have no effect on p requires yij bull 0 From (10) thisj i
amounts to y bull cr that is the industry demand elasticity and the brandshyswitching elasticity are equal If y gt cr gt 0 the industry elasticity yij doninates If y lt cr y lt 0 the brand-switching elasticity dominates ij
making
Page 10
Substituting ( 1 1) now in (6) the Cournot case and in (9) the
Chamberlin ease gives
-(19) Cournot vO a + (y - a) z
Chamberli n vl - y lmiddot th
The 1- element is then
-(20) Cournot vO a+ (y - a) zi i
-Chamberlin vl yi
In (19) and (20) the superscript 0 is used to denote total
independence of decision and the superscript l is used to denote
total interdependence of decision making
The relation between v and vf depends solely on t relation between
y and a f y gt a so that the industry effect dominat 3 vf v with
equality in the trirlal case of z bull 1 Cooperative irms would have ai
higher advertising intensity than non-cooperative fi
If there is equality between the industry elasticity and the
y Whe therbrand-switching elasticity or y bull there is cooperation among the firms has no effect on advertising inten-sity And if brand-switching is more important with y lt a e get
vi lt v cooperative firms will have a lower advertising intensity than
non-cooperative firms The three situations are depicted in Figure L
Given the results for the firm level variables calculation of the
corresponding re sults for the industry is s-traightforward Let
B bull t zf bull z z be the Herfindahl index of concen tration and notethat
z - bull l z bull 1 and that V bull A S bull a i 5 bull (siIS) (asi)i
bull middot Then for the general case from (6) and (9)
bull
Page lOa
y lt1 bull y y
10
f
V0 bull z [ a1 + (y - cr) J bull a + (y - a) H vl bull Y bull y
11Page
( 2 1) vo bull z d -1 r bull r z bull z- d d- i )11
l -1 N N v bull z d r z bull r z bull l y z - ij jibulll jbulll
and for the simplified case from (19)
(22
Since the forms of the equations in (22) are the same as in (20) all
of the analysis of (20) holdS for (22) except that His substituted for
middot z Advertising intensity will be higher for an industry with cooperativei
firms than for one with non-coope rative firms if the industry demand elas-
ticity dominates the brand-switchine lasticity If the ewo elasticities
are equal then so will be advertiSl intensity for the cooperative firm
and non-cooperative firm industries If brand-switching dominates the
cooperative firm industry will have lower advertising intensity
In the industry model context the possibility of a critical concen-
tration level may be easily introduced That is assume that at values
of R below R the firms are non-cooperative but that for alues equal to
or greater than H they are cooperative That is
(23) V bull a + (y - a) H if H lt H bull y if H gtmiddot H
This function is shown in Figure 2 for the three relamptions be eween y and a
IV THE D GRE OF COOEtRATION
The models depicted in Figure 2 are based on a very simple notion about
the relation between concentration and the cooperationnon-cooperation charac-
teristic They use what is essentilly a step function relating concentra-
Figure
I -
J H H
Page l la
2
v v v
a
I J I y a bull Y y ---i
I I I I I I II I
H 10 H H 10 H 1 0
bullbull
bull bull
Page 12
tion and the extent or degree of coo peration as shown in Figure 3 Using
o as a symbol for the degree of cooperation eq uation (23) can be rewritten
as
bull(24) v (1 - o) vo + ovl bull (1 o) [a+ - a) H ] + oy- (y
vhere
(25) o bull o(H) bull 0 if B lt B
Hbull 1 if B gt
The specific relation between concentration (H) and the deg ree of
cooperation ( o) shown in Figure 3 is more restrictive than necessary and
bas less appeal an one which shows a smooth increase of o from 0 to 1 as
B increases fro to 1 Such a smooth function is shown in Figure 4 he
degree of cooper ion is low until H gets close to H increases greatly as
B goes through R attaining a high level for an H larger than B but still
close to it and then goes to 1 as B goes to 1 More formally
(26) 0 bull o(H)
bull 0 if H bull 0
bull 1 if H bull 1
do gt (IdB
d2o- gt 0 if H B dBZ
0 if H H
lt 0 if R gt H-
If (26) is now substituted for (25) in (24) a corresponding smoothing
of the functions shown in Figure 2 is accomplished he smooth functions
Figure 3
I
10
H 10 H
Page 12a
Figure
Page 12b
4
10
H 10 H
l
(27)
Cl - o) o
[ (1 -
[ (1
t y ij z
j
are given in Figure 5
Page 13
The equation in (24) was constructed as a hybrid of two results derived
from first-order equations for the non-cooperative and cooperative models
It may also be constructed directly by using l bull (1 - o) + on as thei i
th 2objective function for the i-- firm Treating o as a constant the first
order equation is
0 an aa
i middot
n i
Using equations (3) and (7) and setting --- bull 0 givesaai
(28) - 0
After collecting terms and converting to matrix notation we get
(29) - Cl - o) + 0 z-1vi yii i
N
j bulll
+ 0
+ o v1
d-1 r z z
-(30) v Cl - o) rd
-
o) r +d
d r J _ -1- d z
and middot
d) r + 0 rjd
l r z
v - z --(31) zv
- (1 0- +d) t r z- d-
If we now impose the simplified demand structure by substituting (11)
we get
- (1 - o) a +( 32) + oy and
Figure
l
-y
v
Y
H
r
I I
v
a
y
H 10 H
v
a
bulla
HH 10
bullbull
(33)
i
Page 14
v - (1 - o) [ a l + (y - a) ] The corresponding equation for Vis given in (24)
If the relation between industry advertising intensity (V) and
concentration (H) is non - linear via the impact of concentration
on the degree of cooperation does the relation take the form of a
auadratic as some other investigators have proposed (Greer ( 1971 ) Martin (1979) Strickland amp Weiss (1976) ) That question can
be approached from two perspectives - conceptual and empirical In
the next section I will try to apply some statistical material to it
here I want to explore the question in the context of the models
developed above
Referring to equation (24) taking the deriviative with respect
to H and letting bull (H) gives
bull( 3 3a ) 3V (y - a) (1 - o) + (1 - H) 3H dH
Given that o and H each are bounded by 0 and 1 and that o is an in-
creasing function of H the term in the brackets in non-negative The
sign of the deriviative then is the same as the sign of (y - a) If the industry effect is dominant (y gt a) then advertising in-
tensity will increase with concentration If the two effects are equal
then advertising intensity will not vbullry with concentration and if the
brand-switching e ffect is larger advertising intensity will decrease
with concentration
For given advertising elasticity parameters then it is not pos-
sible to show an increase in A S up to some H and the n a decrease
he relation may be non - linear but it is monotonic over the permis-
sible range of H
Page 15
One possibility is to introduce a relation between (y - a) and H
If industries characterized as dominated by the brand-switching effect
also were more concentrated then a cross industry comparison would
show an inverted U shape relation between AS and H I know of no
support for such an association between ( y - a) and H
V ERROR SPECUCATION
So far I have implicitly assumed that there are no errors in the
model I now assume that the error
thvariable in the equation for the i- firm is additive that its mean is zero
that its variance is inversely proportionate to its sales and that all error
term covarlances are zero hat is
34) v bull (1 - o) [ a + (y - a) z ] + oy + u i i i
- - -
In vector notation
(35) bull (1 - o) [ at + lt - a) J +
E() bull 0 Cov (u) -
0
If we now ca lculate V bull bull we get
oy + U bull
(36) V bull (1 - o) la + (y - a)R J +
U bull E(U) bull 0 Var (U) bull
P age 16
The error term specifi c ation which follows Hall
wi th the ratio of profits
6 heir analysis seems
as wellbull particularly given
and Weiss was
applied by them to equa tions to equity or
asse ts as dependen t variables to carry the same
weight in the present context the substantial
evidence presented by marketing analysis on the tendency of company de-
cision makers t o use the advertising to sales ratio as the decision
7variable
The zero covariance assumption on the other hand1 is particularly
troublesome Within a probably not true since
many even ts outside the
single industry it is
industry would have
industries
similar effec ts on all the
8firms in the industry Between it is probably not true either
since many companies produce in more than one indus try and even ts in the
firm would tend to effect all of its activities The extension of the
model to allow for non-zero covariances is certainly called for
--- -
Page Ll
Athird characteristic of this specification also deserves some comment
only on equation is shown for the firm or industry Several haveinvestigators
included an advertising equation ina larger system of equations (Comanor and
Wilson (1974) (1979) and Weiss (1976) so in myMartin Strickland Idid
dissertation1 Ihave ewo defenses for presenting only single equation results
here One is the suggestion by Comanor Wilson that simultaneousand equation
verybias may not be important in this context he second is that Iam will-
ing to see the model extended to include a profitability equation and I intend
to move that direction in the near futurein
Onthe inclusion of aconcentration equation on the other hand I amskep-
tical he problem isthat in some fundamental sense equations in models of
industrial organization should be oriented to the decision-making contexts of firms
Ifeach firm in anindustry with Nfirms has only one decision variable then
there can be only Nindependent equations which are based on the first-order
optimization equations It not show large ofis difficult of course to a number
functional relations which follow from the first-order conditions point here he
isthat only Nof them can be independent
Consider for illustrative purposes only a situation in wich teh N
itsquantities say q are set exogeneously and each firm determines advertising1
of and level say ai where a is the true optimal value a a bulla
1 1 i i + i
Ifp a) then the Nq
s and the Na s will determine N p s say pii bullp
i(l i 1 i
Now given Eand 9 we can determine both the vector s andthe vector v since
- bull til
ands We can also determine S s z bull s-1 s-ii
v and we
industry concentration isHbull (alternatively
V and Hare functions of the same
pendent Furthermore if there are
VM) and (H1 H2 bull bull bull )contain
VI STATISTI CAL APPLICATIONS
Page 18
Given the observed values of might be tempted to conclude that we
have a two equation system in which v and z are jointly determined That would i i
be inappropriate however since the 2N random variables (v ) are determined by
onlt N basic random variables t The variance-covariance matrix for ( ) must
be singular bull
The extension of this observation to industry level variables is straightshy
forward Since observed industry advertising intensity is V bull l v and observed 4
C4 bull z ) it follows thati l i
set of random variables t and cannot be indeshy-
M industries then the variables (V1 v2 at most M independent random variables
Given the assumptions made above y a and o are constant within a
given industry That being the case (34) can be written as
(37)
s1 bull (1 - o) (y - a)
If and
So + +
(1 - o) a + oy
81 are esti mated for an industry using a suitable statisticalSo procedure giving and 81 an estimate of y is then available The sumSo
Intra-industry analysis
1974 of Business Re ort -
categories are identified
Page 19
of So and S1 is y so y bull So + S1 is the estimate of y
If o is assumed to be bounded by zero and one the sign of 61 is also
the sign of y - a The estimate of So is of cou se an estimate of
(1 - o ) a + oy Moreove if y gt a then a is less than 6o- (1 - B) a+ 0(
Cor esponding results would
hold if y bull a or y lt a
If 61 gt 0 then we would estimate that a lt 6o
bull
For purposes of estimating this model 32 manufacturing industry categories
were selected The data are fram the Line of Business forms filed with
the Federal Trade Commission they are discussed in detail in the Annual Line
which is available from the FTC The 32 industry
in Table 1
Simple regressions for equ ation (37) were run for each of the 32 industries
with data for each of the firms which filed in the industry constituting an ob-
servation The results are given in Table 2 lines lA 2A
1974
the intercept
For all the A
equations generalized least quares was used th both and market
share bein scaled bv the square root of sales
The 32 industries were taken from the larger set of all manufacturing industry
categories in the 1974 data files Two selection cr iteria were used thereLB
had to be at least ten reporting companies and the industry had to be a consumer
goods industry or a producer goods industry which is strongly associated with a
related consumer goods industry so that adve rtising by the firms in the producer
goods industry impacts on demand in the related industry (eg soft drink syrup
and soft drinks) There were 135 industries with at least ten companies - 103
producer goods and 32 consumer goods
273
274
TABLE 1 INDUSTRY CATEGORIES USED
FTC
COOE OESCRI P T I ON _
2001 MEAT PACKINGt SAUSAGE S ANO OTHER PREPARED
MEAT PROOUCTS
2004 DAIRY PRODUCTS EXC F LUID MILK
2005 CANNED SPECIALTIES
2001 FROZEN SPECIALTIES
OEHYORATFOt AND
I NC LU 0 ING PRE SERVE S
SOUP MIXESbull
SEASO lNGSt ANO
OTHER PET FOOJ
RAIN MILL PROOUCTSt
ANO PREPARED
RELATED PRODUCTS
PRODUCTS
Z008 CANNEOt ORIEOt PICKLED FRUITS
ANO VEGET ABLES JAMS JELLIES DEHYDRATED I G TA3t E
S40CES ANO SALAD ORESSINGS
2010 DOG CAT A O
2012 FlOUR t OTHER RICE
MILLING fiLENOE) FLOUR
2014 B EADt CAKE ANO
20113 CO FECTIONERY
2026 BOTTLED A O CANNED SOFT DRINKS 2021 FLAVORING EXTRACTS AND SYRUPS EC
2029 MISC FOOOS AND KINDRED PROJUCTS EXC
ROASTED COFFEE
2301 MENS AND BOYS SUITS AND COATS
2302 MENS ANO BOYS FURNISHINGS
230 3 kOMENS AND MISSES OUTER EAR
2551 HOUSEHOLD FURNITURE
2702 PERIODICALS
Pqe 19a 1
RELATED 1972
src cooe
2011 f 3
202tX2026
2032
203pound
2033t4t5
2047 2041t4t5
2051 2065 2086
2081
209XZ05
231
232
233
251 272
2703 BOOKS
2704 ISC PVeLISHING
2806 ORGANIC FlEERS
2807 DRUG$ ETHICAL
23234
PT 2 S 3
2808 ORUCSt PROPRIETARY
2809 PERFU ES t COSMETICS ANO OTHER TaiLET
PREPARATIONS
2810 SOAP AND OTHER CLEANING PREPARATIONS 2815
2901
PEST1CIDES AND AGR ICULTURAL
PE OLEUM REF I N ING CHEMICALS EC
3606 HOUSEHOLD COOKING EQUIPMENT
3612 HOUSEHOLD APPLIANC ES NEC INCLUJING ELECshy
rqlC HOUSE ARFS AND FANS ANO SE I G MACHINES
3617 RADIO AND T V RECE I VING SETS 3808 PHOTOG APHIC E UIP amp SUP LfESt EXC
3qo3
PHO TOCOPYING
SPOR TING A D E QU I P M ENT f SUP P L I E S ATHLETTC GOODS NEC
3904 DOLLS GAMESt TOYS AND CHLOqENtS VE ICLES
PT283
2844
234tX2844
287 9
291 3631
3634t6t9
36H
PT386l
3949
15
943
middot
l A6l f 2 R(GRESSION RESULTS
EQ I NO
NR CODE cu 121
lA 2001 A (
2A 2004 r (
3A 2005 a
-R
ClS amp31
16
17
1 z
I TER-CEPT bull t
ooo9t21At 00022311 oooeos uu
OOl49tCAI 0010 2 middot 000960CCI
o049tl1At o04t55C61
MARKET SH ARE bull 5 I
-010670Ctl
-017600181
-001050
-0211AO
-0077 0
AOVER-TIS INti
lbl
000()39 000072
000018 0 0024liAI
-000079
SHARf
4ampgtVf Q bull
111
000140
-000170
RSQ
COl
66 485 862
607 6 3 t ()2
blti6 633
flASTI-
CI T Y PH
-007271(1
000441
-u0288l
clASTSHU (t)Ef
bull 10
tto
-234
210
( OOlltOCCI -126520 000922 181 001790
4A 200 7 18 00590liAI -038670 4e16 -032110 lell u c
OOOS74e 000700 025940
00067018) 0021001AI -02l100CAI
5 26
920
5A 6 c
200 8 26 0 03Sl9CA I 0012 26181 o 024511(
-005540
-0 81420 0001961AI O()Ol90CAI ooll240
441 604 891
-0020 19 274
6A 2010 16 Oe04224CAI o359lOIAI 9oq ll4015dCAI 061 6 -middot 005435CAI 00011 8 1AI 917 c 00535ltAI -009850((1 OOOA51J CAI -001610161 971
7A 2012 001862 000010 351 00187) oo1 a oooq45 ono1o6 439 c 001109 -027050 Oe0067SIAI -OOll60C81 897
- middotmiddot- -
8A 201t ll 000757 Ol4140CCI 814 ott9oocet 095 8 oooqo6 C C J 000069(81 811 c ootl2lampBI -080910CCI Oe009lUBI -OOtlOOCCI 960I
I DIIJIT pound11 1V OTIUTVTfAll-1pound1t- A _ 00 R - qq r - Q
1jl)OQCD
0
-middot middot--
l
_ __ _
I
i
0 00141 c c
91)7
I
T ABLE 2 REG ESSIOH RESULTS CCONTI
EO NR
IND CODE
iR bull
cos
I TEq-
CEPT
t1ARKE T
SHA E
AOVER-
Tl SING
SHARpound-cent
AuVER
RSO ILAST1-
C I T Y ELAST
SH CPtf (lt Cl) bull 11 I 41 I 5I (61 171 I 8 I f9) c 10)
9A 2 0 1 8 19 002397181 -007640 367 -005 246 le46 6 000605 000550CAI 597
-000029 G14180 OOl654CAJ -024li01At
lOA 2026 12 0028901AI -OlZl90 -0114 0 1 1 1 5 759
Ll (
llA 2021
f3 c
12A lOZ9
B
(
13A 2301
B
c
14A 2302
B
c
154 201
n (
16A 2551
8
10
35
10
20
19
24
001897(()
00141 6CCI
Oll9lHCI
010604 middot middot
Oll083
OOJ11JCAt
002059141
001411CIH
-000412
000293(81
000187
001159 (3)
ooo6a4cct
o00792CCI
OOl40dCAt
000006
-000076
OOlltl6fA)
o o0 5 64CCt
-034570
-o1middot o3o
-1659 90
Ol51lO
-046190CAt
Ol5450CA)
-005l60
002940
-030060(()
-0 43l50CCt
0 33650
-019080
00004
bullJ00741CAI
-000069
000466
0001 51CAt
000947CAt
O OOI)70CAt OOl750CA J
000175CCI
000671CAI
OOll5) f A t 002207CBI
0004b7CA I
- 007 1 70
001760
-001410CAt
- 0 16200CAt
-004210
- 1 207 70
705
96 0
542
466 805
460
614
921
832
9 61
996
424
550
9 15
56 1 651
884
-006114
019 037
03504lCA)
oo409Q
-141644((
491 -0 17664
607
871
lell
084
101
072
103
108 11IllOQfl)-- l
-037880fCJ OOl6l4CAJ -040l70fCI 0()
c
I
I
bull 9
9t
middot middot middotbullmiddot bull lraquo
TABLE 2 REGRE SION RESUlTS ICONTt
bull
EO I NO tHl I filER- f4A_K T AOVER- sbulluRE ttSQ flA Jl- iLAS T NR CODE cos CEPT SHARE JISING AOVfR bull (JTY SHR CiJCF
110Cl I f amp Itt 15J tbt bull 1c u caa
llA 2702 ll o079 6 81 Bt -019110 S6s -OlltO 111
a 004679 ooooo1 t69
( oOl26o -015160 001560 -022690 ltJr()I
ldA 270 3 19 0065lt -OIt2l60 220 -Ol57a2 11 8 e o 0 71t6IA t ol08 75tAt ( OOl983CiU -060llOC() OOl7191A t -O lltdOC A) middot 89
lqA 2704 10 005082 -Oil060 22b -l)i1977 lelO 6 -000080 oo2772C4 t 910 c 000907 -011 200 Oe04105CA t -046960 995
20A 2d06 1 II
O OlltStiAt -001120 837 00014 -))th O01380amp At -oooo 11 821 ( OOl526CAt -026700181 0002 5UAt 000530 969
2lA 2807 28 OOt9lt9CAt -00891 0 516 -0 03962 225I8 002095CCt OOOl lliAt 690 ( o02965CA t -056650((1 OOOt50CAJ -003l70ttl 911I
I 22A 2808 15 ot3l06 119730 655 1130) 090
0 - 007526CC t Oe003961At 835- middot-
c I oo7523 -067210 001l 2tA t -0091t80CCt 951
2lA 2809 21 Oe2067lCAt -1218101At 71 2 -10095((1 121 ttlIaOQ8 00797418) OOOI8lltt 686 Ill( 0092lU A t -Olltl70Ct t 000605CAt -OOit8701At 951 _
---- -
32 009506CAt
0 bull 06 981 I A J 006901 cA bull
00 24A 2810 -010960 5t 5 -OOltSO 7 56
8 -000011 s11 c -5811701At oooaoatAJ oot520IAI 888
-- 37 --- middot - -- -1 ) l lbull t5JJ r-l bullhmiddot1 --- - --
l5A
26A
27A
c
2RA
c
2 9A
31A
llA
24
13
597
Ci)55
14
l 6h
RSQ ELASTI- (lASTCOE f-
TABLE 2
EQ I NO NR CODE IU
REGRESSION RESULTS CCONTa
NR INTER- MARKET
(10S CEPT SHARE AOVER- SHARfcent
TIS lNG AOVER
) CITY StiR
c It) bull 5 co a I 1) bull 0 bull 9 c l 0 bull( 2 1
2815 16 003019 -009990 275 -o069bmiddotJ
6 001446 000224 282 c 002178 -054630 o02221CCI -0075 00 646
2901 29 00021t2 CAI 0)0520 592 0007 6 1 068
0 00105C8 t 000012141 60)g( OOOlJ51AJ -00126018) ooou271At -000u80CA) 974
36oa 1 3 003296(4) -oca29o -oo-q94
001755 000203 lt91t 002499((1 -04l820CCI OOl374CAI 0068)0 946
3612 0033281CI 011520 195 01684) oso
00 30 42C OI 00012 2 424 oo579ltAt -2 7l7801At 002 100IAI -003220 819
8 3617 btel 011789 081002259 009530
001448 oOOl OitCCI 732 OOl7 22CAI -0 46160CAI 0004961AI -OOll40181 989c
30A 3808 14 0021t 3 6fAI -000110 628 o023291C) -005 6 middot- _ middot- 0bull 02 04 1 IC I ooooo9 637 ( 001584((1 -0161)1)0 00021 91AI -0 00060
390) 14 002790CCt 006590 h8 t 0 0938 4 070
B OOl89bCCJ 000144 74
0021961(1 -OItlOlO OOl323CAI -004800 915
B
middot--- - middot 4 - middot- - dfll3904 12 004563181 0)8300 879 0428651(1 089 OQIll004 277CA I ooozstcea 919
c 0041)3181 -0-11250 0011881(1 -006270 972 0Ill
- J - bull bull bull bullbullbull middot - - middot - middot gt middotbull L)ofbullflbull middot middotfJltl -tn- middotmiddotmiddotmiddoto +f middot t middot bull
c
is
positive
2
Page 0 determineThe consumer goodproducer good split is sometimes difficult to
Sollle support for the split used in this paper is a comparison of goodness - of - fit
distributions of a
industries
set
advertising are
that industry
a sales increase of
the
highest aD10U3 those
measures for the two groups of industries Table 3 contains
measures for the 32- consumer goods industries and the 103 producer goods
2 Median R is 5 7 for the consumer goods set and 40 for the producer goods
Estimates - of y the elasticity of demand with respect to
given in column of Table 2 The hypothesis that y bull 0 with
alternative was applied for each ittdustry As shown in the
could not be rtjected for 24 of the 32 industries Of the
and 10 are positive For the remalnifg eight three are
and five are significantly positive
Of the threertegative elasetcities the largest (in absolute value)
cosmetics (code 2809) With all elasticity of -1 01 a one
industry advertising would gwerate a decrease in industry sales
more than one percetu
At the other extrema is toys and games (code 39 04) For
a one percent ittcrease in industry advertising would lead to
four-tenths of a pampreentmiddot Though not significantly different from zero
elasticity estimatbull for pt-bpuetary drugs (code 2808)
which are
Both drug industries have
to sales rafios (211 pe1 cent for 28 08 and 13 4 per cent
other hand flaIJo ing ext1ac s (mainly soft-drink SYUlS) also
is for
percent increase in
of slightly
a middottwo - tailed
table that hypothesis
24 14 are negative
significantly negative
9
the codetics and p-ropriacuy very high advertising
for 28 09) bull On the
has a high ind UY
advenisiltg t() sales ratio ( 8 0 per cent) bull but its estimated elasticity is very
small and not signifieantly different uom zero
middot It doe$ not seem unusual to find many elasticities near zero with a few
negative and a few positive If aggregate consumption is insensitive to aggregate
advertising as seems likely ld then what one industry gains must be lost by
others
Page 21
elasticity of demand does
(10) we can
if t 0 then
Finally a determination concerning the industry
not fix the individual firms elasticity Going back to equation
write y bull (1 - ) a + z y Since a gt 0 it follows thatii i i
Y gt 0 If Ylt 0 however y s sign is indeterminateij ij
Turning next to the coefficient of market share 20 of them are negative and
12 are positive with three of each sign being significantly different from zero
The largest positive value which is significant is for pet food (code 2010) and
a close second is mens amp boys suits and coats (code 23 01) The largest signifi
cant negative value is for cosmetics (code 28 09)
In the model some relations among y a 0 and 61 are implied If middota is
negative (1- y) is greater than one Since 61Y bull (1- a) (1 - ) 61Y is
pCsitive but it will be less than one if 0 is large enough Seventeen of the A
ys are negative in each of the seventeen industries 61 is also negative and les
1 (algebraicly) than y so 61Y is greater than one his is not evidence that the
degree of cooperation is low or zero in any industry of course A high o togethegt
with a negative y which is small (in absolute terms) relative to a can give a
value of 61Y greater than one On the other hand if 61Y were less than one
o would have to be greater than zero
For y gt 0 two conditions may hold If y lt a it follows that 61 lt 0 since
(y - a ) lt 0 It also follows that 61Y lt 0 regardless of the value of o bull
In addition the relation is an if-and-only-if one ie if 61Y is negative
then y lt a Three industries have negative values for 61y dairy products
except milk (code 20 04) synthetic fibers (code 2806) and photographic equipment
(code 3808) In all three cases y is quite small so a does not have to be very
large to give a negative value to al
If y gt a the model shows that 61-- gt 0 whatever the value of 0 bull There are
12 cases where both y and 61 are positive with 61 ranging from near zero
(grain milling code 20 12) to 1 20 ( proprietary drugs code 2808) Several
coeffiei-en_rs
--
15
o f to zero
Tab le 3 - Consumer Goods
Indus tr ies
2Distributions of R and Producer Goods
Page 20a
Consumer Producer
Goods Goods
0 - 9 9 0 3
10 - 19 9 0 1 1
20 - 29 9 3 20
30 - 39 9 3 17
40 - 49 9 5 16
50 - 59 9 8
69 9 6 Ii 1060
--f 670 79 9 2
80 - 89 9 4 4
90 - 100 1 1
Totals 32 10 3
Notes
2is used R i s cal-
culated as F F + ((N - K- 1 ) K ] 1 b_ F is the F
statistic Lor the hypothesis that all o f the non - intercept terms are simul taneously equal
1
2
--
-- Since generalized least squ-area regress ion
Page 12
o f these cases have values o f 6 1 Y rea ter than 0 8 (pet food (20 10) bread amp cakes
(20 14 ) miscellaneous processed foods (20 29 ) men s and boys suit s and coats
(23 01 ) p rop riet ary drugs (28 08 ) elect ric housewares (36 1 2) radio and
IV sets 17 and games and toys (39 For be than(36 ) 04)) 6 1 Y to greater 0 8
both o and cry mus t b e less than 0 2 That is in these industries the results
shown in Tab le 2 would hold only with very strong dominance of marke t demand
effects over brand swit ching e ffects and a very low degree o f cooperation
Using the concentration rat ios given in Table 4 the average concentration
ratio for the eight indust ries is 38 1 For the remaining 24 industries it is
43 2 This result would lend some support t o the proposition that concentrat ion
and cooperation are related if the group of eight industries are identi fied
as having low degrees of cooperation
One f inal note on model predict ions S ince both cr and o are assumed to
be non-negative 60 should also be non-negat ive In only one o f the 32 cas es
is 60 negative and it is not signif icant bull
cor pJ
7 1
7 0 4
1 4 7 5 1
)
7 3 z a o a t s 1 1 3 3 5 ( i V Z 8 tJ1 2 1 - 1 1 0 7 3 5 c middot v 2 8 1 0 3 2 -0 0 1 4 3 0 C 1 V 9 8 3 2 8 1 5 1 6 - 0 0 6 t1 6 9 C _ V 1 11 0 2 9 0 1 29 0 00 7 6 1 C l k V 7 3 1 3 6 08 1 3 - o n 4 9 9 4 3 3 3 6 1 2 2 4 0 1 6 4 3
3 1 7 1 3 0 1 1 7 d 7 8 3 3 8 0 8 0 0 2 3 i 2 3 0 6
P 4
3 9 0 3 1 4
1 2
0 09 3 8 4 39 1)1 0 4 l 8 6 5
T A LE 4 e I NCU S T R Y C A T E G O R Y
F TC NOM e tOOE ass
C 1 I c 2 )
2 0 0 1 1 6 2 0 0r 1 7 zo o l Z 2 0 0 7 1 8
z o o 2 6
z o t a 1 6
2 1) 1 2 1 5 2 0 1 4 1 1 2 0 1 a 1 9 2 0 26 1 2 2 0 2 7 1 0 2 0 29 - 3 5
2 3 0 1 10
2 3 02 z o
2 3 03 1 9
2 5 5 1 2 4 Z 1 0 Z 1 1 2 0 3 1 9 2 7 4 1 0 z a o 1 3
2 A 0 7 z s
E LA S T I -C I T Y
c 3 )
- 0 09 7 2 7
0 00 449
- 0 0 2 8 9 3
- 0 3 2 7 70 - 0 OZ il I q
o 4 u t s a 0 0 1 3 7 3
o 1 4 o o-0 0 5 2 4o -0 1 94 0 1
-C l 6 l l 4 0 1 JJ 3 7 0 3 t 0 4 2 0 0 4 0 9
- 0 4 1 8 44 - O l 7 t ) 4 -ll 7 1 4 0 b -0 3 5 7 8 Z
- 0 1 9 7 7 0 1)0 3 3 4
-o n 9 2
IJ A T A
CONV
ON-C ONV
( 4 )
t 1NV
C 1t-V C NV c ) v C - JNV C rlNV c n v c a v C ON V ( i) l V
C CtV
C t1tV C IJNV
C 1 V
c 1 v
Pa ge 2La
HEQ F t Noe x
c 5 )
6 2q
6 89 1 48
6 1 4 ) q 3
1 9 7 1 9
q z t s 5 a e z
l l 9 1
4 o 7 1 2 0
c oqt 1 Z
4 4 )-q 1 z
5 5 9 1 0 64 1 3 3
n 3
Inter industry analysi
Page 23
Equation ( 3 6 ) may be used to fo rmulate a rough test o f the cooperat ion
vs e f ficiency controversy I f we rest ricted ourselves to some subset o f
indus tries where the relat ion be tween y and a i s the same we would be lookingbull
at a sample o f indus tries for which the dependence o f V on H and o is a
quadratic wi th no squared terms
I f we now sub s titute (26) into ( 36 ) we get
(38) v bull a + (y - a) R + (y - a) o (H) - (y - a) H o (H) + u
If it is only d if ferences in marke t shares that affect advertising inten-
s i ty levels i f concentration does not af fect cooperation and i f o bull 0
( 38 ) reduces t o
( 39 ) V bull a + (y - a) H + U
Let this b e the null hyp o thes is and note that V is a linear func-
tion o f H If y gt a and the null hyp othesis holds the expected relat ion
between V and H is as shown by the pos itively sloped straight line in Figure
6 b etween points A and C
In this context the alterna ive hypothesis the degree o f
cooperation ( o) is a positive function o f concentrat ion
is that
(H) I f the func-
tion is as described in (26) and Figure 4 the relation between V and H
would be as shown by the curved line connec t ing po ints A and C in Figure 6 One way to test the null hypothe s is against this al ternative is to test fo r
the hypothesized curvature in the function relating V and H
Fi gure
R
6
v
ID c
H 1 0
I
i I I
sfigned
clasJ s
in def ini g
effec t is the
I
Page middot2lf
One charac teristic of produc ts which may be useful
classes wi th respec t to significanc e of the brand-switching
conveniencenon-convenienc e distinc tion developed by Porter ( 1 97 )
Using his identif ication of IRS indus tries as a s tar t ing point
each of the 32 industries used in this s tudy to one of the two
The ass ignments are not ed in Tab le 4
Using the conveniencenon-convenienc e dis t inc tion the
divided into two sub-samples
sample was
Four regressions were then run bull ene for e ch
sub-sample one for the to tal sample and one for the total sample with
a dummy variable for convenience goods Each regression was run wi th LB s
and wi th indus tries as ob servations The results are given in Tab le 5
For the LB level equations ( 1A - 1D) equation(37) was used ome
rearranging of terms gives v bull y - ( 1-o ) (y - a) ( 1-z ) + ui i i
bull y - 6 1 ( 1 - z ) + ui When observations are pooled across indus trif s yi
is an industry variable In these regressions the values o f y es t imated
in the first stage were used Since industries have been grouped so as
to reduce differences in (y - a) that term is t reated as a cons tant
within each group Finally H the Herfindahl index has been subst ituted
for o the degree of cooperation The resultant equation is linear
y ( 1 - z ) and H x ( l - z ) This is the equation for whi ch results arei i
reported in Table 5 In all regress ions all variables were weighted by the
square roo t of sales
in
raquo
- 34
3 1
i l T 4 8LE S R E G E S S I ON R E SUL T S i X T f N S I ON S T O T Y PE Of
AND L E V E L OF C ONC E kT R A T I O
E O O tt S
tP bull T Y P E NR C ONV O B S tO N-CONV
E L A S T I shy
T I C J T Y
O E - t-41( T SliA R E
H( fl F
1 1 - K O x middot K O X E R F
- H S a 1 1 - H S t 1 8 1 1 9 1
R S O
1 1 0 1
q z
t - bull 9 f 2 1
Z t P D l 6 fC ONV - 1 9 1 s s c 2 1
e - l O I e bull u q a
2 3 1 0 1 1 Nl N- t O V 5 7 1 ( 1 2 19 1 6 1 - J l S 2 5 4
l l l l t z q o a 1 - J o v z I
l ( I NO l l A l l ol u l t o o c bull J u l l
d S
c 3 4 t t r a l
bull Z D 1 0 l l 4 l l 2 3 -gt 2 l l 2 0 - 4 lS 5 4 l C 4 1 o e 6 t l b b 5 1 1 0 1 1 1 c - 1 9 5 1 t d S 2 1 tV
OQ10 T-R A T I O S A E G I VE N l P A RE N T H E S E S l E V E L S f S I G 1 F l C ANC E A - 99 5 B - q 9 t - s NS J UC E NO I Nl E R C ( P T I S U (j f middot) I N f Hf t OUA TIOI S 6 E f 3R E A O J U S T H E N T F OR 1-t E T E R O SIC E O A S T f C I T V pIraquo
1
middot
I
R SO I S C A L C UL A T E D A S f I f bull 1 1 bull 1lt 1 K t t- E E N I S THf U A1 B E R O f fJe S E fi V A T I O S F t S HtE F - S T A T I S
I N DE P E ND E N T y II
T I C F O T H E U Y P O HIE S I S f A l l C JE F F I C H N T A R E QU A L HI l f RO A f40 K I S T H f UUM 8 E R OF Q ( A fL f S
I l l
1 A
1 6
t C
f 1 r R I l l
1 2 1
L Jt
L B
l q
L 1
on s T Y P E bull 2 )
P OUC T
X
H S I l l
l O l A I bull l 0 7 1
O O b O J
1 2 00 1
1 b 2 1 C I l l Z U
l l S b 1 -1 1 0 1
rlt F X l l - fi C f l
bull 1 I
l l
- ()4 2 3 l f H
1 - 2 7 9 1
K O X
1 1 - ti E R F I
l d l
S4 l i A I 2 l b
1 3 00
K O X -t f R F t S J
I 1 - HER F t
)
60 9
z o bull
0
t l l
3 8 8
1 7 6
5 64
64
N f bull
tl S bull
I l l
I 4 1
t ONV
NON-t ONV
A l l
A ll
c o v NO- ( ONV
1 4 1
1 5 1
- l S b
c - 1 1 2 1
5 8 2 C A I 1 6 7 0 1
middot I l 9 H
l 1 il
1 4 1 6 1
E L A C) T J shy
l t C I T Y
I S I
l b l
- 0 2 0
1 - 1 4 9 1
O Z 3 2 1 A I C b 6 b l
0 0 9 2 -J
1 1 b 4 1
o z t t t l 1 2 2 1 1
oru - Ht tt F
1 gt X bull b I
Page 25 bull
Mean val -as ior relevan t variables for the two sub-samples are
Conv Non-Conv
Adv Sales 4 7 2 3
iit Share 2 7 3 2
Ela cicity - 05 3 02 2
Herf indahl 7 3 1 860
That the dist nct ion between coft9enienee goods and non - convenience
goods mat ters is supported not only by these means but also by the regression
results Equat ions lA and lB are different in almost every respec t Only the
concentrat ionmarket share int eract ion term seems to mat ter for convenience
good s Jus t he opposite holds for non - convenience goods both elasticity1
and market sl e are highly significant but the concentrat ionmarket share L
int erac t ion t=rm is not
Given that one minus market share is used these equ3 tions the coeffici t
of market share is the negative of what is sh in col (6) of Table 5 For the
regress ion for non - convenience LB s reported in equation lB then LB s with larger
market shares have lower advertising to sales ratios
On the question of the impact of the degree of cooperation the story is mixed
For convenienc good s the coefficient is pos itive as equation (31) predicts when
there is a po sitive degree of cooperat ion Fo r the non-convenience goods LB s middot
concentration has no effec t
Aggregat ion t o the indus try level and then rer unning the equations $ives the
results shown in (2a-2d) of Tab le 5 As expected R2 goes up and levels of significanc e
fmiddotJr individual coefficients gJ down In add ition the coeffic ient of H ( l -H) inmiddot
quat ion 2A is no t significant though it is po s i tive Since 2A is the indus try level
I
--
all but one of them and that one is not significant For 18 of the 31
wo uld have on its adve rtising to sal es ratio
cognate of LA the variable H ( l-H) pl ays the same role in 2A as H ( l -z 1) p lays
in LA
Vconomies of scale in advert ising
The literature on advert ing contains a number of definitionbull o f economies
of scale and of findings eone rning its presence and magnitude Before I loo k
a t the data with the hope o f mmenting about scale effect s then I wan t to
define the term as I use it
By economies of scale I mean a decrease in the number o f units of adver-
tising which are needed to generate a unit of sales as advert ising is increased
Given the definition of the firm s own elast icity of demand with respect to
advertising (see the text foll wing equation (1) ) the technical definition
is that rii gt 1 Diseconomies of scale are def ined symmetrically i e
Yu lt 1
ith a cro s s - section oi observations on advertising intensity
(v bull a s1) and onmiddot advertising (a ) we may regress the former on the latter 1 i i
The results of such regress ions for the 32 industries are given in Table 2
lines lB ZB bull bull bull 32B For the 32 eases the regression coefficient is positive it
positive coefficients the coefficient i s signif icant
The que tion now is whether these results ar e evidence of pervasive dis-
economies of scale in advertislng I think no t Given my as sumption that the
data I observe are equilibrium results I may ri ghtly conclud e that almost every-
where high levels of advert ising are as sociated in equilibrium with high ratios
of advertising to sales I may no t conclud e anything however about the
impact that a move by some firm away from its equilibrium level o f advertising
middot- Page 27
Given the assumed relation in equation ( 37 ) there is a corre spondshy
ing equilibrium equation relating v and ai The equation is non - linear i
and I have not found a simp le way to characterize it It is po ssib le to
det ermi ne its deriviative with respect to a b y takin$ tne de riviative of ( 3 7 ) i
When that is done the result is
v - 1(40) i bull sl a a z ( 1-z )i i ira-l
Sinc e all the other terms in (40) are non-negative the sign dep ends only
on the sign of 6 1 which is in turn de pend ent only on the sign of (y - a)
That the signs of the coefficients of a in equations lB 2B bull bull bull 32B arei
not always equal to the signs of the coefficients of z in LA 2A bull bull bull 32A i
is still a mystery to me and something to be explored
bull
Summary
My purpose in writing this pa per wa s t o attempt to integrate explicit micro-
economic mode lling reasoned econome tric specification of error terms and high
quality data to exp lore some questions about adver tising by large firms
al quite important improvements could be made in the first two of those areas
and much data ma ssag ing is yet to be done in the third
Concerning the subst an ce of what I have done I think six things are im-
Page 2 8
Severshy
V and Conclusions
portant
1 Explicit modeling is worth the ef for t if for no other reas on than
that it provides a basis for choosing fr om a varie ty of funct ional
forms
2 The same is true for error specif ica t ion I did in fact look at scatshy
ter diagrams for all 32 industry categor ies and they vir tually all show
the het eroscedast icity which I as sumed and for which I correc ted
3 The predicted relation be tween advertis ing in tensity and market share
shows up clearly in on ly 25 of the case s examined I have ye t to
explore why that may b e the cas e
4 Some evidence concerning the presence and impact of coopera tion was
produced but it is not clear ly pervasive
5 The dis tinction between convenience goods and non-c onvenie nce goods is
unambiguously a good one
6 Virtually no evidence concerning econ omies of scale in adver tis ing can
be gleaned fr om this study given its assumpt ions
More work is called for on many of these issues my study seems to have
raised more questions than it has answer ed This is pr obab lly due to the richshy
ne ss of the data source since I had the oppor tunity to address several issue s
at once
- -- - Q --
Foo tnotes
Manager Line o f Business Program Federal Trade Commission The author depended heavily on several staf f members of the LB Pro gram for statistical and clerical support in particular Joe Cholka George Pascoe and Haro lene Jenkins Helpful comments were given b y Richard Caves Dennis Mueller Michael Porter and F M Scherer When the paper was in its early stages o f development both S teve Garber and Jon Rasmussen former economists in the LB Program provided eery useful cri tiques The views expressed here are my own o f course
1 The basic exp licit model of the role of advertising is Dorfman and S teiner ( 1954 ) and the models in this paper may be seen as ano ther extens ion of their work
2 For a review o f models of oligopoly which treat prices and quantities as relevant variables see Shubik ( 1 959) Extens ions to take advertis ing int o account are fairly s traightforward
3 Dorfman and S teiner dealt with a price-setting firm For a monopolis t inverting the demand relation to show Pbull P (Q A) _instead o f Q- (P A) has no effect on the results o f the analysis For oligopolists who face a sys tem of demand relations inversion does have some impac t on the results The dif ferences are not trivial nonetheless they will no t be explored in this study
4 Shubik ( 1 959) p
5 From a formal mathematical point o f view the function is similar to a1 i Legrangian funct ion of the form - r rr
L bull 11 + A ( j - (i i i+j
where the indicates some fixed level of the profits industry Following the same logic which resul ts in and income constraint as the marginal utility of the of the indus tr in the obj ective of firm i
6 Hall amp Weiss ( 1 967) p 323
7 See Schmallensee ( 1 97 2 ) Ch 2 pp 1 6-47
8 On this poin t see Imel amp Helmberger ( 1 9 7 1 )
9 Long ( 1 970) p 130 - 1 39
1 0 See Schmallensee ( 1 972) Ch 3 pp 48-8 7
rrj i+j ) of the rest of the
the identification profits of the rest
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980
Page 7
(15) may be written as
0 (AA)
lx 2x 1x
ith
1y
i
First assume no
that if there are
will be no changes in market shares
- 0a bull
Page 8
Equation (17) is a general statement of the relation betbulleen changes in
advertising levels and changes in market shares which would acompany them It
may be used to make some general observations about brand-switching effects
industry effect eg y bull r t bull 0 and assume further
equal percentage changes in all advertising le =ls then there
Under these further assumptions
0l and dz -1 bull (iA) ltr assumption the second term on the right hand side of this expression is
- By
zero
necessary for rt bull 0Consequently for market shares to be unaffected it is
when y bull r l bull 0 This requirement is satisfied by equation (11)
Secondly consider effects on market shares if one firm inc ases its abull ar-
tising and others do not As a general assumption we require smiddot ch a change o
lead to a decrease in the shares of all the other firms hat is let
0 0aiai gt 0 and a a bull 0 for j + i For any matrix X with N rows and its trans-j j
pose X let X
row of X
bull middotmiddotmiddot where is a column vector which contains the
elements of the Under the assumptions about the iis given just
d-1 0 0above it follows that r becomes (aiai) iimiddot ehFor jzj lt 0 under these conditions then it follows that the j- element
N of (I- l ) lt 0 so middotij - il lt 0 or yij - yik lt 0 If this
kbulll N condition holds for all Y ij where j + i then it follows that yii - z
k Y gta
k- 1 ik since the th firms share must increase to offset the decreases in all other
shares
ei ith
1v ( ) h i
7urning
where is the
leads to z
to the r matrix given in (11) we note that 1 = cr + (y - cr) i i zi elementary vector with 1 in the position and 0 elsewhere This
bull cr + (y - cr) zi Since bull required is thatcr at- zY wYij (y - cr)zi - cr + (y - cr) zilt 0 or cr gt 0
These observations may be summarized by substituting (11) into (17) bull
giving
(18)
i s
d -1 zz -
-
-
+ (y - cr) tz
d -1 0cr(I - l ) a a bull
cr 1 z - (y - cr) 0 a
o
If ai increases and the other a s are const t cr must be positive to havej
a positive effect on zi A positive cr also bull plies a negative ffect of aJ
f all a s in-on zi other a s (including a ) fixed j
eluding a1 change by the same percentage zi remains unchanged Finally k i
if cr bull 0 there is no brand-switching effect
For a to have no effect on p requires yij bull 0 From (10) thisj i
amounts to y bull cr that is the industry demand elasticity and the brandshyswitching elasticity are equal If y gt cr gt 0 the industry elasticity yij doninates If y lt cr y lt 0 the brand-switching elasticity dominates ij
making
Page 10
Substituting ( 1 1) now in (6) the Cournot case and in (9) the
Chamberlin ease gives
-(19) Cournot vO a + (y - a) z
Chamberli n vl - y lmiddot th
The 1- element is then
-(20) Cournot vO a+ (y - a) zi i
-Chamberlin vl yi
In (19) and (20) the superscript 0 is used to denote total
independence of decision and the superscript l is used to denote
total interdependence of decision making
The relation between v and vf depends solely on t relation between
y and a f y gt a so that the industry effect dominat 3 vf v with
equality in the trirlal case of z bull 1 Cooperative irms would have ai
higher advertising intensity than non-cooperative fi
If there is equality between the industry elasticity and the
y Whe therbrand-switching elasticity or y bull there is cooperation among the firms has no effect on advertising inten-sity And if brand-switching is more important with y lt a e get
vi lt v cooperative firms will have a lower advertising intensity than
non-cooperative firms The three situations are depicted in Figure L
Given the results for the firm level variables calculation of the
corresponding re sults for the industry is s-traightforward Let
B bull t zf bull z z be the Herfindahl index of concen tration and notethat
z - bull l z bull 1 and that V bull A S bull a i 5 bull (siIS) (asi)i
bull middot Then for the general case from (6) and (9)
bull
Page lOa
y lt1 bull y y
10
f
V0 bull z [ a1 + (y - cr) J bull a + (y - a) H vl bull Y bull y
11Page
( 2 1) vo bull z d -1 r bull r z bull z- d d- i )11
l -1 N N v bull z d r z bull r z bull l y z - ij jibulll jbulll
and for the simplified case from (19)
(22
Since the forms of the equations in (22) are the same as in (20) all
of the analysis of (20) holdS for (22) except that His substituted for
middot z Advertising intensity will be higher for an industry with cooperativei
firms than for one with non-coope rative firms if the industry demand elas-
ticity dominates the brand-switchine lasticity If the ewo elasticities
are equal then so will be advertiSl intensity for the cooperative firm
and non-cooperative firm industries If brand-switching dominates the
cooperative firm industry will have lower advertising intensity
In the industry model context the possibility of a critical concen-
tration level may be easily introduced That is assume that at values
of R below R the firms are non-cooperative but that for alues equal to
or greater than H they are cooperative That is
(23) V bull a + (y - a) H if H lt H bull y if H gtmiddot H
This function is shown in Figure 2 for the three relamptions be eween y and a
IV THE D GRE OF COOEtRATION
The models depicted in Figure 2 are based on a very simple notion about
the relation between concentration and the cooperationnon-cooperation charac-
teristic They use what is essentilly a step function relating concentra-
Figure
I -
J H H
Page l la
2
v v v
a
I J I y a bull Y y ---i
I I I I I I II I
H 10 H H 10 H 1 0
bullbull
bull bull
Page 12
tion and the extent or degree of coo peration as shown in Figure 3 Using
o as a symbol for the degree of cooperation eq uation (23) can be rewritten
as
bull(24) v (1 - o) vo + ovl bull (1 o) [a+ - a) H ] + oy- (y
vhere
(25) o bull o(H) bull 0 if B lt B
Hbull 1 if B gt
The specific relation between concentration (H) and the deg ree of
cooperation ( o) shown in Figure 3 is more restrictive than necessary and
bas less appeal an one which shows a smooth increase of o from 0 to 1 as
B increases fro to 1 Such a smooth function is shown in Figure 4 he
degree of cooper ion is low until H gets close to H increases greatly as
B goes through R attaining a high level for an H larger than B but still
close to it and then goes to 1 as B goes to 1 More formally
(26) 0 bull o(H)
bull 0 if H bull 0
bull 1 if H bull 1
do gt (IdB
d2o- gt 0 if H B dBZ
0 if H H
lt 0 if R gt H-
If (26) is now substituted for (25) in (24) a corresponding smoothing
of the functions shown in Figure 2 is accomplished he smooth functions
Figure 3
I
10
H 10 H
Page 12a
Figure
Page 12b
4
10
H 10 H
l
(27)
Cl - o) o
[ (1 -
[ (1
t y ij z
j
are given in Figure 5
Page 13
The equation in (24) was constructed as a hybrid of two results derived
from first-order equations for the non-cooperative and cooperative models
It may also be constructed directly by using l bull (1 - o) + on as thei i
th 2objective function for the i-- firm Treating o as a constant the first
order equation is
0 an aa
i middot
n i
Using equations (3) and (7) and setting --- bull 0 givesaai
(28) - 0
After collecting terms and converting to matrix notation we get
(29) - Cl - o) + 0 z-1vi yii i
N
j bulll
+ 0
+ o v1
d-1 r z z
-(30) v Cl - o) rd
-
o) r +d
d r J _ -1- d z
and middot
d) r + 0 rjd
l r z
v - z --(31) zv
- (1 0- +d) t r z- d-
If we now impose the simplified demand structure by substituting (11)
we get
- (1 - o) a +( 32) + oy and
Figure
l
-y
v
Y
H
r
I I
v
a
y
H 10 H
v
a
bulla
HH 10
bullbull
(33)
i
Page 14
v - (1 - o) [ a l + (y - a) ] The corresponding equation for Vis given in (24)
If the relation between industry advertising intensity (V) and
concentration (H) is non - linear via the impact of concentration
on the degree of cooperation does the relation take the form of a
auadratic as some other investigators have proposed (Greer ( 1971 ) Martin (1979) Strickland amp Weiss (1976) ) That question can
be approached from two perspectives - conceptual and empirical In
the next section I will try to apply some statistical material to it
here I want to explore the question in the context of the models
developed above
Referring to equation (24) taking the deriviative with respect
to H and letting bull (H) gives
bull( 3 3a ) 3V (y - a) (1 - o) + (1 - H) 3H dH
Given that o and H each are bounded by 0 and 1 and that o is an in-
creasing function of H the term in the brackets in non-negative The
sign of the deriviative then is the same as the sign of (y - a) If the industry effect is dominant (y gt a) then advertising in-
tensity will increase with concentration If the two effects are equal
then advertising intensity will not vbullry with concentration and if the
brand-switching e ffect is larger advertising intensity will decrease
with concentration
For given advertising elasticity parameters then it is not pos-
sible to show an increase in A S up to some H and the n a decrease
he relation may be non - linear but it is monotonic over the permis-
sible range of H
Page 15
One possibility is to introduce a relation between (y - a) and H
If industries characterized as dominated by the brand-switching effect
also were more concentrated then a cross industry comparison would
show an inverted U shape relation between AS and H I know of no
support for such an association between ( y - a) and H
V ERROR SPECUCATION
So far I have implicitly assumed that there are no errors in the
model I now assume that the error
thvariable in the equation for the i- firm is additive that its mean is zero
that its variance is inversely proportionate to its sales and that all error
term covarlances are zero hat is
34) v bull (1 - o) [ a + (y - a) z ] + oy + u i i i
- - -
In vector notation
(35) bull (1 - o) [ at + lt - a) J +
E() bull 0 Cov (u) -
0
If we now ca lculate V bull bull we get
oy + U bull
(36) V bull (1 - o) la + (y - a)R J +
U bull E(U) bull 0 Var (U) bull
P age 16
The error term specifi c ation which follows Hall
wi th the ratio of profits
6 heir analysis seems
as wellbull particularly given
and Weiss was
applied by them to equa tions to equity or
asse ts as dependen t variables to carry the same
weight in the present context the substantial
evidence presented by marketing analysis on the tendency of company de-
cision makers t o use the advertising to sales ratio as the decision
7variable
The zero covariance assumption on the other hand1 is particularly
troublesome Within a probably not true since
many even ts outside the
single industry it is
industry would have
industries
similar effec ts on all the
8firms in the industry Between it is probably not true either
since many companies produce in more than one indus try and even ts in the
firm would tend to effect all of its activities The extension of the
model to allow for non-zero covariances is certainly called for
--- -
Page Ll
Athird characteristic of this specification also deserves some comment
only on equation is shown for the firm or industry Several haveinvestigators
included an advertising equation ina larger system of equations (Comanor and
Wilson (1974) (1979) and Weiss (1976) so in myMartin Strickland Idid
dissertation1 Ihave ewo defenses for presenting only single equation results
here One is the suggestion by Comanor Wilson that simultaneousand equation
verybias may not be important in this context he second is that Iam will-
ing to see the model extended to include a profitability equation and I intend
to move that direction in the near futurein
Onthe inclusion of aconcentration equation on the other hand I amskep-
tical he problem isthat in some fundamental sense equations in models of
industrial organization should be oriented to the decision-making contexts of firms
Ifeach firm in anindustry with Nfirms has only one decision variable then
there can be only Nindependent equations which are based on the first-order
optimization equations It not show large ofis difficult of course to a number
functional relations which follow from the first-order conditions point here he
isthat only Nof them can be independent
Consider for illustrative purposes only a situation in wich teh N
itsquantities say q are set exogeneously and each firm determines advertising1
of and level say ai where a is the true optimal value a a bulla
1 1 i i + i
Ifp a) then the Nq
s and the Na s will determine N p s say pii bullp
i(l i 1 i
Now given Eand 9 we can determine both the vector s andthe vector v since
- bull til
ands We can also determine S s z bull s-1 s-ii
v and we
industry concentration isHbull (alternatively
V and Hare functions of the same
pendent Furthermore if there are
VM) and (H1 H2 bull bull bull )contain
VI STATISTI CAL APPLICATIONS
Page 18
Given the observed values of might be tempted to conclude that we
have a two equation system in which v and z are jointly determined That would i i
be inappropriate however since the 2N random variables (v ) are determined by
onlt N basic random variables t The variance-covariance matrix for ( ) must
be singular bull
The extension of this observation to industry level variables is straightshy
forward Since observed industry advertising intensity is V bull l v and observed 4
C4 bull z ) it follows thati l i
set of random variables t and cannot be indeshy-
M industries then the variables (V1 v2 at most M independent random variables
Given the assumptions made above y a and o are constant within a
given industry That being the case (34) can be written as
(37)
s1 bull (1 - o) (y - a)
If and
So + +
(1 - o) a + oy
81 are esti mated for an industry using a suitable statisticalSo procedure giving and 81 an estimate of y is then available The sumSo
Intra-industry analysis
1974 of Business Re ort -
categories are identified
Page 19
of So and S1 is y so y bull So + S1 is the estimate of y
If o is assumed to be bounded by zero and one the sign of 61 is also
the sign of y - a The estimate of So is of cou se an estimate of
(1 - o ) a + oy Moreove if y gt a then a is less than 6o- (1 - B) a+ 0(
Cor esponding results would
hold if y bull a or y lt a
If 61 gt 0 then we would estimate that a lt 6o
bull
For purposes of estimating this model 32 manufacturing industry categories
were selected The data are fram the Line of Business forms filed with
the Federal Trade Commission they are discussed in detail in the Annual Line
which is available from the FTC The 32 industry
in Table 1
Simple regressions for equ ation (37) were run for each of the 32 industries
with data for each of the firms which filed in the industry constituting an ob-
servation The results are given in Table 2 lines lA 2A
1974
the intercept
For all the A
equations generalized least quares was used th both and market
share bein scaled bv the square root of sales
The 32 industries were taken from the larger set of all manufacturing industry
categories in the 1974 data files Two selection cr iteria were used thereLB
had to be at least ten reporting companies and the industry had to be a consumer
goods industry or a producer goods industry which is strongly associated with a
related consumer goods industry so that adve rtising by the firms in the producer
goods industry impacts on demand in the related industry (eg soft drink syrup
and soft drinks) There were 135 industries with at least ten companies - 103
producer goods and 32 consumer goods
273
274
TABLE 1 INDUSTRY CATEGORIES USED
FTC
COOE OESCRI P T I ON _
2001 MEAT PACKINGt SAUSAGE S ANO OTHER PREPARED
MEAT PROOUCTS
2004 DAIRY PRODUCTS EXC F LUID MILK
2005 CANNED SPECIALTIES
2001 FROZEN SPECIALTIES
OEHYORATFOt AND
I NC LU 0 ING PRE SERVE S
SOUP MIXESbull
SEASO lNGSt ANO
OTHER PET FOOJ
RAIN MILL PROOUCTSt
ANO PREPARED
RELATED PRODUCTS
PRODUCTS
Z008 CANNEOt ORIEOt PICKLED FRUITS
ANO VEGET ABLES JAMS JELLIES DEHYDRATED I G TA3t E
S40CES ANO SALAD ORESSINGS
2010 DOG CAT A O
2012 FlOUR t OTHER RICE
MILLING fiLENOE) FLOUR
2014 B EADt CAKE ANO
20113 CO FECTIONERY
2026 BOTTLED A O CANNED SOFT DRINKS 2021 FLAVORING EXTRACTS AND SYRUPS EC
2029 MISC FOOOS AND KINDRED PROJUCTS EXC
ROASTED COFFEE
2301 MENS AND BOYS SUITS AND COATS
2302 MENS ANO BOYS FURNISHINGS
230 3 kOMENS AND MISSES OUTER EAR
2551 HOUSEHOLD FURNITURE
2702 PERIODICALS
Pqe 19a 1
RELATED 1972
src cooe
2011 f 3
202tX2026
2032
203pound
2033t4t5
2047 2041t4t5
2051 2065 2086
2081
209XZ05
231
232
233
251 272
2703 BOOKS
2704 ISC PVeLISHING
2806 ORGANIC FlEERS
2807 DRUG$ ETHICAL
23234
PT 2 S 3
2808 ORUCSt PROPRIETARY
2809 PERFU ES t COSMETICS ANO OTHER TaiLET
PREPARATIONS
2810 SOAP AND OTHER CLEANING PREPARATIONS 2815
2901
PEST1CIDES AND AGR ICULTURAL
PE OLEUM REF I N ING CHEMICALS EC
3606 HOUSEHOLD COOKING EQUIPMENT
3612 HOUSEHOLD APPLIANC ES NEC INCLUJING ELECshy
rqlC HOUSE ARFS AND FANS ANO SE I G MACHINES
3617 RADIO AND T V RECE I VING SETS 3808 PHOTOG APHIC E UIP amp SUP LfESt EXC
3qo3
PHO TOCOPYING
SPOR TING A D E QU I P M ENT f SUP P L I E S ATHLETTC GOODS NEC
3904 DOLLS GAMESt TOYS AND CHLOqENtS VE ICLES
PT283
2844
234tX2844
287 9
291 3631
3634t6t9
36H
PT386l
3949
15
943
middot
l A6l f 2 R(GRESSION RESULTS
EQ I NO
NR CODE cu 121
lA 2001 A (
2A 2004 r (
3A 2005 a
-R
ClS amp31
16
17
1 z
I TER-CEPT bull t
ooo9t21At 00022311 oooeos uu
OOl49tCAI 0010 2 middot 000960CCI
o049tl1At o04t55C61
MARKET SH ARE bull 5 I
-010670Ctl
-017600181
-001050
-0211AO
-0077 0
AOVER-TIS INti
lbl
000()39 000072
000018 0 0024liAI
-000079
SHARf
4ampgtVf Q bull
111
000140
-000170
RSQ
COl
66 485 862
607 6 3 t ()2
blti6 633
flASTI-
CI T Y PH
-007271(1
000441
-u0288l
clASTSHU (t)Ef
bull 10
tto
-234
210
( OOlltOCCI -126520 000922 181 001790
4A 200 7 18 00590liAI -038670 4e16 -032110 lell u c
OOOS74e 000700 025940
00067018) 0021001AI -02l100CAI
5 26
920
5A 6 c
200 8 26 0 03Sl9CA I 0012 26181 o 024511(
-005540
-0 81420 0001961AI O()Ol90CAI ooll240
441 604 891
-0020 19 274
6A 2010 16 Oe04224CAI o359lOIAI 9oq ll4015dCAI 061 6 -middot 005435CAI 00011 8 1AI 917 c 00535ltAI -009850((1 OOOA51J CAI -001610161 971
7A 2012 001862 000010 351 00187) oo1 a oooq45 ono1o6 439 c 001109 -027050 Oe0067SIAI -OOll60C81 897
- middotmiddot- -
8A 201t ll 000757 Ol4140CCI 814 ott9oocet 095 8 oooqo6 C C J 000069(81 811 c ootl2lampBI -080910CCI Oe009lUBI -OOtlOOCCI 960I
I DIIJIT pound11 1V OTIUTVTfAll-1pound1t- A _ 00 R - qq r - Q
1jl)OQCD
0
-middot middot--
l
_ __ _
I
i
0 00141 c c
91)7
I
T ABLE 2 REG ESSIOH RESULTS CCONTI
EO NR
IND CODE
iR bull
cos
I TEq-
CEPT
t1ARKE T
SHA E
AOVER-
Tl SING
SHARpound-cent
AuVER
RSO ILAST1-
C I T Y ELAST
SH CPtf (lt Cl) bull 11 I 41 I 5I (61 171 I 8 I f9) c 10)
9A 2 0 1 8 19 002397181 -007640 367 -005 246 le46 6 000605 000550CAI 597
-000029 G14180 OOl654CAJ -024li01At
lOA 2026 12 0028901AI -OlZl90 -0114 0 1 1 1 5 759
Ll (
llA 2021
f3 c
12A lOZ9
B
(
13A 2301
B
c
14A 2302
B
c
154 201
n (
16A 2551
8
10
35
10
20
19
24
001897(()
00141 6CCI
Oll9lHCI
010604 middot middot
Oll083
OOJ11JCAt
002059141
001411CIH
-000412
000293(81
000187
001159 (3)
ooo6a4cct
o00792CCI
OOl40dCAt
000006
-000076
OOlltl6fA)
o o0 5 64CCt
-034570
-o1middot o3o
-1659 90
Ol51lO
-046190CAt
Ol5450CA)
-005l60
002940
-030060(()
-0 43l50CCt
0 33650
-019080
00004
bullJ00741CAI
-000069
000466
0001 51CAt
000947CAt
O OOI)70CAt OOl750CA J
000175CCI
000671CAI
OOll5) f A t 002207CBI
0004b7CA I
- 007 1 70
001760
-001410CAt
- 0 16200CAt
-004210
- 1 207 70
705
96 0
542
466 805
460
614
921
832
9 61
996
424
550
9 15
56 1 651
884
-006114
019 037
03504lCA)
oo409Q
-141644((
491 -0 17664
607
871
lell
084
101
072
103
108 11IllOQfl)-- l
-037880fCJ OOl6l4CAJ -040l70fCI 0()
c
I
I
bull 9
9t
middot middot middotbullmiddot bull lraquo
TABLE 2 REGRE SION RESUlTS ICONTt
bull
EO I NO tHl I filER- f4A_K T AOVER- sbulluRE ttSQ flA Jl- iLAS T NR CODE cos CEPT SHARE JISING AOVfR bull (JTY SHR CiJCF
110Cl I f amp Itt 15J tbt bull 1c u caa
llA 2702 ll o079 6 81 Bt -019110 S6s -OlltO 111
a 004679 ooooo1 t69
( oOl26o -015160 001560 -022690 ltJr()I
ldA 270 3 19 0065lt -OIt2l60 220 -Ol57a2 11 8 e o 0 71t6IA t ol08 75tAt ( OOl983CiU -060llOC() OOl7191A t -O lltdOC A) middot 89
lqA 2704 10 005082 -Oil060 22b -l)i1977 lelO 6 -000080 oo2772C4 t 910 c 000907 -011 200 Oe04105CA t -046960 995
20A 2d06 1 II
O OlltStiAt -001120 837 00014 -))th O01380amp At -oooo 11 821 ( OOl526CAt -026700181 0002 5UAt 000530 969
2lA 2807 28 OOt9lt9CAt -00891 0 516 -0 03962 225I8 002095CCt OOOl lliAt 690 ( o02965CA t -056650((1 OOOt50CAJ -003l70ttl 911I
I 22A 2808 15 ot3l06 119730 655 1130) 090
0 - 007526CC t Oe003961At 835- middot-
c I oo7523 -067210 001l 2tA t -0091t80CCt 951
2lA 2809 21 Oe2067lCAt -1218101At 71 2 -10095((1 121 ttlIaOQ8 00797418) OOOI8lltt 686 Ill( 0092lU A t -Olltl70Ct t 000605CAt -OOit8701At 951 _
---- -
32 009506CAt
0 bull 06 981 I A J 006901 cA bull
00 24A 2810 -010960 5t 5 -OOltSO 7 56
8 -000011 s11 c -5811701At oooaoatAJ oot520IAI 888
-- 37 --- middot - -- -1 ) l lbull t5JJ r-l bullhmiddot1 --- - --
l5A
26A
27A
c
2RA
c
2 9A
31A
llA
24
13
597
Ci)55
14
l 6h
RSQ ELASTI- (lASTCOE f-
TABLE 2
EQ I NO NR CODE IU
REGRESSION RESULTS CCONTa
NR INTER- MARKET
(10S CEPT SHARE AOVER- SHARfcent
TIS lNG AOVER
) CITY StiR
c It) bull 5 co a I 1) bull 0 bull 9 c l 0 bull( 2 1
2815 16 003019 -009990 275 -o069bmiddotJ
6 001446 000224 282 c 002178 -054630 o02221CCI -0075 00 646
2901 29 00021t2 CAI 0)0520 592 0007 6 1 068
0 00105C8 t 000012141 60)g( OOOlJ51AJ -00126018) ooou271At -000u80CA) 974
36oa 1 3 003296(4) -oca29o -oo-q94
001755 000203 lt91t 002499((1 -04l820CCI OOl374CAI 0068)0 946
3612 0033281CI 011520 195 01684) oso
00 30 42C OI 00012 2 424 oo579ltAt -2 7l7801At 002 100IAI -003220 819
8 3617 btel 011789 081002259 009530
001448 oOOl OitCCI 732 OOl7 22CAI -0 46160CAI 0004961AI -OOll40181 989c
30A 3808 14 0021t 3 6fAI -000110 628 o023291C) -005 6 middot- _ middot- 0bull 02 04 1 IC I ooooo9 637 ( 001584((1 -0161)1)0 00021 91AI -0 00060
390) 14 002790CCt 006590 h8 t 0 0938 4 070
B OOl89bCCJ 000144 74
0021961(1 -OItlOlO OOl323CAI -004800 915
B
middot--- - middot 4 - middot- - dfll3904 12 004563181 0)8300 879 0428651(1 089 OQIll004 277CA I ooozstcea 919
c 0041)3181 -0-11250 0011881(1 -006270 972 0Ill
- J - bull bull bull bullbullbull middot - - middot - middot gt middotbull L)ofbullflbull middot middotfJltl -tn- middotmiddotmiddotmiddoto +f middot t middot bull
c
is
positive
2
Page 0 determineThe consumer goodproducer good split is sometimes difficult to
Sollle support for the split used in this paper is a comparison of goodness - of - fit
distributions of a
industries
set
advertising are
that industry
a sales increase of
the
highest aD10U3 those
measures for the two groups of industries Table 3 contains
measures for the 32- consumer goods industries and the 103 producer goods
2 Median R is 5 7 for the consumer goods set and 40 for the producer goods
Estimates - of y the elasticity of demand with respect to
given in column of Table 2 The hypothesis that y bull 0 with
alternative was applied for each ittdustry As shown in the
could not be rtjected for 24 of the 32 industries Of the
and 10 are positive For the remalnifg eight three are
and five are significantly positive
Of the threertegative elasetcities the largest (in absolute value)
cosmetics (code 2809) With all elasticity of -1 01 a one
industry advertising would gwerate a decrease in industry sales
more than one percetu
At the other extrema is toys and games (code 39 04) For
a one percent ittcrease in industry advertising would lead to
four-tenths of a pampreentmiddot Though not significantly different from zero
elasticity estimatbull for pt-bpuetary drugs (code 2808)
which are
Both drug industries have
to sales rafios (211 pe1 cent for 28 08 and 13 4 per cent
other hand flaIJo ing ext1ac s (mainly soft-drink SYUlS) also
is for
percent increase in
of slightly
a middottwo - tailed
table that hypothesis
24 14 are negative
significantly negative
9
the codetics and p-ropriacuy very high advertising
for 28 09) bull On the
has a high ind UY
advenisiltg t() sales ratio ( 8 0 per cent) bull but its estimated elasticity is very
small and not signifieantly different uom zero
middot It doe$ not seem unusual to find many elasticities near zero with a few
negative and a few positive If aggregate consumption is insensitive to aggregate
advertising as seems likely ld then what one industry gains must be lost by
others
Page 21
elasticity of demand does
(10) we can
if t 0 then
Finally a determination concerning the industry
not fix the individual firms elasticity Going back to equation
write y bull (1 - ) a + z y Since a gt 0 it follows thatii i i
Y gt 0 If Ylt 0 however y s sign is indeterminateij ij
Turning next to the coefficient of market share 20 of them are negative and
12 are positive with three of each sign being significantly different from zero
The largest positive value which is significant is for pet food (code 2010) and
a close second is mens amp boys suits and coats (code 23 01) The largest signifi
cant negative value is for cosmetics (code 28 09)
In the model some relations among y a 0 and 61 are implied If middota is
negative (1- y) is greater than one Since 61Y bull (1- a) (1 - ) 61Y is
pCsitive but it will be less than one if 0 is large enough Seventeen of the A
ys are negative in each of the seventeen industries 61 is also negative and les
1 (algebraicly) than y so 61Y is greater than one his is not evidence that the
degree of cooperation is low or zero in any industry of course A high o togethegt
with a negative y which is small (in absolute terms) relative to a can give a
value of 61Y greater than one On the other hand if 61Y were less than one
o would have to be greater than zero
For y gt 0 two conditions may hold If y lt a it follows that 61 lt 0 since
(y - a ) lt 0 It also follows that 61Y lt 0 regardless of the value of o bull
In addition the relation is an if-and-only-if one ie if 61Y is negative
then y lt a Three industries have negative values for 61y dairy products
except milk (code 20 04) synthetic fibers (code 2806) and photographic equipment
(code 3808) In all three cases y is quite small so a does not have to be very
large to give a negative value to al
If y gt a the model shows that 61-- gt 0 whatever the value of 0 bull There are
12 cases where both y and 61 are positive with 61 ranging from near zero
(grain milling code 20 12) to 1 20 ( proprietary drugs code 2808) Several
coeffiei-en_rs
--
15
o f to zero
Tab le 3 - Consumer Goods
Indus tr ies
2Distributions of R and Producer Goods
Page 20a
Consumer Producer
Goods Goods
0 - 9 9 0 3
10 - 19 9 0 1 1
20 - 29 9 3 20
30 - 39 9 3 17
40 - 49 9 5 16
50 - 59 9 8
69 9 6 Ii 1060
--f 670 79 9 2
80 - 89 9 4 4
90 - 100 1 1
Totals 32 10 3
Notes
2is used R i s cal-
culated as F F + ((N - K- 1 ) K ] 1 b_ F is the F
statistic Lor the hypothesis that all o f the non - intercept terms are simul taneously equal
1
2
--
-- Since generalized least squ-area regress ion
Page 12
o f these cases have values o f 6 1 Y rea ter than 0 8 (pet food (20 10) bread amp cakes
(20 14 ) miscellaneous processed foods (20 29 ) men s and boys suit s and coats
(23 01 ) p rop riet ary drugs (28 08 ) elect ric housewares (36 1 2) radio and
IV sets 17 and games and toys (39 For be than(36 ) 04)) 6 1 Y to greater 0 8
both o and cry mus t b e less than 0 2 That is in these industries the results
shown in Tab le 2 would hold only with very strong dominance of marke t demand
effects over brand swit ching e ffects and a very low degree o f cooperation
Using the concentration rat ios given in Table 4 the average concentration
ratio for the eight indust ries is 38 1 For the remaining 24 industries it is
43 2 This result would lend some support t o the proposition that concentrat ion
and cooperation are related if the group of eight industries are identi fied
as having low degrees of cooperation
One f inal note on model predict ions S ince both cr and o are assumed to
be non-negative 60 should also be non-negat ive In only one o f the 32 cas es
is 60 negative and it is not signif icant bull
cor pJ
7 1
7 0 4
1 4 7 5 1
)
7 3 z a o a t s 1 1 3 3 5 ( i V Z 8 tJ1 2 1 - 1 1 0 7 3 5 c middot v 2 8 1 0 3 2 -0 0 1 4 3 0 C 1 V 9 8 3 2 8 1 5 1 6 - 0 0 6 t1 6 9 C _ V 1 11 0 2 9 0 1 29 0 00 7 6 1 C l k V 7 3 1 3 6 08 1 3 - o n 4 9 9 4 3 3 3 6 1 2 2 4 0 1 6 4 3
3 1 7 1 3 0 1 1 7 d 7 8 3 3 8 0 8 0 0 2 3 i 2 3 0 6
P 4
3 9 0 3 1 4
1 2
0 09 3 8 4 39 1)1 0 4 l 8 6 5
T A LE 4 e I NCU S T R Y C A T E G O R Y
F TC NOM e tOOE ass
C 1 I c 2 )
2 0 0 1 1 6 2 0 0r 1 7 zo o l Z 2 0 0 7 1 8
z o o 2 6
z o t a 1 6
2 1) 1 2 1 5 2 0 1 4 1 1 2 0 1 a 1 9 2 0 26 1 2 2 0 2 7 1 0 2 0 29 - 3 5
2 3 0 1 10
2 3 02 z o
2 3 03 1 9
2 5 5 1 2 4 Z 1 0 Z 1 1 2 0 3 1 9 2 7 4 1 0 z a o 1 3
2 A 0 7 z s
E LA S T I -C I T Y
c 3 )
- 0 09 7 2 7
0 00 449
- 0 0 2 8 9 3
- 0 3 2 7 70 - 0 OZ il I q
o 4 u t s a 0 0 1 3 7 3
o 1 4 o o-0 0 5 2 4o -0 1 94 0 1
-C l 6 l l 4 0 1 JJ 3 7 0 3 t 0 4 2 0 0 4 0 9
- 0 4 1 8 44 - O l 7 t ) 4 -ll 7 1 4 0 b -0 3 5 7 8 Z
- 0 1 9 7 7 0 1)0 3 3 4
-o n 9 2
IJ A T A
CONV
ON-C ONV
( 4 )
t 1NV
C 1t-V C NV c ) v C - JNV C rlNV c n v c a v C ON V ( i) l V
C CtV
C t1tV C IJNV
C 1 V
c 1 v
Pa ge 2La
HEQ F t Noe x
c 5 )
6 2q
6 89 1 48
6 1 4 ) q 3
1 9 7 1 9
q z t s 5 a e z
l l 9 1
4 o 7 1 2 0
c oqt 1 Z
4 4 )-q 1 z
5 5 9 1 0 64 1 3 3
n 3
Inter industry analysi
Page 23
Equation ( 3 6 ) may be used to fo rmulate a rough test o f the cooperat ion
vs e f ficiency controversy I f we rest ricted ourselves to some subset o f
indus tries where the relat ion be tween y and a i s the same we would be lookingbull
at a sample o f indus tries for which the dependence o f V on H and o is a
quadratic wi th no squared terms
I f we now sub s titute (26) into ( 36 ) we get
(38) v bull a + (y - a) R + (y - a) o (H) - (y - a) H o (H) + u
If it is only d if ferences in marke t shares that affect advertising inten-
s i ty levels i f concentration does not af fect cooperation and i f o bull 0
( 38 ) reduces t o
( 39 ) V bull a + (y - a) H + U
Let this b e the null hyp o thes is and note that V is a linear func-
tion o f H If y gt a and the null hyp othesis holds the expected relat ion
between V and H is as shown by the pos itively sloped straight line in Figure
6 b etween points A and C
In this context the alterna ive hypothesis the degree o f
cooperation ( o) is a positive function o f concentrat ion
is that
(H) I f the func-
tion is as described in (26) and Figure 4 the relation between V and H
would be as shown by the curved line connec t ing po ints A and C in Figure 6 One way to test the null hypothe s is against this al ternative is to test fo r
the hypothesized curvature in the function relating V and H
Fi gure
R
6
v
ID c
H 1 0
I
i I I
sfigned
clasJ s
in def ini g
effec t is the
I
Page middot2lf
One charac teristic of produc ts which may be useful
classes wi th respec t to significanc e of the brand-switching
conveniencenon-convenienc e distinc tion developed by Porter ( 1 97 )
Using his identif ication of IRS indus tries as a s tar t ing point
each of the 32 industries used in this s tudy to one of the two
The ass ignments are not ed in Tab le 4
Using the conveniencenon-convenienc e dis t inc tion the
divided into two sub-samples
sample was
Four regressions were then run bull ene for e ch
sub-sample one for the to tal sample and one for the total sample with
a dummy variable for convenience goods Each regression was run wi th LB s
and wi th indus tries as ob servations The results are given in Tab le 5
For the LB level equations ( 1A - 1D) equation(37) was used ome
rearranging of terms gives v bull y - ( 1-o ) (y - a) ( 1-z ) + ui i i
bull y - 6 1 ( 1 - z ) + ui When observations are pooled across indus trif s yi
is an industry variable In these regressions the values o f y es t imated
in the first stage were used Since industries have been grouped so as
to reduce differences in (y - a) that term is t reated as a cons tant
within each group Finally H the Herfindahl index has been subst ituted
for o the degree of cooperation The resultant equation is linear
y ( 1 - z ) and H x ( l - z ) This is the equation for whi ch results arei i
reported in Table 5 In all regress ions all variables were weighted by the
square roo t of sales
in
raquo
- 34
3 1
i l T 4 8LE S R E G E S S I ON R E SUL T S i X T f N S I ON S T O T Y PE Of
AND L E V E L OF C ONC E kT R A T I O
E O O tt S
tP bull T Y P E NR C ONV O B S tO N-CONV
E L A S T I shy
T I C J T Y
O E - t-41( T SliA R E
H( fl F
1 1 - K O x middot K O X E R F
- H S a 1 1 - H S t 1 8 1 1 9 1
R S O
1 1 0 1
q z
t - bull 9 f 2 1
Z t P D l 6 fC ONV - 1 9 1 s s c 2 1
e - l O I e bull u q a
2 3 1 0 1 1 Nl N- t O V 5 7 1 ( 1 2 19 1 6 1 - J l S 2 5 4
l l l l t z q o a 1 - J o v z I
l ( I NO l l A l l ol u l t o o c bull J u l l
d S
c 3 4 t t r a l
bull Z D 1 0 l l 4 l l 2 3 -gt 2 l l 2 0 - 4 lS 5 4 l C 4 1 o e 6 t l b b 5 1 1 0 1 1 1 c - 1 9 5 1 t d S 2 1 tV
OQ10 T-R A T I O S A E G I VE N l P A RE N T H E S E S l E V E L S f S I G 1 F l C ANC E A - 99 5 B - q 9 t - s NS J UC E NO I Nl E R C ( P T I S U (j f middot) I N f Hf t OUA TIOI S 6 E f 3R E A O J U S T H E N T F OR 1-t E T E R O SIC E O A S T f C I T V pIraquo
1
middot
I
R SO I S C A L C UL A T E D A S f I f bull 1 1 bull 1lt 1 K t t- E E N I S THf U A1 B E R O f fJe S E fi V A T I O S F t S HtE F - S T A T I S
I N DE P E ND E N T y II
T I C F O T H E U Y P O HIE S I S f A l l C JE F F I C H N T A R E QU A L HI l f RO A f40 K I S T H f UUM 8 E R OF Q ( A fL f S
I l l
1 A
1 6
t C
f 1 r R I l l
1 2 1
L Jt
L B
l q
L 1
on s T Y P E bull 2 )
P OUC T
X
H S I l l
l O l A I bull l 0 7 1
O O b O J
1 2 00 1
1 b 2 1 C I l l Z U
l l S b 1 -1 1 0 1
rlt F X l l - fi C f l
bull 1 I
l l
- ()4 2 3 l f H
1 - 2 7 9 1
K O X
1 1 - ti E R F I
l d l
S4 l i A I 2 l b
1 3 00
K O X -t f R F t S J
I 1 - HER F t
)
60 9
z o bull
0
t l l
3 8 8
1 7 6
5 64
64
N f bull
tl S bull
I l l
I 4 1
t ONV
NON-t ONV
A l l
A ll
c o v NO- ( ONV
1 4 1
1 5 1
- l S b
c - 1 1 2 1
5 8 2 C A I 1 6 7 0 1
middot I l 9 H
l 1 il
1 4 1 6 1
E L A C) T J shy
l t C I T Y
I S I
l b l
- 0 2 0
1 - 1 4 9 1
O Z 3 2 1 A I C b 6 b l
0 0 9 2 -J
1 1 b 4 1
o z t t t l 1 2 2 1 1
oru - Ht tt F
1 gt X bull b I
Page 25 bull
Mean val -as ior relevan t variables for the two sub-samples are
Conv Non-Conv
Adv Sales 4 7 2 3
iit Share 2 7 3 2
Ela cicity - 05 3 02 2
Herf indahl 7 3 1 860
That the dist nct ion between coft9enienee goods and non - convenience
goods mat ters is supported not only by these means but also by the regression
results Equat ions lA and lB are different in almost every respec t Only the
concentrat ionmarket share int eract ion term seems to mat ter for convenience
good s Jus t he opposite holds for non - convenience goods both elasticity1
and market sl e are highly significant but the concentrat ionmarket share L
int erac t ion t=rm is not
Given that one minus market share is used these equ3 tions the coeffici t
of market share is the negative of what is sh in col (6) of Table 5 For the
regress ion for non - convenience LB s reported in equation lB then LB s with larger
market shares have lower advertising to sales ratios
On the question of the impact of the degree of cooperation the story is mixed
For convenienc good s the coefficient is pos itive as equation (31) predicts when
there is a po sitive degree of cooperat ion Fo r the non-convenience goods LB s middot
concentration has no effec t
Aggregat ion t o the indus try level and then rer unning the equations $ives the
results shown in (2a-2d) of Tab le 5 As expected R2 goes up and levels of significanc e
fmiddotJr individual coefficients gJ down In add ition the coeffic ient of H ( l -H) inmiddot
quat ion 2A is no t significant though it is po s i tive Since 2A is the indus try level
I
--
all but one of them and that one is not significant For 18 of the 31
wo uld have on its adve rtising to sal es ratio
cognate of LA the variable H ( l-H) pl ays the same role in 2A as H ( l -z 1) p lays
in LA
Vconomies of scale in advert ising
The literature on advert ing contains a number of definitionbull o f economies
of scale and of findings eone rning its presence and magnitude Before I loo k
a t the data with the hope o f mmenting about scale effect s then I wan t to
define the term as I use it
By economies of scale I mean a decrease in the number o f units of adver-
tising which are needed to generate a unit of sales as advert ising is increased
Given the definition of the firm s own elast icity of demand with respect to
advertising (see the text foll wing equation (1) ) the technical definition
is that rii gt 1 Diseconomies of scale are def ined symmetrically i e
Yu lt 1
ith a cro s s - section oi observations on advertising intensity
(v bull a s1) and onmiddot advertising (a ) we may regress the former on the latter 1 i i
The results of such regress ions for the 32 industries are given in Table 2
lines lB ZB bull bull bull 32B For the 32 eases the regression coefficient is positive it
positive coefficients the coefficient i s signif icant
The que tion now is whether these results ar e evidence of pervasive dis-
economies of scale in advertislng I think no t Given my as sumption that the
data I observe are equilibrium results I may ri ghtly conclud e that almost every-
where high levels of advert ising are as sociated in equilibrium with high ratios
of advertising to sales I may no t conclud e anything however about the
impact that a move by some firm away from its equilibrium level o f advertising
middot- Page 27
Given the assumed relation in equation ( 37 ) there is a corre spondshy
ing equilibrium equation relating v and ai The equation is non - linear i
and I have not found a simp le way to characterize it It is po ssib le to
det ermi ne its deriviative with respect to a b y takin$ tne de riviative of ( 3 7 ) i
When that is done the result is
v - 1(40) i bull sl a a z ( 1-z )i i ira-l
Sinc e all the other terms in (40) are non-negative the sign dep ends only
on the sign of 6 1 which is in turn de pend ent only on the sign of (y - a)
That the signs of the coefficients of a in equations lB 2B bull bull bull 32B arei
not always equal to the signs of the coefficients of z in LA 2A bull bull bull 32A i
is still a mystery to me and something to be explored
bull
Summary
My purpose in writing this pa per wa s t o attempt to integrate explicit micro-
economic mode lling reasoned econome tric specification of error terms and high
quality data to exp lore some questions about adver tising by large firms
al quite important improvements could be made in the first two of those areas
and much data ma ssag ing is yet to be done in the third
Concerning the subst an ce of what I have done I think six things are im-
Page 2 8
Severshy
V and Conclusions
portant
1 Explicit modeling is worth the ef for t if for no other reas on than
that it provides a basis for choosing fr om a varie ty of funct ional
forms
2 The same is true for error specif ica t ion I did in fact look at scatshy
ter diagrams for all 32 industry categor ies and they vir tually all show
the het eroscedast icity which I as sumed and for which I correc ted
3 The predicted relation be tween advertis ing in tensity and market share
shows up clearly in on ly 25 of the case s examined I have ye t to
explore why that may b e the cas e
4 Some evidence concerning the presence and impact of coopera tion was
produced but it is not clear ly pervasive
5 The dis tinction between convenience goods and non-c onvenie nce goods is
unambiguously a good one
6 Virtually no evidence concerning econ omies of scale in adver tis ing can
be gleaned fr om this study given its assumpt ions
More work is called for on many of these issues my study seems to have
raised more questions than it has answer ed This is pr obab lly due to the richshy
ne ss of the data source since I had the oppor tunity to address several issue s
at once
- -- - Q --
Foo tnotes
Manager Line o f Business Program Federal Trade Commission The author depended heavily on several staf f members of the LB Pro gram for statistical and clerical support in particular Joe Cholka George Pascoe and Haro lene Jenkins Helpful comments were given b y Richard Caves Dennis Mueller Michael Porter and F M Scherer When the paper was in its early stages o f development both S teve Garber and Jon Rasmussen former economists in the LB Program provided eery useful cri tiques The views expressed here are my own o f course
1 The basic exp licit model of the role of advertising is Dorfman and S teiner ( 1954 ) and the models in this paper may be seen as ano ther extens ion of their work
2 For a review o f models of oligopoly which treat prices and quantities as relevant variables see Shubik ( 1 959) Extens ions to take advertis ing int o account are fairly s traightforward
3 Dorfman and S teiner dealt with a price-setting firm For a monopolis t inverting the demand relation to show Pbull P (Q A) _instead o f Q- (P A) has no effect on the results o f the analysis For oligopolists who face a sys tem of demand relations inversion does have some impac t on the results The dif ferences are not trivial nonetheless they will no t be explored in this study
4 Shubik ( 1 959) p
5 From a formal mathematical point o f view the function is similar to a1 i Legrangian funct ion of the form - r rr
L bull 11 + A ( j - (i i i+j
where the indicates some fixed level of the profits industry Following the same logic which resul ts in and income constraint as the marginal utility of the of the indus tr in the obj ective of firm i
6 Hall amp Weiss ( 1 967) p 323
7 See Schmallensee ( 1 97 2 ) Ch 2 pp 1 6-47
8 On this poin t see Imel amp Helmberger ( 1 9 7 1 )
9 Long ( 1 970) p 130 - 1 39
1 0 See Schmallensee ( 1 972) Ch 3 pp 48-8 7
rrj i+j ) of the rest of the
the identification profits of the rest
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980
0 (AA)
lx 2x 1x
ith
1y
i
First assume no
that if there are
will be no changes in market shares
- 0a bull
Page 8
Equation (17) is a general statement of the relation betbulleen changes in
advertising levels and changes in market shares which would acompany them It
may be used to make some general observations about brand-switching effects
industry effect eg y bull r t bull 0 and assume further
equal percentage changes in all advertising le =ls then there
Under these further assumptions
0l and dz -1 bull (iA) ltr assumption the second term on the right hand side of this expression is
- By
zero
necessary for rt bull 0Consequently for market shares to be unaffected it is
when y bull r l bull 0 This requirement is satisfied by equation (11)
Secondly consider effects on market shares if one firm inc ases its abull ar-
tising and others do not As a general assumption we require smiddot ch a change o
lead to a decrease in the shares of all the other firms hat is let
0 0aiai gt 0 and a a bull 0 for j + i For any matrix X with N rows and its trans-j j
pose X let X
row of X
bull middotmiddotmiddot where is a column vector which contains the
elements of the Under the assumptions about the iis given just
d-1 0 0above it follows that r becomes (aiai) iimiddot ehFor jzj lt 0 under these conditions then it follows that the j- element
N of (I- l ) lt 0 so middotij - il lt 0 or yij - yik lt 0 If this
kbulll N condition holds for all Y ij where j + i then it follows that yii - z
k Y gta
k- 1 ik since the th firms share must increase to offset the decreases in all other
shares
ei ith
1v ( ) h i
7urning
where is the
leads to z
to the r matrix given in (11) we note that 1 = cr + (y - cr) i i zi elementary vector with 1 in the position and 0 elsewhere This
bull cr + (y - cr) zi Since bull required is thatcr at- zY wYij (y - cr)zi - cr + (y - cr) zilt 0 or cr gt 0
These observations may be summarized by substituting (11) into (17) bull
giving
(18)
i s
d -1 zz -
-
-
+ (y - cr) tz
d -1 0cr(I - l ) a a bull
cr 1 z - (y - cr) 0 a
o
If ai increases and the other a s are const t cr must be positive to havej
a positive effect on zi A positive cr also bull plies a negative ffect of aJ
f all a s in-on zi other a s (including a ) fixed j
eluding a1 change by the same percentage zi remains unchanged Finally k i
if cr bull 0 there is no brand-switching effect
For a to have no effect on p requires yij bull 0 From (10) thisj i
amounts to y bull cr that is the industry demand elasticity and the brandshyswitching elasticity are equal If y gt cr gt 0 the industry elasticity yij doninates If y lt cr y lt 0 the brand-switching elasticity dominates ij
making
Page 10
Substituting ( 1 1) now in (6) the Cournot case and in (9) the
Chamberlin ease gives
-(19) Cournot vO a + (y - a) z
Chamberli n vl - y lmiddot th
The 1- element is then
-(20) Cournot vO a+ (y - a) zi i
-Chamberlin vl yi
In (19) and (20) the superscript 0 is used to denote total
independence of decision and the superscript l is used to denote
total interdependence of decision making
The relation between v and vf depends solely on t relation between
y and a f y gt a so that the industry effect dominat 3 vf v with
equality in the trirlal case of z bull 1 Cooperative irms would have ai
higher advertising intensity than non-cooperative fi
If there is equality between the industry elasticity and the
y Whe therbrand-switching elasticity or y bull there is cooperation among the firms has no effect on advertising inten-sity And if brand-switching is more important with y lt a e get
vi lt v cooperative firms will have a lower advertising intensity than
non-cooperative firms The three situations are depicted in Figure L
Given the results for the firm level variables calculation of the
corresponding re sults for the industry is s-traightforward Let
B bull t zf bull z z be the Herfindahl index of concen tration and notethat
z - bull l z bull 1 and that V bull A S bull a i 5 bull (siIS) (asi)i
bull middot Then for the general case from (6) and (9)
bull
Page lOa
y lt1 bull y y
10
f
V0 bull z [ a1 + (y - cr) J bull a + (y - a) H vl bull Y bull y
11Page
( 2 1) vo bull z d -1 r bull r z bull z- d d- i )11
l -1 N N v bull z d r z bull r z bull l y z - ij jibulll jbulll
and for the simplified case from (19)
(22
Since the forms of the equations in (22) are the same as in (20) all
of the analysis of (20) holdS for (22) except that His substituted for
middot z Advertising intensity will be higher for an industry with cooperativei
firms than for one with non-coope rative firms if the industry demand elas-
ticity dominates the brand-switchine lasticity If the ewo elasticities
are equal then so will be advertiSl intensity for the cooperative firm
and non-cooperative firm industries If brand-switching dominates the
cooperative firm industry will have lower advertising intensity
In the industry model context the possibility of a critical concen-
tration level may be easily introduced That is assume that at values
of R below R the firms are non-cooperative but that for alues equal to
or greater than H they are cooperative That is
(23) V bull a + (y - a) H if H lt H bull y if H gtmiddot H
This function is shown in Figure 2 for the three relamptions be eween y and a
IV THE D GRE OF COOEtRATION
The models depicted in Figure 2 are based on a very simple notion about
the relation between concentration and the cooperationnon-cooperation charac-
teristic They use what is essentilly a step function relating concentra-
Figure
I -
J H H
Page l la
2
v v v
a
I J I y a bull Y y ---i
I I I I I I II I
H 10 H H 10 H 1 0
bullbull
bull bull
Page 12
tion and the extent or degree of coo peration as shown in Figure 3 Using
o as a symbol for the degree of cooperation eq uation (23) can be rewritten
as
bull(24) v (1 - o) vo + ovl bull (1 o) [a+ - a) H ] + oy- (y
vhere
(25) o bull o(H) bull 0 if B lt B
Hbull 1 if B gt
The specific relation between concentration (H) and the deg ree of
cooperation ( o) shown in Figure 3 is more restrictive than necessary and
bas less appeal an one which shows a smooth increase of o from 0 to 1 as
B increases fro to 1 Such a smooth function is shown in Figure 4 he
degree of cooper ion is low until H gets close to H increases greatly as
B goes through R attaining a high level for an H larger than B but still
close to it and then goes to 1 as B goes to 1 More formally
(26) 0 bull o(H)
bull 0 if H bull 0
bull 1 if H bull 1
do gt (IdB
d2o- gt 0 if H B dBZ
0 if H H
lt 0 if R gt H-
If (26) is now substituted for (25) in (24) a corresponding smoothing
of the functions shown in Figure 2 is accomplished he smooth functions
Figure 3
I
10
H 10 H
Page 12a
Figure
Page 12b
4
10
H 10 H
l
(27)
Cl - o) o
[ (1 -
[ (1
t y ij z
j
are given in Figure 5
Page 13
The equation in (24) was constructed as a hybrid of two results derived
from first-order equations for the non-cooperative and cooperative models
It may also be constructed directly by using l bull (1 - o) + on as thei i
th 2objective function for the i-- firm Treating o as a constant the first
order equation is
0 an aa
i middot
n i
Using equations (3) and (7) and setting --- bull 0 givesaai
(28) - 0
After collecting terms and converting to matrix notation we get
(29) - Cl - o) + 0 z-1vi yii i
N
j bulll
+ 0
+ o v1
d-1 r z z
-(30) v Cl - o) rd
-
o) r +d
d r J _ -1- d z
and middot
d) r + 0 rjd
l r z
v - z --(31) zv
- (1 0- +d) t r z- d-
If we now impose the simplified demand structure by substituting (11)
we get
- (1 - o) a +( 32) + oy and
Figure
l
-y
v
Y
H
r
I I
v
a
y
H 10 H
v
a
bulla
HH 10
bullbull
(33)
i
Page 14
v - (1 - o) [ a l + (y - a) ] The corresponding equation for Vis given in (24)
If the relation between industry advertising intensity (V) and
concentration (H) is non - linear via the impact of concentration
on the degree of cooperation does the relation take the form of a
auadratic as some other investigators have proposed (Greer ( 1971 ) Martin (1979) Strickland amp Weiss (1976) ) That question can
be approached from two perspectives - conceptual and empirical In
the next section I will try to apply some statistical material to it
here I want to explore the question in the context of the models
developed above
Referring to equation (24) taking the deriviative with respect
to H and letting bull (H) gives
bull( 3 3a ) 3V (y - a) (1 - o) + (1 - H) 3H dH
Given that o and H each are bounded by 0 and 1 and that o is an in-
creasing function of H the term in the brackets in non-negative The
sign of the deriviative then is the same as the sign of (y - a) If the industry effect is dominant (y gt a) then advertising in-
tensity will increase with concentration If the two effects are equal
then advertising intensity will not vbullry with concentration and if the
brand-switching e ffect is larger advertising intensity will decrease
with concentration
For given advertising elasticity parameters then it is not pos-
sible to show an increase in A S up to some H and the n a decrease
he relation may be non - linear but it is monotonic over the permis-
sible range of H
Page 15
One possibility is to introduce a relation between (y - a) and H
If industries characterized as dominated by the brand-switching effect
also were more concentrated then a cross industry comparison would
show an inverted U shape relation between AS and H I know of no
support for such an association between ( y - a) and H
V ERROR SPECUCATION
So far I have implicitly assumed that there are no errors in the
model I now assume that the error
thvariable in the equation for the i- firm is additive that its mean is zero
that its variance is inversely proportionate to its sales and that all error
term covarlances are zero hat is
34) v bull (1 - o) [ a + (y - a) z ] + oy + u i i i
- - -
In vector notation
(35) bull (1 - o) [ at + lt - a) J +
E() bull 0 Cov (u) -
0
If we now ca lculate V bull bull we get
oy + U bull
(36) V bull (1 - o) la + (y - a)R J +
U bull E(U) bull 0 Var (U) bull
P age 16
The error term specifi c ation which follows Hall
wi th the ratio of profits
6 heir analysis seems
as wellbull particularly given
and Weiss was
applied by them to equa tions to equity or
asse ts as dependen t variables to carry the same
weight in the present context the substantial
evidence presented by marketing analysis on the tendency of company de-
cision makers t o use the advertising to sales ratio as the decision
7variable
The zero covariance assumption on the other hand1 is particularly
troublesome Within a probably not true since
many even ts outside the
single industry it is
industry would have
industries
similar effec ts on all the
8firms in the industry Between it is probably not true either
since many companies produce in more than one indus try and even ts in the
firm would tend to effect all of its activities The extension of the
model to allow for non-zero covariances is certainly called for
--- -
Page Ll
Athird characteristic of this specification also deserves some comment
only on equation is shown for the firm or industry Several haveinvestigators
included an advertising equation ina larger system of equations (Comanor and
Wilson (1974) (1979) and Weiss (1976) so in myMartin Strickland Idid
dissertation1 Ihave ewo defenses for presenting only single equation results
here One is the suggestion by Comanor Wilson that simultaneousand equation
verybias may not be important in this context he second is that Iam will-
ing to see the model extended to include a profitability equation and I intend
to move that direction in the near futurein
Onthe inclusion of aconcentration equation on the other hand I amskep-
tical he problem isthat in some fundamental sense equations in models of
industrial organization should be oriented to the decision-making contexts of firms
Ifeach firm in anindustry with Nfirms has only one decision variable then
there can be only Nindependent equations which are based on the first-order
optimization equations It not show large ofis difficult of course to a number
functional relations which follow from the first-order conditions point here he
isthat only Nof them can be independent
Consider for illustrative purposes only a situation in wich teh N
itsquantities say q are set exogeneously and each firm determines advertising1
of and level say ai where a is the true optimal value a a bulla
1 1 i i + i
Ifp a) then the Nq
s and the Na s will determine N p s say pii bullp
i(l i 1 i
Now given Eand 9 we can determine both the vector s andthe vector v since
- bull til
ands We can also determine S s z bull s-1 s-ii
v and we
industry concentration isHbull (alternatively
V and Hare functions of the same
pendent Furthermore if there are
VM) and (H1 H2 bull bull bull )contain
VI STATISTI CAL APPLICATIONS
Page 18
Given the observed values of might be tempted to conclude that we
have a two equation system in which v and z are jointly determined That would i i
be inappropriate however since the 2N random variables (v ) are determined by
onlt N basic random variables t The variance-covariance matrix for ( ) must
be singular bull
The extension of this observation to industry level variables is straightshy
forward Since observed industry advertising intensity is V bull l v and observed 4
C4 bull z ) it follows thati l i
set of random variables t and cannot be indeshy-
M industries then the variables (V1 v2 at most M independent random variables
Given the assumptions made above y a and o are constant within a
given industry That being the case (34) can be written as
(37)
s1 bull (1 - o) (y - a)
If and
So + +
(1 - o) a + oy
81 are esti mated for an industry using a suitable statisticalSo procedure giving and 81 an estimate of y is then available The sumSo
Intra-industry analysis
1974 of Business Re ort -
categories are identified
Page 19
of So and S1 is y so y bull So + S1 is the estimate of y
If o is assumed to be bounded by zero and one the sign of 61 is also
the sign of y - a The estimate of So is of cou se an estimate of
(1 - o ) a + oy Moreove if y gt a then a is less than 6o- (1 - B) a+ 0(
Cor esponding results would
hold if y bull a or y lt a
If 61 gt 0 then we would estimate that a lt 6o
bull
For purposes of estimating this model 32 manufacturing industry categories
were selected The data are fram the Line of Business forms filed with
the Federal Trade Commission they are discussed in detail in the Annual Line
which is available from the FTC The 32 industry
in Table 1
Simple regressions for equ ation (37) were run for each of the 32 industries
with data for each of the firms which filed in the industry constituting an ob-
servation The results are given in Table 2 lines lA 2A
1974
the intercept
For all the A
equations generalized least quares was used th both and market
share bein scaled bv the square root of sales
The 32 industries were taken from the larger set of all manufacturing industry
categories in the 1974 data files Two selection cr iteria were used thereLB
had to be at least ten reporting companies and the industry had to be a consumer
goods industry or a producer goods industry which is strongly associated with a
related consumer goods industry so that adve rtising by the firms in the producer
goods industry impacts on demand in the related industry (eg soft drink syrup
and soft drinks) There were 135 industries with at least ten companies - 103
producer goods and 32 consumer goods
273
274
TABLE 1 INDUSTRY CATEGORIES USED
FTC
COOE OESCRI P T I ON _
2001 MEAT PACKINGt SAUSAGE S ANO OTHER PREPARED
MEAT PROOUCTS
2004 DAIRY PRODUCTS EXC F LUID MILK
2005 CANNED SPECIALTIES
2001 FROZEN SPECIALTIES
OEHYORATFOt AND
I NC LU 0 ING PRE SERVE S
SOUP MIXESbull
SEASO lNGSt ANO
OTHER PET FOOJ
RAIN MILL PROOUCTSt
ANO PREPARED
RELATED PRODUCTS
PRODUCTS
Z008 CANNEOt ORIEOt PICKLED FRUITS
ANO VEGET ABLES JAMS JELLIES DEHYDRATED I G TA3t E
S40CES ANO SALAD ORESSINGS
2010 DOG CAT A O
2012 FlOUR t OTHER RICE
MILLING fiLENOE) FLOUR
2014 B EADt CAKE ANO
20113 CO FECTIONERY
2026 BOTTLED A O CANNED SOFT DRINKS 2021 FLAVORING EXTRACTS AND SYRUPS EC
2029 MISC FOOOS AND KINDRED PROJUCTS EXC
ROASTED COFFEE
2301 MENS AND BOYS SUITS AND COATS
2302 MENS ANO BOYS FURNISHINGS
230 3 kOMENS AND MISSES OUTER EAR
2551 HOUSEHOLD FURNITURE
2702 PERIODICALS
Pqe 19a 1
RELATED 1972
src cooe
2011 f 3
202tX2026
2032
203pound
2033t4t5
2047 2041t4t5
2051 2065 2086
2081
209XZ05
231
232
233
251 272
2703 BOOKS
2704 ISC PVeLISHING
2806 ORGANIC FlEERS
2807 DRUG$ ETHICAL
23234
PT 2 S 3
2808 ORUCSt PROPRIETARY
2809 PERFU ES t COSMETICS ANO OTHER TaiLET
PREPARATIONS
2810 SOAP AND OTHER CLEANING PREPARATIONS 2815
2901
PEST1CIDES AND AGR ICULTURAL
PE OLEUM REF I N ING CHEMICALS EC
3606 HOUSEHOLD COOKING EQUIPMENT
3612 HOUSEHOLD APPLIANC ES NEC INCLUJING ELECshy
rqlC HOUSE ARFS AND FANS ANO SE I G MACHINES
3617 RADIO AND T V RECE I VING SETS 3808 PHOTOG APHIC E UIP amp SUP LfESt EXC
3qo3
PHO TOCOPYING
SPOR TING A D E QU I P M ENT f SUP P L I E S ATHLETTC GOODS NEC
3904 DOLLS GAMESt TOYS AND CHLOqENtS VE ICLES
PT283
2844
234tX2844
287 9
291 3631
3634t6t9
36H
PT386l
3949
15
943
middot
l A6l f 2 R(GRESSION RESULTS
EQ I NO
NR CODE cu 121
lA 2001 A (
2A 2004 r (
3A 2005 a
-R
ClS amp31
16
17
1 z
I TER-CEPT bull t
ooo9t21At 00022311 oooeos uu
OOl49tCAI 0010 2 middot 000960CCI
o049tl1At o04t55C61
MARKET SH ARE bull 5 I
-010670Ctl
-017600181
-001050
-0211AO
-0077 0
AOVER-TIS INti
lbl
000()39 000072
000018 0 0024liAI
-000079
SHARf
4ampgtVf Q bull
111
000140
-000170
RSQ
COl
66 485 862
607 6 3 t ()2
blti6 633
flASTI-
CI T Y PH
-007271(1
000441
-u0288l
clASTSHU (t)Ef
bull 10
tto
-234
210
( OOlltOCCI -126520 000922 181 001790
4A 200 7 18 00590liAI -038670 4e16 -032110 lell u c
OOOS74e 000700 025940
00067018) 0021001AI -02l100CAI
5 26
920
5A 6 c
200 8 26 0 03Sl9CA I 0012 26181 o 024511(
-005540
-0 81420 0001961AI O()Ol90CAI ooll240
441 604 891
-0020 19 274
6A 2010 16 Oe04224CAI o359lOIAI 9oq ll4015dCAI 061 6 -middot 005435CAI 00011 8 1AI 917 c 00535ltAI -009850((1 OOOA51J CAI -001610161 971
7A 2012 001862 000010 351 00187) oo1 a oooq45 ono1o6 439 c 001109 -027050 Oe0067SIAI -OOll60C81 897
- middotmiddot- -
8A 201t ll 000757 Ol4140CCI 814 ott9oocet 095 8 oooqo6 C C J 000069(81 811 c ootl2lampBI -080910CCI Oe009lUBI -OOtlOOCCI 960I
I DIIJIT pound11 1V OTIUTVTfAll-1pound1t- A _ 00 R - qq r - Q
1jl)OQCD
0
-middot middot--
l
_ __ _
I
i
0 00141 c c
91)7
I
T ABLE 2 REG ESSIOH RESULTS CCONTI
EO NR
IND CODE
iR bull
cos
I TEq-
CEPT
t1ARKE T
SHA E
AOVER-
Tl SING
SHARpound-cent
AuVER
RSO ILAST1-
C I T Y ELAST
SH CPtf (lt Cl) bull 11 I 41 I 5I (61 171 I 8 I f9) c 10)
9A 2 0 1 8 19 002397181 -007640 367 -005 246 le46 6 000605 000550CAI 597
-000029 G14180 OOl654CAJ -024li01At
lOA 2026 12 0028901AI -OlZl90 -0114 0 1 1 1 5 759
Ll (
llA 2021
f3 c
12A lOZ9
B
(
13A 2301
B
c
14A 2302
B
c
154 201
n (
16A 2551
8
10
35
10
20
19
24
001897(()
00141 6CCI
Oll9lHCI
010604 middot middot
Oll083
OOJ11JCAt
002059141
001411CIH
-000412
000293(81
000187
001159 (3)
ooo6a4cct
o00792CCI
OOl40dCAt
000006
-000076
OOlltl6fA)
o o0 5 64CCt
-034570
-o1middot o3o
-1659 90
Ol51lO
-046190CAt
Ol5450CA)
-005l60
002940
-030060(()
-0 43l50CCt
0 33650
-019080
00004
bullJ00741CAI
-000069
000466
0001 51CAt
000947CAt
O OOI)70CAt OOl750CA J
000175CCI
000671CAI
OOll5) f A t 002207CBI
0004b7CA I
- 007 1 70
001760
-001410CAt
- 0 16200CAt
-004210
- 1 207 70
705
96 0
542
466 805
460
614
921
832
9 61
996
424
550
9 15
56 1 651
884
-006114
019 037
03504lCA)
oo409Q
-141644((
491 -0 17664
607
871
lell
084
101
072
103
108 11IllOQfl)-- l
-037880fCJ OOl6l4CAJ -040l70fCI 0()
c
I
I
bull 9
9t
middot middot middotbullmiddot bull lraquo
TABLE 2 REGRE SION RESUlTS ICONTt
bull
EO I NO tHl I filER- f4A_K T AOVER- sbulluRE ttSQ flA Jl- iLAS T NR CODE cos CEPT SHARE JISING AOVfR bull (JTY SHR CiJCF
110Cl I f amp Itt 15J tbt bull 1c u caa
llA 2702 ll o079 6 81 Bt -019110 S6s -OlltO 111
a 004679 ooooo1 t69
( oOl26o -015160 001560 -022690 ltJr()I
ldA 270 3 19 0065lt -OIt2l60 220 -Ol57a2 11 8 e o 0 71t6IA t ol08 75tAt ( OOl983CiU -060llOC() OOl7191A t -O lltdOC A) middot 89
lqA 2704 10 005082 -Oil060 22b -l)i1977 lelO 6 -000080 oo2772C4 t 910 c 000907 -011 200 Oe04105CA t -046960 995
20A 2d06 1 II
O OlltStiAt -001120 837 00014 -))th O01380amp At -oooo 11 821 ( OOl526CAt -026700181 0002 5UAt 000530 969
2lA 2807 28 OOt9lt9CAt -00891 0 516 -0 03962 225I8 002095CCt OOOl lliAt 690 ( o02965CA t -056650((1 OOOt50CAJ -003l70ttl 911I
I 22A 2808 15 ot3l06 119730 655 1130) 090
0 - 007526CC t Oe003961At 835- middot-
c I oo7523 -067210 001l 2tA t -0091t80CCt 951
2lA 2809 21 Oe2067lCAt -1218101At 71 2 -10095((1 121 ttlIaOQ8 00797418) OOOI8lltt 686 Ill( 0092lU A t -Olltl70Ct t 000605CAt -OOit8701At 951 _
---- -
32 009506CAt
0 bull 06 981 I A J 006901 cA bull
00 24A 2810 -010960 5t 5 -OOltSO 7 56
8 -000011 s11 c -5811701At oooaoatAJ oot520IAI 888
-- 37 --- middot - -- -1 ) l lbull t5JJ r-l bullhmiddot1 --- - --
l5A
26A
27A
c
2RA
c
2 9A
31A
llA
24
13
597
Ci)55
14
l 6h
RSQ ELASTI- (lASTCOE f-
TABLE 2
EQ I NO NR CODE IU
REGRESSION RESULTS CCONTa
NR INTER- MARKET
(10S CEPT SHARE AOVER- SHARfcent
TIS lNG AOVER
) CITY StiR
c It) bull 5 co a I 1) bull 0 bull 9 c l 0 bull( 2 1
2815 16 003019 -009990 275 -o069bmiddotJ
6 001446 000224 282 c 002178 -054630 o02221CCI -0075 00 646
2901 29 00021t2 CAI 0)0520 592 0007 6 1 068
0 00105C8 t 000012141 60)g( OOOlJ51AJ -00126018) ooou271At -000u80CA) 974
36oa 1 3 003296(4) -oca29o -oo-q94
001755 000203 lt91t 002499((1 -04l820CCI OOl374CAI 0068)0 946
3612 0033281CI 011520 195 01684) oso
00 30 42C OI 00012 2 424 oo579ltAt -2 7l7801At 002 100IAI -003220 819
8 3617 btel 011789 081002259 009530
001448 oOOl OitCCI 732 OOl7 22CAI -0 46160CAI 0004961AI -OOll40181 989c
30A 3808 14 0021t 3 6fAI -000110 628 o023291C) -005 6 middot- _ middot- 0bull 02 04 1 IC I ooooo9 637 ( 001584((1 -0161)1)0 00021 91AI -0 00060
390) 14 002790CCt 006590 h8 t 0 0938 4 070
B OOl89bCCJ 000144 74
0021961(1 -OItlOlO OOl323CAI -004800 915
B
middot--- - middot 4 - middot- - dfll3904 12 004563181 0)8300 879 0428651(1 089 OQIll004 277CA I ooozstcea 919
c 0041)3181 -0-11250 0011881(1 -006270 972 0Ill
- J - bull bull bull bullbullbull middot - - middot - middot gt middotbull L)ofbullflbull middot middotfJltl -tn- middotmiddotmiddotmiddoto +f middot t middot bull
c
is
positive
2
Page 0 determineThe consumer goodproducer good split is sometimes difficult to
Sollle support for the split used in this paper is a comparison of goodness - of - fit
distributions of a
industries
set
advertising are
that industry
a sales increase of
the
highest aD10U3 those
measures for the two groups of industries Table 3 contains
measures for the 32- consumer goods industries and the 103 producer goods
2 Median R is 5 7 for the consumer goods set and 40 for the producer goods
Estimates - of y the elasticity of demand with respect to
given in column of Table 2 The hypothesis that y bull 0 with
alternative was applied for each ittdustry As shown in the
could not be rtjected for 24 of the 32 industries Of the
and 10 are positive For the remalnifg eight three are
and five are significantly positive
Of the threertegative elasetcities the largest (in absolute value)
cosmetics (code 2809) With all elasticity of -1 01 a one
industry advertising would gwerate a decrease in industry sales
more than one percetu
At the other extrema is toys and games (code 39 04) For
a one percent ittcrease in industry advertising would lead to
four-tenths of a pampreentmiddot Though not significantly different from zero
elasticity estimatbull for pt-bpuetary drugs (code 2808)
which are
Both drug industries have
to sales rafios (211 pe1 cent for 28 08 and 13 4 per cent
other hand flaIJo ing ext1ac s (mainly soft-drink SYUlS) also
is for
percent increase in
of slightly
a middottwo - tailed
table that hypothesis
24 14 are negative
significantly negative
9
the codetics and p-ropriacuy very high advertising
for 28 09) bull On the
has a high ind UY
advenisiltg t() sales ratio ( 8 0 per cent) bull but its estimated elasticity is very
small and not signifieantly different uom zero
middot It doe$ not seem unusual to find many elasticities near zero with a few
negative and a few positive If aggregate consumption is insensitive to aggregate
advertising as seems likely ld then what one industry gains must be lost by
others
Page 21
elasticity of demand does
(10) we can
if t 0 then
Finally a determination concerning the industry
not fix the individual firms elasticity Going back to equation
write y bull (1 - ) a + z y Since a gt 0 it follows thatii i i
Y gt 0 If Ylt 0 however y s sign is indeterminateij ij
Turning next to the coefficient of market share 20 of them are negative and
12 are positive with three of each sign being significantly different from zero
The largest positive value which is significant is for pet food (code 2010) and
a close second is mens amp boys suits and coats (code 23 01) The largest signifi
cant negative value is for cosmetics (code 28 09)
In the model some relations among y a 0 and 61 are implied If middota is
negative (1- y) is greater than one Since 61Y bull (1- a) (1 - ) 61Y is
pCsitive but it will be less than one if 0 is large enough Seventeen of the A
ys are negative in each of the seventeen industries 61 is also negative and les
1 (algebraicly) than y so 61Y is greater than one his is not evidence that the
degree of cooperation is low or zero in any industry of course A high o togethegt
with a negative y which is small (in absolute terms) relative to a can give a
value of 61Y greater than one On the other hand if 61Y were less than one
o would have to be greater than zero
For y gt 0 two conditions may hold If y lt a it follows that 61 lt 0 since
(y - a ) lt 0 It also follows that 61Y lt 0 regardless of the value of o bull
In addition the relation is an if-and-only-if one ie if 61Y is negative
then y lt a Three industries have negative values for 61y dairy products
except milk (code 20 04) synthetic fibers (code 2806) and photographic equipment
(code 3808) In all three cases y is quite small so a does not have to be very
large to give a negative value to al
If y gt a the model shows that 61-- gt 0 whatever the value of 0 bull There are
12 cases where both y and 61 are positive with 61 ranging from near zero
(grain milling code 20 12) to 1 20 ( proprietary drugs code 2808) Several
coeffiei-en_rs
--
15
o f to zero
Tab le 3 - Consumer Goods
Indus tr ies
2Distributions of R and Producer Goods
Page 20a
Consumer Producer
Goods Goods
0 - 9 9 0 3
10 - 19 9 0 1 1
20 - 29 9 3 20
30 - 39 9 3 17
40 - 49 9 5 16
50 - 59 9 8
69 9 6 Ii 1060
--f 670 79 9 2
80 - 89 9 4 4
90 - 100 1 1
Totals 32 10 3
Notes
2is used R i s cal-
culated as F F + ((N - K- 1 ) K ] 1 b_ F is the F
statistic Lor the hypothesis that all o f the non - intercept terms are simul taneously equal
1
2
--
-- Since generalized least squ-area regress ion
Page 12
o f these cases have values o f 6 1 Y rea ter than 0 8 (pet food (20 10) bread amp cakes
(20 14 ) miscellaneous processed foods (20 29 ) men s and boys suit s and coats
(23 01 ) p rop riet ary drugs (28 08 ) elect ric housewares (36 1 2) radio and
IV sets 17 and games and toys (39 For be than(36 ) 04)) 6 1 Y to greater 0 8
both o and cry mus t b e less than 0 2 That is in these industries the results
shown in Tab le 2 would hold only with very strong dominance of marke t demand
effects over brand swit ching e ffects and a very low degree o f cooperation
Using the concentration rat ios given in Table 4 the average concentration
ratio for the eight indust ries is 38 1 For the remaining 24 industries it is
43 2 This result would lend some support t o the proposition that concentrat ion
and cooperation are related if the group of eight industries are identi fied
as having low degrees of cooperation
One f inal note on model predict ions S ince both cr and o are assumed to
be non-negative 60 should also be non-negat ive In only one o f the 32 cas es
is 60 negative and it is not signif icant bull
cor pJ
7 1
7 0 4
1 4 7 5 1
)
7 3 z a o a t s 1 1 3 3 5 ( i V Z 8 tJ1 2 1 - 1 1 0 7 3 5 c middot v 2 8 1 0 3 2 -0 0 1 4 3 0 C 1 V 9 8 3 2 8 1 5 1 6 - 0 0 6 t1 6 9 C _ V 1 11 0 2 9 0 1 29 0 00 7 6 1 C l k V 7 3 1 3 6 08 1 3 - o n 4 9 9 4 3 3 3 6 1 2 2 4 0 1 6 4 3
3 1 7 1 3 0 1 1 7 d 7 8 3 3 8 0 8 0 0 2 3 i 2 3 0 6
P 4
3 9 0 3 1 4
1 2
0 09 3 8 4 39 1)1 0 4 l 8 6 5
T A LE 4 e I NCU S T R Y C A T E G O R Y
F TC NOM e tOOE ass
C 1 I c 2 )
2 0 0 1 1 6 2 0 0r 1 7 zo o l Z 2 0 0 7 1 8
z o o 2 6
z o t a 1 6
2 1) 1 2 1 5 2 0 1 4 1 1 2 0 1 a 1 9 2 0 26 1 2 2 0 2 7 1 0 2 0 29 - 3 5
2 3 0 1 10
2 3 02 z o
2 3 03 1 9
2 5 5 1 2 4 Z 1 0 Z 1 1 2 0 3 1 9 2 7 4 1 0 z a o 1 3
2 A 0 7 z s
E LA S T I -C I T Y
c 3 )
- 0 09 7 2 7
0 00 449
- 0 0 2 8 9 3
- 0 3 2 7 70 - 0 OZ il I q
o 4 u t s a 0 0 1 3 7 3
o 1 4 o o-0 0 5 2 4o -0 1 94 0 1
-C l 6 l l 4 0 1 JJ 3 7 0 3 t 0 4 2 0 0 4 0 9
- 0 4 1 8 44 - O l 7 t ) 4 -ll 7 1 4 0 b -0 3 5 7 8 Z
- 0 1 9 7 7 0 1)0 3 3 4
-o n 9 2
IJ A T A
CONV
ON-C ONV
( 4 )
t 1NV
C 1t-V C NV c ) v C - JNV C rlNV c n v c a v C ON V ( i) l V
C CtV
C t1tV C IJNV
C 1 V
c 1 v
Pa ge 2La
HEQ F t Noe x
c 5 )
6 2q
6 89 1 48
6 1 4 ) q 3
1 9 7 1 9
q z t s 5 a e z
l l 9 1
4 o 7 1 2 0
c oqt 1 Z
4 4 )-q 1 z
5 5 9 1 0 64 1 3 3
n 3
Inter industry analysi
Page 23
Equation ( 3 6 ) may be used to fo rmulate a rough test o f the cooperat ion
vs e f ficiency controversy I f we rest ricted ourselves to some subset o f
indus tries where the relat ion be tween y and a i s the same we would be lookingbull
at a sample o f indus tries for which the dependence o f V on H and o is a
quadratic wi th no squared terms
I f we now sub s titute (26) into ( 36 ) we get
(38) v bull a + (y - a) R + (y - a) o (H) - (y - a) H o (H) + u
If it is only d if ferences in marke t shares that affect advertising inten-
s i ty levels i f concentration does not af fect cooperation and i f o bull 0
( 38 ) reduces t o
( 39 ) V bull a + (y - a) H + U
Let this b e the null hyp o thes is and note that V is a linear func-
tion o f H If y gt a and the null hyp othesis holds the expected relat ion
between V and H is as shown by the pos itively sloped straight line in Figure
6 b etween points A and C
In this context the alterna ive hypothesis the degree o f
cooperation ( o) is a positive function o f concentrat ion
is that
(H) I f the func-
tion is as described in (26) and Figure 4 the relation between V and H
would be as shown by the curved line connec t ing po ints A and C in Figure 6 One way to test the null hypothe s is against this al ternative is to test fo r
the hypothesized curvature in the function relating V and H
Fi gure
R
6
v
ID c
H 1 0
I
i I I
sfigned
clasJ s
in def ini g
effec t is the
I
Page middot2lf
One charac teristic of produc ts which may be useful
classes wi th respec t to significanc e of the brand-switching
conveniencenon-convenienc e distinc tion developed by Porter ( 1 97 )
Using his identif ication of IRS indus tries as a s tar t ing point
each of the 32 industries used in this s tudy to one of the two
The ass ignments are not ed in Tab le 4
Using the conveniencenon-convenienc e dis t inc tion the
divided into two sub-samples
sample was
Four regressions were then run bull ene for e ch
sub-sample one for the to tal sample and one for the total sample with
a dummy variable for convenience goods Each regression was run wi th LB s
and wi th indus tries as ob servations The results are given in Tab le 5
For the LB level equations ( 1A - 1D) equation(37) was used ome
rearranging of terms gives v bull y - ( 1-o ) (y - a) ( 1-z ) + ui i i
bull y - 6 1 ( 1 - z ) + ui When observations are pooled across indus trif s yi
is an industry variable In these regressions the values o f y es t imated
in the first stage were used Since industries have been grouped so as
to reduce differences in (y - a) that term is t reated as a cons tant
within each group Finally H the Herfindahl index has been subst ituted
for o the degree of cooperation The resultant equation is linear
y ( 1 - z ) and H x ( l - z ) This is the equation for whi ch results arei i
reported in Table 5 In all regress ions all variables were weighted by the
square roo t of sales
in
raquo
- 34
3 1
i l T 4 8LE S R E G E S S I ON R E SUL T S i X T f N S I ON S T O T Y PE Of
AND L E V E L OF C ONC E kT R A T I O
E O O tt S
tP bull T Y P E NR C ONV O B S tO N-CONV
E L A S T I shy
T I C J T Y
O E - t-41( T SliA R E
H( fl F
1 1 - K O x middot K O X E R F
- H S a 1 1 - H S t 1 8 1 1 9 1
R S O
1 1 0 1
q z
t - bull 9 f 2 1
Z t P D l 6 fC ONV - 1 9 1 s s c 2 1
e - l O I e bull u q a
2 3 1 0 1 1 Nl N- t O V 5 7 1 ( 1 2 19 1 6 1 - J l S 2 5 4
l l l l t z q o a 1 - J o v z I
l ( I NO l l A l l ol u l t o o c bull J u l l
d S
c 3 4 t t r a l
bull Z D 1 0 l l 4 l l 2 3 -gt 2 l l 2 0 - 4 lS 5 4 l C 4 1 o e 6 t l b b 5 1 1 0 1 1 1 c - 1 9 5 1 t d S 2 1 tV
OQ10 T-R A T I O S A E G I VE N l P A RE N T H E S E S l E V E L S f S I G 1 F l C ANC E A - 99 5 B - q 9 t - s NS J UC E NO I Nl E R C ( P T I S U (j f middot) I N f Hf t OUA TIOI S 6 E f 3R E A O J U S T H E N T F OR 1-t E T E R O SIC E O A S T f C I T V pIraquo
1
middot
I
R SO I S C A L C UL A T E D A S f I f bull 1 1 bull 1lt 1 K t t- E E N I S THf U A1 B E R O f fJe S E fi V A T I O S F t S HtE F - S T A T I S
I N DE P E ND E N T y II
T I C F O T H E U Y P O HIE S I S f A l l C JE F F I C H N T A R E QU A L HI l f RO A f40 K I S T H f UUM 8 E R OF Q ( A fL f S
I l l
1 A
1 6
t C
f 1 r R I l l
1 2 1
L Jt
L B
l q
L 1
on s T Y P E bull 2 )
P OUC T
X
H S I l l
l O l A I bull l 0 7 1
O O b O J
1 2 00 1
1 b 2 1 C I l l Z U
l l S b 1 -1 1 0 1
rlt F X l l - fi C f l
bull 1 I
l l
- ()4 2 3 l f H
1 - 2 7 9 1
K O X
1 1 - ti E R F I
l d l
S4 l i A I 2 l b
1 3 00
K O X -t f R F t S J
I 1 - HER F t
)
60 9
z o bull
0
t l l
3 8 8
1 7 6
5 64
64
N f bull
tl S bull
I l l
I 4 1
t ONV
NON-t ONV
A l l
A ll
c o v NO- ( ONV
1 4 1
1 5 1
- l S b
c - 1 1 2 1
5 8 2 C A I 1 6 7 0 1
middot I l 9 H
l 1 il
1 4 1 6 1
E L A C) T J shy
l t C I T Y
I S I
l b l
- 0 2 0
1 - 1 4 9 1
O Z 3 2 1 A I C b 6 b l
0 0 9 2 -J
1 1 b 4 1
o z t t t l 1 2 2 1 1
oru - Ht tt F
1 gt X bull b I
Page 25 bull
Mean val -as ior relevan t variables for the two sub-samples are
Conv Non-Conv
Adv Sales 4 7 2 3
iit Share 2 7 3 2
Ela cicity - 05 3 02 2
Herf indahl 7 3 1 860
That the dist nct ion between coft9enienee goods and non - convenience
goods mat ters is supported not only by these means but also by the regression
results Equat ions lA and lB are different in almost every respec t Only the
concentrat ionmarket share int eract ion term seems to mat ter for convenience
good s Jus t he opposite holds for non - convenience goods both elasticity1
and market sl e are highly significant but the concentrat ionmarket share L
int erac t ion t=rm is not
Given that one minus market share is used these equ3 tions the coeffici t
of market share is the negative of what is sh in col (6) of Table 5 For the
regress ion for non - convenience LB s reported in equation lB then LB s with larger
market shares have lower advertising to sales ratios
On the question of the impact of the degree of cooperation the story is mixed
For convenienc good s the coefficient is pos itive as equation (31) predicts when
there is a po sitive degree of cooperat ion Fo r the non-convenience goods LB s middot
concentration has no effec t
Aggregat ion t o the indus try level and then rer unning the equations $ives the
results shown in (2a-2d) of Tab le 5 As expected R2 goes up and levels of significanc e
fmiddotJr individual coefficients gJ down In add ition the coeffic ient of H ( l -H) inmiddot
quat ion 2A is no t significant though it is po s i tive Since 2A is the indus try level
I
--
all but one of them and that one is not significant For 18 of the 31
wo uld have on its adve rtising to sal es ratio
cognate of LA the variable H ( l-H) pl ays the same role in 2A as H ( l -z 1) p lays
in LA
Vconomies of scale in advert ising
The literature on advert ing contains a number of definitionbull o f economies
of scale and of findings eone rning its presence and magnitude Before I loo k
a t the data with the hope o f mmenting about scale effect s then I wan t to
define the term as I use it
By economies of scale I mean a decrease in the number o f units of adver-
tising which are needed to generate a unit of sales as advert ising is increased
Given the definition of the firm s own elast icity of demand with respect to
advertising (see the text foll wing equation (1) ) the technical definition
is that rii gt 1 Diseconomies of scale are def ined symmetrically i e
Yu lt 1
ith a cro s s - section oi observations on advertising intensity
(v bull a s1) and onmiddot advertising (a ) we may regress the former on the latter 1 i i
The results of such regress ions for the 32 industries are given in Table 2
lines lB ZB bull bull bull 32B For the 32 eases the regression coefficient is positive it
positive coefficients the coefficient i s signif icant
The que tion now is whether these results ar e evidence of pervasive dis-
economies of scale in advertislng I think no t Given my as sumption that the
data I observe are equilibrium results I may ri ghtly conclud e that almost every-
where high levels of advert ising are as sociated in equilibrium with high ratios
of advertising to sales I may no t conclud e anything however about the
impact that a move by some firm away from its equilibrium level o f advertising
middot- Page 27
Given the assumed relation in equation ( 37 ) there is a corre spondshy
ing equilibrium equation relating v and ai The equation is non - linear i
and I have not found a simp le way to characterize it It is po ssib le to
det ermi ne its deriviative with respect to a b y takin$ tne de riviative of ( 3 7 ) i
When that is done the result is
v - 1(40) i bull sl a a z ( 1-z )i i ira-l
Sinc e all the other terms in (40) are non-negative the sign dep ends only
on the sign of 6 1 which is in turn de pend ent only on the sign of (y - a)
That the signs of the coefficients of a in equations lB 2B bull bull bull 32B arei
not always equal to the signs of the coefficients of z in LA 2A bull bull bull 32A i
is still a mystery to me and something to be explored
bull
Summary
My purpose in writing this pa per wa s t o attempt to integrate explicit micro-
economic mode lling reasoned econome tric specification of error terms and high
quality data to exp lore some questions about adver tising by large firms
al quite important improvements could be made in the first two of those areas
and much data ma ssag ing is yet to be done in the third
Concerning the subst an ce of what I have done I think six things are im-
Page 2 8
Severshy
V and Conclusions
portant
1 Explicit modeling is worth the ef for t if for no other reas on than
that it provides a basis for choosing fr om a varie ty of funct ional
forms
2 The same is true for error specif ica t ion I did in fact look at scatshy
ter diagrams for all 32 industry categor ies and they vir tually all show
the het eroscedast icity which I as sumed and for which I correc ted
3 The predicted relation be tween advertis ing in tensity and market share
shows up clearly in on ly 25 of the case s examined I have ye t to
explore why that may b e the cas e
4 Some evidence concerning the presence and impact of coopera tion was
produced but it is not clear ly pervasive
5 The dis tinction between convenience goods and non-c onvenie nce goods is
unambiguously a good one
6 Virtually no evidence concerning econ omies of scale in adver tis ing can
be gleaned fr om this study given its assumpt ions
More work is called for on many of these issues my study seems to have
raised more questions than it has answer ed This is pr obab lly due to the richshy
ne ss of the data source since I had the oppor tunity to address several issue s
at once
- -- - Q --
Foo tnotes
Manager Line o f Business Program Federal Trade Commission The author depended heavily on several staf f members of the LB Pro gram for statistical and clerical support in particular Joe Cholka George Pascoe and Haro lene Jenkins Helpful comments were given b y Richard Caves Dennis Mueller Michael Porter and F M Scherer When the paper was in its early stages o f development both S teve Garber and Jon Rasmussen former economists in the LB Program provided eery useful cri tiques The views expressed here are my own o f course
1 The basic exp licit model of the role of advertising is Dorfman and S teiner ( 1954 ) and the models in this paper may be seen as ano ther extens ion of their work
2 For a review o f models of oligopoly which treat prices and quantities as relevant variables see Shubik ( 1 959) Extens ions to take advertis ing int o account are fairly s traightforward
3 Dorfman and S teiner dealt with a price-setting firm For a monopolis t inverting the demand relation to show Pbull P (Q A) _instead o f Q- (P A) has no effect on the results o f the analysis For oligopolists who face a sys tem of demand relations inversion does have some impac t on the results The dif ferences are not trivial nonetheless they will no t be explored in this study
4 Shubik ( 1 959) p
5 From a formal mathematical point o f view the function is similar to a1 i Legrangian funct ion of the form - r rr
L bull 11 + A ( j - (i i i+j
where the indicates some fixed level of the profits industry Following the same logic which resul ts in and income constraint as the marginal utility of the of the indus tr in the obj ective of firm i
6 Hall amp Weiss ( 1 967) p 323
7 See Schmallensee ( 1 97 2 ) Ch 2 pp 1 6-47
8 On this poin t see Imel amp Helmberger ( 1 9 7 1 )
9 Long ( 1 970) p 130 - 1 39
1 0 See Schmallensee ( 1 972) Ch 3 pp 48-8 7
rrj i+j ) of the rest of the
the identification profits of the rest
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980
ei ith
1v ( ) h i
7urning
where is the
leads to z
to the r matrix given in (11) we note that 1 = cr + (y - cr) i i zi elementary vector with 1 in the position and 0 elsewhere This
bull cr + (y - cr) zi Since bull required is thatcr at- zY wYij (y - cr)zi - cr + (y - cr) zilt 0 or cr gt 0
These observations may be summarized by substituting (11) into (17) bull
giving
(18)
i s
d -1 zz -
-
-
+ (y - cr) tz
d -1 0cr(I - l ) a a bull
cr 1 z - (y - cr) 0 a
o
If ai increases and the other a s are const t cr must be positive to havej
a positive effect on zi A positive cr also bull plies a negative ffect of aJ
f all a s in-on zi other a s (including a ) fixed j
eluding a1 change by the same percentage zi remains unchanged Finally k i
if cr bull 0 there is no brand-switching effect
For a to have no effect on p requires yij bull 0 From (10) thisj i
amounts to y bull cr that is the industry demand elasticity and the brandshyswitching elasticity are equal If y gt cr gt 0 the industry elasticity yij doninates If y lt cr y lt 0 the brand-switching elasticity dominates ij
making
Page 10
Substituting ( 1 1) now in (6) the Cournot case and in (9) the
Chamberlin ease gives
-(19) Cournot vO a + (y - a) z
Chamberli n vl - y lmiddot th
The 1- element is then
-(20) Cournot vO a+ (y - a) zi i
-Chamberlin vl yi
In (19) and (20) the superscript 0 is used to denote total
independence of decision and the superscript l is used to denote
total interdependence of decision making
The relation between v and vf depends solely on t relation between
y and a f y gt a so that the industry effect dominat 3 vf v with
equality in the trirlal case of z bull 1 Cooperative irms would have ai
higher advertising intensity than non-cooperative fi
If there is equality between the industry elasticity and the
y Whe therbrand-switching elasticity or y bull there is cooperation among the firms has no effect on advertising inten-sity And if brand-switching is more important with y lt a e get
vi lt v cooperative firms will have a lower advertising intensity than
non-cooperative firms The three situations are depicted in Figure L
Given the results for the firm level variables calculation of the
corresponding re sults for the industry is s-traightforward Let
B bull t zf bull z z be the Herfindahl index of concen tration and notethat
z - bull l z bull 1 and that V bull A S bull a i 5 bull (siIS) (asi)i
bull middot Then for the general case from (6) and (9)
bull
Page lOa
y lt1 bull y y
10
f
V0 bull z [ a1 + (y - cr) J bull a + (y - a) H vl bull Y bull y
11Page
( 2 1) vo bull z d -1 r bull r z bull z- d d- i )11
l -1 N N v bull z d r z bull r z bull l y z - ij jibulll jbulll
and for the simplified case from (19)
(22
Since the forms of the equations in (22) are the same as in (20) all
of the analysis of (20) holdS for (22) except that His substituted for
middot z Advertising intensity will be higher for an industry with cooperativei
firms than for one with non-coope rative firms if the industry demand elas-
ticity dominates the brand-switchine lasticity If the ewo elasticities
are equal then so will be advertiSl intensity for the cooperative firm
and non-cooperative firm industries If brand-switching dominates the
cooperative firm industry will have lower advertising intensity
In the industry model context the possibility of a critical concen-
tration level may be easily introduced That is assume that at values
of R below R the firms are non-cooperative but that for alues equal to
or greater than H they are cooperative That is
(23) V bull a + (y - a) H if H lt H bull y if H gtmiddot H
This function is shown in Figure 2 for the three relamptions be eween y and a
IV THE D GRE OF COOEtRATION
The models depicted in Figure 2 are based on a very simple notion about
the relation between concentration and the cooperationnon-cooperation charac-
teristic They use what is essentilly a step function relating concentra-
Figure
I -
J H H
Page l la
2
v v v
a
I J I y a bull Y y ---i
I I I I I I II I
H 10 H H 10 H 1 0
bullbull
bull bull
Page 12
tion and the extent or degree of coo peration as shown in Figure 3 Using
o as a symbol for the degree of cooperation eq uation (23) can be rewritten
as
bull(24) v (1 - o) vo + ovl bull (1 o) [a+ - a) H ] + oy- (y
vhere
(25) o bull o(H) bull 0 if B lt B
Hbull 1 if B gt
The specific relation between concentration (H) and the deg ree of
cooperation ( o) shown in Figure 3 is more restrictive than necessary and
bas less appeal an one which shows a smooth increase of o from 0 to 1 as
B increases fro to 1 Such a smooth function is shown in Figure 4 he
degree of cooper ion is low until H gets close to H increases greatly as
B goes through R attaining a high level for an H larger than B but still
close to it and then goes to 1 as B goes to 1 More formally
(26) 0 bull o(H)
bull 0 if H bull 0
bull 1 if H bull 1
do gt (IdB
d2o- gt 0 if H B dBZ
0 if H H
lt 0 if R gt H-
If (26) is now substituted for (25) in (24) a corresponding smoothing
of the functions shown in Figure 2 is accomplished he smooth functions
Figure 3
I
10
H 10 H
Page 12a
Figure
Page 12b
4
10
H 10 H
l
(27)
Cl - o) o
[ (1 -
[ (1
t y ij z
j
are given in Figure 5
Page 13
The equation in (24) was constructed as a hybrid of two results derived
from first-order equations for the non-cooperative and cooperative models
It may also be constructed directly by using l bull (1 - o) + on as thei i
th 2objective function for the i-- firm Treating o as a constant the first
order equation is
0 an aa
i middot
n i
Using equations (3) and (7) and setting --- bull 0 givesaai
(28) - 0
After collecting terms and converting to matrix notation we get
(29) - Cl - o) + 0 z-1vi yii i
N
j bulll
+ 0
+ o v1
d-1 r z z
-(30) v Cl - o) rd
-
o) r +d
d r J _ -1- d z
and middot
d) r + 0 rjd
l r z
v - z --(31) zv
- (1 0- +d) t r z- d-
If we now impose the simplified demand structure by substituting (11)
we get
- (1 - o) a +( 32) + oy and
Figure
l
-y
v
Y
H
r
I I
v
a
y
H 10 H
v
a
bulla
HH 10
bullbull
(33)
i
Page 14
v - (1 - o) [ a l + (y - a) ] The corresponding equation for Vis given in (24)
If the relation between industry advertising intensity (V) and
concentration (H) is non - linear via the impact of concentration
on the degree of cooperation does the relation take the form of a
auadratic as some other investigators have proposed (Greer ( 1971 ) Martin (1979) Strickland amp Weiss (1976) ) That question can
be approached from two perspectives - conceptual and empirical In
the next section I will try to apply some statistical material to it
here I want to explore the question in the context of the models
developed above
Referring to equation (24) taking the deriviative with respect
to H and letting bull (H) gives
bull( 3 3a ) 3V (y - a) (1 - o) + (1 - H) 3H dH
Given that o and H each are bounded by 0 and 1 and that o is an in-
creasing function of H the term in the brackets in non-negative The
sign of the deriviative then is the same as the sign of (y - a) If the industry effect is dominant (y gt a) then advertising in-
tensity will increase with concentration If the two effects are equal
then advertising intensity will not vbullry with concentration and if the
brand-switching e ffect is larger advertising intensity will decrease
with concentration
For given advertising elasticity parameters then it is not pos-
sible to show an increase in A S up to some H and the n a decrease
he relation may be non - linear but it is monotonic over the permis-
sible range of H
Page 15
One possibility is to introduce a relation between (y - a) and H
If industries characterized as dominated by the brand-switching effect
also were more concentrated then a cross industry comparison would
show an inverted U shape relation between AS and H I know of no
support for such an association between ( y - a) and H
V ERROR SPECUCATION
So far I have implicitly assumed that there are no errors in the
model I now assume that the error
thvariable in the equation for the i- firm is additive that its mean is zero
that its variance is inversely proportionate to its sales and that all error
term covarlances are zero hat is
34) v bull (1 - o) [ a + (y - a) z ] + oy + u i i i
- - -
In vector notation
(35) bull (1 - o) [ at + lt - a) J +
E() bull 0 Cov (u) -
0
If we now ca lculate V bull bull we get
oy + U bull
(36) V bull (1 - o) la + (y - a)R J +
U bull E(U) bull 0 Var (U) bull
P age 16
The error term specifi c ation which follows Hall
wi th the ratio of profits
6 heir analysis seems
as wellbull particularly given
and Weiss was
applied by them to equa tions to equity or
asse ts as dependen t variables to carry the same
weight in the present context the substantial
evidence presented by marketing analysis on the tendency of company de-
cision makers t o use the advertising to sales ratio as the decision
7variable
The zero covariance assumption on the other hand1 is particularly
troublesome Within a probably not true since
many even ts outside the
single industry it is
industry would have
industries
similar effec ts on all the
8firms in the industry Between it is probably not true either
since many companies produce in more than one indus try and even ts in the
firm would tend to effect all of its activities The extension of the
model to allow for non-zero covariances is certainly called for
--- -
Page Ll
Athird characteristic of this specification also deserves some comment
only on equation is shown for the firm or industry Several haveinvestigators
included an advertising equation ina larger system of equations (Comanor and
Wilson (1974) (1979) and Weiss (1976) so in myMartin Strickland Idid
dissertation1 Ihave ewo defenses for presenting only single equation results
here One is the suggestion by Comanor Wilson that simultaneousand equation
verybias may not be important in this context he second is that Iam will-
ing to see the model extended to include a profitability equation and I intend
to move that direction in the near futurein
Onthe inclusion of aconcentration equation on the other hand I amskep-
tical he problem isthat in some fundamental sense equations in models of
industrial organization should be oriented to the decision-making contexts of firms
Ifeach firm in anindustry with Nfirms has only one decision variable then
there can be only Nindependent equations which are based on the first-order
optimization equations It not show large ofis difficult of course to a number
functional relations which follow from the first-order conditions point here he
isthat only Nof them can be independent
Consider for illustrative purposes only a situation in wich teh N
itsquantities say q are set exogeneously and each firm determines advertising1
of and level say ai where a is the true optimal value a a bulla
1 1 i i + i
Ifp a) then the Nq
s and the Na s will determine N p s say pii bullp
i(l i 1 i
Now given Eand 9 we can determine both the vector s andthe vector v since
- bull til
ands We can also determine S s z bull s-1 s-ii
v and we
industry concentration isHbull (alternatively
V and Hare functions of the same
pendent Furthermore if there are
VM) and (H1 H2 bull bull bull )contain
VI STATISTI CAL APPLICATIONS
Page 18
Given the observed values of might be tempted to conclude that we
have a two equation system in which v and z are jointly determined That would i i
be inappropriate however since the 2N random variables (v ) are determined by
onlt N basic random variables t The variance-covariance matrix for ( ) must
be singular bull
The extension of this observation to industry level variables is straightshy
forward Since observed industry advertising intensity is V bull l v and observed 4
C4 bull z ) it follows thati l i
set of random variables t and cannot be indeshy-
M industries then the variables (V1 v2 at most M independent random variables
Given the assumptions made above y a and o are constant within a
given industry That being the case (34) can be written as
(37)
s1 bull (1 - o) (y - a)
If and
So + +
(1 - o) a + oy
81 are esti mated for an industry using a suitable statisticalSo procedure giving and 81 an estimate of y is then available The sumSo
Intra-industry analysis
1974 of Business Re ort -
categories are identified
Page 19
of So and S1 is y so y bull So + S1 is the estimate of y
If o is assumed to be bounded by zero and one the sign of 61 is also
the sign of y - a The estimate of So is of cou se an estimate of
(1 - o ) a + oy Moreove if y gt a then a is less than 6o- (1 - B) a+ 0(
Cor esponding results would
hold if y bull a or y lt a
If 61 gt 0 then we would estimate that a lt 6o
bull
For purposes of estimating this model 32 manufacturing industry categories
were selected The data are fram the Line of Business forms filed with
the Federal Trade Commission they are discussed in detail in the Annual Line
which is available from the FTC The 32 industry
in Table 1
Simple regressions for equ ation (37) were run for each of the 32 industries
with data for each of the firms which filed in the industry constituting an ob-
servation The results are given in Table 2 lines lA 2A
1974
the intercept
For all the A
equations generalized least quares was used th both and market
share bein scaled bv the square root of sales
The 32 industries were taken from the larger set of all manufacturing industry
categories in the 1974 data files Two selection cr iteria were used thereLB
had to be at least ten reporting companies and the industry had to be a consumer
goods industry or a producer goods industry which is strongly associated with a
related consumer goods industry so that adve rtising by the firms in the producer
goods industry impacts on demand in the related industry (eg soft drink syrup
and soft drinks) There were 135 industries with at least ten companies - 103
producer goods and 32 consumer goods
273
274
TABLE 1 INDUSTRY CATEGORIES USED
FTC
COOE OESCRI P T I ON _
2001 MEAT PACKINGt SAUSAGE S ANO OTHER PREPARED
MEAT PROOUCTS
2004 DAIRY PRODUCTS EXC F LUID MILK
2005 CANNED SPECIALTIES
2001 FROZEN SPECIALTIES
OEHYORATFOt AND
I NC LU 0 ING PRE SERVE S
SOUP MIXESbull
SEASO lNGSt ANO
OTHER PET FOOJ
RAIN MILL PROOUCTSt
ANO PREPARED
RELATED PRODUCTS
PRODUCTS
Z008 CANNEOt ORIEOt PICKLED FRUITS
ANO VEGET ABLES JAMS JELLIES DEHYDRATED I G TA3t E
S40CES ANO SALAD ORESSINGS
2010 DOG CAT A O
2012 FlOUR t OTHER RICE
MILLING fiLENOE) FLOUR
2014 B EADt CAKE ANO
20113 CO FECTIONERY
2026 BOTTLED A O CANNED SOFT DRINKS 2021 FLAVORING EXTRACTS AND SYRUPS EC
2029 MISC FOOOS AND KINDRED PROJUCTS EXC
ROASTED COFFEE
2301 MENS AND BOYS SUITS AND COATS
2302 MENS ANO BOYS FURNISHINGS
230 3 kOMENS AND MISSES OUTER EAR
2551 HOUSEHOLD FURNITURE
2702 PERIODICALS
Pqe 19a 1
RELATED 1972
src cooe
2011 f 3
202tX2026
2032
203pound
2033t4t5
2047 2041t4t5
2051 2065 2086
2081
209XZ05
231
232
233
251 272
2703 BOOKS
2704 ISC PVeLISHING
2806 ORGANIC FlEERS
2807 DRUG$ ETHICAL
23234
PT 2 S 3
2808 ORUCSt PROPRIETARY
2809 PERFU ES t COSMETICS ANO OTHER TaiLET
PREPARATIONS
2810 SOAP AND OTHER CLEANING PREPARATIONS 2815
2901
PEST1CIDES AND AGR ICULTURAL
PE OLEUM REF I N ING CHEMICALS EC
3606 HOUSEHOLD COOKING EQUIPMENT
3612 HOUSEHOLD APPLIANC ES NEC INCLUJING ELECshy
rqlC HOUSE ARFS AND FANS ANO SE I G MACHINES
3617 RADIO AND T V RECE I VING SETS 3808 PHOTOG APHIC E UIP amp SUP LfESt EXC
3qo3
PHO TOCOPYING
SPOR TING A D E QU I P M ENT f SUP P L I E S ATHLETTC GOODS NEC
3904 DOLLS GAMESt TOYS AND CHLOqENtS VE ICLES
PT283
2844
234tX2844
287 9
291 3631
3634t6t9
36H
PT386l
3949
15
943
middot
l A6l f 2 R(GRESSION RESULTS
EQ I NO
NR CODE cu 121
lA 2001 A (
2A 2004 r (
3A 2005 a
-R
ClS amp31
16
17
1 z
I TER-CEPT bull t
ooo9t21At 00022311 oooeos uu
OOl49tCAI 0010 2 middot 000960CCI
o049tl1At o04t55C61
MARKET SH ARE bull 5 I
-010670Ctl
-017600181
-001050
-0211AO
-0077 0
AOVER-TIS INti
lbl
000()39 000072
000018 0 0024liAI
-000079
SHARf
4ampgtVf Q bull
111
000140
-000170
RSQ
COl
66 485 862
607 6 3 t ()2
blti6 633
flASTI-
CI T Y PH
-007271(1
000441
-u0288l
clASTSHU (t)Ef
bull 10
tto
-234
210
( OOlltOCCI -126520 000922 181 001790
4A 200 7 18 00590liAI -038670 4e16 -032110 lell u c
OOOS74e 000700 025940
00067018) 0021001AI -02l100CAI
5 26
920
5A 6 c
200 8 26 0 03Sl9CA I 0012 26181 o 024511(
-005540
-0 81420 0001961AI O()Ol90CAI ooll240
441 604 891
-0020 19 274
6A 2010 16 Oe04224CAI o359lOIAI 9oq ll4015dCAI 061 6 -middot 005435CAI 00011 8 1AI 917 c 00535ltAI -009850((1 OOOA51J CAI -001610161 971
7A 2012 001862 000010 351 00187) oo1 a oooq45 ono1o6 439 c 001109 -027050 Oe0067SIAI -OOll60C81 897
- middotmiddot- -
8A 201t ll 000757 Ol4140CCI 814 ott9oocet 095 8 oooqo6 C C J 000069(81 811 c ootl2lampBI -080910CCI Oe009lUBI -OOtlOOCCI 960I
I DIIJIT pound11 1V OTIUTVTfAll-1pound1t- A _ 00 R - qq r - Q
1jl)OQCD
0
-middot middot--
l
_ __ _
I
i
0 00141 c c
91)7
I
T ABLE 2 REG ESSIOH RESULTS CCONTI
EO NR
IND CODE
iR bull
cos
I TEq-
CEPT
t1ARKE T
SHA E
AOVER-
Tl SING
SHARpound-cent
AuVER
RSO ILAST1-
C I T Y ELAST
SH CPtf (lt Cl) bull 11 I 41 I 5I (61 171 I 8 I f9) c 10)
9A 2 0 1 8 19 002397181 -007640 367 -005 246 le46 6 000605 000550CAI 597
-000029 G14180 OOl654CAJ -024li01At
lOA 2026 12 0028901AI -OlZl90 -0114 0 1 1 1 5 759
Ll (
llA 2021
f3 c
12A lOZ9
B
(
13A 2301
B
c
14A 2302
B
c
154 201
n (
16A 2551
8
10
35
10
20
19
24
001897(()
00141 6CCI
Oll9lHCI
010604 middot middot
Oll083
OOJ11JCAt
002059141
001411CIH
-000412
000293(81
000187
001159 (3)
ooo6a4cct
o00792CCI
OOl40dCAt
000006
-000076
OOlltl6fA)
o o0 5 64CCt
-034570
-o1middot o3o
-1659 90
Ol51lO
-046190CAt
Ol5450CA)
-005l60
002940
-030060(()
-0 43l50CCt
0 33650
-019080
00004
bullJ00741CAI
-000069
000466
0001 51CAt
000947CAt
O OOI)70CAt OOl750CA J
000175CCI
000671CAI
OOll5) f A t 002207CBI
0004b7CA I
- 007 1 70
001760
-001410CAt
- 0 16200CAt
-004210
- 1 207 70
705
96 0
542
466 805
460
614
921
832
9 61
996
424
550
9 15
56 1 651
884
-006114
019 037
03504lCA)
oo409Q
-141644((
491 -0 17664
607
871
lell
084
101
072
103
108 11IllOQfl)-- l
-037880fCJ OOl6l4CAJ -040l70fCI 0()
c
I
I
bull 9
9t
middot middot middotbullmiddot bull lraquo
TABLE 2 REGRE SION RESUlTS ICONTt
bull
EO I NO tHl I filER- f4A_K T AOVER- sbulluRE ttSQ flA Jl- iLAS T NR CODE cos CEPT SHARE JISING AOVfR bull (JTY SHR CiJCF
110Cl I f amp Itt 15J tbt bull 1c u caa
llA 2702 ll o079 6 81 Bt -019110 S6s -OlltO 111
a 004679 ooooo1 t69
( oOl26o -015160 001560 -022690 ltJr()I
ldA 270 3 19 0065lt -OIt2l60 220 -Ol57a2 11 8 e o 0 71t6IA t ol08 75tAt ( OOl983CiU -060llOC() OOl7191A t -O lltdOC A) middot 89
lqA 2704 10 005082 -Oil060 22b -l)i1977 lelO 6 -000080 oo2772C4 t 910 c 000907 -011 200 Oe04105CA t -046960 995
20A 2d06 1 II
O OlltStiAt -001120 837 00014 -))th O01380amp At -oooo 11 821 ( OOl526CAt -026700181 0002 5UAt 000530 969
2lA 2807 28 OOt9lt9CAt -00891 0 516 -0 03962 225I8 002095CCt OOOl lliAt 690 ( o02965CA t -056650((1 OOOt50CAJ -003l70ttl 911I
I 22A 2808 15 ot3l06 119730 655 1130) 090
0 - 007526CC t Oe003961At 835- middot-
c I oo7523 -067210 001l 2tA t -0091t80CCt 951
2lA 2809 21 Oe2067lCAt -1218101At 71 2 -10095((1 121 ttlIaOQ8 00797418) OOOI8lltt 686 Ill( 0092lU A t -Olltl70Ct t 000605CAt -OOit8701At 951 _
---- -
32 009506CAt
0 bull 06 981 I A J 006901 cA bull
00 24A 2810 -010960 5t 5 -OOltSO 7 56
8 -000011 s11 c -5811701At oooaoatAJ oot520IAI 888
-- 37 --- middot - -- -1 ) l lbull t5JJ r-l bullhmiddot1 --- - --
l5A
26A
27A
c
2RA
c
2 9A
31A
llA
24
13
597
Ci)55
14
l 6h
RSQ ELASTI- (lASTCOE f-
TABLE 2
EQ I NO NR CODE IU
REGRESSION RESULTS CCONTa
NR INTER- MARKET
(10S CEPT SHARE AOVER- SHARfcent
TIS lNG AOVER
) CITY StiR
c It) bull 5 co a I 1) bull 0 bull 9 c l 0 bull( 2 1
2815 16 003019 -009990 275 -o069bmiddotJ
6 001446 000224 282 c 002178 -054630 o02221CCI -0075 00 646
2901 29 00021t2 CAI 0)0520 592 0007 6 1 068
0 00105C8 t 000012141 60)g( OOOlJ51AJ -00126018) ooou271At -000u80CA) 974
36oa 1 3 003296(4) -oca29o -oo-q94
001755 000203 lt91t 002499((1 -04l820CCI OOl374CAI 0068)0 946
3612 0033281CI 011520 195 01684) oso
00 30 42C OI 00012 2 424 oo579ltAt -2 7l7801At 002 100IAI -003220 819
8 3617 btel 011789 081002259 009530
001448 oOOl OitCCI 732 OOl7 22CAI -0 46160CAI 0004961AI -OOll40181 989c
30A 3808 14 0021t 3 6fAI -000110 628 o023291C) -005 6 middot- _ middot- 0bull 02 04 1 IC I ooooo9 637 ( 001584((1 -0161)1)0 00021 91AI -0 00060
390) 14 002790CCt 006590 h8 t 0 0938 4 070
B OOl89bCCJ 000144 74
0021961(1 -OItlOlO OOl323CAI -004800 915
B
middot--- - middot 4 - middot- - dfll3904 12 004563181 0)8300 879 0428651(1 089 OQIll004 277CA I ooozstcea 919
c 0041)3181 -0-11250 0011881(1 -006270 972 0Ill
- J - bull bull bull bullbullbull middot - - middot - middot gt middotbull L)ofbullflbull middot middotfJltl -tn- middotmiddotmiddotmiddoto +f middot t middot bull
c
is
positive
2
Page 0 determineThe consumer goodproducer good split is sometimes difficult to
Sollle support for the split used in this paper is a comparison of goodness - of - fit
distributions of a
industries
set
advertising are
that industry
a sales increase of
the
highest aD10U3 those
measures for the two groups of industries Table 3 contains
measures for the 32- consumer goods industries and the 103 producer goods
2 Median R is 5 7 for the consumer goods set and 40 for the producer goods
Estimates - of y the elasticity of demand with respect to
given in column of Table 2 The hypothesis that y bull 0 with
alternative was applied for each ittdustry As shown in the
could not be rtjected for 24 of the 32 industries Of the
and 10 are positive For the remalnifg eight three are
and five are significantly positive
Of the threertegative elasetcities the largest (in absolute value)
cosmetics (code 2809) With all elasticity of -1 01 a one
industry advertising would gwerate a decrease in industry sales
more than one percetu
At the other extrema is toys and games (code 39 04) For
a one percent ittcrease in industry advertising would lead to
four-tenths of a pampreentmiddot Though not significantly different from zero
elasticity estimatbull for pt-bpuetary drugs (code 2808)
which are
Both drug industries have
to sales rafios (211 pe1 cent for 28 08 and 13 4 per cent
other hand flaIJo ing ext1ac s (mainly soft-drink SYUlS) also
is for
percent increase in
of slightly
a middottwo - tailed
table that hypothesis
24 14 are negative
significantly negative
9
the codetics and p-ropriacuy very high advertising
for 28 09) bull On the
has a high ind UY
advenisiltg t() sales ratio ( 8 0 per cent) bull but its estimated elasticity is very
small and not signifieantly different uom zero
middot It doe$ not seem unusual to find many elasticities near zero with a few
negative and a few positive If aggregate consumption is insensitive to aggregate
advertising as seems likely ld then what one industry gains must be lost by
others
Page 21
elasticity of demand does
(10) we can
if t 0 then
Finally a determination concerning the industry
not fix the individual firms elasticity Going back to equation
write y bull (1 - ) a + z y Since a gt 0 it follows thatii i i
Y gt 0 If Ylt 0 however y s sign is indeterminateij ij
Turning next to the coefficient of market share 20 of them are negative and
12 are positive with three of each sign being significantly different from zero
The largest positive value which is significant is for pet food (code 2010) and
a close second is mens amp boys suits and coats (code 23 01) The largest signifi
cant negative value is for cosmetics (code 28 09)
In the model some relations among y a 0 and 61 are implied If middota is
negative (1- y) is greater than one Since 61Y bull (1- a) (1 - ) 61Y is
pCsitive but it will be less than one if 0 is large enough Seventeen of the A
ys are negative in each of the seventeen industries 61 is also negative and les
1 (algebraicly) than y so 61Y is greater than one his is not evidence that the
degree of cooperation is low or zero in any industry of course A high o togethegt
with a negative y which is small (in absolute terms) relative to a can give a
value of 61Y greater than one On the other hand if 61Y were less than one
o would have to be greater than zero
For y gt 0 two conditions may hold If y lt a it follows that 61 lt 0 since
(y - a ) lt 0 It also follows that 61Y lt 0 regardless of the value of o bull
In addition the relation is an if-and-only-if one ie if 61Y is negative
then y lt a Three industries have negative values for 61y dairy products
except milk (code 20 04) synthetic fibers (code 2806) and photographic equipment
(code 3808) In all three cases y is quite small so a does not have to be very
large to give a negative value to al
If y gt a the model shows that 61-- gt 0 whatever the value of 0 bull There are
12 cases where both y and 61 are positive with 61 ranging from near zero
(grain milling code 20 12) to 1 20 ( proprietary drugs code 2808) Several
coeffiei-en_rs
--
15
o f to zero
Tab le 3 - Consumer Goods
Indus tr ies
2Distributions of R and Producer Goods
Page 20a
Consumer Producer
Goods Goods
0 - 9 9 0 3
10 - 19 9 0 1 1
20 - 29 9 3 20
30 - 39 9 3 17
40 - 49 9 5 16
50 - 59 9 8
69 9 6 Ii 1060
--f 670 79 9 2
80 - 89 9 4 4
90 - 100 1 1
Totals 32 10 3
Notes
2is used R i s cal-
culated as F F + ((N - K- 1 ) K ] 1 b_ F is the F
statistic Lor the hypothesis that all o f the non - intercept terms are simul taneously equal
1
2
--
-- Since generalized least squ-area regress ion
Page 12
o f these cases have values o f 6 1 Y rea ter than 0 8 (pet food (20 10) bread amp cakes
(20 14 ) miscellaneous processed foods (20 29 ) men s and boys suit s and coats
(23 01 ) p rop riet ary drugs (28 08 ) elect ric housewares (36 1 2) radio and
IV sets 17 and games and toys (39 For be than(36 ) 04)) 6 1 Y to greater 0 8
both o and cry mus t b e less than 0 2 That is in these industries the results
shown in Tab le 2 would hold only with very strong dominance of marke t demand
effects over brand swit ching e ffects and a very low degree o f cooperation
Using the concentration rat ios given in Table 4 the average concentration
ratio for the eight indust ries is 38 1 For the remaining 24 industries it is
43 2 This result would lend some support t o the proposition that concentrat ion
and cooperation are related if the group of eight industries are identi fied
as having low degrees of cooperation
One f inal note on model predict ions S ince both cr and o are assumed to
be non-negative 60 should also be non-negat ive In only one o f the 32 cas es
is 60 negative and it is not signif icant bull
cor pJ
7 1
7 0 4
1 4 7 5 1
)
7 3 z a o a t s 1 1 3 3 5 ( i V Z 8 tJ1 2 1 - 1 1 0 7 3 5 c middot v 2 8 1 0 3 2 -0 0 1 4 3 0 C 1 V 9 8 3 2 8 1 5 1 6 - 0 0 6 t1 6 9 C _ V 1 11 0 2 9 0 1 29 0 00 7 6 1 C l k V 7 3 1 3 6 08 1 3 - o n 4 9 9 4 3 3 3 6 1 2 2 4 0 1 6 4 3
3 1 7 1 3 0 1 1 7 d 7 8 3 3 8 0 8 0 0 2 3 i 2 3 0 6
P 4
3 9 0 3 1 4
1 2
0 09 3 8 4 39 1)1 0 4 l 8 6 5
T A LE 4 e I NCU S T R Y C A T E G O R Y
F TC NOM e tOOE ass
C 1 I c 2 )
2 0 0 1 1 6 2 0 0r 1 7 zo o l Z 2 0 0 7 1 8
z o o 2 6
z o t a 1 6
2 1) 1 2 1 5 2 0 1 4 1 1 2 0 1 a 1 9 2 0 26 1 2 2 0 2 7 1 0 2 0 29 - 3 5
2 3 0 1 10
2 3 02 z o
2 3 03 1 9
2 5 5 1 2 4 Z 1 0 Z 1 1 2 0 3 1 9 2 7 4 1 0 z a o 1 3
2 A 0 7 z s
E LA S T I -C I T Y
c 3 )
- 0 09 7 2 7
0 00 449
- 0 0 2 8 9 3
- 0 3 2 7 70 - 0 OZ il I q
o 4 u t s a 0 0 1 3 7 3
o 1 4 o o-0 0 5 2 4o -0 1 94 0 1
-C l 6 l l 4 0 1 JJ 3 7 0 3 t 0 4 2 0 0 4 0 9
- 0 4 1 8 44 - O l 7 t ) 4 -ll 7 1 4 0 b -0 3 5 7 8 Z
- 0 1 9 7 7 0 1)0 3 3 4
-o n 9 2
IJ A T A
CONV
ON-C ONV
( 4 )
t 1NV
C 1t-V C NV c ) v C - JNV C rlNV c n v c a v C ON V ( i) l V
C CtV
C t1tV C IJNV
C 1 V
c 1 v
Pa ge 2La
HEQ F t Noe x
c 5 )
6 2q
6 89 1 48
6 1 4 ) q 3
1 9 7 1 9
q z t s 5 a e z
l l 9 1
4 o 7 1 2 0
c oqt 1 Z
4 4 )-q 1 z
5 5 9 1 0 64 1 3 3
n 3
Inter industry analysi
Page 23
Equation ( 3 6 ) may be used to fo rmulate a rough test o f the cooperat ion
vs e f ficiency controversy I f we rest ricted ourselves to some subset o f
indus tries where the relat ion be tween y and a i s the same we would be lookingbull
at a sample o f indus tries for which the dependence o f V on H and o is a
quadratic wi th no squared terms
I f we now sub s titute (26) into ( 36 ) we get
(38) v bull a + (y - a) R + (y - a) o (H) - (y - a) H o (H) + u
If it is only d if ferences in marke t shares that affect advertising inten-
s i ty levels i f concentration does not af fect cooperation and i f o bull 0
( 38 ) reduces t o
( 39 ) V bull a + (y - a) H + U
Let this b e the null hyp o thes is and note that V is a linear func-
tion o f H If y gt a and the null hyp othesis holds the expected relat ion
between V and H is as shown by the pos itively sloped straight line in Figure
6 b etween points A and C
In this context the alterna ive hypothesis the degree o f
cooperation ( o) is a positive function o f concentrat ion
is that
(H) I f the func-
tion is as described in (26) and Figure 4 the relation between V and H
would be as shown by the curved line connec t ing po ints A and C in Figure 6 One way to test the null hypothe s is against this al ternative is to test fo r
the hypothesized curvature in the function relating V and H
Fi gure
R
6
v
ID c
H 1 0
I
i I I
sfigned
clasJ s
in def ini g
effec t is the
I
Page middot2lf
One charac teristic of produc ts which may be useful
classes wi th respec t to significanc e of the brand-switching
conveniencenon-convenienc e distinc tion developed by Porter ( 1 97 )
Using his identif ication of IRS indus tries as a s tar t ing point
each of the 32 industries used in this s tudy to one of the two
The ass ignments are not ed in Tab le 4
Using the conveniencenon-convenienc e dis t inc tion the
divided into two sub-samples
sample was
Four regressions were then run bull ene for e ch
sub-sample one for the to tal sample and one for the total sample with
a dummy variable for convenience goods Each regression was run wi th LB s
and wi th indus tries as ob servations The results are given in Tab le 5
For the LB level equations ( 1A - 1D) equation(37) was used ome
rearranging of terms gives v bull y - ( 1-o ) (y - a) ( 1-z ) + ui i i
bull y - 6 1 ( 1 - z ) + ui When observations are pooled across indus trif s yi
is an industry variable In these regressions the values o f y es t imated
in the first stage were used Since industries have been grouped so as
to reduce differences in (y - a) that term is t reated as a cons tant
within each group Finally H the Herfindahl index has been subst ituted
for o the degree of cooperation The resultant equation is linear
y ( 1 - z ) and H x ( l - z ) This is the equation for whi ch results arei i
reported in Table 5 In all regress ions all variables were weighted by the
square roo t of sales
in
raquo
- 34
3 1
i l T 4 8LE S R E G E S S I ON R E SUL T S i X T f N S I ON S T O T Y PE Of
AND L E V E L OF C ONC E kT R A T I O
E O O tt S
tP bull T Y P E NR C ONV O B S tO N-CONV
E L A S T I shy
T I C J T Y
O E - t-41( T SliA R E
H( fl F
1 1 - K O x middot K O X E R F
- H S a 1 1 - H S t 1 8 1 1 9 1
R S O
1 1 0 1
q z
t - bull 9 f 2 1
Z t P D l 6 fC ONV - 1 9 1 s s c 2 1
e - l O I e bull u q a
2 3 1 0 1 1 Nl N- t O V 5 7 1 ( 1 2 19 1 6 1 - J l S 2 5 4
l l l l t z q o a 1 - J o v z I
l ( I NO l l A l l ol u l t o o c bull J u l l
d S
c 3 4 t t r a l
bull Z D 1 0 l l 4 l l 2 3 -gt 2 l l 2 0 - 4 lS 5 4 l C 4 1 o e 6 t l b b 5 1 1 0 1 1 1 c - 1 9 5 1 t d S 2 1 tV
OQ10 T-R A T I O S A E G I VE N l P A RE N T H E S E S l E V E L S f S I G 1 F l C ANC E A - 99 5 B - q 9 t - s NS J UC E NO I Nl E R C ( P T I S U (j f middot) I N f Hf t OUA TIOI S 6 E f 3R E A O J U S T H E N T F OR 1-t E T E R O SIC E O A S T f C I T V pIraquo
1
middot
I
R SO I S C A L C UL A T E D A S f I f bull 1 1 bull 1lt 1 K t t- E E N I S THf U A1 B E R O f fJe S E fi V A T I O S F t S HtE F - S T A T I S
I N DE P E ND E N T y II
T I C F O T H E U Y P O HIE S I S f A l l C JE F F I C H N T A R E QU A L HI l f RO A f40 K I S T H f UUM 8 E R OF Q ( A fL f S
I l l
1 A
1 6
t C
f 1 r R I l l
1 2 1
L Jt
L B
l q
L 1
on s T Y P E bull 2 )
P OUC T
X
H S I l l
l O l A I bull l 0 7 1
O O b O J
1 2 00 1
1 b 2 1 C I l l Z U
l l S b 1 -1 1 0 1
rlt F X l l - fi C f l
bull 1 I
l l
- ()4 2 3 l f H
1 - 2 7 9 1
K O X
1 1 - ti E R F I
l d l
S4 l i A I 2 l b
1 3 00
K O X -t f R F t S J
I 1 - HER F t
)
60 9
z o bull
0
t l l
3 8 8
1 7 6
5 64
64
N f bull
tl S bull
I l l
I 4 1
t ONV
NON-t ONV
A l l
A ll
c o v NO- ( ONV
1 4 1
1 5 1
- l S b
c - 1 1 2 1
5 8 2 C A I 1 6 7 0 1
middot I l 9 H
l 1 il
1 4 1 6 1
E L A C) T J shy
l t C I T Y
I S I
l b l
- 0 2 0
1 - 1 4 9 1
O Z 3 2 1 A I C b 6 b l
0 0 9 2 -J
1 1 b 4 1
o z t t t l 1 2 2 1 1
oru - Ht tt F
1 gt X bull b I
Page 25 bull
Mean val -as ior relevan t variables for the two sub-samples are
Conv Non-Conv
Adv Sales 4 7 2 3
iit Share 2 7 3 2
Ela cicity - 05 3 02 2
Herf indahl 7 3 1 860
That the dist nct ion between coft9enienee goods and non - convenience
goods mat ters is supported not only by these means but also by the regression
results Equat ions lA and lB are different in almost every respec t Only the
concentrat ionmarket share int eract ion term seems to mat ter for convenience
good s Jus t he opposite holds for non - convenience goods both elasticity1
and market sl e are highly significant but the concentrat ionmarket share L
int erac t ion t=rm is not
Given that one minus market share is used these equ3 tions the coeffici t
of market share is the negative of what is sh in col (6) of Table 5 For the
regress ion for non - convenience LB s reported in equation lB then LB s with larger
market shares have lower advertising to sales ratios
On the question of the impact of the degree of cooperation the story is mixed
For convenienc good s the coefficient is pos itive as equation (31) predicts when
there is a po sitive degree of cooperat ion Fo r the non-convenience goods LB s middot
concentration has no effec t
Aggregat ion t o the indus try level and then rer unning the equations $ives the
results shown in (2a-2d) of Tab le 5 As expected R2 goes up and levels of significanc e
fmiddotJr individual coefficients gJ down In add ition the coeffic ient of H ( l -H) inmiddot
quat ion 2A is no t significant though it is po s i tive Since 2A is the indus try level
I
--
all but one of them and that one is not significant For 18 of the 31
wo uld have on its adve rtising to sal es ratio
cognate of LA the variable H ( l-H) pl ays the same role in 2A as H ( l -z 1) p lays
in LA
Vconomies of scale in advert ising
The literature on advert ing contains a number of definitionbull o f economies
of scale and of findings eone rning its presence and magnitude Before I loo k
a t the data with the hope o f mmenting about scale effect s then I wan t to
define the term as I use it
By economies of scale I mean a decrease in the number o f units of adver-
tising which are needed to generate a unit of sales as advert ising is increased
Given the definition of the firm s own elast icity of demand with respect to
advertising (see the text foll wing equation (1) ) the technical definition
is that rii gt 1 Diseconomies of scale are def ined symmetrically i e
Yu lt 1
ith a cro s s - section oi observations on advertising intensity
(v bull a s1) and onmiddot advertising (a ) we may regress the former on the latter 1 i i
The results of such regress ions for the 32 industries are given in Table 2
lines lB ZB bull bull bull 32B For the 32 eases the regression coefficient is positive it
positive coefficients the coefficient i s signif icant
The que tion now is whether these results ar e evidence of pervasive dis-
economies of scale in advertislng I think no t Given my as sumption that the
data I observe are equilibrium results I may ri ghtly conclud e that almost every-
where high levels of advert ising are as sociated in equilibrium with high ratios
of advertising to sales I may no t conclud e anything however about the
impact that a move by some firm away from its equilibrium level o f advertising
middot- Page 27
Given the assumed relation in equation ( 37 ) there is a corre spondshy
ing equilibrium equation relating v and ai The equation is non - linear i
and I have not found a simp le way to characterize it It is po ssib le to
det ermi ne its deriviative with respect to a b y takin$ tne de riviative of ( 3 7 ) i
When that is done the result is
v - 1(40) i bull sl a a z ( 1-z )i i ira-l
Sinc e all the other terms in (40) are non-negative the sign dep ends only
on the sign of 6 1 which is in turn de pend ent only on the sign of (y - a)
That the signs of the coefficients of a in equations lB 2B bull bull bull 32B arei
not always equal to the signs of the coefficients of z in LA 2A bull bull bull 32A i
is still a mystery to me and something to be explored
bull
Summary
My purpose in writing this pa per wa s t o attempt to integrate explicit micro-
economic mode lling reasoned econome tric specification of error terms and high
quality data to exp lore some questions about adver tising by large firms
al quite important improvements could be made in the first two of those areas
and much data ma ssag ing is yet to be done in the third
Concerning the subst an ce of what I have done I think six things are im-
Page 2 8
Severshy
V and Conclusions
portant
1 Explicit modeling is worth the ef for t if for no other reas on than
that it provides a basis for choosing fr om a varie ty of funct ional
forms
2 The same is true for error specif ica t ion I did in fact look at scatshy
ter diagrams for all 32 industry categor ies and they vir tually all show
the het eroscedast icity which I as sumed and for which I correc ted
3 The predicted relation be tween advertis ing in tensity and market share
shows up clearly in on ly 25 of the case s examined I have ye t to
explore why that may b e the cas e
4 Some evidence concerning the presence and impact of coopera tion was
produced but it is not clear ly pervasive
5 The dis tinction between convenience goods and non-c onvenie nce goods is
unambiguously a good one
6 Virtually no evidence concerning econ omies of scale in adver tis ing can
be gleaned fr om this study given its assumpt ions
More work is called for on many of these issues my study seems to have
raised more questions than it has answer ed This is pr obab lly due to the richshy
ne ss of the data source since I had the oppor tunity to address several issue s
at once
- -- - Q --
Foo tnotes
Manager Line o f Business Program Federal Trade Commission The author depended heavily on several staf f members of the LB Pro gram for statistical and clerical support in particular Joe Cholka George Pascoe and Haro lene Jenkins Helpful comments were given b y Richard Caves Dennis Mueller Michael Porter and F M Scherer When the paper was in its early stages o f development both S teve Garber and Jon Rasmussen former economists in the LB Program provided eery useful cri tiques The views expressed here are my own o f course
1 The basic exp licit model of the role of advertising is Dorfman and S teiner ( 1954 ) and the models in this paper may be seen as ano ther extens ion of their work
2 For a review o f models of oligopoly which treat prices and quantities as relevant variables see Shubik ( 1 959) Extens ions to take advertis ing int o account are fairly s traightforward
3 Dorfman and S teiner dealt with a price-setting firm For a monopolis t inverting the demand relation to show Pbull P (Q A) _instead o f Q- (P A) has no effect on the results o f the analysis For oligopolists who face a sys tem of demand relations inversion does have some impac t on the results The dif ferences are not trivial nonetheless they will no t be explored in this study
4 Shubik ( 1 959) p
5 From a formal mathematical point o f view the function is similar to a1 i Legrangian funct ion of the form - r rr
L bull 11 + A ( j - (i i i+j
where the indicates some fixed level of the profits industry Following the same logic which resul ts in and income constraint as the marginal utility of the of the indus tr in the obj ective of firm i
6 Hall amp Weiss ( 1 967) p 323
7 See Schmallensee ( 1 97 2 ) Ch 2 pp 1 6-47
8 On this poin t see Imel amp Helmberger ( 1 9 7 1 )
9 Long ( 1 970) p 130 - 1 39
1 0 See Schmallensee ( 1 972) Ch 3 pp 48-8 7
rrj i+j ) of the rest of the
the identification profits of the rest
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980
making
Page 10
Substituting ( 1 1) now in (6) the Cournot case and in (9) the
Chamberlin ease gives
-(19) Cournot vO a + (y - a) z
Chamberli n vl - y lmiddot th
The 1- element is then
-(20) Cournot vO a+ (y - a) zi i
-Chamberlin vl yi
In (19) and (20) the superscript 0 is used to denote total
independence of decision and the superscript l is used to denote
total interdependence of decision making
The relation between v and vf depends solely on t relation between
y and a f y gt a so that the industry effect dominat 3 vf v with
equality in the trirlal case of z bull 1 Cooperative irms would have ai
higher advertising intensity than non-cooperative fi
If there is equality between the industry elasticity and the
y Whe therbrand-switching elasticity or y bull there is cooperation among the firms has no effect on advertising inten-sity And if brand-switching is more important with y lt a e get
vi lt v cooperative firms will have a lower advertising intensity than
non-cooperative firms The three situations are depicted in Figure L
Given the results for the firm level variables calculation of the
corresponding re sults for the industry is s-traightforward Let
B bull t zf bull z z be the Herfindahl index of concen tration and notethat
z - bull l z bull 1 and that V bull A S bull a i 5 bull (siIS) (asi)i
bull middot Then for the general case from (6) and (9)
bull
Page lOa
y lt1 bull y y
10
f
V0 bull z [ a1 + (y - cr) J bull a + (y - a) H vl bull Y bull y
11Page
( 2 1) vo bull z d -1 r bull r z bull z- d d- i )11
l -1 N N v bull z d r z bull r z bull l y z - ij jibulll jbulll
and for the simplified case from (19)
(22
Since the forms of the equations in (22) are the same as in (20) all
of the analysis of (20) holdS for (22) except that His substituted for
middot z Advertising intensity will be higher for an industry with cooperativei
firms than for one with non-coope rative firms if the industry demand elas-
ticity dominates the brand-switchine lasticity If the ewo elasticities
are equal then so will be advertiSl intensity for the cooperative firm
and non-cooperative firm industries If brand-switching dominates the
cooperative firm industry will have lower advertising intensity
In the industry model context the possibility of a critical concen-
tration level may be easily introduced That is assume that at values
of R below R the firms are non-cooperative but that for alues equal to
or greater than H they are cooperative That is
(23) V bull a + (y - a) H if H lt H bull y if H gtmiddot H
This function is shown in Figure 2 for the three relamptions be eween y and a
IV THE D GRE OF COOEtRATION
The models depicted in Figure 2 are based on a very simple notion about
the relation between concentration and the cooperationnon-cooperation charac-
teristic They use what is essentilly a step function relating concentra-
Figure
I -
J H H
Page l la
2
v v v
a
I J I y a bull Y y ---i
I I I I I I II I
H 10 H H 10 H 1 0
bullbull
bull bull
Page 12
tion and the extent or degree of coo peration as shown in Figure 3 Using
o as a symbol for the degree of cooperation eq uation (23) can be rewritten
as
bull(24) v (1 - o) vo + ovl bull (1 o) [a+ - a) H ] + oy- (y
vhere
(25) o bull o(H) bull 0 if B lt B
Hbull 1 if B gt
The specific relation between concentration (H) and the deg ree of
cooperation ( o) shown in Figure 3 is more restrictive than necessary and
bas less appeal an one which shows a smooth increase of o from 0 to 1 as
B increases fro to 1 Such a smooth function is shown in Figure 4 he
degree of cooper ion is low until H gets close to H increases greatly as
B goes through R attaining a high level for an H larger than B but still
close to it and then goes to 1 as B goes to 1 More formally
(26) 0 bull o(H)
bull 0 if H bull 0
bull 1 if H bull 1
do gt (IdB
d2o- gt 0 if H B dBZ
0 if H H
lt 0 if R gt H-
If (26) is now substituted for (25) in (24) a corresponding smoothing
of the functions shown in Figure 2 is accomplished he smooth functions
Figure 3
I
10
H 10 H
Page 12a
Figure
Page 12b
4
10
H 10 H
l
(27)
Cl - o) o
[ (1 -
[ (1
t y ij z
j
are given in Figure 5
Page 13
The equation in (24) was constructed as a hybrid of two results derived
from first-order equations for the non-cooperative and cooperative models
It may also be constructed directly by using l bull (1 - o) + on as thei i
th 2objective function for the i-- firm Treating o as a constant the first
order equation is
0 an aa
i middot
n i
Using equations (3) and (7) and setting --- bull 0 givesaai
(28) - 0
After collecting terms and converting to matrix notation we get
(29) - Cl - o) + 0 z-1vi yii i
N
j bulll
+ 0
+ o v1
d-1 r z z
-(30) v Cl - o) rd
-
o) r +d
d r J _ -1- d z
and middot
d) r + 0 rjd
l r z
v - z --(31) zv
- (1 0- +d) t r z- d-
If we now impose the simplified demand structure by substituting (11)
we get
- (1 - o) a +( 32) + oy and
Figure
l
-y
v
Y
H
r
I I
v
a
y
H 10 H
v
a
bulla
HH 10
bullbull
(33)
i
Page 14
v - (1 - o) [ a l + (y - a) ] The corresponding equation for Vis given in (24)
If the relation between industry advertising intensity (V) and
concentration (H) is non - linear via the impact of concentration
on the degree of cooperation does the relation take the form of a
auadratic as some other investigators have proposed (Greer ( 1971 ) Martin (1979) Strickland amp Weiss (1976) ) That question can
be approached from two perspectives - conceptual and empirical In
the next section I will try to apply some statistical material to it
here I want to explore the question in the context of the models
developed above
Referring to equation (24) taking the deriviative with respect
to H and letting bull (H) gives
bull( 3 3a ) 3V (y - a) (1 - o) + (1 - H) 3H dH
Given that o and H each are bounded by 0 and 1 and that o is an in-
creasing function of H the term in the brackets in non-negative The
sign of the deriviative then is the same as the sign of (y - a) If the industry effect is dominant (y gt a) then advertising in-
tensity will increase with concentration If the two effects are equal
then advertising intensity will not vbullry with concentration and if the
brand-switching e ffect is larger advertising intensity will decrease
with concentration
For given advertising elasticity parameters then it is not pos-
sible to show an increase in A S up to some H and the n a decrease
he relation may be non - linear but it is monotonic over the permis-
sible range of H
Page 15
One possibility is to introduce a relation between (y - a) and H
If industries characterized as dominated by the brand-switching effect
also were more concentrated then a cross industry comparison would
show an inverted U shape relation between AS and H I know of no
support for such an association between ( y - a) and H
V ERROR SPECUCATION
So far I have implicitly assumed that there are no errors in the
model I now assume that the error
thvariable in the equation for the i- firm is additive that its mean is zero
that its variance is inversely proportionate to its sales and that all error
term covarlances are zero hat is
34) v bull (1 - o) [ a + (y - a) z ] + oy + u i i i
- - -
In vector notation
(35) bull (1 - o) [ at + lt - a) J +
E() bull 0 Cov (u) -
0
If we now ca lculate V bull bull we get
oy + U bull
(36) V bull (1 - o) la + (y - a)R J +
U bull E(U) bull 0 Var (U) bull
P age 16
The error term specifi c ation which follows Hall
wi th the ratio of profits
6 heir analysis seems
as wellbull particularly given
and Weiss was
applied by them to equa tions to equity or
asse ts as dependen t variables to carry the same
weight in the present context the substantial
evidence presented by marketing analysis on the tendency of company de-
cision makers t o use the advertising to sales ratio as the decision
7variable
The zero covariance assumption on the other hand1 is particularly
troublesome Within a probably not true since
many even ts outside the
single industry it is
industry would have
industries
similar effec ts on all the
8firms in the industry Between it is probably not true either
since many companies produce in more than one indus try and even ts in the
firm would tend to effect all of its activities The extension of the
model to allow for non-zero covariances is certainly called for
--- -
Page Ll
Athird characteristic of this specification also deserves some comment
only on equation is shown for the firm or industry Several haveinvestigators
included an advertising equation ina larger system of equations (Comanor and
Wilson (1974) (1979) and Weiss (1976) so in myMartin Strickland Idid
dissertation1 Ihave ewo defenses for presenting only single equation results
here One is the suggestion by Comanor Wilson that simultaneousand equation
verybias may not be important in this context he second is that Iam will-
ing to see the model extended to include a profitability equation and I intend
to move that direction in the near futurein
Onthe inclusion of aconcentration equation on the other hand I amskep-
tical he problem isthat in some fundamental sense equations in models of
industrial organization should be oriented to the decision-making contexts of firms
Ifeach firm in anindustry with Nfirms has only one decision variable then
there can be only Nindependent equations which are based on the first-order
optimization equations It not show large ofis difficult of course to a number
functional relations which follow from the first-order conditions point here he
isthat only Nof them can be independent
Consider for illustrative purposes only a situation in wich teh N
itsquantities say q are set exogeneously and each firm determines advertising1
of and level say ai where a is the true optimal value a a bulla
1 1 i i + i
Ifp a) then the Nq
s and the Na s will determine N p s say pii bullp
i(l i 1 i
Now given Eand 9 we can determine both the vector s andthe vector v since
- bull til
ands We can also determine S s z bull s-1 s-ii
v and we
industry concentration isHbull (alternatively
V and Hare functions of the same
pendent Furthermore if there are
VM) and (H1 H2 bull bull bull )contain
VI STATISTI CAL APPLICATIONS
Page 18
Given the observed values of might be tempted to conclude that we
have a two equation system in which v and z are jointly determined That would i i
be inappropriate however since the 2N random variables (v ) are determined by
onlt N basic random variables t The variance-covariance matrix for ( ) must
be singular bull
The extension of this observation to industry level variables is straightshy
forward Since observed industry advertising intensity is V bull l v and observed 4
C4 bull z ) it follows thati l i
set of random variables t and cannot be indeshy-
M industries then the variables (V1 v2 at most M independent random variables
Given the assumptions made above y a and o are constant within a
given industry That being the case (34) can be written as
(37)
s1 bull (1 - o) (y - a)
If and
So + +
(1 - o) a + oy
81 are esti mated for an industry using a suitable statisticalSo procedure giving and 81 an estimate of y is then available The sumSo
Intra-industry analysis
1974 of Business Re ort -
categories are identified
Page 19
of So and S1 is y so y bull So + S1 is the estimate of y
If o is assumed to be bounded by zero and one the sign of 61 is also
the sign of y - a The estimate of So is of cou se an estimate of
(1 - o ) a + oy Moreove if y gt a then a is less than 6o- (1 - B) a+ 0(
Cor esponding results would
hold if y bull a or y lt a
If 61 gt 0 then we would estimate that a lt 6o
bull
For purposes of estimating this model 32 manufacturing industry categories
were selected The data are fram the Line of Business forms filed with
the Federal Trade Commission they are discussed in detail in the Annual Line
which is available from the FTC The 32 industry
in Table 1
Simple regressions for equ ation (37) were run for each of the 32 industries
with data for each of the firms which filed in the industry constituting an ob-
servation The results are given in Table 2 lines lA 2A
1974
the intercept
For all the A
equations generalized least quares was used th both and market
share bein scaled bv the square root of sales
The 32 industries were taken from the larger set of all manufacturing industry
categories in the 1974 data files Two selection cr iteria were used thereLB
had to be at least ten reporting companies and the industry had to be a consumer
goods industry or a producer goods industry which is strongly associated with a
related consumer goods industry so that adve rtising by the firms in the producer
goods industry impacts on demand in the related industry (eg soft drink syrup
and soft drinks) There were 135 industries with at least ten companies - 103
producer goods and 32 consumer goods
273
274
TABLE 1 INDUSTRY CATEGORIES USED
FTC
COOE OESCRI P T I ON _
2001 MEAT PACKINGt SAUSAGE S ANO OTHER PREPARED
MEAT PROOUCTS
2004 DAIRY PRODUCTS EXC F LUID MILK
2005 CANNED SPECIALTIES
2001 FROZEN SPECIALTIES
OEHYORATFOt AND
I NC LU 0 ING PRE SERVE S
SOUP MIXESbull
SEASO lNGSt ANO
OTHER PET FOOJ
RAIN MILL PROOUCTSt
ANO PREPARED
RELATED PRODUCTS
PRODUCTS
Z008 CANNEOt ORIEOt PICKLED FRUITS
ANO VEGET ABLES JAMS JELLIES DEHYDRATED I G TA3t E
S40CES ANO SALAD ORESSINGS
2010 DOG CAT A O
2012 FlOUR t OTHER RICE
MILLING fiLENOE) FLOUR
2014 B EADt CAKE ANO
20113 CO FECTIONERY
2026 BOTTLED A O CANNED SOFT DRINKS 2021 FLAVORING EXTRACTS AND SYRUPS EC
2029 MISC FOOOS AND KINDRED PROJUCTS EXC
ROASTED COFFEE
2301 MENS AND BOYS SUITS AND COATS
2302 MENS ANO BOYS FURNISHINGS
230 3 kOMENS AND MISSES OUTER EAR
2551 HOUSEHOLD FURNITURE
2702 PERIODICALS
Pqe 19a 1
RELATED 1972
src cooe
2011 f 3
202tX2026
2032
203pound
2033t4t5
2047 2041t4t5
2051 2065 2086
2081
209XZ05
231
232
233
251 272
2703 BOOKS
2704 ISC PVeLISHING
2806 ORGANIC FlEERS
2807 DRUG$ ETHICAL
23234
PT 2 S 3
2808 ORUCSt PROPRIETARY
2809 PERFU ES t COSMETICS ANO OTHER TaiLET
PREPARATIONS
2810 SOAP AND OTHER CLEANING PREPARATIONS 2815
2901
PEST1CIDES AND AGR ICULTURAL
PE OLEUM REF I N ING CHEMICALS EC
3606 HOUSEHOLD COOKING EQUIPMENT
3612 HOUSEHOLD APPLIANC ES NEC INCLUJING ELECshy
rqlC HOUSE ARFS AND FANS ANO SE I G MACHINES
3617 RADIO AND T V RECE I VING SETS 3808 PHOTOG APHIC E UIP amp SUP LfESt EXC
3qo3
PHO TOCOPYING
SPOR TING A D E QU I P M ENT f SUP P L I E S ATHLETTC GOODS NEC
3904 DOLLS GAMESt TOYS AND CHLOqENtS VE ICLES
PT283
2844
234tX2844
287 9
291 3631
3634t6t9
36H
PT386l
3949
15
943
middot
l A6l f 2 R(GRESSION RESULTS
EQ I NO
NR CODE cu 121
lA 2001 A (
2A 2004 r (
3A 2005 a
-R
ClS amp31
16
17
1 z
I TER-CEPT bull t
ooo9t21At 00022311 oooeos uu
OOl49tCAI 0010 2 middot 000960CCI
o049tl1At o04t55C61
MARKET SH ARE bull 5 I
-010670Ctl
-017600181
-001050
-0211AO
-0077 0
AOVER-TIS INti
lbl
000()39 000072
000018 0 0024liAI
-000079
SHARf
4ampgtVf Q bull
111
000140
-000170
RSQ
COl
66 485 862
607 6 3 t ()2
blti6 633
flASTI-
CI T Y PH
-007271(1
000441
-u0288l
clASTSHU (t)Ef
bull 10
tto
-234
210
( OOlltOCCI -126520 000922 181 001790
4A 200 7 18 00590liAI -038670 4e16 -032110 lell u c
OOOS74e 000700 025940
00067018) 0021001AI -02l100CAI
5 26
920
5A 6 c
200 8 26 0 03Sl9CA I 0012 26181 o 024511(
-005540
-0 81420 0001961AI O()Ol90CAI ooll240
441 604 891
-0020 19 274
6A 2010 16 Oe04224CAI o359lOIAI 9oq ll4015dCAI 061 6 -middot 005435CAI 00011 8 1AI 917 c 00535ltAI -009850((1 OOOA51J CAI -001610161 971
7A 2012 001862 000010 351 00187) oo1 a oooq45 ono1o6 439 c 001109 -027050 Oe0067SIAI -OOll60C81 897
- middotmiddot- -
8A 201t ll 000757 Ol4140CCI 814 ott9oocet 095 8 oooqo6 C C J 000069(81 811 c ootl2lampBI -080910CCI Oe009lUBI -OOtlOOCCI 960I
I DIIJIT pound11 1V OTIUTVTfAll-1pound1t- A _ 00 R - qq r - Q
1jl)OQCD
0
-middot middot--
l
_ __ _
I
i
0 00141 c c
91)7
I
T ABLE 2 REG ESSIOH RESULTS CCONTI
EO NR
IND CODE
iR bull
cos
I TEq-
CEPT
t1ARKE T
SHA E
AOVER-
Tl SING
SHARpound-cent
AuVER
RSO ILAST1-
C I T Y ELAST
SH CPtf (lt Cl) bull 11 I 41 I 5I (61 171 I 8 I f9) c 10)
9A 2 0 1 8 19 002397181 -007640 367 -005 246 le46 6 000605 000550CAI 597
-000029 G14180 OOl654CAJ -024li01At
lOA 2026 12 0028901AI -OlZl90 -0114 0 1 1 1 5 759
Ll (
llA 2021
f3 c
12A lOZ9
B
(
13A 2301
B
c
14A 2302
B
c
154 201
n (
16A 2551
8
10
35
10
20
19
24
001897(()
00141 6CCI
Oll9lHCI
010604 middot middot
Oll083
OOJ11JCAt
002059141
001411CIH
-000412
000293(81
000187
001159 (3)
ooo6a4cct
o00792CCI
OOl40dCAt
000006
-000076
OOlltl6fA)
o o0 5 64CCt
-034570
-o1middot o3o
-1659 90
Ol51lO
-046190CAt
Ol5450CA)
-005l60
002940
-030060(()
-0 43l50CCt
0 33650
-019080
00004
bullJ00741CAI
-000069
000466
0001 51CAt
000947CAt
O OOI)70CAt OOl750CA J
000175CCI
000671CAI
OOll5) f A t 002207CBI
0004b7CA I
- 007 1 70
001760
-001410CAt
- 0 16200CAt
-004210
- 1 207 70
705
96 0
542
466 805
460
614
921
832
9 61
996
424
550
9 15
56 1 651
884
-006114
019 037
03504lCA)
oo409Q
-141644((
491 -0 17664
607
871
lell
084
101
072
103
108 11IllOQfl)-- l
-037880fCJ OOl6l4CAJ -040l70fCI 0()
c
I
I
bull 9
9t
middot middot middotbullmiddot bull lraquo
TABLE 2 REGRE SION RESUlTS ICONTt
bull
EO I NO tHl I filER- f4A_K T AOVER- sbulluRE ttSQ flA Jl- iLAS T NR CODE cos CEPT SHARE JISING AOVfR bull (JTY SHR CiJCF
110Cl I f amp Itt 15J tbt bull 1c u caa
llA 2702 ll o079 6 81 Bt -019110 S6s -OlltO 111
a 004679 ooooo1 t69
( oOl26o -015160 001560 -022690 ltJr()I
ldA 270 3 19 0065lt -OIt2l60 220 -Ol57a2 11 8 e o 0 71t6IA t ol08 75tAt ( OOl983CiU -060llOC() OOl7191A t -O lltdOC A) middot 89
lqA 2704 10 005082 -Oil060 22b -l)i1977 lelO 6 -000080 oo2772C4 t 910 c 000907 -011 200 Oe04105CA t -046960 995
20A 2d06 1 II
O OlltStiAt -001120 837 00014 -))th O01380amp At -oooo 11 821 ( OOl526CAt -026700181 0002 5UAt 000530 969
2lA 2807 28 OOt9lt9CAt -00891 0 516 -0 03962 225I8 002095CCt OOOl lliAt 690 ( o02965CA t -056650((1 OOOt50CAJ -003l70ttl 911I
I 22A 2808 15 ot3l06 119730 655 1130) 090
0 - 007526CC t Oe003961At 835- middot-
c I oo7523 -067210 001l 2tA t -0091t80CCt 951
2lA 2809 21 Oe2067lCAt -1218101At 71 2 -10095((1 121 ttlIaOQ8 00797418) OOOI8lltt 686 Ill( 0092lU A t -Olltl70Ct t 000605CAt -OOit8701At 951 _
---- -
32 009506CAt
0 bull 06 981 I A J 006901 cA bull
00 24A 2810 -010960 5t 5 -OOltSO 7 56
8 -000011 s11 c -5811701At oooaoatAJ oot520IAI 888
-- 37 --- middot - -- -1 ) l lbull t5JJ r-l bullhmiddot1 --- - --
l5A
26A
27A
c
2RA
c
2 9A
31A
llA
24
13
597
Ci)55
14
l 6h
RSQ ELASTI- (lASTCOE f-
TABLE 2
EQ I NO NR CODE IU
REGRESSION RESULTS CCONTa
NR INTER- MARKET
(10S CEPT SHARE AOVER- SHARfcent
TIS lNG AOVER
) CITY StiR
c It) bull 5 co a I 1) bull 0 bull 9 c l 0 bull( 2 1
2815 16 003019 -009990 275 -o069bmiddotJ
6 001446 000224 282 c 002178 -054630 o02221CCI -0075 00 646
2901 29 00021t2 CAI 0)0520 592 0007 6 1 068
0 00105C8 t 000012141 60)g( OOOlJ51AJ -00126018) ooou271At -000u80CA) 974
36oa 1 3 003296(4) -oca29o -oo-q94
001755 000203 lt91t 002499((1 -04l820CCI OOl374CAI 0068)0 946
3612 0033281CI 011520 195 01684) oso
00 30 42C OI 00012 2 424 oo579ltAt -2 7l7801At 002 100IAI -003220 819
8 3617 btel 011789 081002259 009530
001448 oOOl OitCCI 732 OOl7 22CAI -0 46160CAI 0004961AI -OOll40181 989c
30A 3808 14 0021t 3 6fAI -000110 628 o023291C) -005 6 middot- _ middot- 0bull 02 04 1 IC I ooooo9 637 ( 001584((1 -0161)1)0 00021 91AI -0 00060
390) 14 002790CCt 006590 h8 t 0 0938 4 070
B OOl89bCCJ 000144 74
0021961(1 -OItlOlO OOl323CAI -004800 915
B
middot--- - middot 4 - middot- - dfll3904 12 004563181 0)8300 879 0428651(1 089 OQIll004 277CA I ooozstcea 919
c 0041)3181 -0-11250 0011881(1 -006270 972 0Ill
- J - bull bull bull bullbullbull middot - - middot - middot gt middotbull L)ofbullflbull middot middotfJltl -tn- middotmiddotmiddotmiddoto +f middot t middot bull
c
is
positive
2
Page 0 determineThe consumer goodproducer good split is sometimes difficult to
Sollle support for the split used in this paper is a comparison of goodness - of - fit
distributions of a
industries
set
advertising are
that industry
a sales increase of
the
highest aD10U3 those
measures for the two groups of industries Table 3 contains
measures for the 32- consumer goods industries and the 103 producer goods
2 Median R is 5 7 for the consumer goods set and 40 for the producer goods
Estimates - of y the elasticity of demand with respect to
given in column of Table 2 The hypothesis that y bull 0 with
alternative was applied for each ittdustry As shown in the
could not be rtjected for 24 of the 32 industries Of the
and 10 are positive For the remalnifg eight three are
and five are significantly positive
Of the threertegative elasetcities the largest (in absolute value)
cosmetics (code 2809) With all elasticity of -1 01 a one
industry advertising would gwerate a decrease in industry sales
more than one percetu
At the other extrema is toys and games (code 39 04) For
a one percent ittcrease in industry advertising would lead to
four-tenths of a pampreentmiddot Though not significantly different from zero
elasticity estimatbull for pt-bpuetary drugs (code 2808)
which are
Both drug industries have
to sales rafios (211 pe1 cent for 28 08 and 13 4 per cent
other hand flaIJo ing ext1ac s (mainly soft-drink SYUlS) also
is for
percent increase in
of slightly
a middottwo - tailed
table that hypothesis
24 14 are negative
significantly negative
9
the codetics and p-ropriacuy very high advertising
for 28 09) bull On the
has a high ind UY
advenisiltg t() sales ratio ( 8 0 per cent) bull but its estimated elasticity is very
small and not signifieantly different uom zero
middot It doe$ not seem unusual to find many elasticities near zero with a few
negative and a few positive If aggregate consumption is insensitive to aggregate
advertising as seems likely ld then what one industry gains must be lost by
others
Page 21
elasticity of demand does
(10) we can
if t 0 then
Finally a determination concerning the industry
not fix the individual firms elasticity Going back to equation
write y bull (1 - ) a + z y Since a gt 0 it follows thatii i i
Y gt 0 If Ylt 0 however y s sign is indeterminateij ij
Turning next to the coefficient of market share 20 of them are negative and
12 are positive with three of each sign being significantly different from zero
The largest positive value which is significant is for pet food (code 2010) and
a close second is mens amp boys suits and coats (code 23 01) The largest signifi
cant negative value is for cosmetics (code 28 09)
In the model some relations among y a 0 and 61 are implied If middota is
negative (1- y) is greater than one Since 61Y bull (1- a) (1 - ) 61Y is
pCsitive but it will be less than one if 0 is large enough Seventeen of the A
ys are negative in each of the seventeen industries 61 is also negative and les
1 (algebraicly) than y so 61Y is greater than one his is not evidence that the
degree of cooperation is low or zero in any industry of course A high o togethegt
with a negative y which is small (in absolute terms) relative to a can give a
value of 61Y greater than one On the other hand if 61Y were less than one
o would have to be greater than zero
For y gt 0 two conditions may hold If y lt a it follows that 61 lt 0 since
(y - a ) lt 0 It also follows that 61Y lt 0 regardless of the value of o bull
In addition the relation is an if-and-only-if one ie if 61Y is negative
then y lt a Three industries have negative values for 61y dairy products
except milk (code 20 04) synthetic fibers (code 2806) and photographic equipment
(code 3808) In all three cases y is quite small so a does not have to be very
large to give a negative value to al
If y gt a the model shows that 61-- gt 0 whatever the value of 0 bull There are
12 cases where both y and 61 are positive with 61 ranging from near zero
(grain milling code 20 12) to 1 20 ( proprietary drugs code 2808) Several
coeffiei-en_rs
--
15
o f to zero
Tab le 3 - Consumer Goods
Indus tr ies
2Distributions of R and Producer Goods
Page 20a
Consumer Producer
Goods Goods
0 - 9 9 0 3
10 - 19 9 0 1 1
20 - 29 9 3 20
30 - 39 9 3 17
40 - 49 9 5 16
50 - 59 9 8
69 9 6 Ii 1060
--f 670 79 9 2
80 - 89 9 4 4
90 - 100 1 1
Totals 32 10 3
Notes
2is used R i s cal-
culated as F F + ((N - K- 1 ) K ] 1 b_ F is the F
statistic Lor the hypothesis that all o f the non - intercept terms are simul taneously equal
1
2
--
-- Since generalized least squ-area regress ion
Page 12
o f these cases have values o f 6 1 Y rea ter than 0 8 (pet food (20 10) bread amp cakes
(20 14 ) miscellaneous processed foods (20 29 ) men s and boys suit s and coats
(23 01 ) p rop riet ary drugs (28 08 ) elect ric housewares (36 1 2) radio and
IV sets 17 and games and toys (39 For be than(36 ) 04)) 6 1 Y to greater 0 8
both o and cry mus t b e less than 0 2 That is in these industries the results
shown in Tab le 2 would hold only with very strong dominance of marke t demand
effects over brand swit ching e ffects and a very low degree o f cooperation
Using the concentration rat ios given in Table 4 the average concentration
ratio for the eight indust ries is 38 1 For the remaining 24 industries it is
43 2 This result would lend some support t o the proposition that concentrat ion
and cooperation are related if the group of eight industries are identi fied
as having low degrees of cooperation
One f inal note on model predict ions S ince both cr and o are assumed to
be non-negative 60 should also be non-negat ive In only one o f the 32 cas es
is 60 negative and it is not signif icant bull
cor pJ
7 1
7 0 4
1 4 7 5 1
)
7 3 z a o a t s 1 1 3 3 5 ( i V Z 8 tJ1 2 1 - 1 1 0 7 3 5 c middot v 2 8 1 0 3 2 -0 0 1 4 3 0 C 1 V 9 8 3 2 8 1 5 1 6 - 0 0 6 t1 6 9 C _ V 1 11 0 2 9 0 1 29 0 00 7 6 1 C l k V 7 3 1 3 6 08 1 3 - o n 4 9 9 4 3 3 3 6 1 2 2 4 0 1 6 4 3
3 1 7 1 3 0 1 1 7 d 7 8 3 3 8 0 8 0 0 2 3 i 2 3 0 6
P 4
3 9 0 3 1 4
1 2
0 09 3 8 4 39 1)1 0 4 l 8 6 5
T A LE 4 e I NCU S T R Y C A T E G O R Y
F TC NOM e tOOE ass
C 1 I c 2 )
2 0 0 1 1 6 2 0 0r 1 7 zo o l Z 2 0 0 7 1 8
z o o 2 6
z o t a 1 6
2 1) 1 2 1 5 2 0 1 4 1 1 2 0 1 a 1 9 2 0 26 1 2 2 0 2 7 1 0 2 0 29 - 3 5
2 3 0 1 10
2 3 02 z o
2 3 03 1 9
2 5 5 1 2 4 Z 1 0 Z 1 1 2 0 3 1 9 2 7 4 1 0 z a o 1 3
2 A 0 7 z s
E LA S T I -C I T Y
c 3 )
- 0 09 7 2 7
0 00 449
- 0 0 2 8 9 3
- 0 3 2 7 70 - 0 OZ il I q
o 4 u t s a 0 0 1 3 7 3
o 1 4 o o-0 0 5 2 4o -0 1 94 0 1
-C l 6 l l 4 0 1 JJ 3 7 0 3 t 0 4 2 0 0 4 0 9
- 0 4 1 8 44 - O l 7 t ) 4 -ll 7 1 4 0 b -0 3 5 7 8 Z
- 0 1 9 7 7 0 1)0 3 3 4
-o n 9 2
IJ A T A
CONV
ON-C ONV
( 4 )
t 1NV
C 1t-V C NV c ) v C - JNV C rlNV c n v c a v C ON V ( i) l V
C CtV
C t1tV C IJNV
C 1 V
c 1 v
Pa ge 2La
HEQ F t Noe x
c 5 )
6 2q
6 89 1 48
6 1 4 ) q 3
1 9 7 1 9
q z t s 5 a e z
l l 9 1
4 o 7 1 2 0
c oqt 1 Z
4 4 )-q 1 z
5 5 9 1 0 64 1 3 3
n 3
Inter industry analysi
Page 23
Equation ( 3 6 ) may be used to fo rmulate a rough test o f the cooperat ion
vs e f ficiency controversy I f we rest ricted ourselves to some subset o f
indus tries where the relat ion be tween y and a i s the same we would be lookingbull
at a sample o f indus tries for which the dependence o f V on H and o is a
quadratic wi th no squared terms
I f we now sub s titute (26) into ( 36 ) we get
(38) v bull a + (y - a) R + (y - a) o (H) - (y - a) H o (H) + u
If it is only d if ferences in marke t shares that affect advertising inten-
s i ty levels i f concentration does not af fect cooperation and i f o bull 0
( 38 ) reduces t o
( 39 ) V bull a + (y - a) H + U
Let this b e the null hyp o thes is and note that V is a linear func-
tion o f H If y gt a and the null hyp othesis holds the expected relat ion
between V and H is as shown by the pos itively sloped straight line in Figure
6 b etween points A and C
In this context the alterna ive hypothesis the degree o f
cooperation ( o) is a positive function o f concentrat ion
is that
(H) I f the func-
tion is as described in (26) and Figure 4 the relation between V and H
would be as shown by the curved line connec t ing po ints A and C in Figure 6 One way to test the null hypothe s is against this al ternative is to test fo r
the hypothesized curvature in the function relating V and H
Fi gure
R
6
v
ID c
H 1 0
I
i I I
sfigned
clasJ s
in def ini g
effec t is the
I
Page middot2lf
One charac teristic of produc ts which may be useful
classes wi th respec t to significanc e of the brand-switching
conveniencenon-convenienc e distinc tion developed by Porter ( 1 97 )
Using his identif ication of IRS indus tries as a s tar t ing point
each of the 32 industries used in this s tudy to one of the two
The ass ignments are not ed in Tab le 4
Using the conveniencenon-convenienc e dis t inc tion the
divided into two sub-samples
sample was
Four regressions were then run bull ene for e ch
sub-sample one for the to tal sample and one for the total sample with
a dummy variable for convenience goods Each regression was run wi th LB s
and wi th indus tries as ob servations The results are given in Tab le 5
For the LB level equations ( 1A - 1D) equation(37) was used ome
rearranging of terms gives v bull y - ( 1-o ) (y - a) ( 1-z ) + ui i i
bull y - 6 1 ( 1 - z ) + ui When observations are pooled across indus trif s yi
is an industry variable In these regressions the values o f y es t imated
in the first stage were used Since industries have been grouped so as
to reduce differences in (y - a) that term is t reated as a cons tant
within each group Finally H the Herfindahl index has been subst ituted
for o the degree of cooperation The resultant equation is linear
y ( 1 - z ) and H x ( l - z ) This is the equation for whi ch results arei i
reported in Table 5 In all regress ions all variables were weighted by the
square roo t of sales
in
raquo
- 34
3 1
i l T 4 8LE S R E G E S S I ON R E SUL T S i X T f N S I ON S T O T Y PE Of
AND L E V E L OF C ONC E kT R A T I O
E O O tt S
tP bull T Y P E NR C ONV O B S tO N-CONV
E L A S T I shy
T I C J T Y
O E - t-41( T SliA R E
H( fl F
1 1 - K O x middot K O X E R F
- H S a 1 1 - H S t 1 8 1 1 9 1
R S O
1 1 0 1
q z
t - bull 9 f 2 1
Z t P D l 6 fC ONV - 1 9 1 s s c 2 1
e - l O I e bull u q a
2 3 1 0 1 1 Nl N- t O V 5 7 1 ( 1 2 19 1 6 1 - J l S 2 5 4
l l l l t z q o a 1 - J o v z I
l ( I NO l l A l l ol u l t o o c bull J u l l
d S
c 3 4 t t r a l
bull Z D 1 0 l l 4 l l 2 3 -gt 2 l l 2 0 - 4 lS 5 4 l C 4 1 o e 6 t l b b 5 1 1 0 1 1 1 c - 1 9 5 1 t d S 2 1 tV
OQ10 T-R A T I O S A E G I VE N l P A RE N T H E S E S l E V E L S f S I G 1 F l C ANC E A - 99 5 B - q 9 t - s NS J UC E NO I Nl E R C ( P T I S U (j f middot) I N f Hf t OUA TIOI S 6 E f 3R E A O J U S T H E N T F OR 1-t E T E R O SIC E O A S T f C I T V pIraquo
1
middot
I
R SO I S C A L C UL A T E D A S f I f bull 1 1 bull 1lt 1 K t t- E E N I S THf U A1 B E R O f fJe S E fi V A T I O S F t S HtE F - S T A T I S
I N DE P E ND E N T y II
T I C F O T H E U Y P O HIE S I S f A l l C JE F F I C H N T A R E QU A L HI l f RO A f40 K I S T H f UUM 8 E R OF Q ( A fL f S
I l l
1 A
1 6
t C
f 1 r R I l l
1 2 1
L Jt
L B
l q
L 1
on s T Y P E bull 2 )
P OUC T
X
H S I l l
l O l A I bull l 0 7 1
O O b O J
1 2 00 1
1 b 2 1 C I l l Z U
l l S b 1 -1 1 0 1
rlt F X l l - fi C f l
bull 1 I
l l
- ()4 2 3 l f H
1 - 2 7 9 1
K O X
1 1 - ti E R F I
l d l
S4 l i A I 2 l b
1 3 00
K O X -t f R F t S J
I 1 - HER F t
)
60 9
z o bull
0
t l l
3 8 8
1 7 6
5 64
64
N f bull
tl S bull
I l l
I 4 1
t ONV
NON-t ONV
A l l
A ll
c o v NO- ( ONV
1 4 1
1 5 1
- l S b
c - 1 1 2 1
5 8 2 C A I 1 6 7 0 1
middot I l 9 H
l 1 il
1 4 1 6 1
E L A C) T J shy
l t C I T Y
I S I
l b l
- 0 2 0
1 - 1 4 9 1
O Z 3 2 1 A I C b 6 b l
0 0 9 2 -J
1 1 b 4 1
o z t t t l 1 2 2 1 1
oru - Ht tt F
1 gt X bull b I
Page 25 bull
Mean val -as ior relevan t variables for the two sub-samples are
Conv Non-Conv
Adv Sales 4 7 2 3
iit Share 2 7 3 2
Ela cicity - 05 3 02 2
Herf indahl 7 3 1 860
That the dist nct ion between coft9enienee goods and non - convenience
goods mat ters is supported not only by these means but also by the regression
results Equat ions lA and lB are different in almost every respec t Only the
concentrat ionmarket share int eract ion term seems to mat ter for convenience
good s Jus t he opposite holds for non - convenience goods both elasticity1
and market sl e are highly significant but the concentrat ionmarket share L
int erac t ion t=rm is not
Given that one minus market share is used these equ3 tions the coeffici t
of market share is the negative of what is sh in col (6) of Table 5 For the
regress ion for non - convenience LB s reported in equation lB then LB s with larger
market shares have lower advertising to sales ratios
On the question of the impact of the degree of cooperation the story is mixed
For convenienc good s the coefficient is pos itive as equation (31) predicts when
there is a po sitive degree of cooperat ion Fo r the non-convenience goods LB s middot
concentration has no effec t
Aggregat ion t o the indus try level and then rer unning the equations $ives the
results shown in (2a-2d) of Tab le 5 As expected R2 goes up and levels of significanc e
fmiddotJr individual coefficients gJ down In add ition the coeffic ient of H ( l -H) inmiddot
quat ion 2A is no t significant though it is po s i tive Since 2A is the indus try level
I
--
all but one of them and that one is not significant For 18 of the 31
wo uld have on its adve rtising to sal es ratio
cognate of LA the variable H ( l-H) pl ays the same role in 2A as H ( l -z 1) p lays
in LA
Vconomies of scale in advert ising
The literature on advert ing contains a number of definitionbull o f economies
of scale and of findings eone rning its presence and magnitude Before I loo k
a t the data with the hope o f mmenting about scale effect s then I wan t to
define the term as I use it
By economies of scale I mean a decrease in the number o f units of adver-
tising which are needed to generate a unit of sales as advert ising is increased
Given the definition of the firm s own elast icity of demand with respect to
advertising (see the text foll wing equation (1) ) the technical definition
is that rii gt 1 Diseconomies of scale are def ined symmetrically i e
Yu lt 1
ith a cro s s - section oi observations on advertising intensity
(v bull a s1) and onmiddot advertising (a ) we may regress the former on the latter 1 i i
The results of such regress ions for the 32 industries are given in Table 2
lines lB ZB bull bull bull 32B For the 32 eases the regression coefficient is positive it
positive coefficients the coefficient i s signif icant
The que tion now is whether these results ar e evidence of pervasive dis-
economies of scale in advertislng I think no t Given my as sumption that the
data I observe are equilibrium results I may ri ghtly conclud e that almost every-
where high levels of advert ising are as sociated in equilibrium with high ratios
of advertising to sales I may no t conclud e anything however about the
impact that a move by some firm away from its equilibrium level o f advertising
middot- Page 27
Given the assumed relation in equation ( 37 ) there is a corre spondshy
ing equilibrium equation relating v and ai The equation is non - linear i
and I have not found a simp le way to characterize it It is po ssib le to
det ermi ne its deriviative with respect to a b y takin$ tne de riviative of ( 3 7 ) i
When that is done the result is
v - 1(40) i bull sl a a z ( 1-z )i i ira-l
Sinc e all the other terms in (40) are non-negative the sign dep ends only
on the sign of 6 1 which is in turn de pend ent only on the sign of (y - a)
That the signs of the coefficients of a in equations lB 2B bull bull bull 32B arei
not always equal to the signs of the coefficients of z in LA 2A bull bull bull 32A i
is still a mystery to me and something to be explored
bull
Summary
My purpose in writing this pa per wa s t o attempt to integrate explicit micro-
economic mode lling reasoned econome tric specification of error terms and high
quality data to exp lore some questions about adver tising by large firms
al quite important improvements could be made in the first two of those areas
and much data ma ssag ing is yet to be done in the third
Concerning the subst an ce of what I have done I think six things are im-
Page 2 8
Severshy
V and Conclusions
portant
1 Explicit modeling is worth the ef for t if for no other reas on than
that it provides a basis for choosing fr om a varie ty of funct ional
forms
2 The same is true for error specif ica t ion I did in fact look at scatshy
ter diagrams for all 32 industry categor ies and they vir tually all show
the het eroscedast icity which I as sumed and for which I correc ted
3 The predicted relation be tween advertis ing in tensity and market share
shows up clearly in on ly 25 of the case s examined I have ye t to
explore why that may b e the cas e
4 Some evidence concerning the presence and impact of coopera tion was
produced but it is not clear ly pervasive
5 The dis tinction between convenience goods and non-c onvenie nce goods is
unambiguously a good one
6 Virtually no evidence concerning econ omies of scale in adver tis ing can
be gleaned fr om this study given its assumpt ions
More work is called for on many of these issues my study seems to have
raised more questions than it has answer ed This is pr obab lly due to the richshy
ne ss of the data source since I had the oppor tunity to address several issue s
at once
- -- - Q --
Foo tnotes
Manager Line o f Business Program Federal Trade Commission The author depended heavily on several staf f members of the LB Pro gram for statistical and clerical support in particular Joe Cholka George Pascoe and Haro lene Jenkins Helpful comments were given b y Richard Caves Dennis Mueller Michael Porter and F M Scherer When the paper was in its early stages o f development both S teve Garber and Jon Rasmussen former economists in the LB Program provided eery useful cri tiques The views expressed here are my own o f course
1 The basic exp licit model of the role of advertising is Dorfman and S teiner ( 1954 ) and the models in this paper may be seen as ano ther extens ion of their work
2 For a review o f models of oligopoly which treat prices and quantities as relevant variables see Shubik ( 1 959) Extens ions to take advertis ing int o account are fairly s traightforward
3 Dorfman and S teiner dealt with a price-setting firm For a monopolis t inverting the demand relation to show Pbull P (Q A) _instead o f Q- (P A) has no effect on the results o f the analysis For oligopolists who face a sys tem of demand relations inversion does have some impac t on the results The dif ferences are not trivial nonetheless they will no t be explored in this study
4 Shubik ( 1 959) p
5 From a formal mathematical point o f view the function is similar to a1 i Legrangian funct ion of the form - r rr
L bull 11 + A ( j - (i i i+j
where the indicates some fixed level of the profits industry Following the same logic which resul ts in and income constraint as the marginal utility of the of the indus tr in the obj ective of firm i
6 Hall amp Weiss ( 1 967) p 323
7 See Schmallensee ( 1 97 2 ) Ch 2 pp 1 6-47
8 On this poin t see Imel amp Helmberger ( 1 9 7 1 )
9 Long ( 1 970) p 130 - 1 39
1 0 See Schmallensee ( 1 972) Ch 3 pp 48-8 7
rrj i+j ) of the rest of the
the identification profits of the rest
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980
bull
Page lOa
y lt1 bull y y
10
f
V0 bull z [ a1 + (y - cr) J bull a + (y - a) H vl bull Y bull y
11Page
( 2 1) vo bull z d -1 r bull r z bull z- d d- i )11
l -1 N N v bull z d r z bull r z bull l y z - ij jibulll jbulll
and for the simplified case from (19)
(22
Since the forms of the equations in (22) are the same as in (20) all
of the analysis of (20) holdS for (22) except that His substituted for
middot z Advertising intensity will be higher for an industry with cooperativei
firms than for one with non-coope rative firms if the industry demand elas-
ticity dominates the brand-switchine lasticity If the ewo elasticities
are equal then so will be advertiSl intensity for the cooperative firm
and non-cooperative firm industries If brand-switching dominates the
cooperative firm industry will have lower advertising intensity
In the industry model context the possibility of a critical concen-
tration level may be easily introduced That is assume that at values
of R below R the firms are non-cooperative but that for alues equal to
or greater than H they are cooperative That is
(23) V bull a + (y - a) H if H lt H bull y if H gtmiddot H
This function is shown in Figure 2 for the three relamptions be eween y and a
IV THE D GRE OF COOEtRATION
The models depicted in Figure 2 are based on a very simple notion about
the relation between concentration and the cooperationnon-cooperation charac-
teristic They use what is essentilly a step function relating concentra-
Figure
I -
J H H
Page l la
2
v v v
a
I J I y a bull Y y ---i
I I I I I I II I
H 10 H H 10 H 1 0
bullbull
bull bull
Page 12
tion and the extent or degree of coo peration as shown in Figure 3 Using
o as a symbol for the degree of cooperation eq uation (23) can be rewritten
as
bull(24) v (1 - o) vo + ovl bull (1 o) [a+ - a) H ] + oy- (y
vhere
(25) o bull o(H) bull 0 if B lt B
Hbull 1 if B gt
The specific relation between concentration (H) and the deg ree of
cooperation ( o) shown in Figure 3 is more restrictive than necessary and
bas less appeal an one which shows a smooth increase of o from 0 to 1 as
B increases fro to 1 Such a smooth function is shown in Figure 4 he
degree of cooper ion is low until H gets close to H increases greatly as
B goes through R attaining a high level for an H larger than B but still
close to it and then goes to 1 as B goes to 1 More formally
(26) 0 bull o(H)
bull 0 if H bull 0
bull 1 if H bull 1
do gt (IdB
d2o- gt 0 if H B dBZ
0 if H H
lt 0 if R gt H-
If (26) is now substituted for (25) in (24) a corresponding smoothing
of the functions shown in Figure 2 is accomplished he smooth functions
Figure 3
I
10
H 10 H
Page 12a
Figure
Page 12b
4
10
H 10 H
l
(27)
Cl - o) o
[ (1 -
[ (1
t y ij z
j
are given in Figure 5
Page 13
The equation in (24) was constructed as a hybrid of two results derived
from first-order equations for the non-cooperative and cooperative models
It may also be constructed directly by using l bull (1 - o) + on as thei i
th 2objective function for the i-- firm Treating o as a constant the first
order equation is
0 an aa
i middot
n i
Using equations (3) and (7) and setting --- bull 0 givesaai
(28) - 0
After collecting terms and converting to matrix notation we get
(29) - Cl - o) + 0 z-1vi yii i
N
j bulll
+ 0
+ o v1
d-1 r z z
-(30) v Cl - o) rd
-
o) r +d
d r J _ -1- d z
and middot
d) r + 0 rjd
l r z
v - z --(31) zv
- (1 0- +d) t r z- d-
If we now impose the simplified demand structure by substituting (11)
we get
- (1 - o) a +( 32) + oy and
Figure
l
-y
v
Y
H
r
I I
v
a
y
H 10 H
v
a
bulla
HH 10
bullbull
(33)
i
Page 14
v - (1 - o) [ a l + (y - a) ] The corresponding equation for Vis given in (24)
If the relation between industry advertising intensity (V) and
concentration (H) is non - linear via the impact of concentration
on the degree of cooperation does the relation take the form of a
auadratic as some other investigators have proposed (Greer ( 1971 ) Martin (1979) Strickland amp Weiss (1976) ) That question can
be approached from two perspectives - conceptual and empirical In
the next section I will try to apply some statistical material to it
here I want to explore the question in the context of the models
developed above
Referring to equation (24) taking the deriviative with respect
to H and letting bull (H) gives
bull( 3 3a ) 3V (y - a) (1 - o) + (1 - H) 3H dH
Given that o and H each are bounded by 0 and 1 and that o is an in-
creasing function of H the term in the brackets in non-negative The
sign of the deriviative then is the same as the sign of (y - a) If the industry effect is dominant (y gt a) then advertising in-
tensity will increase with concentration If the two effects are equal
then advertising intensity will not vbullry with concentration and if the
brand-switching e ffect is larger advertising intensity will decrease
with concentration
For given advertising elasticity parameters then it is not pos-
sible to show an increase in A S up to some H and the n a decrease
he relation may be non - linear but it is monotonic over the permis-
sible range of H
Page 15
One possibility is to introduce a relation between (y - a) and H
If industries characterized as dominated by the brand-switching effect
also were more concentrated then a cross industry comparison would
show an inverted U shape relation between AS and H I know of no
support for such an association between ( y - a) and H
V ERROR SPECUCATION
So far I have implicitly assumed that there are no errors in the
model I now assume that the error
thvariable in the equation for the i- firm is additive that its mean is zero
that its variance is inversely proportionate to its sales and that all error
term covarlances are zero hat is
34) v bull (1 - o) [ a + (y - a) z ] + oy + u i i i
- - -
In vector notation
(35) bull (1 - o) [ at + lt - a) J +
E() bull 0 Cov (u) -
0
If we now ca lculate V bull bull we get
oy + U bull
(36) V bull (1 - o) la + (y - a)R J +
U bull E(U) bull 0 Var (U) bull
P age 16
The error term specifi c ation which follows Hall
wi th the ratio of profits
6 heir analysis seems
as wellbull particularly given
and Weiss was
applied by them to equa tions to equity or
asse ts as dependen t variables to carry the same
weight in the present context the substantial
evidence presented by marketing analysis on the tendency of company de-
cision makers t o use the advertising to sales ratio as the decision
7variable
The zero covariance assumption on the other hand1 is particularly
troublesome Within a probably not true since
many even ts outside the
single industry it is
industry would have
industries
similar effec ts on all the
8firms in the industry Between it is probably not true either
since many companies produce in more than one indus try and even ts in the
firm would tend to effect all of its activities The extension of the
model to allow for non-zero covariances is certainly called for
--- -
Page Ll
Athird characteristic of this specification also deserves some comment
only on equation is shown for the firm or industry Several haveinvestigators
included an advertising equation ina larger system of equations (Comanor and
Wilson (1974) (1979) and Weiss (1976) so in myMartin Strickland Idid
dissertation1 Ihave ewo defenses for presenting only single equation results
here One is the suggestion by Comanor Wilson that simultaneousand equation
verybias may not be important in this context he second is that Iam will-
ing to see the model extended to include a profitability equation and I intend
to move that direction in the near futurein
Onthe inclusion of aconcentration equation on the other hand I amskep-
tical he problem isthat in some fundamental sense equations in models of
industrial organization should be oriented to the decision-making contexts of firms
Ifeach firm in anindustry with Nfirms has only one decision variable then
there can be only Nindependent equations which are based on the first-order
optimization equations It not show large ofis difficult of course to a number
functional relations which follow from the first-order conditions point here he
isthat only Nof them can be independent
Consider for illustrative purposes only a situation in wich teh N
itsquantities say q are set exogeneously and each firm determines advertising1
of and level say ai where a is the true optimal value a a bulla
1 1 i i + i
Ifp a) then the Nq
s and the Na s will determine N p s say pii bullp
i(l i 1 i
Now given Eand 9 we can determine both the vector s andthe vector v since
- bull til
ands We can also determine S s z bull s-1 s-ii
v and we
industry concentration isHbull (alternatively
V and Hare functions of the same
pendent Furthermore if there are
VM) and (H1 H2 bull bull bull )contain
VI STATISTI CAL APPLICATIONS
Page 18
Given the observed values of might be tempted to conclude that we
have a two equation system in which v and z are jointly determined That would i i
be inappropriate however since the 2N random variables (v ) are determined by
onlt N basic random variables t The variance-covariance matrix for ( ) must
be singular bull
The extension of this observation to industry level variables is straightshy
forward Since observed industry advertising intensity is V bull l v and observed 4
C4 bull z ) it follows thati l i
set of random variables t and cannot be indeshy-
M industries then the variables (V1 v2 at most M independent random variables
Given the assumptions made above y a and o are constant within a
given industry That being the case (34) can be written as
(37)
s1 bull (1 - o) (y - a)
If and
So + +
(1 - o) a + oy
81 are esti mated for an industry using a suitable statisticalSo procedure giving and 81 an estimate of y is then available The sumSo
Intra-industry analysis
1974 of Business Re ort -
categories are identified
Page 19
of So and S1 is y so y bull So + S1 is the estimate of y
If o is assumed to be bounded by zero and one the sign of 61 is also
the sign of y - a The estimate of So is of cou se an estimate of
(1 - o ) a + oy Moreove if y gt a then a is less than 6o- (1 - B) a+ 0(
Cor esponding results would
hold if y bull a or y lt a
If 61 gt 0 then we would estimate that a lt 6o
bull
For purposes of estimating this model 32 manufacturing industry categories
were selected The data are fram the Line of Business forms filed with
the Federal Trade Commission they are discussed in detail in the Annual Line
which is available from the FTC The 32 industry
in Table 1
Simple regressions for equ ation (37) were run for each of the 32 industries
with data for each of the firms which filed in the industry constituting an ob-
servation The results are given in Table 2 lines lA 2A
1974
the intercept
For all the A
equations generalized least quares was used th both and market
share bein scaled bv the square root of sales
The 32 industries were taken from the larger set of all manufacturing industry
categories in the 1974 data files Two selection cr iteria were used thereLB
had to be at least ten reporting companies and the industry had to be a consumer
goods industry or a producer goods industry which is strongly associated with a
related consumer goods industry so that adve rtising by the firms in the producer
goods industry impacts on demand in the related industry (eg soft drink syrup
and soft drinks) There were 135 industries with at least ten companies - 103
producer goods and 32 consumer goods
273
274
TABLE 1 INDUSTRY CATEGORIES USED
FTC
COOE OESCRI P T I ON _
2001 MEAT PACKINGt SAUSAGE S ANO OTHER PREPARED
MEAT PROOUCTS
2004 DAIRY PRODUCTS EXC F LUID MILK
2005 CANNED SPECIALTIES
2001 FROZEN SPECIALTIES
OEHYORATFOt AND
I NC LU 0 ING PRE SERVE S
SOUP MIXESbull
SEASO lNGSt ANO
OTHER PET FOOJ
RAIN MILL PROOUCTSt
ANO PREPARED
RELATED PRODUCTS
PRODUCTS
Z008 CANNEOt ORIEOt PICKLED FRUITS
ANO VEGET ABLES JAMS JELLIES DEHYDRATED I G TA3t E
S40CES ANO SALAD ORESSINGS
2010 DOG CAT A O
2012 FlOUR t OTHER RICE
MILLING fiLENOE) FLOUR
2014 B EADt CAKE ANO
20113 CO FECTIONERY
2026 BOTTLED A O CANNED SOFT DRINKS 2021 FLAVORING EXTRACTS AND SYRUPS EC
2029 MISC FOOOS AND KINDRED PROJUCTS EXC
ROASTED COFFEE
2301 MENS AND BOYS SUITS AND COATS
2302 MENS ANO BOYS FURNISHINGS
230 3 kOMENS AND MISSES OUTER EAR
2551 HOUSEHOLD FURNITURE
2702 PERIODICALS
Pqe 19a 1
RELATED 1972
src cooe
2011 f 3
202tX2026
2032
203pound
2033t4t5
2047 2041t4t5
2051 2065 2086
2081
209XZ05
231
232
233
251 272
2703 BOOKS
2704 ISC PVeLISHING
2806 ORGANIC FlEERS
2807 DRUG$ ETHICAL
23234
PT 2 S 3
2808 ORUCSt PROPRIETARY
2809 PERFU ES t COSMETICS ANO OTHER TaiLET
PREPARATIONS
2810 SOAP AND OTHER CLEANING PREPARATIONS 2815
2901
PEST1CIDES AND AGR ICULTURAL
PE OLEUM REF I N ING CHEMICALS EC
3606 HOUSEHOLD COOKING EQUIPMENT
3612 HOUSEHOLD APPLIANC ES NEC INCLUJING ELECshy
rqlC HOUSE ARFS AND FANS ANO SE I G MACHINES
3617 RADIO AND T V RECE I VING SETS 3808 PHOTOG APHIC E UIP amp SUP LfESt EXC
3qo3
PHO TOCOPYING
SPOR TING A D E QU I P M ENT f SUP P L I E S ATHLETTC GOODS NEC
3904 DOLLS GAMESt TOYS AND CHLOqENtS VE ICLES
PT283
2844
234tX2844
287 9
291 3631
3634t6t9
36H
PT386l
3949
15
943
middot
l A6l f 2 R(GRESSION RESULTS
EQ I NO
NR CODE cu 121
lA 2001 A (
2A 2004 r (
3A 2005 a
-R
ClS amp31
16
17
1 z
I TER-CEPT bull t
ooo9t21At 00022311 oooeos uu
OOl49tCAI 0010 2 middot 000960CCI
o049tl1At o04t55C61
MARKET SH ARE bull 5 I
-010670Ctl
-017600181
-001050
-0211AO
-0077 0
AOVER-TIS INti
lbl
000()39 000072
000018 0 0024liAI
-000079
SHARf
4ampgtVf Q bull
111
000140
-000170
RSQ
COl
66 485 862
607 6 3 t ()2
blti6 633
flASTI-
CI T Y PH
-007271(1
000441
-u0288l
clASTSHU (t)Ef
bull 10
tto
-234
210
( OOlltOCCI -126520 000922 181 001790
4A 200 7 18 00590liAI -038670 4e16 -032110 lell u c
OOOS74e 000700 025940
00067018) 0021001AI -02l100CAI
5 26
920
5A 6 c
200 8 26 0 03Sl9CA I 0012 26181 o 024511(
-005540
-0 81420 0001961AI O()Ol90CAI ooll240
441 604 891
-0020 19 274
6A 2010 16 Oe04224CAI o359lOIAI 9oq ll4015dCAI 061 6 -middot 005435CAI 00011 8 1AI 917 c 00535ltAI -009850((1 OOOA51J CAI -001610161 971
7A 2012 001862 000010 351 00187) oo1 a oooq45 ono1o6 439 c 001109 -027050 Oe0067SIAI -OOll60C81 897
- middotmiddot- -
8A 201t ll 000757 Ol4140CCI 814 ott9oocet 095 8 oooqo6 C C J 000069(81 811 c ootl2lampBI -080910CCI Oe009lUBI -OOtlOOCCI 960I
I DIIJIT pound11 1V OTIUTVTfAll-1pound1t- A _ 00 R - qq r - Q
1jl)OQCD
0
-middot middot--
l
_ __ _
I
i
0 00141 c c
91)7
I
T ABLE 2 REG ESSIOH RESULTS CCONTI
EO NR
IND CODE
iR bull
cos
I TEq-
CEPT
t1ARKE T
SHA E
AOVER-
Tl SING
SHARpound-cent
AuVER
RSO ILAST1-
C I T Y ELAST
SH CPtf (lt Cl) bull 11 I 41 I 5I (61 171 I 8 I f9) c 10)
9A 2 0 1 8 19 002397181 -007640 367 -005 246 le46 6 000605 000550CAI 597
-000029 G14180 OOl654CAJ -024li01At
lOA 2026 12 0028901AI -OlZl90 -0114 0 1 1 1 5 759
Ll (
llA 2021
f3 c
12A lOZ9
B
(
13A 2301
B
c
14A 2302
B
c
154 201
n (
16A 2551
8
10
35
10
20
19
24
001897(()
00141 6CCI
Oll9lHCI
010604 middot middot
Oll083
OOJ11JCAt
002059141
001411CIH
-000412
000293(81
000187
001159 (3)
ooo6a4cct
o00792CCI
OOl40dCAt
000006
-000076
OOlltl6fA)
o o0 5 64CCt
-034570
-o1middot o3o
-1659 90
Ol51lO
-046190CAt
Ol5450CA)
-005l60
002940
-030060(()
-0 43l50CCt
0 33650
-019080
00004
bullJ00741CAI
-000069
000466
0001 51CAt
000947CAt
O OOI)70CAt OOl750CA J
000175CCI
000671CAI
OOll5) f A t 002207CBI
0004b7CA I
- 007 1 70
001760
-001410CAt
- 0 16200CAt
-004210
- 1 207 70
705
96 0
542
466 805
460
614
921
832
9 61
996
424
550
9 15
56 1 651
884
-006114
019 037
03504lCA)
oo409Q
-141644((
491 -0 17664
607
871
lell
084
101
072
103
108 11IllOQfl)-- l
-037880fCJ OOl6l4CAJ -040l70fCI 0()
c
I
I
bull 9
9t
middot middot middotbullmiddot bull lraquo
TABLE 2 REGRE SION RESUlTS ICONTt
bull
EO I NO tHl I filER- f4A_K T AOVER- sbulluRE ttSQ flA Jl- iLAS T NR CODE cos CEPT SHARE JISING AOVfR bull (JTY SHR CiJCF
110Cl I f amp Itt 15J tbt bull 1c u caa
llA 2702 ll o079 6 81 Bt -019110 S6s -OlltO 111
a 004679 ooooo1 t69
( oOl26o -015160 001560 -022690 ltJr()I
ldA 270 3 19 0065lt -OIt2l60 220 -Ol57a2 11 8 e o 0 71t6IA t ol08 75tAt ( OOl983CiU -060llOC() OOl7191A t -O lltdOC A) middot 89
lqA 2704 10 005082 -Oil060 22b -l)i1977 lelO 6 -000080 oo2772C4 t 910 c 000907 -011 200 Oe04105CA t -046960 995
20A 2d06 1 II
O OlltStiAt -001120 837 00014 -))th O01380amp At -oooo 11 821 ( OOl526CAt -026700181 0002 5UAt 000530 969
2lA 2807 28 OOt9lt9CAt -00891 0 516 -0 03962 225I8 002095CCt OOOl lliAt 690 ( o02965CA t -056650((1 OOOt50CAJ -003l70ttl 911I
I 22A 2808 15 ot3l06 119730 655 1130) 090
0 - 007526CC t Oe003961At 835- middot-
c I oo7523 -067210 001l 2tA t -0091t80CCt 951
2lA 2809 21 Oe2067lCAt -1218101At 71 2 -10095((1 121 ttlIaOQ8 00797418) OOOI8lltt 686 Ill( 0092lU A t -Olltl70Ct t 000605CAt -OOit8701At 951 _
---- -
32 009506CAt
0 bull 06 981 I A J 006901 cA bull
00 24A 2810 -010960 5t 5 -OOltSO 7 56
8 -000011 s11 c -5811701At oooaoatAJ oot520IAI 888
-- 37 --- middot - -- -1 ) l lbull t5JJ r-l bullhmiddot1 --- - --
l5A
26A
27A
c
2RA
c
2 9A
31A
llA
24
13
597
Ci)55
14
l 6h
RSQ ELASTI- (lASTCOE f-
TABLE 2
EQ I NO NR CODE IU
REGRESSION RESULTS CCONTa
NR INTER- MARKET
(10S CEPT SHARE AOVER- SHARfcent
TIS lNG AOVER
) CITY StiR
c It) bull 5 co a I 1) bull 0 bull 9 c l 0 bull( 2 1
2815 16 003019 -009990 275 -o069bmiddotJ
6 001446 000224 282 c 002178 -054630 o02221CCI -0075 00 646
2901 29 00021t2 CAI 0)0520 592 0007 6 1 068
0 00105C8 t 000012141 60)g( OOOlJ51AJ -00126018) ooou271At -000u80CA) 974
36oa 1 3 003296(4) -oca29o -oo-q94
001755 000203 lt91t 002499((1 -04l820CCI OOl374CAI 0068)0 946
3612 0033281CI 011520 195 01684) oso
00 30 42C OI 00012 2 424 oo579ltAt -2 7l7801At 002 100IAI -003220 819
8 3617 btel 011789 081002259 009530
001448 oOOl OitCCI 732 OOl7 22CAI -0 46160CAI 0004961AI -OOll40181 989c
30A 3808 14 0021t 3 6fAI -000110 628 o023291C) -005 6 middot- _ middot- 0bull 02 04 1 IC I ooooo9 637 ( 001584((1 -0161)1)0 00021 91AI -0 00060
390) 14 002790CCt 006590 h8 t 0 0938 4 070
B OOl89bCCJ 000144 74
0021961(1 -OItlOlO OOl323CAI -004800 915
B
middot--- - middot 4 - middot- - dfll3904 12 004563181 0)8300 879 0428651(1 089 OQIll004 277CA I ooozstcea 919
c 0041)3181 -0-11250 0011881(1 -006270 972 0Ill
- J - bull bull bull bullbullbull middot - - middot - middot gt middotbull L)ofbullflbull middot middotfJltl -tn- middotmiddotmiddotmiddoto +f middot t middot bull
c
is
positive
2
Page 0 determineThe consumer goodproducer good split is sometimes difficult to
Sollle support for the split used in this paper is a comparison of goodness - of - fit
distributions of a
industries
set
advertising are
that industry
a sales increase of
the
highest aD10U3 those
measures for the two groups of industries Table 3 contains
measures for the 32- consumer goods industries and the 103 producer goods
2 Median R is 5 7 for the consumer goods set and 40 for the producer goods
Estimates - of y the elasticity of demand with respect to
given in column of Table 2 The hypothesis that y bull 0 with
alternative was applied for each ittdustry As shown in the
could not be rtjected for 24 of the 32 industries Of the
and 10 are positive For the remalnifg eight three are
and five are significantly positive
Of the threertegative elasetcities the largest (in absolute value)
cosmetics (code 2809) With all elasticity of -1 01 a one
industry advertising would gwerate a decrease in industry sales
more than one percetu
At the other extrema is toys and games (code 39 04) For
a one percent ittcrease in industry advertising would lead to
four-tenths of a pampreentmiddot Though not significantly different from zero
elasticity estimatbull for pt-bpuetary drugs (code 2808)
which are
Both drug industries have
to sales rafios (211 pe1 cent for 28 08 and 13 4 per cent
other hand flaIJo ing ext1ac s (mainly soft-drink SYUlS) also
is for
percent increase in
of slightly
a middottwo - tailed
table that hypothesis
24 14 are negative
significantly negative
9
the codetics and p-ropriacuy very high advertising
for 28 09) bull On the
has a high ind UY
advenisiltg t() sales ratio ( 8 0 per cent) bull but its estimated elasticity is very
small and not signifieantly different uom zero
middot It doe$ not seem unusual to find many elasticities near zero with a few
negative and a few positive If aggregate consumption is insensitive to aggregate
advertising as seems likely ld then what one industry gains must be lost by
others
Page 21
elasticity of demand does
(10) we can
if t 0 then
Finally a determination concerning the industry
not fix the individual firms elasticity Going back to equation
write y bull (1 - ) a + z y Since a gt 0 it follows thatii i i
Y gt 0 If Ylt 0 however y s sign is indeterminateij ij
Turning next to the coefficient of market share 20 of them are negative and
12 are positive with three of each sign being significantly different from zero
The largest positive value which is significant is for pet food (code 2010) and
a close second is mens amp boys suits and coats (code 23 01) The largest signifi
cant negative value is for cosmetics (code 28 09)
In the model some relations among y a 0 and 61 are implied If middota is
negative (1- y) is greater than one Since 61Y bull (1- a) (1 - ) 61Y is
pCsitive but it will be less than one if 0 is large enough Seventeen of the A
ys are negative in each of the seventeen industries 61 is also negative and les
1 (algebraicly) than y so 61Y is greater than one his is not evidence that the
degree of cooperation is low or zero in any industry of course A high o togethegt
with a negative y which is small (in absolute terms) relative to a can give a
value of 61Y greater than one On the other hand if 61Y were less than one
o would have to be greater than zero
For y gt 0 two conditions may hold If y lt a it follows that 61 lt 0 since
(y - a ) lt 0 It also follows that 61Y lt 0 regardless of the value of o bull
In addition the relation is an if-and-only-if one ie if 61Y is negative
then y lt a Three industries have negative values for 61y dairy products
except milk (code 20 04) synthetic fibers (code 2806) and photographic equipment
(code 3808) In all three cases y is quite small so a does not have to be very
large to give a negative value to al
If y gt a the model shows that 61-- gt 0 whatever the value of 0 bull There are
12 cases where both y and 61 are positive with 61 ranging from near zero
(grain milling code 20 12) to 1 20 ( proprietary drugs code 2808) Several
coeffiei-en_rs
--
15
o f to zero
Tab le 3 - Consumer Goods
Indus tr ies
2Distributions of R and Producer Goods
Page 20a
Consumer Producer
Goods Goods
0 - 9 9 0 3
10 - 19 9 0 1 1
20 - 29 9 3 20
30 - 39 9 3 17
40 - 49 9 5 16
50 - 59 9 8
69 9 6 Ii 1060
--f 670 79 9 2
80 - 89 9 4 4
90 - 100 1 1
Totals 32 10 3
Notes
2is used R i s cal-
culated as F F + ((N - K- 1 ) K ] 1 b_ F is the F
statistic Lor the hypothesis that all o f the non - intercept terms are simul taneously equal
1
2
--
-- Since generalized least squ-area regress ion
Page 12
o f these cases have values o f 6 1 Y rea ter than 0 8 (pet food (20 10) bread amp cakes
(20 14 ) miscellaneous processed foods (20 29 ) men s and boys suit s and coats
(23 01 ) p rop riet ary drugs (28 08 ) elect ric housewares (36 1 2) radio and
IV sets 17 and games and toys (39 For be than(36 ) 04)) 6 1 Y to greater 0 8
both o and cry mus t b e less than 0 2 That is in these industries the results
shown in Tab le 2 would hold only with very strong dominance of marke t demand
effects over brand swit ching e ffects and a very low degree o f cooperation
Using the concentration rat ios given in Table 4 the average concentration
ratio for the eight indust ries is 38 1 For the remaining 24 industries it is
43 2 This result would lend some support t o the proposition that concentrat ion
and cooperation are related if the group of eight industries are identi fied
as having low degrees of cooperation
One f inal note on model predict ions S ince both cr and o are assumed to
be non-negative 60 should also be non-negat ive In only one o f the 32 cas es
is 60 negative and it is not signif icant bull
cor pJ
7 1
7 0 4
1 4 7 5 1
)
7 3 z a o a t s 1 1 3 3 5 ( i V Z 8 tJ1 2 1 - 1 1 0 7 3 5 c middot v 2 8 1 0 3 2 -0 0 1 4 3 0 C 1 V 9 8 3 2 8 1 5 1 6 - 0 0 6 t1 6 9 C _ V 1 11 0 2 9 0 1 29 0 00 7 6 1 C l k V 7 3 1 3 6 08 1 3 - o n 4 9 9 4 3 3 3 6 1 2 2 4 0 1 6 4 3
3 1 7 1 3 0 1 1 7 d 7 8 3 3 8 0 8 0 0 2 3 i 2 3 0 6
P 4
3 9 0 3 1 4
1 2
0 09 3 8 4 39 1)1 0 4 l 8 6 5
T A LE 4 e I NCU S T R Y C A T E G O R Y
F TC NOM e tOOE ass
C 1 I c 2 )
2 0 0 1 1 6 2 0 0r 1 7 zo o l Z 2 0 0 7 1 8
z o o 2 6
z o t a 1 6
2 1) 1 2 1 5 2 0 1 4 1 1 2 0 1 a 1 9 2 0 26 1 2 2 0 2 7 1 0 2 0 29 - 3 5
2 3 0 1 10
2 3 02 z o
2 3 03 1 9
2 5 5 1 2 4 Z 1 0 Z 1 1 2 0 3 1 9 2 7 4 1 0 z a o 1 3
2 A 0 7 z s
E LA S T I -C I T Y
c 3 )
- 0 09 7 2 7
0 00 449
- 0 0 2 8 9 3
- 0 3 2 7 70 - 0 OZ il I q
o 4 u t s a 0 0 1 3 7 3
o 1 4 o o-0 0 5 2 4o -0 1 94 0 1
-C l 6 l l 4 0 1 JJ 3 7 0 3 t 0 4 2 0 0 4 0 9
- 0 4 1 8 44 - O l 7 t ) 4 -ll 7 1 4 0 b -0 3 5 7 8 Z
- 0 1 9 7 7 0 1)0 3 3 4
-o n 9 2
IJ A T A
CONV
ON-C ONV
( 4 )
t 1NV
C 1t-V C NV c ) v C - JNV C rlNV c n v c a v C ON V ( i) l V
C CtV
C t1tV C IJNV
C 1 V
c 1 v
Pa ge 2La
HEQ F t Noe x
c 5 )
6 2q
6 89 1 48
6 1 4 ) q 3
1 9 7 1 9
q z t s 5 a e z
l l 9 1
4 o 7 1 2 0
c oqt 1 Z
4 4 )-q 1 z
5 5 9 1 0 64 1 3 3
n 3
Inter industry analysi
Page 23
Equation ( 3 6 ) may be used to fo rmulate a rough test o f the cooperat ion
vs e f ficiency controversy I f we rest ricted ourselves to some subset o f
indus tries where the relat ion be tween y and a i s the same we would be lookingbull
at a sample o f indus tries for which the dependence o f V on H and o is a
quadratic wi th no squared terms
I f we now sub s titute (26) into ( 36 ) we get
(38) v bull a + (y - a) R + (y - a) o (H) - (y - a) H o (H) + u
If it is only d if ferences in marke t shares that affect advertising inten-
s i ty levels i f concentration does not af fect cooperation and i f o bull 0
( 38 ) reduces t o
( 39 ) V bull a + (y - a) H + U
Let this b e the null hyp o thes is and note that V is a linear func-
tion o f H If y gt a and the null hyp othesis holds the expected relat ion
between V and H is as shown by the pos itively sloped straight line in Figure
6 b etween points A and C
In this context the alterna ive hypothesis the degree o f
cooperation ( o) is a positive function o f concentrat ion
is that
(H) I f the func-
tion is as described in (26) and Figure 4 the relation between V and H
would be as shown by the curved line connec t ing po ints A and C in Figure 6 One way to test the null hypothe s is against this al ternative is to test fo r
the hypothesized curvature in the function relating V and H
Fi gure
R
6
v
ID c
H 1 0
I
i I I
sfigned
clasJ s
in def ini g
effec t is the
I
Page middot2lf
One charac teristic of produc ts which may be useful
classes wi th respec t to significanc e of the brand-switching
conveniencenon-convenienc e distinc tion developed by Porter ( 1 97 )
Using his identif ication of IRS indus tries as a s tar t ing point
each of the 32 industries used in this s tudy to one of the two
The ass ignments are not ed in Tab le 4
Using the conveniencenon-convenienc e dis t inc tion the
divided into two sub-samples
sample was
Four regressions were then run bull ene for e ch
sub-sample one for the to tal sample and one for the total sample with
a dummy variable for convenience goods Each regression was run wi th LB s
and wi th indus tries as ob servations The results are given in Tab le 5
For the LB level equations ( 1A - 1D) equation(37) was used ome
rearranging of terms gives v bull y - ( 1-o ) (y - a) ( 1-z ) + ui i i
bull y - 6 1 ( 1 - z ) + ui When observations are pooled across indus trif s yi
is an industry variable In these regressions the values o f y es t imated
in the first stage were used Since industries have been grouped so as
to reduce differences in (y - a) that term is t reated as a cons tant
within each group Finally H the Herfindahl index has been subst ituted
for o the degree of cooperation The resultant equation is linear
y ( 1 - z ) and H x ( l - z ) This is the equation for whi ch results arei i
reported in Table 5 In all regress ions all variables were weighted by the
square roo t of sales
in
raquo
- 34
3 1
i l T 4 8LE S R E G E S S I ON R E SUL T S i X T f N S I ON S T O T Y PE Of
AND L E V E L OF C ONC E kT R A T I O
E O O tt S
tP bull T Y P E NR C ONV O B S tO N-CONV
E L A S T I shy
T I C J T Y
O E - t-41( T SliA R E
H( fl F
1 1 - K O x middot K O X E R F
- H S a 1 1 - H S t 1 8 1 1 9 1
R S O
1 1 0 1
q z
t - bull 9 f 2 1
Z t P D l 6 fC ONV - 1 9 1 s s c 2 1
e - l O I e bull u q a
2 3 1 0 1 1 Nl N- t O V 5 7 1 ( 1 2 19 1 6 1 - J l S 2 5 4
l l l l t z q o a 1 - J o v z I
l ( I NO l l A l l ol u l t o o c bull J u l l
d S
c 3 4 t t r a l
bull Z D 1 0 l l 4 l l 2 3 -gt 2 l l 2 0 - 4 lS 5 4 l C 4 1 o e 6 t l b b 5 1 1 0 1 1 1 c - 1 9 5 1 t d S 2 1 tV
OQ10 T-R A T I O S A E G I VE N l P A RE N T H E S E S l E V E L S f S I G 1 F l C ANC E A - 99 5 B - q 9 t - s NS J UC E NO I Nl E R C ( P T I S U (j f middot) I N f Hf t OUA TIOI S 6 E f 3R E A O J U S T H E N T F OR 1-t E T E R O SIC E O A S T f C I T V pIraquo
1
middot
I
R SO I S C A L C UL A T E D A S f I f bull 1 1 bull 1lt 1 K t t- E E N I S THf U A1 B E R O f fJe S E fi V A T I O S F t S HtE F - S T A T I S
I N DE P E ND E N T y II
T I C F O T H E U Y P O HIE S I S f A l l C JE F F I C H N T A R E QU A L HI l f RO A f40 K I S T H f UUM 8 E R OF Q ( A fL f S
I l l
1 A
1 6
t C
f 1 r R I l l
1 2 1
L Jt
L B
l q
L 1
on s T Y P E bull 2 )
P OUC T
X
H S I l l
l O l A I bull l 0 7 1
O O b O J
1 2 00 1
1 b 2 1 C I l l Z U
l l S b 1 -1 1 0 1
rlt F X l l - fi C f l
bull 1 I
l l
- ()4 2 3 l f H
1 - 2 7 9 1
K O X
1 1 - ti E R F I
l d l
S4 l i A I 2 l b
1 3 00
K O X -t f R F t S J
I 1 - HER F t
)
60 9
z o bull
0
t l l
3 8 8
1 7 6
5 64
64
N f bull
tl S bull
I l l
I 4 1
t ONV
NON-t ONV
A l l
A ll
c o v NO- ( ONV
1 4 1
1 5 1
- l S b
c - 1 1 2 1
5 8 2 C A I 1 6 7 0 1
middot I l 9 H
l 1 il
1 4 1 6 1
E L A C) T J shy
l t C I T Y
I S I
l b l
- 0 2 0
1 - 1 4 9 1
O Z 3 2 1 A I C b 6 b l
0 0 9 2 -J
1 1 b 4 1
o z t t t l 1 2 2 1 1
oru - Ht tt F
1 gt X bull b I
Page 25 bull
Mean val -as ior relevan t variables for the two sub-samples are
Conv Non-Conv
Adv Sales 4 7 2 3
iit Share 2 7 3 2
Ela cicity - 05 3 02 2
Herf indahl 7 3 1 860
That the dist nct ion between coft9enienee goods and non - convenience
goods mat ters is supported not only by these means but also by the regression
results Equat ions lA and lB are different in almost every respec t Only the
concentrat ionmarket share int eract ion term seems to mat ter for convenience
good s Jus t he opposite holds for non - convenience goods both elasticity1
and market sl e are highly significant but the concentrat ionmarket share L
int erac t ion t=rm is not
Given that one minus market share is used these equ3 tions the coeffici t
of market share is the negative of what is sh in col (6) of Table 5 For the
regress ion for non - convenience LB s reported in equation lB then LB s with larger
market shares have lower advertising to sales ratios
On the question of the impact of the degree of cooperation the story is mixed
For convenienc good s the coefficient is pos itive as equation (31) predicts when
there is a po sitive degree of cooperat ion Fo r the non-convenience goods LB s middot
concentration has no effec t
Aggregat ion t o the indus try level and then rer unning the equations $ives the
results shown in (2a-2d) of Tab le 5 As expected R2 goes up and levels of significanc e
fmiddotJr individual coefficients gJ down In add ition the coeffic ient of H ( l -H) inmiddot
quat ion 2A is no t significant though it is po s i tive Since 2A is the indus try level
I
--
all but one of them and that one is not significant For 18 of the 31
wo uld have on its adve rtising to sal es ratio
cognate of LA the variable H ( l-H) pl ays the same role in 2A as H ( l -z 1) p lays
in LA
Vconomies of scale in advert ising
The literature on advert ing contains a number of definitionbull o f economies
of scale and of findings eone rning its presence and magnitude Before I loo k
a t the data with the hope o f mmenting about scale effect s then I wan t to
define the term as I use it
By economies of scale I mean a decrease in the number o f units of adver-
tising which are needed to generate a unit of sales as advert ising is increased
Given the definition of the firm s own elast icity of demand with respect to
advertising (see the text foll wing equation (1) ) the technical definition
is that rii gt 1 Diseconomies of scale are def ined symmetrically i e
Yu lt 1
ith a cro s s - section oi observations on advertising intensity
(v bull a s1) and onmiddot advertising (a ) we may regress the former on the latter 1 i i
The results of such regress ions for the 32 industries are given in Table 2
lines lB ZB bull bull bull 32B For the 32 eases the regression coefficient is positive it
positive coefficients the coefficient i s signif icant
The que tion now is whether these results ar e evidence of pervasive dis-
economies of scale in advertislng I think no t Given my as sumption that the
data I observe are equilibrium results I may ri ghtly conclud e that almost every-
where high levels of advert ising are as sociated in equilibrium with high ratios
of advertising to sales I may no t conclud e anything however about the
impact that a move by some firm away from its equilibrium level o f advertising
middot- Page 27
Given the assumed relation in equation ( 37 ) there is a corre spondshy
ing equilibrium equation relating v and ai The equation is non - linear i
and I have not found a simp le way to characterize it It is po ssib le to
det ermi ne its deriviative with respect to a b y takin$ tne de riviative of ( 3 7 ) i
When that is done the result is
v - 1(40) i bull sl a a z ( 1-z )i i ira-l
Sinc e all the other terms in (40) are non-negative the sign dep ends only
on the sign of 6 1 which is in turn de pend ent only on the sign of (y - a)
That the signs of the coefficients of a in equations lB 2B bull bull bull 32B arei
not always equal to the signs of the coefficients of z in LA 2A bull bull bull 32A i
is still a mystery to me and something to be explored
bull
Summary
My purpose in writing this pa per wa s t o attempt to integrate explicit micro-
economic mode lling reasoned econome tric specification of error terms and high
quality data to exp lore some questions about adver tising by large firms
al quite important improvements could be made in the first two of those areas
and much data ma ssag ing is yet to be done in the third
Concerning the subst an ce of what I have done I think six things are im-
Page 2 8
Severshy
V and Conclusions
portant
1 Explicit modeling is worth the ef for t if for no other reas on than
that it provides a basis for choosing fr om a varie ty of funct ional
forms
2 The same is true for error specif ica t ion I did in fact look at scatshy
ter diagrams for all 32 industry categor ies and they vir tually all show
the het eroscedast icity which I as sumed and for which I correc ted
3 The predicted relation be tween advertis ing in tensity and market share
shows up clearly in on ly 25 of the case s examined I have ye t to
explore why that may b e the cas e
4 Some evidence concerning the presence and impact of coopera tion was
produced but it is not clear ly pervasive
5 The dis tinction between convenience goods and non-c onvenie nce goods is
unambiguously a good one
6 Virtually no evidence concerning econ omies of scale in adver tis ing can
be gleaned fr om this study given its assumpt ions
More work is called for on many of these issues my study seems to have
raised more questions than it has answer ed This is pr obab lly due to the richshy
ne ss of the data source since I had the oppor tunity to address several issue s
at once
- -- - Q --
Foo tnotes
Manager Line o f Business Program Federal Trade Commission The author depended heavily on several staf f members of the LB Pro gram for statistical and clerical support in particular Joe Cholka George Pascoe and Haro lene Jenkins Helpful comments were given b y Richard Caves Dennis Mueller Michael Porter and F M Scherer When the paper was in its early stages o f development both S teve Garber and Jon Rasmussen former economists in the LB Program provided eery useful cri tiques The views expressed here are my own o f course
1 The basic exp licit model of the role of advertising is Dorfman and S teiner ( 1954 ) and the models in this paper may be seen as ano ther extens ion of their work
2 For a review o f models of oligopoly which treat prices and quantities as relevant variables see Shubik ( 1 959) Extens ions to take advertis ing int o account are fairly s traightforward
3 Dorfman and S teiner dealt with a price-setting firm For a monopolis t inverting the demand relation to show Pbull P (Q A) _instead o f Q- (P A) has no effect on the results o f the analysis For oligopolists who face a sys tem of demand relations inversion does have some impac t on the results The dif ferences are not trivial nonetheless they will no t be explored in this study
4 Shubik ( 1 959) p
5 From a formal mathematical point o f view the function is similar to a1 i Legrangian funct ion of the form - r rr
L bull 11 + A ( j - (i i i+j
where the indicates some fixed level of the profits industry Following the same logic which resul ts in and income constraint as the marginal utility of the of the indus tr in the obj ective of firm i
6 Hall amp Weiss ( 1 967) p 323
7 See Schmallensee ( 1 97 2 ) Ch 2 pp 1 6-47
8 On this poin t see Imel amp Helmberger ( 1 9 7 1 )
9 Long ( 1 970) p 130 - 1 39
1 0 See Schmallensee ( 1 972) Ch 3 pp 48-8 7
rrj i+j ) of the rest of the
the identification profits of the rest
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980
V0 bull z [ a1 + (y - cr) J bull a + (y - a) H vl bull Y bull y
11Page
( 2 1) vo bull z d -1 r bull r z bull z- d d- i )11
l -1 N N v bull z d r z bull r z bull l y z - ij jibulll jbulll
and for the simplified case from (19)
(22
Since the forms of the equations in (22) are the same as in (20) all
of the analysis of (20) holdS for (22) except that His substituted for
middot z Advertising intensity will be higher for an industry with cooperativei
firms than for one with non-coope rative firms if the industry demand elas-
ticity dominates the brand-switchine lasticity If the ewo elasticities
are equal then so will be advertiSl intensity for the cooperative firm
and non-cooperative firm industries If brand-switching dominates the
cooperative firm industry will have lower advertising intensity
In the industry model context the possibility of a critical concen-
tration level may be easily introduced That is assume that at values
of R below R the firms are non-cooperative but that for alues equal to
or greater than H they are cooperative That is
(23) V bull a + (y - a) H if H lt H bull y if H gtmiddot H
This function is shown in Figure 2 for the three relamptions be eween y and a
IV THE D GRE OF COOEtRATION
The models depicted in Figure 2 are based on a very simple notion about
the relation between concentration and the cooperationnon-cooperation charac-
teristic They use what is essentilly a step function relating concentra-
Figure
I -
J H H
Page l la
2
v v v
a
I J I y a bull Y y ---i
I I I I I I II I
H 10 H H 10 H 1 0
bullbull
bull bull
Page 12
tion and the extent or degree of coo peration as shown in Figure 3 Using
o as a symbol for the degree of cooperation eq uation (23) can be rewritten
as
bull(24) v (1 - o) vo + ovl bull (1 o) [a+ - a) H ] + oy- (y
vhere
(25) o bull o(H) bull 0 if B lt B
Hbull 1 if B gt
The specific relation between concentration (H) and the deg ree of
cooperation ( o) shown in Figure 3 is more restrictive than necessary and
bas less appeal an one which shows a smooth increase of o from 0 to 1 as
B increases fro to 1 Such a smooth function is shown in Figure 4 he
degree of cooper ion is low until H gets close to H increases greatly as
B goes through R attaining a high level for an H larger than B but still
close to it and then goes to 1 as B goes to 1 More formally
(26) 0 bull o(H)
bull 0 if H bull 0
bull 1 if H bull 1
do gt (IdB
d2o- gt 0 if H B dBZ
0 if H H
lt 0 if R gt H-
If (26) is now substituted for (25) in (24) a corresponding smoothing
of the functions shown in Figure 2 is accomplished he smooth functions
Figure 3
I
10
H 10 H
Page 12a
Figure
Page 12b
4
10
H 10 H
l
(27)
Cl - o) o
[ (1 -
[ (1
t y ij z
j
are given in Figure 5
Page 13
The equation in (24) was constructed as a hybrid of two results derived
from first-order equations for the non-cooperative and cooperative models
It may also be constructed directly by using l bull (1 - o) + on as thei i
th 2objective function for the i-- firm Treating o as a constant the first
order equation is
0 an aa
i middot
n i
Using equations (3) and (7) and setting --- bull 0 givesaai
(28) - 0
After collecting terms and converting to matrix notation we get
(29) - Cl - o) + 0 z-1vi yii i
N
j bulll
+ 0
+ o v1
d-1 r z z
-(30) v Cl - o) rd
-
o) r +d
d r J _ -1- d z
and middot
d) r + 0 rjd
l r z
v - z --(31) zv
- (1 0- +d) t r z- d-
If we now impose the simplified demand structure by substituting (11)
we get
- (1 - o) a +( 32) + oy and
Figure
l
-y
v
Y
H
r
I I
v
a
y
H 10 H
v
a
bulla
HH 10
bullbull
(33)
i
Page 14
v - (1 - o) [ a l + (y - a) ] The corresponding equation for Vis given in (24)
If the relation between industry advertising intensity (V) and
concentration (H) is non - linear via the impact of concentration
on the degree of cooperation does the relation take the form of a
auadratic as some other investigators have proposed (Greer ( 1971 ) Martin (1979) Strickland amp Weiss (1976) ) That question can
be approached from two perspectives - conceptual and empirical In
the next section I will try to apply some statistical material to it
here I want to explore the question in the context of the models
developed above
Referring to equation (24) taking the deriviative with respect
to H and letting bull (H) gives
bull( 3 3a ) 3V (y - a) (1 - o) + (1 - H) 3H dH
Given that o and H each are bounded by 0 and 1 and that o is an in-
creasing function of H the term in the brackets in non-negative The
sign of the deriviative then is the same as the sign of (y - a) If the industry effect is dominant (y gt a) then advertising in-
tensity will increase with concentration If the two effects are equal
then advertising intensity will not vbullry with concentration and if the
brand-switching e ffect is larger advertising intensity will decrease
with concentration
For given advertising elasticity parameters then it is not pos-
sible to show an increase in A S up to some H and the n a decrease
he relation may be non - linear but it is monotonic over the permis-
sible range of H
Page 15
One possibility is to introduce a relation between (y - a) and H
If industries characterized as dominated by the brand-switching effect
also were more concentrated then a cross industry comparison would
show an inverted U shape relation between AS and H I know of no
support for such an association between ( y - a) and H
V ERROR SPECUCATION
So far I have implicitly assumed that there are no errors in the
model I now assume that the error
thvariable in the equation for the i- firm is additive that its mean is zero
that its variance is inversely proportionate to its sales and that all error
term covarlances are zero hat is
34) v bull (1 - o) [ a + (y - a) z ] + oy + u i i i
- - -
In vector notation
(35) bull (1 - o) [ at + lt - a) J +
E() bull 0 Cov (u) -
0
If we now ca lculate V bull bull we get
oy + U bull
(36) V bull (1 - o) la + (y - a)R J +
U bull E(U) bull 0 Var (U) bull
P age 16
The error term specifi c ation which follows Hall
wi th the ratio of profits
6 heir analysis seems
as wellbull particularly given
and Weiss was
applied by them to equa tions to equity or
asse ts as dependen t variables to carry the same
weight in the present context the substantial
evidence presented by marketing analysis on the tendency of company de-
cision makers t o use the advertising to sales ratio as the decision
7variable
The zero covariance assumption on the other hand1 is particularly
troublesome Within a probably not true since
many even ts outside the
single industry it is
industry would have
industries
similar effec ts on all the
8firms in the industry Between it is probably not true either
since many companies produce in more than one indus try and even ts in the
firm would tend to effect all of its activities The extension of the
model to allow for non-zero covariances is certainly called for
--- -
Page Ll
Athird characteristic of this specification also deserves some comment
only on equation is shown for the firm or industry Several haveinvestigators
included an advertising equation ina larger system of equations (Comanor and
Wilson (1974) (1979) and Weiss (1976) so in myMartin Strickland Idid
dissertation1 Ihave ewo defenses for presenting only single equation results
here One is the suggestion by Comanor Wilson that simultaneousand equation
verybias may not be important in this context he second is that Iam will-
ing to see the model extended to include a profitability equation and I intend
to move that direction in the near futurein
Onthe inclusion of aconcentration equation on the other hand I amskep-
tical he problem isthat in some fundamental sense equations in models of
industrial organization should be oriented to the decision-making contexts of firms
Ifeach firm in anindustry with Nfirms has only one decision variable then
there can be only Nindependent equations which are based on the first-order
optimization equations It not show large ofis difficult of course to a number
functional relations which follow from the first-order conditions point here he
isthat only Nof them can be independent
Consider for illustrative purposes only a situation in wich teh N
itsquantities say q are set exogeneously and each firm determines advertising1
of and level say ai where a is the true optimal value a a bulla
1 1 i i + i
Ifp a) then the Nq
s and the Na s will determine N p s say pii bullp
i(l i 1 i
Now given Eand 9 we can determine both the vector s andthe vector v since
- bull til
ands We can also determine S s z bull s-1 s-ii
v and we
industry concentration isHbull (alternatively
V and Hare functions of the same
pendent Furthermore if there are
VM) and (H1 H2 bull bull bull )contain
VI STATISTI CAL APPLICATIONS
Page 18
Given the observed values of might be tempted to conclude that we
have a two equation system in which v and z are jointly determined That would i i
be inappropriate however since the 2N random variables (v ) are determined by
onlt N basic random variables t The variance-covariance matrix for ( ) must
be singular bull
The extension of this observation to industry level variables is straightshy
forward Since observed industry advertising intensity is V bull l v and observed 4
C4 bull z ) it follows thati l i
set of random variables t and cannot be indeshy-
M industries then the variables (V1 v2 at most M independent random variables
Given the assumptions made above y a and o are constant within a
given industry That being the case (34) can be written as
(37)
s1 bull (1 - o) (y - a)
If and
So + +
(1 - o) a + oy
81 are esti mated for an industry using a suitable statisticalSo procedure giving and 81 an estimate of y is then available The sumSo
Intra-industry analysis
1974 of Business Re ort -
categories are identified
Page 19
of So and S1 is y so y bull So + S1 is the estimate of y
If o is assumed to be bounded by zero and one the sign of 61 is also
the sign of y - a The estimate of So is of cou se an estimate of
(1 - o ) a + oy Moreove if y gt a then a is less than 6o- (1 - B) a+ 0(
Cor esponding results would
hold if y bull a or y lt a
If 61 gt 0 then we would estimate that a lt 6o
bull
For purposes of estimating this model 32 manufacturing industry categories
were selected The data are fram the Line of Business forms filed with
the Federal Trade Commission they are discussed in detail in the Annual Line
which is available from the FTC The 32 industry
in Table 1
Simple regressions for equ ation (37) were run for each of the 32 industries
with data for each of the firms which filed in the industry constituting an ob-
servation The results are given in Table 2 lines lA 2A
1974
the intercept
For all the A
equations generalized least quares was used th both and market
share bein scaled bv the square root of sales
The 32 industries were taken from the larger set of all manufacturing industry
categories in the 1974 data files Two selection cr iteria were used thereLB
had to be at least ten reporting companies and the industry had to be a consumer
goods industry or a producer goods industry which is strongly associated with a
related consumer goods industry so that adve rtising by the firms in the producer
goods industry impacts on demand in the related industry (eg soft drink syrup
and soft drinks) There were 135 industries with at least ten companies - 103
producer goods and 32 consumer goods
273
274
TABLE 1 INDUSTRY CATEGORIES USED
FTC
COOE OESCRI P T I ON _
2001 MEAT PACKINGt SAUSAGE S ANO OTHER PREPARED
MEAT PROOUCTS
2004 DAIRY PRODUCTS EXC F LUID MILK
2005 CANNED SPECIALTIES
2001 FROZEN SPECIALTIES
OEHYORATFOt AND
I NC LU 0 ING PRE SERVE S
SOUP MIXESbull
SEASO lNGSt ANO
OTHER PET FOOJ
RAIN MILL PROOUCTSt
ANO PREPARED
RELATED PRODUCTS
PRODUCTS
Z008 CANNEOt ORIEOt PICKLED FRUITS
ANO VEGET ABLES JAMS JELLIES DEHYDRATED I G TA3t E
S40CES ANO SALAD ORESSINGS
2010 DOG CAT A O
2012 FlOUR t OTHER RICE
MILLING fiLENOE) FLOUR
2014 B EADt CAKE ANO
20113 CO FECTIONERY
2026 BOTTLED A O CANNED SOFT DRINKS 2021 FLAVORING EXTRACTS AND SYRUPS EC
2029 MISC FOOOS AND KINDRED PROJUCTS EXC
ROASTED COFFEE
2301 MENS AND BOYS SUITS AND COATS
2302 MENS ANO BOYS FURNISHINGS
230 3 kOMENS AND MISSES OUTER EAR
2551 HOUSEHOLD FURNITURE
2702 PERIODICALS
Pqe 19a 1
RELATED 1972
src cooe
2011 f 3
202tX2026
2032
203pound
2033t4t5
2047 2041t4t5
2051 2065 2086
2081
209XZ05
231
232
233
251 272
2703 BOOKS
2704 ISC PVeLISHING
2806 ORGANIC FlEERS
2807 DRUG$ ETHICAL
23234
PT 2 S 3
2808 ORUCSt PROPRIETARY
2809 PERFU ES t COSMETICS ANO OTHER TaiLET
PREPARATIONS
2810 SOAP AND OTHER CLEANING PREPARATIONS 2815
2901
PEST1CIDES AND AGR ICULTURAL
PE OLEUM REF I N ING CHEMICALS EC
3606 HOUSEHOLD COOKING EQUIPMENT
3612 HOUSEHOLD APPLIANC ES NEC INCLUJING ELECshy
rqlC HOUSE ARFS AND FANS ANO SE I G MACHINES
3617 RADIO AND T V RECE I VING SETS 3808 PHOTOG APHIC E UIP amp SUP LfESt EXC
3qo3
PHO TOCOPYING
SPOR TING A D E QU I P M ENT f SUP P L I E S ATHLETTC GOODS NEC
3904 DOLLS GAMESt TOYS AND CHLOqENtS VE ICLES
PT283
2844
234tX2844
287 9
291 3631
3634t6t9
36H
PT386l
3949
15
943
middot
l A6l f 2 R(GRESSION RESULTS
EQ I NO
NR CODE cu 121
lA 2001 A (
2A 2004 r (
3A 2005 a
-R
ClS amp31
16
17
1 z
I TER-CEPT bull t
ooo9t21At 00022311 oooeos uu
OOl49tCAI 0010 2 middot 000960CCI
o049tl1At o04t55C61
MARKET SH ARE bull 5 I
-010670Ctl
-017600181
-001050
-0211AO
-0077 0
AOVER-TIS INti
lbl
000()39 000072
000018 0 0024liAI
-000079
SHARf
4ampgtVf Q bull
111
000140
-000170
RSQ
COl
66 485 862
607 6 3 t ()2
blti6 633
flASTI-
CI T Y PH
-007271(1
000441
-u0288l
clASTSHU (t)Ef
bull 10
tto
-234
210
( OOlltOCCI -126520 000922 181 001790
4A 200 7 18 00590liAI -038670 4e16 -032110 lell u c
OOOS74e 000700 025940
00067018) 0021001AI -02l100CAI
5 26
920
5A 6 c
200 8 26 0 03Sl9CA I 0012 26181 o 024511(
-005540
-0 81420 0001961AI O()Ol90CAI ooll240
441 604 891
-0020 19 274
6A 2010 16 Oe04224CAI o359lOIAI 9oq ll4015dCAI 061 6 -middot 005435CAI 00011 8 1AI 917 c 00535ltAI -009850((1 OOOA51J CAI -001610161 971
7A 2012 001862 000010 351 00187) oo1 a oooq45 ono1o6 439 c 001109 -027050 Oe0067SIAI -OOll60C81 897
- middotmiddot- -
8A 201t ll 000757 Ol4140CCI 814 ott9oocet 095 8 oooqo6 C C J 000069(81 811 c ootl2lampBI -080910CCI Oe009lUBI -OOtlOOCCI 960I
I DIIJIT pound11 1V OTIUTVTfAll-1pound1t- A _ 00 R - qq r - Q
1jl)OQCD
0
-middot middot--
l
_ __ _
I
i
0 00141 c c
91)7
I
T ABLE 2 REG ESSIOH RESULTS CCONTI
EO NR
IND CODE
iR bull
cos
I TEq-
CEPT
t1ARKE T
SHA E
AOVER-
Tl SING
SHARpound-cent
AuVER
RSO ILAST1-
C I T Y ELAST
SH CPtf (lt Cl) bull 11 I 41 I 5I (61 171 I 8 I f9) c 10)
9A 2 0 1 8 19 002397181 -007640 367 -005 246 le46 6 000605 000550CAI 597
-000029 G14180 OOl654CAJ -024li01At
lOA 2026 12 0028901AI -OlZl90 -0114 0 1 1 1 5 759
Ll (
llA 2021
f3 c
12A lOZ9
B
(
13A 2301
B
c
14A 2302
B
c
154 201
n (
16A 2551
8
10
35
10
20
19
24
001897(()
00141 6CCI
Oll9lHCI
010604 middot middot
Oll083
OOJ11JCAt
002059141
001411CIH
-000412
000293(81
000187
001159 (3)
ooo6a4cct
o00792CCI
OOl40dCAt
000006
-000076
OOlltl6fA)
o o0 5 64CCt
-034570
-o1middot o3o
-1659 90
Ol51lO
-046190CAt
Ol5450CA)
-005l60
002940
-030060(()
-0 43l50CCt
0 33650
-019080
00004
bullJ00741CAI
-000069
000466
0001 51CAt
000947CAt
O OOI)70CAt OOl750CA J
000175CCI
000671CAI
OOll5) f A t 002207CBI
0004b7CA I
- 007 1 70
001760
-001410CAt
- 0 16200CAt
-004210
- 1 207 70
705
96 0
542
466 805
460
614
921
832
9 61
996
424
550
9 15
56 1 651
884
-006114
019 037
03504lCA)
oo409Q
-141644((
491 -0 17664
607
871
lell
084
101
072
103
108 11IllOQfl)-- l
-037880fCJ OOl6l4CAJ -040l70fCI 0()
c
I
I
bull 9
9t
middot middot middotbullmiddot bull lraquo
TABLE 2 REGRE SION RESUlTS ICONTt
bull
EO I NO tHl I filER- f4A_K T AOVER- sbulluRE ttSQ flA Jl- iLAS T NR CODE cos CEPT SHARE JISING AOVfR bull (JTY SHR CiJCF
110Cl I f amp Itt 15J tbt bull 1c u caa
llA 2702 ll o079 6 81 Bt -019110 S6s -OlltO 111
a 004679 ooooo1 t69
( oOl26o -015160 001560 -022690 ltJr()I
ldA 270 3 19 0065lt -OIt2l60 220 -Ol57a2 11 8 e o 0 71t6IA t ol08 75tAt ( OOl983CiU -060llOC() OOl7191A t -O lltdOC A) middot 89
lqA 2704 10 005082 -Oil060 22b -l)i1977 lelO 6 -000080 oo2772C4 t 910 c 000907 -011 200 Oe04105CA t -046960 995
20A 2d06 1 II
O OlltStiAt -001120 837 00014 -))th O01380amp At -oooo 11 821 ( OOl526CAt -026700181 0002 5UAt 000530 969
2lA 2807 28 OOt9lt9CAt -00891 0 516 -0 03962 225I8 002095CCt OOOl lliAt 690 ( o02965CA t -056650((1 OOOt50CAJ -003l70ttl 911I
I 22A 2808 15 ot3l06 119730 655 1130) 090
0 - 007526CC t Oe003961At 835- middot-
c I oo7523 -067210 001l 2tA t -0091t80CCt 951
2lA 2809 21 Oe2067lCAt -1218101At 71 2 -10095((1 121 ttlIaOQ8 00797418) OOOI8lltt 686 Ill( 0092lU A t -Olltl70Ct t 000605CAt -OOit8701At 951 _
---- -
32 009506CAt
0 bull 06 981 I A J 006901 cA bull
00 24A 2810 -010960 5t 5 -OOltSO 7 56
8 -000011 s11 c -5811701At oooaoatAJ oot520IAI 888
-- 37 --- middot - -- -1 ) l lbull t5JJ r-l bullhmiddot1 --- - --
l5A
26A
27A
c
2RA
c
2 9A
31A
llA
24
13
597
Ci)55
14
l 6h
RSQ ELASTI- (lASTCOE f-
TABLE 2
EQ I NO NR CODE IU
REGRESSION RESULTS CCONTa
NR INTER- MARKET
(10S CEPT SHARE AOVER- SHARfcent
TIS lNG AOVER
) CITY StiR
c It) bull 5 co a I 1) bull 0 bull 9 c l 0 bull( 2 1
2815 16 003019 -009990 275 -o069bmiddotJ
6 001446 000224 282 c 002178 -054630 o02221CCI -0075 00 646
2901 29 00021t2 CAI 0)0520 592 0007 6 1 068
0 00105C8 t 000012141 60)g( OOOlJ51AJ -00126018) ooou271At -000u80CA) 974
36oa 1 3 003296(4) -oca29o -oo-q94
001755 000203 lt91t 002499((1 -04l820CCI OOl374CAI 0068)0 946
3612 0033281CI 011520 195 01684) oso
00 30 42C OI 00012 2 424 oo579ltAt -2 7l7801At 002 100IAI -003220 819
8 3617 btel 011789 081002259 009530
001448 oOOl OitCCI 732 OOl7 22CAI -0 46160CAI 0004961AI -OOll40181 989c
30A 3808 14 0021t 3 6fAI -000110 628 o023291C) -005 6 middot- _ middot- 0bull 02 04 1 IC I ooooo9 637 ( 001584((1 -0161)1)0 00021 91AI -0 00060
390) 14 002790CCt 006590 h8 t 0 0938 4 070
B OOl89bCCJ 000144 74
0021961(1 -OItlOlO OOl323CAI -004800 915
B
middot--- - middot 4 - middot- - dfll3904 12 004563181 0)8300 879 0428651(1 089 OQIll004 277CA I ooozstcea 919
c 0041)3181 -0-11250 0011881(1 -006270 972 0Ill
- J - bull bull bull bullbullbull middot - - middot - middot gt middotbull L)ofbullflbull middot middotfJltl -tn- middotmiddotmiddotmiddoto +f middot t middot bull
c
is
positive
2
Page 0 determineThe consumer goodproducer good split is sometimes difficult to
Sollle support for the split used in this paper is a comparison of goodness - of - fit
distributions of a
industries
set
advertising are
that industry
a sales increase of
the
highest aD10U3 those
measures for the two groups of industries Table 3 contains
measures for the 32- consumer goods industries and the 103 producer goods
2 Median R is 5 7 for the consumer goods set and 40 for the producer goods
Estimates - of y the elasticity of demand with respect to
given in column of Table 2 The hypothesis that y bull 0 with
alternative was applied for each ittdustry As shown in the
could not be rtjected for 24 of the 32 industries Of the
and 10 are positive For the remalnifg eight three are
and five are significantly positive
Of the threertegative elasetcities the largest (in absolute value)
cosmetics (code 2809) With all elasticity of -1 01 a one
industry advertising would gwerate a decrease in industry sales
more than one percetu
At the other extrema is toys and games (code 39 04) For
a one percent ittcrease in industry advertising would lead to
four-tenths of a pampreentmiddot Though not significantly different from zero
elasticity estimatbull for pt-bpuetary drugs (code 2808)
which are
Both drug industries have
to sales rafios (211 pe1 cent for 28 08 and 13 4 per cent
other hand flaIJo ing ext1ac s (mainly soft-drink SYUlS) also
is for
percent increase in
of slightly
a middottwo - tailed
table that hypothesis
24 14 are negative
significantly negative
9
the codetics and p-ropriacuy very high advertising
for 28 09) bull On the
has a high ind UY
advenisiltg t() sales ratio ( 8 0 per cent) bull but its estimated elasticity is very
small and not signifieantly different uom zero
middot It doe$ not seem unusual to find many elasticities near zero with a few
negative and a few positive If aggregate consumption is insensitive to aggregate
advertising as seems likely ld then what one industry gains must be lost by
others
Page 21
elasticity of demand does
(10) we can
if t 0 then
Finally a determination concerning the industry
not fix the individual firms elasticity Going back to equation
write y bull (1 - ) a + z y Since a gt 0 it follows thatii i i
Y gt 0 If Ylt 0 however y s sign is indeterminateij ij
Turning next to the coefficient of market share 20 of them are negative and
12 are positive with three of each sign being significantly different from zero
The largest positive value which is significant is for pet food (code 2010) and
a close second is mens amp boys suits and coats (code 23 01) The largest signifi
cant negative value is for cosmetics (code 28 09)
In the model some relations among y a 0 and 61 are implied If middota is
negative (1- y) is greater than one Since 61Y bull (1- a) (1 - ) 61Y is
pCsitive but it will be less than one if 0 is large enough Seventeen of the A
ys are negative in each of the seventeen industries 61 is also negative and les
1 (algebraicly) than y so 61Y is greater than one his is not evidence that the
degree of cooperation is low or zero in any industry of course A high o togethegt
with a negative y which is small (in absolute terms) relative to a can give a
value of 61Y greater than one On the other hand if 61Y were less than one
o would have to be greater than zero
For y gt 0 two conditions may hold If y lt a it follows that 61 lt 0 since
(y - a ) lt 0 It also follows that 61Y lt 0 regardless of the value of o bull
In addition the relation is an if-and-only-if one ie if 61Y is negative
then y lt a Three industries have negative values for 61y dairy products
except milk (code 20 04) synthetic fibers (code 2806) and photographic equipment
(code 3808) In all three cases y is quite small so a does not have to be very
large to give a negative value to al
If y gt a the model shows that 61-- gt 0 whatever the value of 0 bull There are
12 cases where both y and 61 are positive with 61 ranging from near zero
(grain milling code 20 12) to 1 20 ( proprietary drugs code 2808) Several
coeffiei-en_rs
--
15
o f to zero
Tab le 3 - Consumer Goods
Indus tr ies
2Distributions of R and Producer Goods
Page 20a
Consumer Producer
Goods Goods
0 - 9 9 0 3
10 - 19 9 0 1 1
20 - 29 9 3 20
30 - 39 9 3 17
40 - 49 9 5 16
50 - 59 9 8
69 9 6 Ii 1060
--f 670 79 9 2
80 - 89 9 4 4
90 - 100 1 1
Totals 32 10 3
Notes
2is used R i s cal-
culated as F F + ((N - K- 1 ) K ] 1 b_ F is the F
statistic Lor the hypothesis that all o f the non - intercept terms are simul taneously equal
1
2
--
-- Since generalized least squ-area regress ion
Page 12
o f these cases have values o f 6 1 Y rea ter than 0 8 (pet food (20 10) bread amp cakes
(20 14 ) miscellaneous processed foods (20 29 ) men s and boys suit s and coats
(23 01 ) p rop riet ary drugs (28 08 ) elect ric housewares (36 1 2) radio and
IV sets 17 and games and toys (39 For be than(36 ) 04)) 6 1 Y to greater 0 8
both o and cry mus t b e less than 0 2 That is in these industries the results
shown in Tab le 2 would hold only with very strong dominance of marke t demand
effects over brand swit ching e ffects and a very low degree o f cooperation
Using the concentration rat ios given in Table 4 the average concentration
ratio for the eight indust ries is 38 1 For the remaining 24 industries it is
43 2 This result would lend some support t o the proposition that concentrat ion
and cooperation are related if the group of eight industries are identi fied
as having low degrees of cooperation
One f inal note on model predict ions S ince both cr and o are assumed to
be non-negative 60 should also be non-negat ive In only one o f the 32 cas es
is 60 negative and it is not signif icant bull
cor pJ
7 1
7 0 4
1 4 7 5 1
)
7 3 z a o a t s 1 1 3 3 5 ( i V Z 8 tJ1 2 1 - 1 1 0 7 3 5 c middot v 2 8 1 0 3 2 -0 0 1 4 3 0 C 1 V 9 8 3 2 8 1 5 1 6 - 0 0 6 t1 6 9 C _ V 1 11 0 2 9 0 1 29 0 00 7 6 1 C l k V 7 3 1 3 6 08 1 3 - o n 4 9 9 4 3 3 3 6 1 2 2 4 0 1 6 4 3
3 1 7 1 3 0 1 1 7 d 7 8 3 3 8 0 8 0 0 2 3 i 2 3 0 6
P 4
3 9 0 3 1 4
1 2
0 09 3 8 4 39 1)1 0 4 l 8 6 5
T A LE 4 e I NCU S T R Y C A T E G O R Y
F TC NOM e tOOE ass
C 1 I c 2 )
2 0 0 1 1 6 2 0 0r 1 7 zo o l Z 2 0 0 7 1 8
z o o 2 6
z o t a 1 6
2 1) 1 2 1 5 2 0 1 4 1 1 2 0 1 a 1 9 2 0 26 1 2 2 0 2 7 1 0 2 0 29 - 3 5
2 3 0 1 10
2 3 02 z o
2 3 03 1 9
2 5 5 1 2 4 Z 1 0 Z 1 1 2 0 3 1 9 2 7 4 1 0 z a o 1 3
2 A 0 7 z s
E LA S T I -C I T Y
c 3 )
- 0 09 7 2 7
0 00 449
- 0 0 2 8 9 3
- 0 3 2 7 70 - 0 OZ il I q
o 4 u t s a 0 0 1 3 7 3
o 1 4 o o-0 0 5 2 4o -0 1 94 0 1
-C l 6 l l 4 0 1 JJ 3 7 0 3 t 0 4 2 0 0 4 0 9
- 0 4 1 8 44 - O l 7 t ) 4 -ll 7 1 4 0 b -0 3 5 7 8 Z
- 0 1 9 7 7 0 1)0 3 3 4
-o n 9 2
IJ A T A
CONV
ON-C ONV
( 4 )
t 1NV
C 1t-V C NV c ) v C - JNV C rlNV c n v c a v C ON V ( i) l V
C CtV
C t1tV C IJNV
C 1 V
c 1 v
Pa ge 2La
HEQ F t Noe x
c 5 )
6 2q
6 89 1 48
6 1 4 ) q 3
1 9 7 1 9
q z t s 5 a e z
l l 9 1
4 o 7 1 2 0
c oqt 1 Z
4 4 )-q 1 z
5 5 9 1 0 64 1 3 3
n 3
Inter industry analysi
Page 23
Equation ( 3 6 ) may be used to fo rmulate a rough test o f the cooperat ion
vs e f ficiency controversy I f we rest ricted ourselves to some subset o f
indus tries where the relat ion be tween y and a i s the same we would be lookingbull
at a sample o f indus tries for which the dependence o f V on H and o is a
quadratic wi th no squared terms
I f we now sub s titute (26) into ( 36 ) we get
(38) v bull a + (y - a) R + (y - a) o (H) - (y - a) H o (H) + u
If it is only d if ferences in marke t shares that affect advertising inten-
s i ty levels i f concentration does not af fect cooperation and i f o bull 0
( 38 ) reduces t o
( 39 ) V bull a + (y - a) H + U
Let this b e the null hyp o thes is and note that V is a linear func-
tion o f H If y gt a and the null hyp othesis holds the expected relat ion
between V and H is as shown by the pos itively sloped straight line in Figure
6 b etween points A and C
In this context the alterna ive hypothesis the degree o f
cooperation ( o) is a positive function o f concentrat ion
is that
(H) I f the func-
tion is as described in (26) and Figure 4 the relation between V and H
would be as shown by the curved line connec t ing po ints A and C in Figure 6 One way to test the null hypothe s is against this al ternative is to test fo r
the hypothesized curvature in the function relating V and H
Fi gure
R
6
v
ID c
H 1 0
I
i I I
sfigned
clasJ s
in def ini g
effec t is the
I
Page middot2lf
One charac teristic of produc ts which may be useful
classes wi th respec t to significanc e of the brand-switching
conveniencenon-convenienc e distinc tion developed by Porter ( 1 97 )
Using his identif ication of IRS indus tries as a s tar t ing point
each of the 32 industries used in this s tudy to one of the two
The ass ignments are not ed in Tab le 4
Using the conveniencenon-convenienc e dis t inc tion the
divided into two sub-samples
sample was
Four regressions were then run bull ene for e ch
sub-sample one for the to tal sample and one for the total sample with
a dummy variable for convenience goods Each regression was run wi th LB s
and wi th indus tries as ob servations The results are given in Tab le 5
For the LB level equations ( 1A - 1D) equation(37) was used ome
rearranging of terms gives v bull y - ( 1-o ) (y - a) ( 1-z ) + ui i i
bull y - 6 1 ( 1 - z ) + ui When observations are pooled across indus trif s yi
is an industry variable In these regressions the values o f y es t imated
in the first stage were used Since industries have been grouped so as
to reduce differences in (y - a) that term is t reated as a cons tant
within each group Finally H the Herfindahl index has been subst ituted
for o the degree of cooperation The resultant equation is linear
y ( 1 - z ) and H x ( l - z ) This is the equation for whi ch results arei i
reported in Table 5 In all regress ions all variables were weighted by the
square roo t of sales
in
raquo
- 34
3 1
i l T 4 8LE S R E G E S S I ON R E SUL T S i X T f N S I ON S T O T Y PE Of
AND L E V E L OF C ONC E kT R A T I O
E O O tt S
tP bull T Y P E NR C ONV O B S tO N-CONV
E L A S T I shy
T I C J T Y
O E - t-41( T SliA R E
H( fl F
1 1 - K O x middot K O X E R F
- H S a 1 1 - H S t 1 8 1 1 9 1
R S O
1 1 0 1
q z
t - bull 9 f 2 1
Z t P D l 6 fC ONV - 1 9 1 s s c 2 1
e - l O I e bull u q a
2 3 1 0 1 1 Nl N- t O V 5 7 1 ( 1 2 19 1 6 1 - J l S 2 5 4
l l l l t z q o a 1 - J o v z I
l ( I NO l l A l l ol u l t o o c bull J u l l
d S
c 3 4 t t r a l
bull Z D 1 0 l l 4 l l 2 3 -gt 2 l l 2 0 - 4 lS 5 4 l C 4 1 o e 6 t l b b 5 1 1 0 1 1 1 c - 1 9 5 1 t d S 2 1 tV
OQ10 T-R A T I O S A E G I VE N l P A RE N T H E S E S l E V E L S f S I G 1 F l C ANC E A - 99 5 B - q 9 t - s NS J UC E NO I Nl E R C ( P T I S U (j f middot) I N f Hf t OUA TIOI S 6 E f 3R E A O J U S T H E N T F OR 1-t E T E R O SIC E O A S T f C I T V pIraquo
1
middot
I
R SO I S C A L C UL A T E D A S f I f bull 1 1 bull 1lt 1 K t t- E E N I S THf U A1 B E R O f fJe S E fi V A T I O S F t S HtE F - S T A T I S
I N DE P E ND E N T y II
T I C F O T H E U Y P O HIE S I S f A l l C JE F F I C H N T A R E QU A L HI l f RO A f40 K I S T H f UUM 8 E R OF Q ( A fL f S
I l l
1 A
1 6
t C
f 1 r R I l l
1 2 1
L Jt
L B
l q
L 1
on s T Y P E bull 2 )
P OUC T
X
H S I l l
l O l A I bull l 0 7 1
O O b O J
1 2 00 1
1 b 2 1 C I l l Z U
l l S b 1 -1 1 0 1
rlt F X l l - fi C f l
bull 1 I
l l
- ()4 2 3 l f H
1 - 2 7 9 1
K O X
1 1 - ti E R F I
l d l
S4 l i A I 2 l b
1 3 00
K O X -t f R F t S J
I 1 - HER F t
)
60 9
z o bull
0
t l l
3 8 8
1 7 6
5 64
64
N f bull
tl S bull
I l l
I 4 1
t ONV
NON-t ONV
A l l
A ll
c o v NO- ( ONV
1 4 1
1 5 1
- l S b
c - 1 1 2 1
5 8 2 C A I 1 6 7 0 1
middot I l 9 H
l 1 il
1 4 1 6 1
E L A C) T J shy
l t C I T Y
I S I
l b l
- 0 2 0
1 - 1 4 9 1
O Z 3 2 1 A I C b 6 b l
0 0 9 2 -J
1 1 b 4 1
o z t t t l 1 2 2 1 1
oru - Ht tt F
1 gt X bull b I
Page 25 bull
Mean val -as ior relevan t variables for the two sub-samples are
Conv Non-Conv
Adv Sales 4 7 2 3
iit Share 2 7 3 2
Ela cicity - 05 3 02 2
Herf indahl 7 3 1 860
That the dist nct ion between coft9enienee goods and non - convenience
goods mat ters is supported not only by these means but also by the regression
results Equat ions lA and lB are different in almost every respec t Only the
concentrat ionmarket share int eract ion term seems to mat ter for convenience
good s Jus t he opposite holds for non - convenience goods both elasticity1
and market sl e are highly significant but the concentrat ionmarket share L
int erac t ion t=rm is not
Given that one minus market share is used these equ3 tions the coeffici t
of market share is the negative of what is sh in col (6) of Table 5 For the
regress ion for non - convenience LB s reported in equation lB then LB s with larger
market shares have lower advertising to sales ratios
On the question of the impact of the degree of cooperation the story is mixed
For convenienc good s the coefficient is pos itive as equation (31) predicts when
there is a po sitive degree of cooperat ion Fo r the non-convenience goods LB s middot
concentration has no effec t
Aggregat ion t o the indus try level and then rer unning the equations $ives the
results shown in (2a-2d) of Tab le 5 As expected R2 goes up and levels of significanc e
fmiddotJr individual coefficients gJ down In add ition the coeffic ient of H ( l -H) inmiddot
quat ion 2A is no t significant though it is po s i tive Since 2A is the indus try level
I
--
all but one of them and that one is not significant For 18 of the 31
wo uld have on its adve rtising to sal es ratio
cognate of LA the variable H ( l-H) pl ays the same role in 2A as H ( l -z 1) p lays
in LA
Vconomies of scale in advert ising
The literature on advert ing contains a number of definitionbull o f economies
of scale and of findings eone rning its presence and magnitude Before I loo k
a t the data with the hope o f mmenting about scale effect s then I wan t to
define the term as I use it
By economies of scale I mean a decrease in the number o f units of adver-
tising which are needed to generate a unit of sales as advert ising is increased
Given the definition of the firm s own elast icity of demand with respect to
advertising (see the text foll wing equation (1) ) the technical definition
is that rii gt 1 Diseconomies of scale are def ined symmetrically i e
Yu lt 1
ith a cro s s - section oi observations on advertising intensity
(v bull a s1) and onmiddot advertising (a ) we may regress the former on the latter 1 i i
The results of such regress ions for the 32 industries are given in Table 2
lines lB ZB bull bull bull 32B For the 32 eases the regression coefficient is positive it
positive coefficients the coefficient i s signif icant
The que tion now is whether these results ar e evidence of pervasive dis-
economies of scale in advertislng I think no t Given my as sumption that the
data I observe are equilibrium results I may ri ghtly conclud e that almost every-
where high levels of advert ising are as sociated in equilibrium with high ratios
of advertising to sales I may no t conclud e anything however about the
impact that a move by some firm away from its equilibrium level o f advertising
middot- Page 27
Given the assumed relation in equation ( 37 ) there is a corre spondshy
ing equilibrium equation relating v and ai The equation is non - linear i
and I have not found a simp le way to characterize it It is po ssib le to
det ermi ne its deriviative with respect to a b y takin$ tne de riviative of ( 3 7 ) i
When that is done the result is
v - 1(40) i bull sl a a z ( 1-z )i i ira-l
Sinc e all the other terms in (40) are non-negative the sign dep ends only
on the sign of 6 1 which is in turn de pend ent only on the sign of (y - a)
That the signs of the coefficients of a in equations lB 2B bull bull bull 32B arei
not always equal to the signs of the coefficients of z in LA 2A bull bull bull 32A i
is still a mystery to me and something to be explored
bull
Summary
My purpose in writing this pa per wa s t o attempt to integrate explicit micro-
economic mode lling reasoned econome tric specification of error terms and high
quality data to exp lore some questions about adver tising by large firms
al quite important improvements could be made in the first two of those areas
and much data ma ssag ing is yet to be done in the third
Concerning the subst an ce of what I have done I think six things are im-
Page 2 8
Severshy
V and Conclusions
portant
1 Explicit modeling is worth the ef for t if for no other reas on than
that it provides a basis for choosing fr om a varie ty of funct ional
forms
2 The same is true for error specif ica t ion I did in fact look at scatshy
ter diagrams for all 32 industry categor ies and they vir tually all show
the het eroscedast icity which I as sumed and for which I correc ted
3 The predicted relation be tween advertis ing in tensity and market share
shows up clearly in on ly 25 of the case s examined I have ye t to
explore why that may b e the cas e
4 Some evidence concerning the presence and impact of coopera tion was
produced but it is not clear ly pervasive
5 The dis tinction between convenience goods and non-c onvenie nce goods is
unambiguously a good one
6 Virtually no evidence concerning econ omies of scale in adver tis ing can
be gleaned fr om this study given its assumpt ions
More work is called for on many of these issues my study seems to have
raised more questions than it has answer ed This is pr obab lly due to the richshy
ne ss of the data source since I had the oppor tunity to address several issue s
at once
- -- - Q --
Foo tnotes
Manager Line o f Business Program Federal Trade Commission The author depended heavily on several staf f members of the LB Pro gram for statistical and clerical support in particular Joe Cholka George Pascoe and Haro lene Jenkins Helpful comments were given b y Richard Caves Dennis Mueller Michael Porter and F M Scherer When the paper was in its early stages o f development both S teve Garber and Jon Rasmussen former economists in the LB Program provided eery useful cri tiques The views expressed here are my own o f course
1 The basic exp licit model of the role of advertising is Dorfman and S teiner ( 1954 ) and the models in this paper may be seen as ano ther extens ion of their work
2 For a review o f models of oligopoly which treat prices and quantities as relevant variables see Shubik ( 1 959) Extens ions to take advertis ing int o account are fairly s traightforward
3 Dorfman and S teiner dealt with a price-setting firm For a monopolis t inverting the demand relation to show Pbull P (Q A) _instead o f Q- (P A) has no effect on the results o f the analysis For oligopolists who face a sys tem of demand relations inversion does have some impac t on the results The dif ferences are not trivial nonetheless they will no t be explored in this study
4 Shubik ( 1 959) p
5 From a formal mathematical point o f view the function is similar to a1 i Legrangian funct ion of the form - r rr
L bull 11 + A ( j - (i i i+j
where the indicates some fixed level of the profits industry Following the same logic which resul ts in and income constraint as the marginal utility of the of the indus tr in the obj ective of firm i
6 Hall amp Weiss ( 1 967) p 323
7 See Schmallensee ( 1 97 2 ) Ch 2 pp 1 6-47
8 On this poin t see Imel amp Helmberger ( 1 9 7 1 )
9 Long ( 1 970) p 130 - 1 39
1 0 See Schmallensee ( 1 972) Ch 3 pp 48-8 7
rrj i+j ) of the rest of the
the identification profits of the rest
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980
Figure
I -
J H H
Page l la
2
v v v
a
I J I y a bull Y y ---i
I I I I I I II I
H 10 H H 10 H 1 0
bullbull
bull bull
Page 12
tion and the extent or degree of coo peration as shown in Figure 3 Using
o as a symbol for the degree of cooperation eq uation (23) can be rewritten
as
bull(24) v (1 - o) vo + ovl bull (1 o) [a+ - a) H ] + oy- (y
vhere
(25) o bull o(H) bull 0 if B lt B
Hbull 1 if B gt
The specific relation between concentration (H) and the deg ree of
cooperation ( o) shown in Figure 3 is more restrictive than necessary and
bas less appeal an one which shows a smooth increase of o from 0 to 1 as
B increases fro to 1 Such a smooth function is shown in Figure 4 he
degree of cooper ion is low until H gets close to H increases greatly as
B goes through R attaining a high level for an H larger than B but still
close to it and then goes to 1 as B goes to 1 More formally
(26) 0 bull o(H)
bull 0 if H bull 0
bull 1 if H bull 1
do gt (IdB
d2o- gt 0 if H B dBZ
0 if H H
lt 0 if R gt H-
If (26) is now substituted for (25) in (24) a corresponding smoothing
of the functions shown in Figure 2 is accomplished he smooth functions
Figure 3
I
10
H 10 H
Page 12a
Figure
Page 12b
4
10
H 10 H
l
(27)
Cl - o) o
[ (1 -
[ (1
t y ij z
j
are given in Figure 5
Page 13
The equation in (24) was constructed as a hybrid of two results derived
from first-order equations for the non-cooperative and cooperative models
It may also be constructed directly by using l bull (1 - o) + on as thei i
th 2objective function for the i-- firm Treating o as a constant the first
order equation is
0 an aa
i middot
n i
Using equations (3) and (7) and setting --- bull 0 givesaai
(28) - 0
After collecting terms and converting to matrix notation we get
(29) - Cl - o) + 0 z-1vi yii i
N
j bulll
+ 0
+ o v1
d-1 r z z
-(30) v Cl - o) rd
-
o) r +d
d r J _ -1- d z
and middot
d) r + 0 rjd
l r z
v - z --(31) zv
- (1 0- +d) t r z- d-
If we now impose the simplified demand structure by substituting (11)
we get
- (1 - o) a +( 32) + oy and
Figure
l
-y
v
Y
H
r
I I
v
a
y
H 10 H
v
a
bulla
HH 10
bullbull
(33)
i
Page 14
v - (1 - o) [ a l + (y - a) ] The corresponding equation for Vis given in (24)
If the relation between industry advertising intensity (V) and
concentration (H) is non - linear via the impact of concentration
on the degree of cooperation does the relation take the form of a
auadratic as some other investigators have proposed (Greer ( 1971 ) Martin (1979) Strickland amp Weiss (1976) ) That question can
be approached from two perspectives - conceptual and empirical In
the next section I will try to apply some statistical material to it
here I want to explore the question in the context of the models
developed above
Referring to equation (24) taking the deriviative with respect
to H and letting bull (H) gives
bull( 3 3a ) 3V (y - a) (1 - o) + (1 - H) 3H dH
Given that o and H each are bounded by 0 and 1 and that o is an in-
creasing function of H the term in the brackets in non-negative The
sign of the deriviative then is the same as the sign of (y - a) If the industry effect is dominant (y gt a) then advertising in-
tensity will increase with concentration If the two effects are equal
then advertising intensity will not vbullry with concentration and if the
brand-switching e ffect is larger advertising intensity will decrease
with concentration
For given advertising elasticity parameters then it is not pos-
sible to show an increase in A S up to some H and the n a decrease
he relation may be non - linear but it is monotonic over the permis-
sible range of H
Page 15
One possibility is to introduce a relation between (y - a) and H
If industries characterized as dominated by the brand-switching effect
also were more concentrated then a cross industry comparison would
show an inverted U shape relation between AS and H I know of no
support for such an association between ( y - a) and H
V ERROR SPECUCATION
So far I have implicitly assumed that there are no errors in the
model I now assume that the error
thvariable in the equation for the i- firm is additive that its mean is zero
that its variance is inversely proportionate to its sales and that all error
term covarlances are zero hat is
34) v bull (1 - o) [ a + (y - a) z ] + oy + u i i i
- - -
In vector notation
(35) bull (1 - o) [ at + lt - a) J +
E() bull 0 Cov (u) -
0
If we now ca lculate V bull bull we get
oy + U bull
(36) V bull (1 - o) la + (y - a)R J +
U bull E(U) bull 0 Var (U) bull
P age 16
The error term specifi c ation which follows Hall
wi th the ratio of profits
6 heir analysis seems
as wellbull particularly given
and Weiss was
applied by them to equa tions to equity or
asse ts as dependen t variables to carry the same
weight in the present context the substantial
evidence presented by marketing analysis on the tendency of company de-
cision makers t o use the advertising to sales ratio as the decision
7variable
The zero covariance assumption on the other hand1 is particularly
troublesome Within a probably not true since
many even ts outside the
single industry it is
industry would have
industries
similar effec ts on all the
8firms in the industry Between it is probably not true either
since many companies produce in more than one indus try and even ts in the
firm would tend to effect all of its activities The extension of the
model to allow for non-zero covariances is certainly called for
--- -
Page Ll
Athird characteristic of this specification also deserves some comment
only on equation is shown for the firm or industry Several haveinvestigators
included an advertising equation ina larger system of equations (Comanor and
Wilson (1974) (1979) and Weiss (1976) so in myMartin Strickland Idid
dissertation1 Ihave ewo defenses for presenting only single equation results
here One is the suggestion by Comanor Wilson that simultaneousand equation
verybias may not be important in this context he second is that Iam will-
ing to see the model extended to include a profitability equation and I intend
to move that direction in the near futurein
Onthe inclusion of aconcentration equation on the other hand I amskep-
tical he problem isthat in some fundamental sense equations in models of
industrial organization should be oriented to the decision-making contexts of firms
Ifeach firm in anindustry with Nfirms has only one decision variable then
there can be only Nindependent equations which are based on the first-order
optimization equations It not show large ofis difficult of course to a number
functional relations which follow from the first-order conditions point here he
isthat only Nof them can be independent
Consider for illustrative purposes only a situation in wich teh N
itsquantities say q are set exogeneously and each firm determines advertising1
of and level say ai where a is the true optimal value a a bulla
1 1 i i + i
Ifp a) then the Nq
s and the Na s will determine N p s say pii bullp
i(l i 1 i
Now given Eand 9 we can determine both the vector s andthe vector v since
- bull til
ands We can also determine S s z bull s-1 s-ii
v and we
industry concentration isHbull (alternatively
V and Hare functions of the same
pendent Furthermore if there are
VM) and (H1 H2 bull bull bull )contain
VI STATISTI CAL APPLICATIONS
Page 18
Given the observed values of might be tempted to conclude that we
have a two equation system in which v and z are jointly determined That would i i
be inappropriate however since the 2N random variables (v ) are determined by
onlt N basic random variables t The variance-covariance matrix for ( ) must
be singular bull
The extension of this observation to industry level variables is straightshy
forward Since observed industry advertising intensity is V bull l v and observed 4
C4 bull z ) it follows thati l i
set of random variables t and cannot be indeshy-
M industries then the variables (V1 v2 at most M independent random variables
Given the assumptions made above y a and o are constant within a
given industry That being the case (34) can be written as
(37)
s1 bull (1 - o) (y - a)
If and
So + +
(1 - o) a + oy
81 are esti mated for an industry using a suitable statisticalSo procedure giving and 81 an estimate of y is then available The sumSo
Intra-industry analysis
1974 of Business Re ort -
categories are identified
Page 19
of So and S1 is y so y bull So + S1 is the estimate of y
If o is assumed to be bounded by zero and one the sign of 61 is also
the sign of y - a The estimate of So is of cou se an estimate of
(1 - o ) a + oy Moreove if y gt a then a is less than 6o- (1 - B) a+ 0(
Cor esponding results would
hold if y bull a or y lt a
If 61 gt 0 then we would estimate that a lt 6o
bull
For purposes of estimating this model 32 manufacturing industry categories
were selected The data are fram the Line of Business forms filed with
the Federal Trade Commission they are discussed in detail in the Annual Line
which is available from the FTC The 32 industry
in Table 1
Simple regressions for equ ation (37) were run for each of the 32 industries
with data for each of the firms which filed in the industry constituting an ob-
servation The results are given in Table 2 lines lA 2A
1974
the intercept
For all the A
equations generalized least quares was used th both and market
share bein scaled bv the square root of sales
The 32 industries were taken from the larger set of all manufacturing industry
categories in the 1974 data files Two selection cr iteria were used thereLB
had to be at least ten reporting companies and the industry had to be a consumer
goods industry or a producer goods industry which is strongly associated with a
related consumer goods industry so that adve rtising by the firms in the producer
goods industry impacts on demand in the related industry (eg soft drink syrup
and soft drinks) There were 135 industries with at least ten companies - 103
producer goods and 32 consumer goods
273
274
TABLE 1 INDUSTRY CATEGORIES USED
FTC
COOE OESCRI P T I ON _
2001 MEAT PACKINGt SAUSAGE S ANO OTHER PREPARED
MEAT PROOUCTS
2004 DAIRY PRODUCTS EXC F LUID MILK
2005 CANNED SPECIALTIES
2001 FROZEN SPECIALTIES
OEHYORATFOt AND
I NC LU 0 ING PRE SERVE S
SOUP MIXESbull
SEASO lNGSt ANO
OTHER PET FOOJ
RAIN MILL PROOUCTSt
ANO PREPARED
RELATED PRODUCTS
PRODUCTS
Z008 CANNEOt ORIEOt PICKLED FRUITS
ANO VEGET ABLES JAMS JELLIES DEHYDRATED I G TA3t E
S40CES ANO SALAD ORESSINGS
2010 DOG CAT A O
2012 FlOUR t OTHER RICE
MILLING fiLENOE) FLOUR
2014 B EADt CAKE ANO
20113 CO FECTIONERY
2026 BOTTLED A O CANNED SOFT DRINKS 2021 FLAVORING EXTRACTS AND SYRUPS EC
2029 MISC FOOOS AND KINDRED PROJUCTS EXC
ROASTED COFFEE
2301 MENS AND BOYS SUITS AND COATS
2302 MENS ANO BOYS FURNISHINGS
230 3 kOMENS AND MISSES OUTER EAR
2551 HOUSEHOLD FURNITURE
2702 PERIODICALS
Pqe 19a 1
RELATED 1972
src cooe
2011 f 3
202tX2026
2032
203pound
2033t4t5
2047 2041t4t5
2051 2065 2086
2081
209XZ05
231
232
233
251 272
2703 BOOKS
2704 ISC PVeLISHING
2806 ORGANIC FlEERS
2807 DRUG$ ETHICAL
23234
PT 2 S 3
2808 ORUCSt PROPRIETARY
2809 PERFU ES t COSMETICS ANO OTHER TaiLET
PREPARATIONS
2810 SOAP AND OTHER CLEANING PREPARATIONS 2815
2901
PEST1CIDES AND AGR ICULTURAL
PE OLEUM REF I N ING CHEMICALS EC
3606 HOUSEHOLD COOKING EQUIPMENT
3612 HOUSEHOLD APPLIANC ES NEC INCLUJING ELECshy
rqlC HOUSE ARFS AND FANS ANO SE I G MACHINES
3617 RADIO AND T V RECE I VING SETS 3808 PHOTOG APHIC E UIP amp SUP LfESt EXC
3qo3
PHO TOCOPYING
SPOR TING A D E QU I P M ENT f SUP P L I E S ATHLETTC GOODS NEC
3904 DOLLS GAMESt TOYS AND CHLOqENtS VE ICLES
PT283
2844
234tX2844
287 9
291 3631
3634t6t9
36H
PT386l
3949
15
943
middot
l A6l f 2 R(GRESSION RESULTS
EQ I NO
NR CODE cu 121
lA 2001 A (
2A 2004 r (
3A 2005 a
-R
ClS amp31
16
17
1 z
I TER-CEPT bull t
ooo9t21At 00022311 oooeos uu
OOl49tCAI 0010 2 middot 000960CCI
o049tl1At o04t55C61
MARKET SH ARE bull 5 I
-010670Ctl
-017600181
-001050
-0211AO
-0077 0
AOVER-TIS INti
lbl
000()39 000072
000018 0 0024liAI
-000079
SHARf
4ampgtVf Q bull
111
000140
-000170
RSQ
COl
66 485 862
607 6 3 t ()2
blti6 633
flASTI-
CI T Y PH
-007271(1
000441
-u0288l
clASTSHU (t)Ef
bull 10
tto
-234
210
( OOlltOCCI -126520 000922 181 001790
4A 200 7 18 00590liAI -038670 4e16 -032110 lell u c
OOOS74e 000700 025940
00067018) 0021001AI -02l100CAI
5 26
920
5A 6 c
200 8 26 0 03Sl9CA I 0012 26181 o 024511(
-005540
-0 81420 0001961AI O()Ol90CAI ooll240
441 604 891
-0020 19 274
6A 2010 16 Oe04224CAI o359lOIAI 9oq ll4015dCAI 061 6 -middot 005435CAI 00011 8 1AI 917 c 00535ltAI -009850((1 OOOA51J CAI -001610161 971
7A 2012 001862 000010 351 00187) oo1 a oooq45 ono1o6 439 c 001109 -027050 Oe0067SIAI -OOll60C81 897
- middotmiddot- -
8A 201t ll 000757 Ol4140CCI 814 ott9oocet 095 8 oooqo6 C C J 000069(81 811 c ootl2lampBI -080910CCI Oe009lUBI -OOtlOOCCI 960I
I DIIJIT pound11 1V OTIUTVTfAll-1pound1t- A _ 00 R - qq r - Q
1jl)OQCD
0
-middot middot--
l
_ __ _
I
i
0 00141 c c
91)7
I
T ABLE 2 REG ESSIOH RESULTS CCONTI
EO NR
IND CODE
iR bull
cos
I TEq-
CEPT
t1ARKE T
SHA E
AOVER-
Tl SING
SHARpound-cent
AuVER
RSO ILAST1-
C I T Y ELAST
SH CPtf (lt Cl) bull 11 I 41 I 5I (61 171 I 8 I f9) c 10)
9A 2 0 1 8 19 002397181 -007640 367 -005 246 le46 6 000605 000550CAI 597
-000029 G14180 OOl654CAJ -024li01At
lOA 2026 12 0028901AI -OlZl90 -0114 0 1 1 1 5 759
Ll (
llA 2021
f3 c
12A lOZ9
B
(
13A 2301
B
c
14A 2302
B
c
154 201
n (
16A 2551
8
10
35
10
20
19
24
001897(()
00141 6CCI
Oll9lHCI
010604 middot middot
Oll083
OOJ11JCAt
002059141
001411CIH
-000412
000293(81
000187
001159 (3)
ooo6a4cct
o00792CCI
OOl40dCAt
000006
-000076
OOlltl6fA)
o o0 5 64CCt
-034570
-o1middot o3o
-1659 90
Ol51lO
-046190CAt
Ol5450CA)
-005l60
002940
-030060(()
-0 43l50CCt
0 33650
-019080
00004
bullJ00741CAI
-000069
000466
0001 51CAt
000947CAt
O OOI)70CAt OOl750CA J
000175CCI
000671CAI
OOll5) f A t 002207CBI
0004b7CA I
- 007 1 70
001760
-001410CAt
- 0 16200CAt
-004210
- 1 207 70
705
96 0
542
466 805
460
614
921
832
9 61
996
424
550
9 15
56 1 651
884
-006114
019 037
03504lCA)
oo409Q
-141644((
491 -0 17664
607
871
lell
084
101
072
103
108 11IllOQfl)-- l
-037880fCJ OOl6l4CAJ -040l70fCI 0()
c
I
I
bull 9
9t
middot middot middotbullmiddot bull lraquo
TABLE 2 REGRE SION RESUlTS ICONTt
bull
EO I NO tHl I filER- f4A_K T AOVER- sbulluRE ttSQ flA Jl- iLAS T NR CODE cos CEPT SHARE JISING AOVfR bull (JTY SHR CiJCF
110Cl I f amp Itt 15J tbt bull 1c u caa
llA 2702 ll o079 6 81 Bt -019110 S6s -OlltO 111
a 004679 ooooo1 t69
( oOl26o -015160 001560 -022690 ltJr()I
ldA 270 3 19 0065lt -OIt2l60 220 -Ol57a2 11 8 e o 0 71t6IA t ol08 75tAt ( OOl983CiU -060llOC() OOl7191A t -O lltdOC A) middot 89
lqA 2704 10 005082 -Oil060 22b -l)i1977 lelO 6 -000080 oo2772C4 t 910 c 000907 -011 200 Oe04105CA t -046960 995
20A 2d06 1 II
O OlltStiAt -001120 837 00014 -))th O01380amp At -oooo 11 821 ( OOl526CAt -026700181 0002 5UAt 000530 969
2lA 2807 28 OOt9lt9CAt -00891 0 516 -0 03962 225I8 002095CCt OOOl lliAt 690 ( o02965CA t -056650((1 OOOt50CAJ -003l70ttl 911I
I 22A 2808 15 ot3l06 119730 655 1130) 090
0 - 007526CC t Oe003961At 835- middot-
c I oo7523 -067210 001l 2tA t -0091t80CCt 951
2lA 2809 21 Oe2067lCAt -1218101At 71 2 -10095((1 121 ttlIaOQ8 00797418) OOOI8lltt 686 Ill( 0092lU A t -Olltl70Ct t 000605CAt -OOit8701At 951 _
---- -
32 009506CAt
0 bull 06 981 I A J 006901 cA bull
00 24A 2810 -010960 5t 5 -OOltSO 7 56
8 -000011 s11 c -5811701At oooaoatAJ oot520IAI 888
-- 37 --- middot - -- -1 ) l lbull t5JJ r-l bullhmiddot1 --- - --
l5A
26A
27A
c
2RA
c
2 9A
31A
llA
24
13
597
Ci)55
14
l 6h
RSQ ELASTI- (lASTCOE f-
TABLE 2
EQ I NO NR CODE IU
REGRESSION RESULTS CCONTa
NR INTER- MARKET
(10S CEPT SHARE AOVER- SHARfcent
TIS lNG AOVER
) CITY StiR
c It) bull 5 co a I 1) bull 0 bull 9 c l 0 bull( 2 1
2815 16 003019 -009990 275 -o069bmiddotJ
6 001446 000224 282 c 002178 -054630 o02221CCI -0075 00 646
2901 29 00021t2 CAI 0)0520 592 0007 6 1 068
0 00105C8 t 000012141 60)g( OOOlJ51AJ -00126018) ooou271At -000u80CA) 974
36oa 1 3 003296(4) -oca29o -oo-q94
001755 000203 lt91t 002499((1 -04l820CCI OOl374CAI 0068)0 946
3612 0033281CI 011520 195 01684) oso
00 30 42C OI 00012 2 424 oo579ltAt -2 7l7801At 002 100IAI -003220 819
8 3617 btel 011789 081002259 009530
001448 oOOl OitCCI 732 OOl7 22CAI -0 46160CAI 0004961AI -OOll40181 989c
30A 3808 14 0021t 3 6fAI -000110 628 o023291C) -005 6 middot- _ middot- 0bull 02 04 1 IC I ooooo9 637 ( 001584((1 -0161)1)0 00021 91AI -0 00060
390) 14 002790CCt 006590 h8 t 0 0938 4 070
B OOl89bCCJ 000144 74
0021961(1 -OItlOlO OOl323CAI -004800 915
B
middot--- - middot 4 - middot- - dfll3904 12 004563181 0)8300 879 0428651(1 089 OQIll004 277CA I ooozstcea 919
c 0041)3181 -0-11250 0011881(1 -006270 972 0Ill
- J - bull bull bull bullbullbull middot - - middot - middot gt middotbull L)ofbullflbull middot middotfJltl -tn- middotmiddotmiddotmiddoto +f middot t middot bull
c
is
positive
2
Page 0 determineThe consumer goodproducer good split is sometimes difficult to
Sollle support for the split used in this paper is a comparison of goodness - of - fit
distributions of a
industries
set
advertising are
that industry
a sales increase of
the
highest aD10U3 those
measures for the two groups of industries Table 3 contains
measures for the 32- consumer goods industries and the 103 producer goods
2 Median R is 5 7 for the consumer goods set and 40 for the producer goods
Estimates - of y the elasticity of demand with respect to
given in column of Table 2 The hypothesis that y bull 0 with
alternative was applied for each ittdustry As shown in the
could not be rtjected for 24 of the 32 industries Of the
and 10 are positive For the remalnifg eight three are
and five are significantly positive
Of the threertegative elasetcities the largest (in absolute value)
cosmetics (code 2809) With all elasticity of -1 01 a one
industry advertising would gwerate a decrease in industry sales
more than one percetu
At the other extrema is toys and games (code 39 04) For
a one percent ittcrease in industry advertising would lead to
four-tenths of a pampreentmiddot Though not significantly different from zero
elasticity estimatbull for pt-bpuetary drugs (code 2808)
which are
Both drug industries have
to sales rafios (211 pe1 cent for 28 08 and 13 4 per cent
other hand flaIJo ing ext1ac s (mainly soft-drink SYUlS) also
is for
percent increase in
of slightly
a middottwo - tailed
table that hypothesis
24 14 are negative
significantly negative
9
the codetics and p-ropriacuy very high advertising
for 28 09) bull On the
has a high ind UY
advenisiltg t() sales ratio ( 8 0 per cent) bull but its estimated elasticity is very
small and not signifieantly different uom zero
middot It doe$ not seem unusual to find many elasticities near zero with a few
negative and a few positive If aggregate consumption is insensitive to aggregate
advertising as seems likely ld then what one industry gains must be lost by
others
Page 21
elasticity of demand does
(10) we can
if t 0 then
Finally a determination concerning the industry
not fix the individual firms elasticity Going back to equation
write y bull (1 - ) a + z y Since a gt 0 it follows thatii i i
Y gt 0 If Ylt 0 however y s sign is indeterminateij ij
Turning next to the coefficient of market share 20 of them are negative and
12 are positive with three of each sign being significantly different from zero
The largest positive value which is significant is for pet food (code 2010) and
a close second is mens amp boys suits and coats (code 23 01) The largest signifi
cant negative value is for cosmetics (code 28 09)
In the model some relations among y a 0 and 61 are implied If middota is
negative (1- y) is greater than one Since 61Y bull (1- a) (1 - ) 61Y is
pCsitive but it will be less than one if 0 is large enough Seventeen of the A
ys are negative in each of the seventeen industries 61 is also negative and les
1 (algebraicly) than y so 61Y is greater than one his is not evidence that the
degree of cooperation is low or zero in any industry of course A high o togethegt
with a negative y which is small (in absolute terms) relative to a can give a
value of 61Y greater than one On the other hand if 61Y were less than one
o would have to be greater than zero
For y gt 0 two conditions may hold If y lt a it follows that 61 lt 0 since
(y - a ) lt 0 It also follows that 61Y lt 0 regardless of the value of o bull
In addition the relation is an if-and-only-if one ie if 61Y is negative
then y lt a Three industries have negative values for 61y dairy products
except milk (code 20 04) synthetic fibers (code 2806) and photographic equipment
(code 3808) In all three cases y is quite small so a does not have to be very
large to give a negative value to al
If y gt a the model shows that 61-- gt 0 whatever the value of 0 bull There are
12 cases where both y and 61 are positive with 61 ranging from near zero
(grain milling code 20 12) to 1 20 ( proprietary drugs code 2808) Several
coeffiei-en_rs
--
15
o f to zero
Tab le 3 - Consumer Goods
Indus tr ies
2Distributions of R and Producer Goods
Page 20a
Consumer Producer
Goods Goods
0 - 9 9 0 3
10 - 19 9 0 1 1
20 - 29 9 3 20
30 - 39 9 3 17
40 - 49 9 5 16
50 - 59 9 8
69 9 6 Ii 1060
--f 670 79 9 2
80 - 89 9 4 4
90 - 100 1 1
Totals 32 10 3
Notes
2is used R i s cal-
culated as F F + ((N - K- 1 ) K ] 1 b_ F is the F
statistic Lor the hypothesis that all o f the non - intercept terms are simul taneously equal
1
2
--
-- Since generalized least squ-area regress ion
Page 12
o f these cases have values o f 6 1 Y rea ter than 0 8 (pet food (20 10) bread amp cakes
(20 14 ) miscellaneous processed foods (20 29 ) men s and boys suit s and coats
(23 01 ) p rop riet ary drugs (28 08 ) elect ric housewares (36 1 2) radio and
IV sets 17 and games and toys (39 For be than(36 ) 04)) 6 1 Y to greater 0 8
both o and cry mus t b e less than 0 2 That is in these industries the results
shown in Tab le 2 would hold only with very strong dominance of marke t demand
effects over brand swit ching e ffects and a very low degree o f cooperation
Using the concentration rat ios given in Table 4 the average concentration
ratio for the eight indust ries is 38 1 For the remaining 24 industries it is
43 2 This result would lend some support t o the proposition that concentrat ion
and cooperation are related if the group of eight industries are identi fied
as having low degrees of cooperation
One f inal note on model predict ions S ince both cr and o are assumed to
be non-negative 60 should also be non-negat ive In only one o f the 32 cas es
is 60 negative and it is not signif icant bull
cor pJ
7 1
7 0 4
1 4 7 5 1
)
7 3 z a o a t s 1 1 3 3 5 ( i V Z 8 tJ1 2 1 - 1 1 0 7 3 5 c middot v 2 8 1 0 3 2 -0 0 1 4 3 0 C 1 V 9 8 3 2 8 1 5 1 6 - 0 0 6 t1 6 9 C _ V 1 11 0 2 9 0 1 29 0 00 7 6 1 C l k V 7 3 1 3 6 08 1 3 - o n 4 9 9 4 3 3 3 6 1 2 2 4 0 1 6 4 3
3 1 7 1 3 0 1 1 7 d 7 8 3 3 8 0 8 0 0 2 3 i 2 3 0 6
P 4
3 9 0 3 1 4
1 2
0 09 3 8 4 39 1)1 0 4 l 8 6 5
T A LE 4 e I NCU S T R Y C A T E G O R Y
F TC NOM e tOOE ass
C 1 I c 2 )
2 0 0 1 1 6 2 0 0r 1 7 zo o l Z 2 0 0 7 1 8
z o o 2 6
z o t a 1 6
2 1) 1 2 1 5 2 0 1 4 1 1 2 0 1 a 1 9 2 0 26 1 2 2 0 2 7 1 0 2 0 29 - 3 5
2 3 0 1 10
2 3 02 z o
2 3 03 1 9
2 5 5 1 2 4 Z 1 0 Z 1 1 2 0 3 1 9 2 7 4 1 0 z a o 1 3
2 A 0 7 z s
E LA S T I -C I T Y
c 3 )
- 0 09 7 2 7
0 00 449
- 0 0 2 8 9 3
- 0 3 2 7 70 - 0 OZ il I q
o 4 u t s a 0 0 1 3 7 3
o 1 4 o o-0 0 5 2 4o -0 1 94 0 1
-C l 6 l l 4 0 1 JJ 3 7 0 3 t 0 4 2 0 0 4 0 9
- 0 4 1 8 44 - O l 7 t ) 4 -ll 7 1 4 0 b -0 3 5 7 8 Z
- 0 1 9 7 7 0 1)0 3 3 4
-o n 9 2
IJ A T A
CONV
ON-C ONV
( 4 )
t 1NV
C 1t-V C NV c ) v C - JNV C rlNV c n v c a v C ON V ( i) l V
C CtV
C t1tV C IJNV
C 1 V
c 1 v
Pa ge 2La
HEQ F t Noe x
c 5 )
6 2q
6 89 1 48
6 1 4 ) q 3
1 9 7 1 9
q z t s 5 a e z
l l 9 1
4 o 7 1 2 0
c oqt 1 Z
4 4 )-q 1 z
5 5 9 1 0 64 1 3 3
n 3
Inter industry analysi
Page 23
Equation ( 3 6 ) may be used to fo rmulate a rough test o f the cooperat ion
vs e f ficiency controversy I f we rest ricted ourselves to some subset o f
indus tries where the relat ion be tween y and a i s the same we would be lookingbull
at a sample o f indus tries for which the dependence o f V on H and o is a
quadratic wi th no squared terms
I f we now sub s titute (26) into ( 36 ) we get
(38) v bull a + (y - a) R + (y - a) o (H) - (y - a) H o (H) + u
If it is only d if ferences in marke t shares that affect advertising inten-
s i ty levels i f concentration does not af fect cooperation and i f o bull 0
( 38 ) reduces t o
( 39 ) V bull a + (y - a) H + U
Let this b e the null hyp o thes is and note that V is a linear func-
tion o f H If y gt a and the null hyp othesis holds the expected relat ion
between V and H is as shown by the pos itively sloped straight line in Figure
6 b etween points A and C
In this context the alterna ive hypothesis the degree o f
cooperation ( o) is a positive function o f concentrat ion
is that
(H) I f the func-
tion is as described in (26) and Figure 4 the relation between V and H
would be as shown by the curved line connec t ing po ints A and C in Figure 6 One way to test the null hypothe s is against this al ternative is to test fo r
the hypothesized curvature in the function relating V and H
Fi gure
R
6
v
ID c
H 1 0
I
i I I
sfigned
clasJ s
in def ini g
effec t is the
I
Page middot2lf
One charac teristic of produc ts which may be useful
classes wi th respec t to significanc e of the brand-switching
conveniencenon-convenienc e distinc tion developed by Porter ( 1 97 )
Using his identif ication of IRS indus tries as a s tar t ing point
each of the 32 industries used in this s tudy to one of the two
The ass ignments are not ed in Tab le 4
Using the conveniencenon-convenienc e dis t inc tion the
divided into two sub-samples
sample was
Four regressions were then run bull ene for e ch
sub-sample one for the to tal sample and one for the total sample with
a dummy variable for convenience goods Each regression was run wi th LB s
and wi th indus tries as ob servations The results are given in Tab le 5
For the LB level equations ( 1A - 1D) equation(37) was used ome
rearranging of terms gives v bull y - ( 1-o ) (y - a) ( 1-z ) + ui i i
bull y - 6 1 ( 1 - z ) + ui When observations are pooled across indus trif s yi
is an industry variable In these regressions the values o f y es t imated
in the first stage were used Since industries have been grouped so as
to reduce differences in (y - a) that term is t reated as a cons tant
within each group Finally H the Herfindahl index has been subst ituted
for o the degree of cooperation The resultant equation is linear
y ( 1 - z ) and H x ( l - z ) This is the equation for whi ch results arei i
reported in Table 5 In all regress ions all variables were weighted by the
square roo t of sales
in
raquo
- 34
3 1
i l T 4 8LE S R E G E S S I ON R E SUL T S i X T f N S I ON S T O T Y PE Of
AND L E V E L OF C ONC E kT R A T I O
E O O tt S
tP bull T Y P E NR C ONV O B S tO N-CONV
E L A S T I shy
T I C J T Y
O E - t-41( T SliA R E
H( fl F
1 1 - K O x middot K O X E R F
- H S a 1 1 - H S t 1 8 1 1 9 1
R S O
1 1 0 1
q z
t - bull 9 f 2 1
Z t P D l 6 fC ONV - 1 9 1 s s c 2 1
e - l O I e bull u q a
2 3 1 0 1 1 Nl N- t O V 5 7 1 ( 1 2 19 1 6 1 - J l S 2 5 4
l l l l t z q o a 1 - J o v z I
l ( I NO l l A l l ol u l t o o c bull J u l l
d S
c 3 4 t t r a l
bull Z D 1 0 l l 4 l l 2 3 -gt 2 l l 2 0 - 4 lS 5 4 l C 4 1 o e 6 t l b b 5 1 1 0 1 1 1 c - 1 9 5 1 t d S 2 1 tV
OQ10 T-R A T I O S A E G I VE N l P A RE N T H E S E S l E V E L S f S I G 1 F l C ANC E A - 99 5 B - q 9 t - s NS J UC E NO I Nl E R C ( P T I S U (j f middot) I N f Hf t OUA TIOI S 6 E f 3R E A O J U S T H E N T F OR 1-t E T E R O SIC E O A S T f C I T V pIraquo
1
middot
I
R SO I S C A L C UL A T E D A S f I f bull 1 1 bull 1lt 1 K t t- E E N I S THf U A1 B E R O f fJe S E fi V A T I O S F t S HtE F - S T A T I S
I N DE P E ND E N T y II
T I C F O T H E U Y P O HIE S I S f A l l C JE F F I C H N T A R E QU A L HI l f RO A f40 K I S T H f UUM 8 E R OF Q ( A fL f S
I l l
1 A
1 6
t C
f 1 r R I l l
1 2 1
L Jt
L B
l q
L 1
on s T Y P E bull 2 )
P OUC T
X
H S I l l
l O l A I bull l 0 7 1
O O b O J
1 2 00 1
1 b 2 1 C I l l Z U
l l S b 1 -1 1 0 1
rlt F X l l - fi C f l
bull 1 I
l l
- ()4 2 3 l f H
1 - 2 7 9 1
K O X
1 1 - ti E R F I
l d l
S4 l i A I 2 l b
1 3 00
K O X -t f R F t S J
I 1 - HER F t
)
60 9
z o bull
0
t l l
3 8 8
1 7 6
5 64
64
N f bull
tl S bull
I l l
I 4 1
t ONV
NON-t ONV
A l l
A ll
c o v NO- ( ONV
1 4 1
1 5 1
- l S b
c - 1 1 2 1
5 8 2 C A I 1 6 7 0 1
middot I l 9 H
l 1 il
1 4 1 6 1
E L A C) T J shy
l t C I T Y
I S I
l b l
- 0 2 0
1 - 1 4 9 1
O Z 3 2 1 A I C b 6 b l
0 0 9 2 -J
1 1 b 4 1
o z t t t l 1 2 2 1 1
oru - Ht tt F
1 gt X bull b I
Page 25 bull
Mean val -as ior relevan t variables for the two sub-samples are
Conv Non-Conv
Adv Sales 4 7 2 3
iit Share 2 7 3 2
Ela cicity - 05 3 02 2
Herf indahl 7 3 1 860
That the dist nct ion between coft9enienee goods and non - convenience
goods mat ters is supported not only by these means but also by the regression
results Equat ions lA and lB are different in almost every respec t Only the
concentrat ionmarket share int eract ion term seems to mat ter for convenience
good s Jus t he opposite holds for non - convenience goods both elasticity1
and market sl e are highly significant but the concentrat ionmarket share L
int erac t ion t=rm is not
Given that one minus market share is used these equ3 tions the coeffici t
of market share is the negative of what is sh in col (6) of Table 5 For the
regress ion for non - convenience LB s reported in equation lB then LB s with larger
market shares have lower advertising to sales ratios
On the question of the impact of the degree of cooperation the story is mixed
For convenienc good s the coefficient is pos itive as equation (31) predicts when
there is a po sitive degree of cooperat ion Fo r the non-convenience goods LB s middot
concentration has no effec t
Aggregat ion t o the indus try level and then rer unning the equations $ives the
results shown in (2a-2d) of Tab le 5 As expected R2 goes up and levels of significanc e
fmiddotJr individual coefficients gJ down In add ition the coeffic ient of H ( l -H) inmiddot
quat ion 2A is no t significant though it is po s i tive Since 2A is the indus try level
I
--
all but one of them and that one is not significant For 18 of the 31
wo uld have on its adve rtising to sal es ratio
cognate of LA the variable H ( l-H) pl ays the same role in 2A as H ( l -z 1) p lays
in LA
Vconomies of scale in advert ising
The literature on advert ing contains a number of definitionbull o f economies
of scale and of findings eone rning its presence and magnitude Before I loo k
a t the data with the hope o f mmenting about scale effect s then I wan t to
define the term as I use it
By economies of scale I mean a decrease in the number o f units of adver-
tising which are needed to generate a unit of sales as advert ising is increased
Given the definition of the firm s own elast icity of demand with respect to
advertising (see the text foll wing equation (1) ) the technical definition
is that rii gt 1 Diseconomies of scale are def ined symmetrically i e
Yu lt 1
ith a cro s s - section oi observations on advertising intensity
(v bull a s1) and onmiddot advertising (a ) we may regress the former on the latter 1 i i
The results of such regress ions for the 32 industries are given in Table 2
lines lB ZB bull bull bull 32B For the 32 eases the regression coefficient is positive it
positive coefficients the coefficient i s signif icant
The que tion now is whether these results ar e evidence of pervasive dis-
economies of scale in advertislng I think no t Given my as sumption that the
data I observe are equilibrium results I may ri ghtly conclud e that almost every-
where high levels of advert ising are as sociated in equilibrium with high ratios
of advertising to sales I may no t conclud e anything however about the
impact that a move by some firm away from its equilibrium level o f advertising
middot- Page 27
Given the assumed relation in equation ( 37 ) there is a corre spondshy
ing equilibrium equation relating v and ai The equation is non - linear i
and I have not found a simp le way to characterize it It is po ssib le to
det ermi ne its deriviative with respect to a b y takin$ tne de riviative of ( 3 7 ) i
When that is done the result is
v - 1(40) i bull sl a a z ( 1-z )i i ira-l
Sinc e all the other terms in (40) are non-negative the sign dep ends only
on the sign of 6 1 which is in turn de pend ent only on the sign of (y - a)
That the signs of the coefficients of a in equations lB 2B bull bull bull 32B arei
not always equal to the signs of the coefficients of z in LA 2A bull bull bull 32A i
is still a mystery to me and something to be explored
bull
Summary
My purpose in writing this pa per wa s t o attempt to integrate explicit micro-
economic mode lling reasoned econome tric specification of error terms and high
quality data to exp lore some questions about adver tising by large firms
al quite important improvements could be made in the first two of those areas
and much data ma ssag ing is yet to be done in the third
Concerning the subst an ce of what I have done I think six things are im-
Page 2 8
Severshy
V and Conclusions
portant
1 Explicit modeling is worth the ef for t if for no other reas on than
that it provides a basis for choosing fr om a varie ty of funct ional
forms
2 The same is true for error specif ica t ion I did in fact look at scatshy
ter diagrams for all 32 industry categor ies and they vir tually all show
the het eroscedast icity which I as sumed and for which I correc ted
3 The predicted relation be tween advertis ing in tensity and market share
shows up clearly in on ly 25 of the case s examined I have ye t to
explore why that may b e the cas e
4 Some evidence concerning the presence and impact of coopera tion was
produced but it is not clear ly pervasive
5 The dis tinction between convenience goods and non-c onvenie nce goods is
unambiguously a good one
6 Virtually no evidence concerning econ omies of scale in adver tis ing can
be gleaned fr om this study given its assumpt ions
More work is called for on many of these issues my study seems to have
raised more questions than it has answer ed This is pr obab lly due to the richshy
ne ss of the data source since I had the oppor tunity to address several issue s
at once
- -- - Q --
Foo tnotes
Manager Line o f Business Program Federal Trade Commission The author depended heavily on several staf f members of the LB Pro gram for statistical and clerical support in particular Joe Cholka George Pascoe and Haro lene Jenkins Helpful comments were given b y Richard Caves Dennis Mueller Michael Porter and F M Scherer When the paper was in its early stages o f development both S teve Garber and Jon Rasmussen former economists in the LB Program provided eery useful cri tiques The views expressed here are my own o f course
1 The basic exp licit model of the role of advertising is Dorfman and S teiner ( 1954 ) and the models in this paper may be seen as ano ther extens ion of their work
2 For a review o f models of oligopoly which treat prices and quantities as relevant variables see Shubik ( 1 959) Extens ions to take advertis ing int o account are fairly s traightforward
3 Dorfman and S teiner dealt with a price-setting firm For a monopolis t inverting the demand relation to show Pbull P (Q A) _instead o f Q- (P A) has no effect on the results o f the analysis For oligopolists who face a sys tem of demand relations inversion does have some impac t on the results The dif ferences are not trivial nonetheless they will no t be explored in this study
4 Shubik ( 1 959) p
5 From a formal mathematical point o f view the function is similar to a1 i Legrangian funct ion of the form - r rr
L bull 11 + A ( j - (i i i+j
where the indicates some fixed level of the profits industry Following the same logic which resul ts in and income constraint as the marginal utility of the of the indus tr in the obj ective of firm i
6 Hall amp Weiss ( 1 967) p 323
7 See Schmallensee ( 1 97 2 ) Ch 2 pp 1 6-47
8 On this poin t see Imel amp Helmberger ( 1 9 7 1 )
9 Long ( 1 970) p 130 - 1 39
1 0 See Schmallensee ( 1 972) Ch 3 pp 48-8 7
rrj i+j ) of the rest of the
the identification profits of the rest
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980
bull bull
Page 12
tion and the extent or degree of coo peration as shown in Figure 3 Using
o as a symbol for the degree of cooperation eq uation (23) can be rewritten
as
bull(24) v (1 - o) vo + ovl bull (1 o) [a+ - a) H ] + oy- (y
vhere
(25) o bull o(H) bull 0 if B lt B
Hbull 1 if B gt
The specific relation between concentration (H) and the deg ree of
cooperation ( o) shown in Figure 3 is more restrictive than necessary and
bas less appeal an one which shows a smooth increase of o from 0 to 1 as
B increases fro to 1 Such a smooth function is shown in Figure 4 he
degree of cooper ion is low until H gets close to H increases greatly as
B goes through R attaining a high level for an H larger than B but still
close to it and then goes to 1 as B goes to 1 More formally
(26) 0 bull o(H)
bull 0 if H bull 0
bull 1 if H bull 1
do gt (IdB
d2o- gt 0 if H B dBZ
0 if H H
lt 0 if R gt H-
If (26) is now substituted for (25) in (24) a corresponding smoothing
of the functions shown in Figure 2 is accomplished he smooth functions
Figure 3
I
10
H 10 H
Page 12a
Figure
Page 12b
4
10
H 10 H
l
(27)
Cl - o) o
[ (1 -
[ (1
t y ij z
j
are given in Figure 5
Page 13
The equation in (24) was constructed as a hybrid of two results derived
from first-order equations for the non-cooperative and cooperative models
It may also be constructed directly by using l bull (1 - o) + on as thei i
th 2objective function for the i-- firm Treating o as a constant the first
order equation is
0 an aa
i middot
n i
Using equations (3) and (7) and setting --- bull 0 givesaai
(28) - 0
After collecting terms and converting to matrix notation we get
(29) - Cl - o) + 0 z-1vi yii i
N
j bulll
+ 0
+ o v1
d-1 r z z
-(30) v Cl - o) rd
-
o) r +d
d r J _ -1- d z
and middot
d) r + 0 rjd
l r z
v - z --(31) zv
- (1 0- +d) t r z- d-
If we now impose the simplified demand structure by substituting (11)
we get
- (1 - o) a +( 32) + oy and
Figure
l
-y
v
Y
H
r
I I
v
a
y
H 10 H
v
a
bulla
HH 10
bullbull
(33)
i
Page 14
v - (1 - o) [ a l + (y - a) ] The corresponding equation for Vis given in (24)
If the relation between industry advertising intensity (V) and
concentration (H) is non - linear via the impact of concentration
on the degree of cooperation does the relation take the form of a
auadratic as some other investigators have proposed (Greer ( 1971 ) Martin (1979) Strickland amp Weiss (1976) ) That question can
be approached from two perspectives - conceptual and empirical In
the next section I will try to apply some statistical material to it
here I want to explore the question in the context of the models
developed above
Referring to equation (24) taking the deriviative with respect
to H and letting bull (H) gives
bull( 3 3a ) 3V (y - a) (1 - o) + (1 - H) 3H dH
Given that o and H each are bounded by 0 and 1 and that o is an in-
creasing function of H the term in the brackets in non-negative The
sign of the deriviative then is the same as the sign of (y - a) If the industry effect is dominant (y gt a) then advertising in-
tensity will increase with concentration If the two effects are equal
then advertising intensity will not vbullry with concentration and if the
brand-switching e ffect is larger advertising intensity will decrease
with concentration
For given advertising elasticity parameters then it is not pos-
sible to show an increase in A S up to some H and the n a decrease
he relation may be non - linear but it is monotonic over the permis-
sible range of H
Page 15
One possibility is to introduce a relation between (y - a) and H
If industries characterized as dominated by the brand-switching effect
also were more concentrated then a cross industry comparison would
show an inverted U shape relation between AS and H I know of no
support for such an association between ( y - a) and H
V ERROR SPECUCATION
So far I have implicitly assumed that there are no errors in the
model I now assume that the error
thvariable in the equation for the i- firm is additive that its mean is zero
that its variance is inversely proportionate to its sales and that all error
term covarlances are zero hat is
34) v bull (1 - o) [ a + (y - a) z ] + oy + u i i i
- - -
In vector notation
(35) bull (1 - o) [ at + lt - a) J +
E() bull 0 Cov (u) -
0
If we now ca lculate V bull bull we get
oy + U bull
(36) V bull (1 - o) la + (y - a)R J +
U bull E(U) bull 0 Var (U) bull
P age 16
The error term specifi c ation which follows Hall
wi th the ratio of profits
6 heir analysis seems
as wellbull particularly given
and Weiss was
applied by them to equa tions to equity or
asse ts as dependen t variables to carry the same
weight in the present context the substantial
evidence presented by marketing analysis on the tendency of company de-
cision makers t o use the advertising to sales ratio as the decision
7variable
The zero covariance assumption on the other hand1 is particularly
troublesome Within a probably not true since
many even ts outside the
single industry it is
industry would have
industries
similar effec ts on all the
8firms in the industry Between it is probably not true either
since many companies produce in more than one indus try and even ts in the
firm would tend to effect all of its activities The extension of the
model to allow for non-zero covariances is certainly called for
--- -
Page Ll
Athird characteristic of this specification also deserves some comment
only on equation is shown for the firm or industry Several haveinvestigators
included an advertising equation ina larger system of equations (Comanor and
Wilson (1974) (1979) and Weiss (1976) so in myMartin Strickland Idid
dissertation1 Ihave ewo defenses for presenting only single equation results
here One is the suggestion by Comanor Wilson that simultaneousand equation
verybias may not be important in this context he second is that Iam will-
ing to see the model extended to include a profitability equation and I intend
to move that direction in the near futurein
Onthe inclusion of aconcentration equation on the other hand I amskep-
tical he problem isthat in some fundamental sense equations in models of
industrial organization should be oriented to the decision-making contexts of firms
Ifeach firm in anindustry with Nfirms has only one decision variable then
there can be only Nindependent equations which are based on the first-order
optimization equations It not show large ofis difficult of course to a number
functional relations which follow from the first-order conditions point here he
isthat only Nof them can be independent
Consider for illustrative purposes only a situation in wich teh N
itsquantities say q are set exogeneously and each firm determines advertising1
of and level say ai where a is the true optimal value a a bulla
1 1 i i + i
Ifp a) then the Nq
s and the Na s will determine N p s say pii bullp
i(l i 1 i
Now given Eand 9 we can determine both the vector s andthe vector v since
- bull til
ands We can also determine S s z bull s-1 s-ii
v and we
industry concentration isHbull (alternatively
V and Hare functions of the same
pendent Furthermore if there are
VM) and (H1 H2 bull bull bull )contain
VI STATISTI CAL APPLICATIONS
Page 18
Given the observed values of might be tempted to conclude that we
have a two equation system in which v and z are jointly determined That would i i
be inappropriate however since the 2N random variables (v ) are determined by
onlt N basic random variables t The variance-covariance matrix for ( ) must
be singular bull
The extension of this observation to industry level variables is straightshy
forward Since observed industry advertising intensity is V bull l v and observed 4
C4 bull z ) it follows thati l i
set of random variables t and cannot be indeshy-
M industries then the variables (V1 v2 at most M independent random variables
Given the assumptions made above y a and o are constant within a
given industry That being the case (34) can be written as
(37)
s1 bull (1 - o) (y - a)
If and
So + +
(1 - o) a + oy
81 are esti mated for an industry using a suitable statisticalSo procedure giving and 81 an estimate of y is then available The sumSo
Intra-industry analysis
1974 of Business Re ort -
categories are identified
Page 19
of So and S1 is y so y bull So + S1 is the estimate of y
If o is assumed to be bounded by zero and one the sign of 61 is also
the sign of y - a The estimate of So is of cou se an estimate of
(1 - o ) a + oy Moreove if y gt a then a is less than 6o- (1 - B) a+ 0(
Cor esponding results would
hold if y bull a or y lt a
If 61 gt 0 then we would estimate that a lt 6o
bull
For purposes of estimating this model 32 manufacturing industry categories
were selected The data are fram the Line of Business forms filed with
the Federal Trade Commission they are discussed in detail in the Annual Line
which is available from the FTC The 32 industry
in Table 1
Simple regressions for equ ation (37) were run for each of the 32 industries
with data for each of the firms which filed in the industry constituting an ob-
servation The results are given in Table 2 lines lA 2A
1974
the intercept
For all the A
equations generalized least quares was used th both and market
share bein scaled bv the square root of sales
The 32 industries were taken from the larger set of all manufacturing industry
categories in the 1974 data files Two selection cr iteria were used thereLB
had to be at least ten reporting companies and the industry had to be a consumer
goods industry or a producer goods industry which is strongly associated with a
related consumer goods industry so that adve rtising by the firms in the producer
goods industry impacts on demand in the related industry (eg soft drink syrup
and soft drinks) There were 135 industries with at least ten companies - 103
producer goods and 32 consumer goods
273
274
TABLE 1 INDUSTRY CATEGORIES USED
FTC
COOE OESCRI P T I ON _
2001 MEAT PACKINGt SAUSAGE S ANO OTHER PREPARED
MEAT PROOUCTS
2004 DAIRY PRODUCTS EXC F LUID MILK
2005 CANNED SPECIALTIES
2001 FROZEN SPECIALTIES
OEHYORATFOt AND
I NC LU 0 ING PRE SERVE S
SOUP MIXESbull
SEASO lNGSt ANO
OTHER PET FOOJ
RAIN MILL PROOUCTSt
ANO PREPARED
RELATED PRODUCTS
PRODUCTS
Z008 CANNEOt ORIEOt PICKLED FRUITS
ANO VEGET ABLES JAMS JELLIES DEHYDRATED I G TA3t E
S40CES ANO SALAD ORESSINGS
2010 DOG CAT A O
2012 FlOUR t OTHER RICE
MILLING fiLENOE) FLOUR
2014 B EADt CAKE ANO
20113 CO FECTIONERY
2026 BOTTLED A O CANNED SOFT DRINKS 2021 FLAVORING EXTRACTS AND SYRUPS EC
2029 MISC FOOOS AND KINDRED PROJUCTS EXC
ROASTED COFFEE
2301 MENS AND BOYS SUITS AND COATS
2302 MENS ANO BOYS FURNISHINGS
230 3 kOMENS AND MISSES OUTER EAR
2551 HOUSEHOLD FURNITURE
2702 PERIODICALS
Pqe 19a 1
RELATED 1972
src cooe
2011 f 3
202tX2026
2032
203pound
2033t4t5
2047 2041t4t5
2051 2065 2086
2081
209XZ05
231
232
233
251 272
2703 BOOKS
2704 ISC PVeLISHING
2806 ORGANIC FlEERS
2807 DRUG$ ETHICAL
23234
PT 2 S 3
2808 ORUCSt PROPRIETARY
2809 PERFU ES t COSMETICS ANO OTHER TaiLET
PREPARATIONS
2810 SOAP AND OTHER CLEANING PREPARATIONS 2815
2901
PEST1CIDES AND AGR ICULTURAL
PE OLEUM REF I N ING CHEMICALS EC
3606 HOUSEHOLD COOKING EQUIPMENT
3612 HOUSEHOLD APPLIANC ES NEC INCLUJING ELECshy
rqlC HOUSE ARFS AND FANS ANO SE I G MACHINES
3617 RADIO AND T V RECE I VING SETS 3808 PHOTOG APHIC E UIP amp SUP LfESt EXC
3qo3
PHO TOCOPYING
SPOR TING A D E QU I P M ENT f SUP P L I E S ATHLETTC GOODS NEC
3904 DOLLS GAMESt TOYS AND CHLOqENtS VE ICLES
PT283
2844
234tX2844
287 9
291 3631
3634t6t9
36H
PT386l
3949
15
943
middot
l A6l f 2 R(GRESSION RESULTS
EQ I NO
NR CODE cu 121
lA 2001 A (
2A 2004 r (
3A 2005 a
-R
ClS amp31
16
17
1 z
I TER-CEPT bull t
ooo9t21At 00022311 oooeos uu
OOl49tCAI 0010 2 middot 000960CCI
o049tl1At o04t55C61
MARKET SH ARE bull 5 I
-010670Ctl
-017600181
-001050
-0211AO
-0077 0
AOVER-TIS INti
lbl
000()39 000072
000018 0 0024liAI
-000079
SHARf
4ampgtVf Q bull
111
000140
-000170
RSQ
COl
66 485 862
607 6 3 t ()2
blti6 633
flASTI-
CI T Y PH
-007271(1
000441
-u0288l
clASTSHU (t)Ef
bull 10
tto
-234
210
( OOlltOCCI -126520 000922 181 001790
4A 200 7 18 00590liAI -038670 4e16 -032110 lell u c
OOOS74e 000700 025940
00067018) 0021001AI -02l100CAI
5 26
920
5A 6 c
200 8 26 0 03Sl9CA I 0012 26181 o 024511(
-005540
-0 81420 0001961AI O()Ol90CAI ooll240
441 604 891
-0020 19 274
6A 2010 16 Oe04224CAI o359lOIAI 9oq ll4015dCAI 061 6 -middot 005435CAI 00011 8 1AI 917 c 00535ltAI -009850((1 OOOA51J CAI -001610161 971
7A 2012 001862 000010 351 00187) oo1 a oooq45 ono1o6 439 c 001109 -027050 Oe0067SIAI -OOll60C81 897
- middotmiddot- -
8A 201t ll 000757 Ol4140CCI 814 ott9oocet 095 8 oooqo6 C C J 000069(81 811 c ootl2lampBI -080910CCI Oe009lUBI -OOtlOOCCI 960I
I DIIJIT pound11 1V OTIUTVTfAll-1pound1t- A _ 00 R - qq r - Q
1jl)OQCD
0
-middot middot--
l
_ __ _
I
i
0 00141 c c
91)7
I
T ABLE 2 REG ESSIOH RESULTS CCONTI
EO NR
IND CODE
iR bull
cos
I TEq-
CEPT
t1ARKE T
SHA E
AOVER-
Tl SING
SHARpound-cent
AuVER
RSO ILAST1-
C I T Y ELAST
SH CPtf (lt Cl) bull 11 I 41 I 5I (61 171 I 8 I f9) c 10)
9A 2 0 1 8 19 002397181 -007640 367 -005 246 le46 6 000605 000550CAI 597
-000029 G14180 OOl654CAJ -024li01At
lOA 2026 12 0028901AI -OlZl90 -0114 0 1 1 1 5 759
Ll (
llA 2021
f3 c
12A lOZ9
B
(
13A 2301
B
c
14A 2302
B
c
154 201
n (
16A 2551
8
10
35
10
20
19
24
001897(()
00141 6CCI
Oll9lHCI
010604 middot middot
Oll083
OOJ11JCAt
002059141
001411CIH
-000412
000293(81
000187
001159 (3)
ooo6a4cct
o00792CCI
OOl40dCAt
000006
-000076
OOlltl6fA)
o o0 5 64CCt
-034570
-o1middot o3o
-1659 90
Ol51lO
-046190CAt
Ol5450CA)
-005l60
002940
-030060(()
-0 43l50CCt
0 33650
-019080
00004
bullJ00741CAI
-000069
000466
0001 51CAt
000947CAt
O OOI)70CAt OOl750CA J
000175CCI
000671CAI
OOll5) f A t 002207CBI
0004b7CA I
- 007 1 70
001760
-001410CAt
- 0 16200CAt
-004210
- 1 207 70
705
96 0
542
466 805
460
614
921
832
9 61
996
424
550
9 15
56 1 651
884
-006114
019 037
03504lCA)
oo409Q
-141644((
491 -0 17664
607
871
lell
084
101
072
103
108 11IllOQfl)-- l
-037880fCJ OOl6l4CAJ -040l70fCI 0()
c
I
I
bull 9
9t
middot middot middotbullmiddot bull lraquo
TABLE 2 REGRE SION RESUlTS ICONTt
bull
EO I NO tHl I filER- f4A_K T AOVER- sbulluRE ttSQ flA Jl- iLAS T NR CODE cos CEPT SHARE JISING AOVfR bull (JTY SHR CiJCF
110Cl I f amp Itt 15J tbt bull 1c u caa
llA 2702 ll o079 6 81 Bt -019110 S6s -OlltO 111
a 004679 ooooo1 t69
( oOl26o -015160 001560 -022690 ltJr()I
ldA 270 3 19 0065lt -OIt2l60 220 -Ol57a2 11 8 e o 0 71t6IA t ol08 75tAt ( OOl983CiU -060llOC() OOl7191A t -O lltdOC A) middot 89
lqA 2704 10 005082 -Oil060 22b -l)i1977 lelO 6 -000080 oo2772C4 t 910 c 000907 -011 200 Oe04105CA t -046960 995
20A 2d06 1 II
O OlltStiAt -001120 837 00014 -))th O01380amp At -oooo 11 821 ( OOl526CAt -026700181 0002 5UAt 000530 969
2lA 2807 28 OOt9lt9CAt -00891 0 516 -0 03962 225I8 002095CCt OOOl lliAt 690 ( o02965CA t -056650((1 OOOt50CAJ -003l70ttl 911I
I 22A 2808 15 ot3l06 119730 655 1130) 090
0 - 007526CC t Oe003961At 835- middot-
c I oo7523 -067210 001l 2tA t -0091t80CCt 951
2lA 2809 21 Oe2067lCAt -1218101At 71 2 -10095((1 121 ttlIaOQ8 00797418) OOOI8lltt 686 Ill( 0092lU A t -Olltl70Ct t 000605CAt -OOit8701At 951 _
---- -
32 009506CAt
0 bull 06 981 I A J 006901 cA bull
00 24A 2810 -010960 5t 5 -OOltSO 7 56
8 -000011 s11 c -5811701At oooaoatAJ oot520IAI 888
-- 37 --- middot - -- -1 ) l lbull t5JJ r-l bullhmiddot1 --- - --
l5A
26A
27A
c
2RA
c
2 9A
31A
llA
24
13
597
Ci)55
14
l 6h
RSQ ELASTI- (lASTCOE f-
TABLE 2
EQ I NO NR CODE IU
REGRESSION RESULTS CCONTa
NR INTER- MARKET
(10S CEPT SHARE AOVER- SHARfcent
TIS lNG AOVER
) CITY StiR
c It) bull 5 co a I 1) bull 0 bull 9 c l 0 bull( 2 1
2815 16 003019 -009990 275 -o069bmiddotJ
6 001446 000224 282 c 002178 -054630 o02221CCI -0075 00 646
2901 29 00021t2 CAI 0)0520 592 0007 6 1 068
0 00105C8 t 000012141 60)g( OOOlJ51AJ -00126018) ooou271At -000u80CA) 974
36oa 1 3 003296(4) -oca29o -oo-q94
001755 000203 lt91t 002499((1 -04l820CCI OOl374CAI 0068)0 946
3612 0033281CI 011520 195 01684) oso
00 30 42C OI 00012 2 424 oo579ltAt -2 7l7801At 002 100IAI -003220 819
8 3617 btel 011789 081002259 009530
001448 oOOl OitCCI 732 OOl7 22CAI -0 46160CAI 0004961AI -OOll40181 989c
30A 3808 14 0021t 3 6fAI -000110 628 o023291C) -005 6 middot- _ middot- 0bull 02 04 1 IC I ooooo9 637 ( 001584((1 -0161)1)0 00021 91AI -0 00060
390) 14 002790CCt 006590 h8 t 0 0938 4 070
B OOl89bCCJ 000144 74
0021961(1 -OItlOlO OOl323CAI -004800 915
B
middot--- - middot 4 - middot- - dfll3904 12 004563181 0)8300 879 0428651(1 089 OQIll004 277CA I ooozstcea 919
c 0041)3181 -0-11250 0011881(1 -006270 972 0Ill
- J - bull bull bull bullbullbull middot - - middot - middot gt middotbull L)ofbullflbull middot middotfJltl -tn- middotmiddotmiddotmiddoto +f middot t middot bull
c
is
positive
2
Page 0 determineThe consumer goodproducer good split is sometimes difficult to
Sollle support for the split used in this paper is a comparison of goodness - of - fit
distributions of a
industries
set
advertising are
that industry
a sales increase of
the
highest aD10U3 those
measures for the two groups of industries Table 3 contains
measures for the 32- consumer goods industries and the 103 producer goods
2 Median R is 5 7 for the consumer goods set and 40 for the producer goods
Estimates - of y the elasticity of demand with respect to
given in column of Table 2 The hypothesis that y bull 0 with
alternative was applied for each ittdustry As shown in the
could not be rtjected for 24 of the 32 industries Of the
and 10 are positive For the remalnifg eight three are
and five are significantly positive
Of the threertegative elasetcities the largest (in absolute value)
cosmetics (code 2809) With all elasticity of -1 01 a one
industry advertising would gwerate a decrease in industry sales
more than one percetu
At the other extrema is toys and games (code 39 04) For
a one percent ittcrease in industry advertising would lead to
four-tenths of a pampreentmiddot Though not significantly different from zero
elasticity estimatbull for pt-bpuetary drugs (code 2808)
which are
Both drug industries have
to sales rafios (211 pe1 cent for 28 08 and 13 4 per cent
other hand flaIJo ing ext1ac s (mainly soft-drink SYUlS) also
is for
percent increase in
of slightly
a middottwo - tailed
table that hypothesis
24 14 are negative
significantly negative
9
the codetics and p-ropriacuy very high advertising
for 28 09) bull On the
has a high ind UY
advenisiltg t() sales ratio ( 8 0 per cent) bull but its estimated elasticity is very
small and not signifieantly different uom zero
middot It doe$ not seem unusual to find many elasticities near zero with a few
negative and a few positive If aggregate consumption is insensitive to aggregate
advertising as seems likely ld then what one industry gains must be lost by
others
Page 21
elasticity of demand does
(10) we can
if t 0 then
Finally a determination concerning the industry
not fix the individual firms elasticity Going back to equation
write y bull (1 - ) a + z y Since a gt 0 it follows thatii i i
Y gt 0 If Ylt 0 however y s sign is indeterminateij ij
Turning next to the coefficient of market share 20 of them are negative and
12 are positive with three of each sign being significantly different from zero
The largest positive value which is significant is for pet food (code 2010) and
a close second is mens amp boys suits and coats (code 23 01) The largest signifi
cant negative value is for cosmetics (code 28 09)
In the model some relations among y a 0 and 61 are implied If middota is
negative (1- y) is greater than one Since 61Y bull (1- a) (1 - ) 61Y is
pCsitive but it will be less than one if 0 is large enough Seventeen of the A
ys are negative in each of the seventeen industries 61 is also negative and les
1 (algebraicly) than y so 61Y is greater than one his is not evidence that the
degree of cooperation is low or zero in any industry of course A high o togethegt
with a negative y which is small (in absolute terms) relative to a can give a
value of 61Y greater than one On the other hand if 61Y were less than one
o would have to be greater than zero
For y gt 0 two conditions may hold If y lt a it follows that 61 lt 0 since
(y - a ) lt 0 It also follows that 61Y lt 0 regardless of the value of o bull
In addition the relation is an if-and-only-if one ie if 61Y is negative
then y lt a Three industries have negative values for 61y dairy products
except milk (code 20 04) synthetic fibers (code 2806) and photographic equipment
(code 3808) In all three cases y is quite small so a does not have to be very
large to give a negative value to al
If y gt a the model shows that 61-- gt 0 whatever the value of 0 bull There are
12 cases where both y and 61 are positive with 61 ranging from near zero
(grain milling code 20 12) to 1 20 ( proprietary drugs code 2808) Several
coeffiei-en_rs
--
15
o f to zero
Tab le 3 - Consumer Goods
Indus tr ies
2Distributions of R and Producer Goods
Page 20a
Consumer Producer
Goods Goods
0 - 9 9 0 3
10 - 19 9 0 1 1
20 - 29 9 3 20
30 - 39 9 3 17
40 - 49 9 5 16
50 - 59 9 8
69 9 6 Ii 1060
--f 670 79 9 2
80 - 89 9 4 4
90 - 100 1 1
Totals 32 10 3
Notes
2is used R i s cal-
culated as F F + ((N - K- 1 ) K ] 1 b_ F is the F
statistic Lor the hypothesis that all o f the non - intercept terms are simul taneously equal
1
2
--
-- Since generalized least squ-area regress ion
Page 12
o f these cases have values o f 6 1 Y rea ter than 0 8 (pet food (20 10) bread amp cakes
(20 14 ) miscellaneous processed foods (20 29 ) men s and boys suit s and coats
(23 01 ) p rop riet ary drugs (28 08 ) elect ric housewares (36 1 2) radio and
IV sets 17 and games and toys (39 For be than(36 ) 04)) 6 1 Y to greater 0 8
both o and cry mus t b e less than 0 2 That is in these industries the results
shown in Tab le 2 would hold only with very strong dominance of marke t demand
effects over brand swit ching e ffects and a very low degree o f cooperation
Using the concentration rat ios given in Table 4 the average concentration
ratio for the eight indust ries is 38 1 For the remaining 24 industries it is
43 2 This result would lend some support t o the proposition that concentrat ion
and cooperation are related if the group of eight industries are identi fied
as having low degrees of cooperation
One f inal note on model predict ions S ince both cr and o are assumed to
be non-negative 60 should also be non-negat ive In only one o f the 32 cas es
is 60 negative and it is not signif icant bull
cor pJ
7 1
7 0 4
1 4 7 5 1
)
7 3 z a o a t s 1 1 3 3 5 ( i V Z 8 tJ1 2 1 - 1 1 0 7 3 5 c middot v 2 8 1 0 3 2 -0 0 1 4 3 0 C 1 V 9 8 3 2 8 1 5 1 6 - 0 0 6 t1 6 9 C _ V 1 11 0 2 9 0 1 29 0 00 7 6 1 C l k V 7 3 1 3 6 08 1 3 - o n 4 9 9 4 3 3 3 6 1 2 2 4 0 1 6 4 3
3 1 7 1 3 0 1 1 7 d 7 8 3 3 8 0 8 0 0 2 3 i 2 3 0 6
P 4
3 9 0 3 1 4
1 2
0 09 3 8 4 39 1)1 0 4 l 8 6 5
T A LE 4 e I NCU S T R Y C A T E G O R Y
F TC NOM e tOOE ass
C 1 I c 2 )
2 0 0 1 1 6 2 0 0r 1 7 zo o l Z 2 0 0 7 1 8
z o o 2 6
z o t a 1 6
2 1) 1 2 1 5 2 0 1 4 1 1 2 0 1 a 1 9 2 0 26 1 2 2 0 2 7 1 0 2 0 29 - 3 5
2 3 0 1 10
2 3 02 z o
2 3 03 1 9
2 5 5 1 2 4 Z 1 0 Z 1 1 2 0 3 1 9 2 7 4 1 0 z a o 1 3
2 A 0 7 z s
E LA S T I -C I T Y
c 3 )
- 0 09 7 2 7
0 00 449
- 0 0 2 8 9 3
- 0 3 2 7 70 - 0 OZ il I q
o 4 u t s a 0 0 1 3 7 3
o 1 4 o o-0 0 5 2 4o -0 1 94 0 1
-C l 6 l l 4 0 1 JJ 3 7 0 3 t 0 4 2 0 0 4 0 9
- 0 4 1 8 44 - O l 7 t ) 4 -ll 7 1 4 0 b -0 3 5 7 8 Z
- 0 1 9 7 7 0 1)0 3 3 4
-o n 9 2
IJ A T A
CONV
ON-C ONV
( 4 )
t 1NV
C 1t-V C NV c ) v C - JNV C rlNV c n v c a v C ON V ( i) l V
C CtV
C t1tV C IJNV
C 1 V
c 1 v
Pa ge 2La
HEQ F t Noe x
c 5 )
6 2q
6 89 1 48
6 1 4 ) q 3
1 9 7 1 9
q z t s 5 a e z
l l 9 1
4 o 7 1 2 0
c oqt 1 Z
4 4 )-q 1 z
5 5 9 1 0 64 1 3 3
n 3
Inter industry analysi
Page 23
Equation ( 3 6 ) may be used to fo rmulate a rough test o f the cooperat ion
vs e f ficiency controversy I f we rest ricted ourselves to some subset o f
indus tries where the relat ion be tween y and a i s the same we would be lookingbull
at a sample o f indus tries for which the dependence o f V on H and o is a
quadratic wi th no squared terms
I f we now sub s titute (26) into ( 36 ) we get
(38) v bull a + (y - a) R + (y - a) o (H) - (y - a) H o (H) + u
If it is only d if ferences in marke t shares that affect advertising inten-
s i ty levels i f concentration does not af fect cooperation and i f o bull 0
( 38 ) reduces t o
( 39 ) V bull a + (y - a) H + U
Let this b e the null hyp o thes is and note that V is a linear func-
tion o f H If y gt a and the null hyp othesis holds the expected relat ion
between V and H is as shown by the pos itively sloped straight line in Figure
6 b etween points A and C
In this context the alterna ive hypothesis the degree o f
cooperation ( o) is a positive function o f concentrat ion
is that
(H) I f the func-
tion is as described in (26) and Figure 4 the relation between V and H
would be as shown by the curved line connec t ing po ints A and C in Figure 6 One way to test the null hypothe s is against this al ternative is to test fo r
the hypothesized curvature in the function relating V and H
Fi gure
R
6
v
ID c
H 1 0
I
i I I
sfigned
clasJ s
in def ini g
effec t is the
I
Page middot2lf
One charac teristic of produc ts which may be useful
classes wi th respec t to significanc e of the brand-switching
conveniencenon-convenienc e distinc tion developed by Porter ( 1 97 )
Using his identif ication of IRS indus tries as a s tar t ing point
each of the 32 industries used in this s tudy to one of the two
The ass ignments are not ed in Tab le 4
Using the conveniencenon-convenienc e dis t inc tion the
divided into two sub-samples
sample was
Four regressions were then run bull ene for e ch
sub-sample one for the to tal sample and one for the total sample with
a dummy variable for convenience goods Each regression was run wi th LB s
and wi th indus tries as ob servations The results are given in Tab le 5
For the LB level equations ( 1A - 1D) equation(37) was used ome
rearranging of terms gives v bull y - ( 1-o ) (y - a) ( 1-z ) + ui i i
bull y - 6 1 ( 1 - z ) + ui When observations are pooled across indus trif s yi
is an industry variable In these regressions the values o f y es t imated
in the first stage were used Since industries have been grouped so as
to reduce differences in (y - a) that term is t reated as a cons tant
within each group Finally H the Herfindahl index has been subst ituted
for o the degree of cooperation The resultant equation is linear
y ( 1 - z ) and H x ( l - z ) This is the equation for whi ch results arei i
reported in Table 5 In all regress ions all variables were weighted by the
square roo t of sales
in
raquo
- 34
3 1
i l T 4 8LE S R E G E S S I ON R E SUL T S i X T f N S I ON S T O T Y PE Of
AND L E V E L OF C ONC E kT R A T I O
E O O tt S
tP bull T Y P E NR C ONV O B S tO N-CONV
E L A S T I shy
T I C J T Y
O E - t-41( T SliA R E
H( fl F
1 1 - K O x middot K O X E R F
- H S a 1 1 - H S t 1 8 1 1 9 1
R S O
1 1 0 1
q z
t - bull 9 f 2 1
Z t P D l 6 fC ONV - 1 9 1 s s c 2 1
e - l O I e bull u q a
2 3 1 0 1 1 Nl N- t O V 5 7 1 ( 1 2 19 1 6 1 - J l S 2 5 4
l l l l t z q o a 1 - J o v z I
l ( I NO l l A l l ol u l t o o c bull J u l l
d S
c 3 4 t t r a l
bull Z D 1 0 l l 4 l l 2 3 -gt 2 l l 2 0 - 4 lS 5 4 l C 4 1 o e 6 t l b b 5 1 1 0 1 1 1 c - 1 9 5 1 t d S 2 1 tV
OQ10 T-R A T I O S A E G I VE N l P A RE N T H E S E S l E V E L S f S I G 1 F l C ANC E A - 99 5 B - q 9 t - s NS J UC E NO I Nl E R C ( P T I S U (j f middot) I N f Hf t OUA TIOI S 6 E f 3R E A O J U S T H E N T F OR 1-t E T E R O SIC E O A S T f C I T V pIraquo
1
middot
I
R SO I S C A L C UL A T E D A S f I f bull 1 1 bull 1lt 1 K t t- E E N I S THf U A1 B E R O f fJe S E fi V A T I O S F t S HtE F - S T A T I S
I N DE P E ND E N T y II
T I C F O T H E U Y P O HIE S I S f A l l C JE F F I C H N T A R E QU A L HI l f RO A f40 K I S T H f UUM 8 E R OF Q ( A fL f S
I l l
1 A
1 6
t C
f 1 r R I l l
1 2 1
L Jt
L B
l q
L 1
on s T Y P E bull 2 )
P OUC T
X
H S I l l
l O l A I bull l 0 7 1
O O b O J
1 2 00 1
1 b 2 1 C I l l Z U
l l S b 1 -1 1 0 1
rlt F X l l - fi C f l
bull 1 I
l l
- ()4 2 3 l f H
1 - 2 7 9 1
K O X
1 1 - ti E R F I
l d l
S4 l i A I 2 l b
1 3 00
K O X -t f R F t S J
I 1 - HER F t
)
60 9
z o bull
0
t l l
3 8 8
1 7 6
5 64
64
N f bull
tl S bull
I l l
I 4 1
t ONV
NON-t ONV
A l l
A ll
c o v NO- ( ONV
1 4 1
1 5 1
- l S b
c - 1 1 2 1
5 8 2 C A I 1 6 7 0 1
middot I l 9 H
l 1 il
1 4 1 6 1
E L A C) T J shy
l t C I T Y
I S I
l b l
- 0 2 0
1 - 1 4 9 1
O Z 3 2 1 A I C b 6 b l
0 0 9 2 -J
1 1 b 4 1
o z t t t l 1 2 2 1 1
oru - Ht tt F
1 gt X bull b I
Page 25 bull
Mean val -as ior relevan t variables for the two sub-samples are
Conv Non-Conv
Adv Sales 4 7 2 3
iit Share 2 7 3 2
Ela cicity - 05 3 02 2
Herf indahl 7 3 1 860
That the dist nct ion between coft9enienee goods and non - convenience
goods mat ters is supported not only by these means but also by the regression
results Equat ions lA and lB are different in almost every respec t Only the
concentrat ionmarket share int eract ion term seems to mat ter for convenience
good s Jus t he opposite holds for non - convenience goods both elasticity1
and market sl e are highly significant but the concentrat ionmarket share L
int erac t ion t=rm is not
Given that one minus market share is used these equ3 tions the coeffici t
of market share is the negative of what is sh in col (6) of Table 5 For the
regress ion for non - convenience LB s reported in equation lB then LB s with larger
market shares have lower advertising to sales ratios
On the question of the impact of the degree of cooperation the story is mixed
For convenienc good s the coefficient is pos itive as equation (31) predicts when
there is a po sitive degree of cooperat ion Fo r the non-convenience goods LB s middot
concentration has no effec t
Aggregat ion t o the indus try level and then rer unning the equations $ives the
results shown in (2a-2d) of Tab le 5 As expected R2 goes up and levels of significanc e
fmiddotJr individual coefficients gJ down In add ition the coeffic ient of H ( l -H) inmiddot
quat ion 2A is no t significant though it is po s i tive Since 2A is the indus try level
I
--
all but one of them and that one is not significant For 18 of the 31
wo uld have on its adve rtising to sal es ratio
cognate of LA the variable H ( l-H) pl ays the same role in 2A as H ( l -z 1) p lays
in LA
Vconomies of scale in advert ising
The literature on advert ing contains a number of definitionbull o f economies
of scale and of findings eone rning its presence and magnitude Before I loo k
a t the data with the hope o f mmenting about scale effect s then I wan t to
define the term as I use it
By economies of scale I mean a decrease in the number o f units of adver-
tising which are needed to generate a unit of sales as advert ising is increased
Given the definition of the firm s own elast icity of demand with respect to
advertising (see the text foll wing equation (1) ) the technical definition
is that rii gt 1 Diseconomies of scale are def ined symmetrically i e
Yu lt 1
ith a cro s s - section oi observations on advertising intensity
(v bull a s1) and onmiddot advertising (a ) we may regress the former on the latter 1 i i
The results of such regress ions for the 32 industries are given in Table 2
lines lB ZB bull bull bull 32B For the 32 eases the regression coefficient is positive it
positive coefficients the coefficient i s signif icant
The que tion now is whether these results ar e evidence of pervasive dis-
economies of scale in advertislng I think no t Given my as sumption that the
data I observe are equilibrium results I may ri ghtly conclud e that almost every-
where high levels of advert ising are as sociated in equilibrium with high ratios
of advertising to sales I may no t conclud e anything however about the
impact that a move by some firm away from its equilibrium level o f advertising
middot- Page 27
Given the assumed relation in equation ( 37 ) there is a corre spondshy
ing equilibrium equation relating v and ai The equation is non - linear i
and I have not found a simp le way to characterize it It is po ssib le to
det ermi ne its deriviative with respect to a b y takin$ tne de riviative of ( 3 7 ) i
When that is done the result is
v - 1(40) i bull sl a a z ( 1-z )i i ira-l
Sinc e all the other terms in (40) are non-negative the sign dep ends only
on the sign of 6 1 which is in turn de pend ent only on the sign of (y - a)
That the signs of the coefficients of a in equations lB 2B bull bull bull 32B arei
not always equal to the signs of the coefficients of z in LA 2A bull bull bull 32A i
is still a mystery to me and something to be explored
bull
Summary
My purpose in writing this pa per wa s t o attempt to integrate explicit micro-
economic mode lling reasoned econome tric specification of error terms and high
quality data to exp lore some questions about adver tising by large firms
al quite important improvements could be made in the first two of those areas
and much data ma ssag ing is yet to be done in the third
Concerning the subst an ce of what I have done I think six things are im-
Page 2 8
Severshy
V and Conclusions
portant
1 Explicit modeling is worth the ef for t if for no other reas on than
that it provides a basis for choosing fr om a varie ty of funct ional
forms
2 The same is true for error specif ica t ion I did in fact look at scatshy
ter diagrams for all 32 industry categor ies and they vir tually all show
the het eroscedast icity which I as sumed and for which I correc ted
3 The predicted relation be tween advertis ing in tensity and market share
shows up clearly in on ly 25 of the case s examined I have ye t to
explore why that may b e the cas e
4 Some evidence concerning the presence and impact of coopera tion was
produced but it is not clear ly pervasive
5 The dis tinction between convenience goods and non-c onvenie nce goods is
unambiguously a good one
6 Virtually no evidence concerning econ omies of scale in adver tis ing can
be gleaned fr om this study given its assumpt ions
More work is called for on many of these issues my study seems to have
raised more questions than it has answer ed This is pr obab lly due to the richshy
ne ss of the data source since I had the oppor tunity to address several issue s
at once
- -- - Q --
Foo tnotes
Manager Line o f Business Program Federal Trade Commission The author depended heavily on several staf f members of the LB Pro gram for statistical and clerical support in particular Joe Cholka George Pascoe and Haro lene Jenkins Helpful comments were given b y Richard Caves Dennis Mueller Michael Porter and F M Scherer When the paper was in its early stages o f development both S teve Garber and Jon Rasmussen former economists in the LB Program provided eery useful cri tiques The views expressed here are my own o f course
1 The basic exp licit model of the role of advertising is Dorfman and S teiner ( 1954 ) and the models in this paper may be seen as ano ther extens ion of their work
2 For a review o f models of oligopoly which treat prices and quantities as relevant variables see Shubik ( 1 959) Extens ions to take advertis ing int o account are fairly s traightforward
3 Dorfman and S teiner dealt with a price-setting firm For a monopolis t inverting the demand relation to show Pbull P (Q A) _instead o f Q- (P A) has no effect on the results o f the analysis For oligopolists who face a sys tem of demand relations inversion does have some impac t on the results The dif ferences are not trivial nonetheless they will no t be explored in this study
4 Shubik ( 1 959) p
5 From a formal mathematical point o f view the function is similar to a1 i Legrangian funct ion of the form - r rr
L bull 11 + A ( j - (i i i+j
where the indicates some fixed level of the profits industry Following the same logic which resul ts in and income constraint as the marginal utility of the of the indus tr in the obj ective of firm i
6 Hall amp Weiss ( 1 967) p 323
7 See Schmallensee ( 1 97 2 ) Ch 2 pp 1 6-47
8 On this poin t see Imel amp Helmberger ( 1 9 7 1 )
9 Long ( 1 970) p 130 - 1 39
1 0 See Schmallensee ( 1 972) Ch 3 pp 48-8 7
rrj i+j ) of the rest of the
the identification profits of the rest
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980
Figure 3
I
10
H 10 H
Page 12a
Figure
Page 12b
4
10
H 10 H
l
(27)
Cl - o) o
[ (1 -
[ (1
t y ij z
j
are given in Figure 5
Page 13
The equation in (24) was constructed as a hybrid of two results derived
from first-order equations for the non-cooperative and cooperative models
It may also be constructed directly by using l bull (1 - o) + on as thei i
th 2objective function for the i-- firm Treating o as a constant the first
order equation is
0 an aa
i middot
n i
Using equations (3) and (7) and setting --- bull 0 givesaai
(28) - 0
After collecting terms and converting to matrix notation we get
(29) - Cl - o) + 0 z-1vi yii i
N
j bulll
+ 0
+ o v1
d-1 r z z
-(30) v Cl - o) rd
-
o) r +d
d r J _ -1- d z
and middot
d) r + 0 rjd
l r z
v - z --(31) zv
- (1 0- +d) t r z- d-
If we now impose the simplified demand structure by substituting (11)
we get
- (1 - o) a +( 32) + oy and
Figure
l
-y
v
Y
H
r
I I
v
a
y
H 10 H
v
a
bulla
HH 10
bullbull
(33)
i
Page 14
v - (1 - o) [ a l + (y - a) ] The corresponding equation for Vis given in (24)
If the relation between industry advertising intensity (V) and
concentration (H) is non - linear via the impact of concentration
on the degree of cooperation does the relation take the form of a
auadratic as some other investigators have proposed (Greer ( 1971 ) Martin (1979) Strickland amp Weiss (1976) ) That question can
be approached from two perspectives - conceptual and empirical In
the next section I will try to apply some statistical material to it
here I want to explore the question in the context of the models
developed above
Referring to equation (24) taking the deriviative with respect
to H and letting bull (H) gives
bull( 3 3a ) 3V (y - a) (1 - o) + (1 - H) 3H dH
Given that o and H each are bounded by 0 and 1 and that o is an in-
creasing function of H the term in the brackets in non-negative The
sign of the deriviative then is the same as the sign of (y - a) If the industry effect is dominant (y gt a) then advertising in-
tensity will increase with concentration If the two effects are equal
then advertising intensity will not vbullry with concentration and if the
brand-switching e ffect is larger advertising intensity will decrease
with concentration
For given advertising elasticity parameters then it is not pos-
sible to show an increase in A S up to some H and the n a decrease
he relation may be non - linear but it is monotonic over the permis-
sible range of H
Page 15
One possibility is to introduce a relation between (y - a) and H
If industries characterized as dominated by the brand-switching effect
also were more concentrated then a cross industry comparison would
show an inverted U shape relation between AS and H I know of no
support for such an association between ( y - a) and H
V ERROR SPECUCATION
So far I have implicitly assumed that there are no errors in the
model I now assume that the error
thvariable in the equation for the i- firm is additive that its mean is zero
that its variance is inversely proportionate to its sales and that all error
term covarlances are zero hat is
34) v bull (1 - o) [ a + (y - a) z ] + oy + u i i i
- - -
In vector notation
(35) bull (1 - o) [ at + lt - a) J +
E() bull 0 Cov (u) -
0
If we now ca lculate V bull bull we get
oy + U bull
(36) V bull (1 - o) la + (y - a)R J +
U bull E(U) bull 0 Var (U) bull
P age 16
The error term specifi c ation which follows Hall
wi th the ratio of profits
6 heir analysis seems
as wellbull particularly given
and Weiss was
applied by them to equa tions to equity or
asse ts as dependen t variables to carry the same
weight in the present context the substantial
evidence presented by marketing analysis on the tendency of company de-
cision makers t o use the advertising to sales ratio as the decision
7variable
The zero covariance assumption on the other hand1 is particularly
troublesome Within a probably not true since
many even ts outside the
single industry it is
industry would have
industries
similar effec ts on all the
8firms in the industry Between it is probably not true either
since many companies produce in more than one indus try and even ts in the
firm would tend to effect all of its activities The extension of the
model to allow for non-zero covariances is certainly called for
--- -
Page Ll
Athird characteristic of this specification also deserves some comment
only on equation is shown for the firm or industry Several haveinvestigators
included an advertising equation ina larger system of equations (Comanor and
Wilson (1974) (1979) and Weiss (1976) so in myMartin Strickland Idid
dissertation1 Ihave ewo defenses for presenting only single equation results
here One is the suggestion by Comanor Wilson that simultaneousand equation
verybias may not be important in this context he second is that Iam will-
ing to see the model extended to include a profitability equation and I intend
to move that direction in the near futurein
Onthe inclusion of aconcentration equation on the other hand I amskep-
tical he problem isthat in some fundamental sense equations in models of
industrial organization should be oriented to the decision-making contexts of firms
Ifeach firm in anindustry with Nfirms has only one decision variable then
there can be only Nindependent equations which are based on the first-order
optimization equations It not show large ofis difficult of course to a number
functional relations which follow from the first-order conditions point here he
isthat only Nof them can be independent
Consider for illustrative purposes only a situation in wich teh N
itsquantities say q are set exogeneously and each firm determines advertising1
of and level say ai where a is the true optimal value a a bulla
1 1 i i + i
Ifp a) then the Nq
s and the Na s will determine N p s say pii bullp
i(l i 1 i
Now given Eand 9 we can determine both the vector s andthe vector v since
- bull til
ands We can also determine S s z bull s-1 s-ii
v and we
industry concentration isHbull (alternatively
V and Hare functions of the same
pendent Furthermore if there are
VM) and (H1 H2 bull bull bull )contain
VI STATISTI CAL APPLICATIONS
Page 18
Given the observed values of might be tempted to conclude that we
have a two equation system in which v and z are jointly determined That would i i
be inappropriate however since the 2N random variables (v ) are determined by
onlt N basic random variables t The variance-covariance matrix for ( ) must
be singular bull
The extension of this observation to industry level variables is straightshy
forward Since observed industry advertising intensity is V bull l v and observed 4
C4 bull z ) it follows thati l i
set of random variables t and cannot be indeshy-
M industries then the variables (V1 v2 at most M independent random variables
Given the assumptions made above y a and o are constant within a
given industry That being the case (34) can be written as
(37)
s1 bull (1 - o) (y - a)
If and
So + +
(1 - o) a + oy
81 are esti mated for an industry using a suitable statisticalSo procedure giving and 81 an estimate of y is then available The sumSo
Intra-industry analysis
1974 of Business Re ort -
categories are identified
Page 19
of So and S1 is y so y bull So + S1 is the estimate of y
If o is assumed to be bounded by zero and one the sign of 61 is also
the sign of y - a The estimate of So is of cou se an estimate of
(1 - o ) a + oy Moreove if y gt a then a is less than 6o- (1 - B) a+ 0(
Cor esponding results would
hold if y bull a or y lt a
If 61 gt 0 then we would estimate that a lt 6o
bull
For purposes of estimating this model 32 manufacturing industry categories
were selected The data are fram the Line of Business forms filed with
the Federal Trade Commission they are discussed in detail in the Annual Line
which is available from the FTC The 32 industry
in Table 1
Simple regressions for equ ation (37) were run for each of the 32 industries
with data for each of the firms which filed in the industry constituting an ob-
servation The results are given in Table 2 lines lA 2A
1974
the intercept
For all the A
equations generalized least quares was used th both and market
share bein scaled bv the square root of sales
The 32 industries were taken from the larger set of all manufacturing industry
categories in the 1974 data files Two selection cr iteria were used thereLB
had to be at least ten reporting companies and the industry had to be a consumer
goods industry or a producer goods industry which is strongly associated with a
related consumer goods industry so that adve rtising by the firms in the producer
goods industry impacts on demand in the related industry (eg soft drink syrup
and soft drinks) There were 135 industries with at least ten companies - 103
producer goods and 32 consumer goods
273
274
TABLE 1 INDUSTRY CATEGORIES USED
FTC
COOE OESCRI P T I ON _
2001 MEAT PACKINGt SAUSAGE S ANO OTHER PREPARED
MEAT PROOUCTS
2004 DAIRY PRODUCTS EXC F LUID MILK
2005 CANNED SPECIALTIES
2001 FROZEN SPECIALTIES
OEHYORATFOt AND
I NC LU 0 ING PRE SERVE S
SOUP MIXESbull
SEASO lNGSt ANO
OTHER PET FOOJ
RAIN MILL PROOUCTSt
ANO PREPARED
RELATED PRODUCTS
PRODUCTS
Z008 CANNEOt ORIEOt PICKLED FRUITS
ANO VEGET ABLES JAMS JELLIES DEHYDRATED I G TA3t E
S40CES ANO SALAD ORESSINGS
2010 DOG CAT A O
2012 FlOUR t OTHER RICE
MILLING fiLENOE) FLOUR
2014 B EADt CAKE ANO
20113 CO FECTIONERY
2026 BOTTLED A O CANNED SOFT DRINKS 2021 FLAVORING EXTRACTS AND SYRUPS EC
2029 MISC FOOOS AND KINDRED PROJUCTS EXC
ROASTED COFFEE
2301 MENS AND BOYS SUITS AND COATS
2302 MENS ANO BOYS FURNISHINGS
230 3 kOMENS AND MISSES OUTER EAR
2551 HOUSEHOLD FURNITURE
2702 PERIODICALS
Pqe 19a 1
RELATED 1972
src cooe
2011 f 3
202tX2026
2032
203pound
2033t4t5
2047 2041t4t5
2051 2065 2086
2081
209XZ05
231
232
233
251 272
2703 BOOKS
2704 ISC PVeLISHING
2806 ORGANIC FlEERS
2807 DRUG$ ETHICAL
23234
PT 2 S 3
2808 ORUCSt PROPRIETARY
2809 PERFU ES t COSMETICS ANO OTHER TaiLET
PREPARATIONS
2810 SOAP AND OTHER CLEANING PREPARATIONS 2815
2901
PEST1CIDES AND AGR ICULTURAL
PE OLEUM REF I N ING CHEMICALS EC
3606 HOUSEHOLD COOKING EQUIPMENT
3612 HOUSEHOLD APPLIANC ES NEC INCLUJING ELECshy
rqlC HOUSE ARFS AND FANS ANO SE I G MACHINES
3617 RADIO AND T V RECE I VING SETS 3808 PHOTOG APHIC E UIP amp SUP LfESt EXC
3qo3
PHO TOCOPYING
SPOR TING A D E QU I P M ENT f SUP P L I E S ATHLETTC GOODS NEC
3904 DOLLS GAMESt TOYS AND CHLOqENtS VE ICLES
PT283
2844
234tX2844
287 9
291 3631
3634t6t9
36H
PT386l
3949
15
943
middot
l A6l f 2 R(GRESSION RESULTS
EQ I NO
NR CODE cu 121
lA 2001 A (
2A 2004 r (
3A 2005 a
-R
ClS amp31
16
17
1 z
I TER-CEPT bull t
ooo9t21At 00022311 oooeos uu
OOl49tCAI 0010 2 middot 000960CCI
o049tl1At o04t55C61
MARKET SH ARE bull 5 I
-010670Ctl
-017600181
-001050
-0211AO
-0077 0
AOVER-TIS INti
lbl
000()39 000072
000018 0 0024liAI
-000079
SHARf
4ampgtVf Q bull
111
000140
-000170
RSQ
COl
66 485 862
607 6 3 t ()2
blti6 633
flASTI-
CI T Y PH
-007271(1
000441
-u0288l
clASTSHU (t)Ef
bull 10
tto
-234
210
( OOlltOCCI -126520 000922 181 001790
4A 200 7 18 00590liAI -038670 4e16 -032110 lell u c
OOOS74e 000700 025940
00067018) 0021001AI -02l100CAI
5 26
920
5A 6 c
200 8 26 0 03Sl9CA I 0012 26181 o 024511(
-005540
-0 81420 0001961AI O()Ol90CAI ooll240
441 604 891
-0020 19 274
6A 2010 16 Oe04224CAI o359lOIAI 9oq ll4015dCAI 061 6 -middot 005435CAI 00011 8 1AI 917 c 00535ltAI -009850((1 OOOA51J CAI -001610161 971
7A 2012 001862 000010 351 00187) oo1 a oooq45 ono1o6 439 c 001109 -027050 Oe0067SIAI -OOll60C81 897
- middotmiddot- -
8A 201t ll 000757 Ol4140CCI 814 ott9oocet 095 8 oooqo6 C C J 000069(81 811 c ootl2lampBI -080910CCI Oe009lUBI -OOtlOOCCI 960I
I DIIJIT pound11 1V OTIUTVTfAll-1pound1t- A _ 00 R - qq r - Q
1jl)OQCD
0
-middot middot--
l
_ __ _
I
i
0 00141 c c
91)7
I
T ABLE 2 REG ESSIOH RESULTS CCONTI
EO NR
IND CODE
iR bull
cos
I TEq-
CEPT
t1ARKE T
SHA E
AOVER-
Tl SING
SHARpound-cent
AuVER
RSO ILAST1-
C I T Y ELAST
SH CPtf (lt Cl) bull 11 I 41 I 5I (61 171 I 8 I f9) c 10)
9A 2 0 1 8 19 002397181 -007640 367 -005 246 le46 6 000605 000550CAI 597
-000029 G14180 OOl654CAJ -024li01At
lOA 2026 12 0028901AI -OlZl90 -0114 0 1 1 1 5 759
Ll (
llA 2021
f3 c
12A lOZ9
B
(
13A 2301
B
c
14A 2302
B
c
154 201
n (
16A 2551
8
10
35
10
20
19
24
001897(()
00141 6CCI
Oll9lHCI
010604 middot middot
Oll083
OOJ11JCAt
002059141
001411CIH
-000412
000293(81
000187
001159 (3)
ooo6a4cct
o00792CCI
OOl40dCAt
000006
-000076
OOlltl6fA)
o o0 5 64CCt
-034570
-o1middot o3o
-1659 90
Ol51lO
-046190CAt
Ol5450CA)
-005l60
002940
-030060(()
-0 43l50CCt
0 33650
-019080
00004
bullJ00741CAI
-000069
000466
0001 51CAt
000947CAt
O OOI)70CAt OOl750CA J
000175CCI
000671CAI
OOll5) f A t 002207CBI
0004b7CA I
- 007 1 70
001760
-001410CAt
- 0 16200CAt
-004210
- 1 207 70
705
96 0
542
466 805
460
614
921
832
9 61
996
424
550
9 15
56 1 651
884
-006114
019 037
03504lCA)
oo409Q
-141644((
491 -0 17664
607
871
lell
084
101
072
103
108 11IllOQfl)-- l
-037880fCJ OOl6l4CAJ -040l70fCI 0()
c
I
I
bull 9
9t
middot middot middotbullmiddot bull lraquo
TABLE 2 REGRE SION RESUlTS ICONTt
bull
EO I NO tHl I filER- f4A_K T AOVER- sbulluRE ttSQ flA Jl- iLAS T NR CODE cos CEPT SHARE JISING AOVfR bull (JTY SHR CiJCF
110Cl I f amp Itt 15J tbt bull 1c u caa
llA 2702 ll o079 6 81 Bt -019110 S6s -OlltO 111
a 004679 ooooo1 t69
( oOl26o -015160 001560 -022690 ltJr()I
ldA 270 3 19 0065lt -OIt2l60 220 -Ol57a2 11 8 e o 0 71t6IA t ol08 75tAt ( OOl983CiU -060llOC() OOl7191A t -O lltdOC A) middot 89
lqA 2704 10 005082 -Oil060 22b -l)i1977 lelO 6 -000080 oo2772C4 t 910 c 000907 -011 200 Oe04105CA t -046960 995
20A 2d06 1 II
O OlltStiAt -001120 837 00014 -))th O01380amp At -oooo 11 821 ( OOl526CAt -026700181 0002 5UAt 000530 969
2lA 2807 28 OOt9lt9CAt -00891 0 516 -0 03962 225I8 002095CCt OOOl lliAt 690 ( o02965CA t -056650((1 OOOt50CAJ -003l70ttl 911I
I 22A 2808 15 ot3l06 119730 655 1130) 090
0 - 007526CC t Oe003961At 835- middot-
c I oo7523 -067210 001l 2tA t -0091t80CCt 951
2lA 2809 21 Oe2067lCAt -1218101At 71 2 -10095((1 121 ttlIaOQ8 00797418) OOOI8lltt 686 Ill( 0092lU A t -Olltl70Ct t 000605CAt -OOit8701At 951 _
---- -
32 009506CAt
0 bull 06 981 I A J 006901 cA bull
00 24A 2810 -010960 5t 5 -OOltSO 7 56
8 -000011 s11 c -5811701At oooaoatAJ oot520IAI 888
-- 37 --- middot - -- -1 ) l lbull t5JJ r-l bullhmiddot1 --- - --
l5A
26A
27A
c
2RA
c
2 9A
31A
llA
24
13
597
Ci)55
14
l 6h
RSQ ELASTI- (lASTCOE f-
TABLE 2
EQ I NO NR CODE IU
REGRESSION RESULTS CCONTa
NR INTER- MARKET
(10S CEPT SHARE AOVER- SHARfcent
TIS lNG AOVER
) CITY StiR
c It) bull 5 co a I 1) bull 0 bull 9 c l 0 bull( 2 1
2815 16 003019 -009990 275 -o069bmiddotJ
6 001446 000224 282 c 002178 -054630 o02221CCI -0075 00 646
2901 29 00021t2 CAI 0)0520 592 0007 6 1 068
0 00105C8 t 000012141 60)g( OOOlJ51AJ -00126018) ooou271At -000u80CA) 974
36oa 1 3 003296(4) -oca29o -oo-q94
001755 000203 lt91t 002499((1 -04l820CCI OOl374CAI 0068)0 946
3612 0033281CI 011520 195 01684) oso
00 30 42C OI 00012 2 424 oo579ltAt -2 7l7801At 002 100IAI -003220 819
8 3617 btel 011789 081002259 009530
001448 oOOl OitCCI 732 OOl7 22CAI -0 46160CAI 0004961AI -OOll40181 989c
30A 3808 14 0021t 3 6fAI -000110 628 o023291C) -005 6 middot- _ middot- 0bull 02 04 1 IC I ooooo9 637 ( 001584((1 -0161)1)0 00021 91AI -0 00060
390) 14 002790CCt 006590 h8 t 0 0938 4 070
B OOl89bCCJ 000144 74
0021961(1 -OItlOlO OOl323CAI -004800 915
B
middot--- - middot 4 - middot- - dfll3904 12 004563181 0)8300 879 0428651(1 089 OQIll004 277CA I ooozstcea 919
c 0041)3181 -0-11250 0011881(1 -006270 972 0Ill
- J - bull bull bull bullbullbull middot - - middot - middot gt middotbull L)ofbullflbull middot middotfJltl -tn- middotmiddotmiddotmiddoto +f middot t middot bull
c
is
positive
2
Page 0 determineThe consumer goodproducer good split is sometimes difficult to
Sollle support for the split used in this paper is a comparison of goodness - of - fit
distributions of a
industries
set
advertising are
that industry
a sales increase of
the
highest aD10U3 those
measures for the two groups of industries Table 3 contains
measures for the 32- consumer goods industries and the 103 producer goods
2 Median R is 5 7 for the consumer goods set and 40 for the producer goods
Estimates - of y the elasticity of demand with respect to
given in column of Table 2 The hypothesis that y bull 0 with
alternative was applied for each ittdustry As shown in the
could not be rtjected for 24 of the 32 industries Of the
and 10 are positive For the remalnifg eight three are
and five are significantly positive
Of the threertegative elasetcities the largest (in absolute value)
cosmetics (code 2809) With all elasticity of -1 01 a one
industry advertising would gwerate a decrease in industry sales
more than one percetu
At the other extrema is toys and games (code 39 04) For
a one percent ittcrease in industry advertising would lead to
four-tenths of a pampreentmiddot Though not significantly different from zero
elasticity estimatbull for pt-bpuetary drugs (code 2808)
which are
Both drug industries have
to sales rafios (211 pe1 cent for 28 08 and 13 4 per cent
other hand flaIJo ing ext1ac s (mainly soft-drink SYUlS) also
is for
percent increase in
of slightly
a middottwo - tailed
table that hypothesis
24 14 are negative
significantly negative
9
the codetics and p-ropriacuy very high advertising
for 28 09) bull On the
has a high ind UY
advenisiltg t() sales ratio ( 8 0 per cent) bull but its estimated elasticity is very
small and not signifieantly different uom zero
middot It doe$ not seem unusual to find many elasticities near zero with a few
negative and a few positive If aggregate consumption is insensitive to aggregate
advertising as seems likely ld then what one industry gains must be lost by
others
Page 21
elasticity of demand does
(10) we can
if t 0 then
Finally a determination concerning the industry
not fix the individual firms elasticity Going back to equation
write y bull (1 - ) a + z y Since a gt 0 it follows thatii i i
Y gt 0 If Ylt 0 however y s sign is indeterminateij ij
Turning next to the coefficient of market share 20 of them are negative and
12 are positive with three of each sign being significantly different from zero
The largest positive value which is significant is for pet food (code 2010) and
a close second is mens amp boys suits and coats (code 23 01) The largest signifi
cant negative value is for cosmetics (code 28 09)
In the model some relations among y a 0 and 61 are implied If middota is
negative (1- y) is greater than one Since 61Y bull (1- a) (1 - ) 61Y is
pCsitive but it will be less than one if 0 is large enough Seventeen of the A
ys are negative in each of the seventeen industries 61 is also negative and les
1 (algebraicly) than y so 61Y is greater than one his is not evidence that the
degree of cooperation is low or zero in any industry of course A high o togethegt
with a negative y which is small (in absolute terms) relative to a can give a
value of 61Y greater than one On the other hand if 61Y were less than one
o would have to be greater than zero
For y gt 0 two conditions may hold If y lt a it follows that 61 lt 0 since
(y - a ) lt 0 It also follows that 61Y lt 0 regardless of the value of o bull
In addition the relation is an if-and-only-if one ie if 61Y is negative
then y lt a Three industries have negative values for 61y dairy products
except milk (code 20 04) synthetic fibers (code 2806) and photographic equipment
(code 3808) In all three cases y is quite small so a does not have to be very
large to give a negative value to al
If y gt a the model shows that 61-- gt 0 whatever the value of 0 bull There are
12 cases where both y and 61 are positive with 61 ranging from near zero
(grain milling code 20 12) to 1 20 ( proprietary drugs code 2808) Several
coeffiei-en_rs
--
15
o f to zero
Tab le 3 - Consumer Goods
Indus tr ies
2Distributions of R and Producer Goods
Page 20a
Consumer Producer
Goods Goods
0 - 9 9 0 3
10 - 19 9 0 1 1
20 - 29 9 3 20
30 - 39 9 3 17
40 - 49 9 5 16
50 - 59 9 8
69 9 6 Ii 1060
--f 670 79 9 2
80 - 89 9 4 4
90 - 100 1 1
Totals 32 10 3
Notes
2is used R i s cal-
culated as F F + ((N - K- 1 ) K ] 1 b_ F is the F
statistic Lor the hypothesis that all o f the non - intercept terms are simul taneously equal
1
2
--
-- Since generalized least squ-area regress ion
Page 12
o f these cases have values o f 6 1 Y rea ter than 0 8 (pet food (20 10) bread amp cakes
(20 14 ) miscellaneous processed foods (20 29 ) men s and boys suit s and coats
(23 01 ) p rop riet ary drugs (28 08 ) elect ric housewares (36 1 2) radio and
IV sets 17 and games and toys (39 For be than(36 ) 04)) 6 1 Y to greater 0 8
both o and cry mus t b e less than 0 2 That is in these industries the results
shown in Tab le 2 would hold only with very strong dominance of marke t demand
effects over brand swit ching e ffects and a very low degree o f cooperation
Using the concentration rat ios given in Table 4 the average concentration
ratio for the eight indust ries is 38 1 For the remaining 24 industries it is
43 2 This result would lend some support t o the proposition that concentrat ion
and cooperation are related if the group of eight industries are identi fied
as having low degrees of cooperation
One f inal note on model predict ions S ince both cr and o are assumed to
be non-negative 60 should also be non-negat ive In only one o f the 32 cas es
is 60 negative and it is not signif icant bull
cor pJ
7 1
7 0 4
1 4 7 5 1
)
7 3 z a o a t s 1 1 3 3 5 ( i V Z 8 tJ1 2 1 - 1 1 0 7 3 5 c middot v 2 8 1 0 3 2 -0 0 1 4 3 0 C 1 V 9 8 3 2 8 1 5 1 6 - 0 0 6 t1 6 9 C _ V 1 11 0 2 9 0 1 29 0 00 7 6 1 C l k V 7 3 1 3 6 08 1 3 - o n 4 9 9 4 3 3 3 6 1 2 2 4 0 1 6 4 3
3 1 7 1 3 0 1 1 7 d 7 8 3 3 8 0 8 0 0 2 3 i 2 3 0 6
P 4
3 9 0 3 1 4
1 2
0 09 3 8 4 39 1)1 0 4 l 8 6 5
T A LE 4 e I NCU S T R Y C A T E G O R Y
F TC NOM e tOOE ass
C 1 I c 2 )
2 0 0 1 1 6 2 0 0r 1 7 zo o l Z 2 0 0 7 1 8
z o o 2 6
z o t a 1 6
2 1) 1 2 1 5 2 0 1 4 1 1 2 0 1 a 1 9 2 0 26 1 2 2 0 2 7 1 0 2 0 29 - 3 5
2 3 0 1 10
2 3 02 z o
2 3 03 1 9
2 5 5 1 2 4 Z 1 0 Z 1 1 2 0 3 1 9 2 7 4 1 0 z a o 1 3
2 A 0 7 z s
E LA S T I -C I T Y
c 3 )
- 0 09 7 2 7
0 00 449
- 0 0 2 8 9 3
- 0 3 2 7 70 - 0 OZ il I q
o 4 u t s a 0 0 1 3 7 3
o 1 4 o o-0 0 5 2 4o -0 1 94 0 1
-C l 6 l l 4 0 1 JJ 3 7 0 3 t 0 4 2 0 0 4 0 9
- 0 4 1 8 44 - O l 7 t ) 4 -ll 7 1 4 0 b -0 3 5 7 8 Z
- 0 1 9 7 7 0 1)0 3 3 4
-o n 9 2
IJ A T A
CONV
ON-C ONV
( 4 )
t 1NV
C 1t-V C NV c ) v C - JNV C rlNV c n v c a v C ON V ( i) l V
C CtV
C t1tV C IJNV
C 1 V
c 1 v
Pa ge 2La
HEQ F t Noe x
c 5 )
6 2q
6 89 1 48
6 1 4 ) q 3
1 9 7 1 9
q z t s 5 a e z
l l 9 1
4 o 7 1 2 0
c oqt 1 Z
4 4 )-q 1 z
5 5 9 1 0 64 1 3 3
n 3
Inter industry analysi
Page 23
Equation ( 3 6 ) may be used to fo rmulate a rough test o f the cooperat ion
vs e f ficiency controversy I f we rest ricted ourselves to some subset o f
indus tries where the relat ion be tween y and a i s the same we would be lookingbull
at a sample o f indus tries for which the dependence o f V on H and o is a
quadratic wi th no squared terms
I f we now sub s titute (26) into ( 36 ) we get
(38) v bull a + (y - a) R + (y - a) o (H) - (y - a) H o (H) + u
If it is only d if ferences in marke t shares that affect advertising inten-
s i ty levels i f concentration does not af fect cooperation and i f o bull 0
( 38 ) reduces t o
( 39 ) V bull a + (y - a) H + U
Let this b e the null hyp o thes is and note that V is a linear func-
tion o f H If y gt a and the null hyp othesis holds the expected relat ion
between V and H is as shown by the pos itively sloped straight line in Figure
6 b etween points A and C
In this context the alterna ive hypothesis the degree o f
cooperation ( o) is a positive function o f concentrat ion
is that
(H) I f the func-
tion is as described in (26) and Figure 4 the relation between V and H
would be as shown by the curved line connec t ing po ints A and C in Figure 6 One way to test the null hypothe s is against this al ternative is to test fo r
the hypothesized curvature in the function relating V and H
Fi gure
R
6
v
ID c
H 1 0
I
i I I
sfigned
clasJ s
in def ini g
effec t is the
I
Page middot2lf
One charac teristic of produc ts which may be useful
classes wi th respec t to significanc e of the brand-switching
conveniencenon-convenienc e distinc tion developed by Porter ( 1 97 )
Using his identif ication of IRS indus tries as a s tar t ing point
each of the 32 industries used in this s tudy to one of the two
The ass ignments are not ed in Tab le 4
Using the conveniencenon-convenienc e dis t inc tion the
divided into two sub-samples
sample was
Four regressions were then run bull ene for e ch
sub-sample one for the to tal sample and one for the total sample with
a dummy variable for convenience goods Each regression was run wi th LB s
and wi th indus tries as ob servations The results are given in Tab le 5
For the LB level equations ( 1A - 1D) equation(37) was used ome
rearranging of terms gives v bull y - ( 1-o ) (y - a) ( 1-z ) + ui i i
bull y - 6 1 ( 1 - z ) + ui When observations are pooled across indus trif s yi
is an industry variable In these regressions the values o f y es t imated
in the first stage were used Since industries have been grouped so as
to reduce differences in (y - a) that term is t reated as a cons tant
within each group Finally H the Herfindahl index has been subst ituted
for o the degree of cooperation The resultant equation is linear
y ( 1 - z ) and H x ( l - z ) This is the equation for whi ch results arei i
reported in Table 5 In all regress ions all variables were weighted by the
square roo t of sales
in
raquo
- 34
3 1
i l T 4 8LE S R E G E S S I ON R E SUL T S i X T f N S I ON S T O T Y PE Of
AND L E V E L OF C ONC E kT R A T I O
E O O tt S
tP bull T Y P E NR C ONV O B S tO N-CONV
E L A S T I shy
T I C J T Y
O E - t-41( T SliA R E
H( fl F
1 1 - K O x middot K O X E R F
- H S a 1 1 - H S t 1 8 1 1 9 1
R S O
1 1 0 1
q z
t - bull 9 f 2 1
Z t P D l 6 fC ONV - 1 9 1 s s c 2 1
e - l O I e bull u q a
2 3 1 0 1 1 Nl N- t O V 5 7 1 ( 1 2 19 1 6 1 - J l S 2 5 4
l l l l t z q o a 1 - J o v z I
l ( I NO l l A l l ol u l t o o c bull J u l l
d S
c 3 4 t t r a l
bull Z D 1 0 l l 4 l l 2 3 -gt 2 l l 2 0 - 4 lS 5 4 l C 4 1 o e 6 t l b b 5 1 1 0 1 1 1 c - 1 9 5 1 t d S 2 1 tV
OQ10 T-R A T I O S A E G I VE N l P A RE N T H E S E S l E V E L S f S I G 1 F l C ANC E A - 99 5 B - q 9 t - s NS J UC E NO I Nl E R C ( P T I S U (j f middot) I N f Hf t OUA TIOI S 6 E f 3R E A O J U S T H E N T F OR 1-t E T E R O SIC E O A S T f C I T V pIraquo
1
middot
I
R SO I S C A L C UL A T E D A S f I f bull 1 1 bull 1lt 1 K t t- E E N I S THf U A1 B E R O f fJe S E fi V A T I O S F t S HtE F - S T A T I S
I N DE P E ND E N T y II
T I C F O T H E U Y P O HIE S I S f A l l C JE F F I C H N T A R E QU A L HI l f RO A f40 K I S T H f UUM 8 E R OF Q ( A fL f S
I l l
1 A
1 6
t C
f 1 r R I l l
1 2 1
L Jt
L B
l q
L 1
on s T Y P E bull 2 )
P OUC T
X
H S I l l
l O l A I bull l 0 7 1
O O b O J
1 2 00 1
1 b 2 1 C I l l Z U
l l S b 1 -1 1 0 1
rlt F X l l - fi C f l
bull 1 I
l l
- ()4 2 3 l f H
1 - 2 7 9 1
K O X
1 1 - ti E R F I
l d l
S4 l i A I 2 l b
1 3 00
K O X -t f R F t S J
I 1 - HER F t
)
60 9
z o bull
0
t l l
3 8 8
1 7 6
5 64
64
N f bull
tl S bull
I l l
I 4 1
t ONV
NON-t ONV
A l l
A ll
c o v NO- ( ONV
1 4 1
1 5 1
- l S b
c - 1 1 2 1
5 8 2 C A I 1 6 7 0 1
middot I l 9 H
l 1 il
1 4 1 6 1
E L A C) T J shy
l t C I T Y
I S I
l b l
- 0 2 0
1 - 1 4 9 1
O Z 3 2 1 A I C b 6 b l
0 0 9 2 -J
1 1 b 4 1
o z t t t l 1 2 2 1 1
oru - Ht tt F
1 gt X bull b I
Page 25 bull
Mean val -as ior relevan t variables for the two sub-samples are
Conv Non-Conv
Adv Sales 4 7 2 3
iit Share 2 7 3 2
Ela cicity - 05 3 02 2
Herf indahl 7 3 1 860
That the dist nct ion between coft9enienee goods and non - convenience
goods mat ters is supported not only by these means but also by the regression
results Equat ions lA and lB are different in almost every respec t Only the
concentrat ionmarket share int eract ion term seems to mat ter for convenience
good s Jus t he opposite holds for non - convenience goods both elasticity1
and market sl e are highly significant but the concentrat ionmarket share L
int erac t ion t=rm is not
Given that one minus market share is used these equ3 tions the coeffici t
of market share is the negative of what is sh in col (6) of Table 5 For the
regress ion for non - convenience LB s reported in equation lB then LB s with larger
market shares have lower advertising to sales ratios
On the question of the impact of the degree of cooperation the story is mixed
For convenienc good s the coefficient is pos itive as equation (31) predicts when
there is a po sitive degree of cooperat ion Fo r the non-convenience goods LB s middot
concentration has no effec t
Aggregat ion t o the indus try level and then rer unning the equations $ives the
results shown in (2a-2d) of Tab le 5 As expected R2 goes up and levels of significanc e
fmiddotJr individual coefficients gJ down In add ition the coeffic ient of H ( l -H) inmiddot
quat ion 2A is no t significant though it is po s i tive Since 2A is the indus try level
I
--
all but one of them and that one is not significant For 18 of the 31
wo uld have on its adve rtising to sal es ratio
cognate of LA the variable H ( l-H) pl ays the same role in 2A as H ( l -z 1) p lays
in LA
Vconomies of scale in advert ising
The literature on advert ing contains a number of definitionbull o f economies
of scale and of findings eone rning its presence and magnitude Before I loo k
a t the data with the hope o f mmenting about scale effect s then I wan t to
define the term as I use it
By economies of scale I mean a decrease in the number o f units of adver-
tising which are needed to generate a unit of sales as advert ising is increased
Given the definition of the firm s own elast icity of demand with respect to
advertising (see the text foll wing equation (1) ) the technical definition
is that rii gt 1 Diseconomies of scale are def ined symmetrically i e
Yu lt 1
ith a cro s s - section oi observations on advertising intensity
(v bull a s1) and onmiddot advertising (a ) we may regress the former on the latter 1 i i
The results of such regress ions for the 32 industries are given in Table 2
lines lB ZB bull bull bull 32B For the 32 eases the regression coefficient is positive it
positive coefficients the coefficient i s signif icant
The que tion now is whether these results ar e evidence of pervasive dis-
economies of scale in advertislng I think no t Given my as sumption that the
data I observe are equilibrium results I may ri ghtly conclud e that almost every-
where high levels of advert ising are as sociated in equilibrium with high ratios
of advertising to sales I may no t conclud e anything however about the
impact that a move by some firm away from its equilibrium level o f advertising
middot- Page 27
Given the assumed relation in equation ( 37 ) there is a corre spondshy
ing equilibrium equation relating v and ai The equation is non - linear i
and I have not found a simp le way to characterize it It is po ssib le to
det ermi ne its deriviative with respect to a b y takin$ tne de riviative of ( 3 7 ) i
When that is done the result is
v - 1(40) i bull sl a a z ( 1-z )i i ira-l
Sinc e all the other terms in (40) are non-negative the sign dep ends only
on the sign of 6 1 which is in turn de pend ent only on the sign of (y - a)
That the signs of the coefficients of a in equations lB 2B bull bull bull 32B arei
not always equal to the signs of the coefficients of z in LA 2A bull bull bull 32A i
is still a mystery to me and something to be explored
bull
Summary
My purpose in writing this pa per wa s t o attempt to integrate explicit micro-
economic mode lling reasoned econome tric specification of error terms and high
quality data to exp lore some questions about adver tising by large firms
al quite important improvements could be made in the first two of those areas
and much data ma ssag ing is yet to be done in the third
Concerning the subst an ce of what I have done I think six things are im-
Page 2 8
Severshy
V and Conclusions
portant
1 Explicit modeling is worth the ef for t if for no other reas on than
that it provides a basis for choosing fr om a varie ty of funct ional
forms
2 The same is true for error specif ica t ion I did in fact look at scatshy
ter diagrams for all 32 industry categor ies and they vir tually all show
the het eroscedast icity which I as sumed and for which I correc ted
3 The predicted relation be tween advertis ing in tensity and market share
shows up clearly in on ly 25 of the case s examined I have ye t to
explore why that may b e the cas e
4 Some evidence concerning the presence and impact of coopera tion was
produced but it is not clear ly pervasive
5 The dis tinction between convenience goods and non-c onvenie nce goods is
unambiguously a good one
6 Virtually no evidence concerning econ omies of scale in adver tis ing can
be gleaned fr om this study given its assumpt ions
More work is called for on many of these issues my study seems to have
raised more questions than it has answer ed This is pr obab lly due to the richshy
ne ss of the data source since I had the oppor tunity to address several issue s
at once
- -- - Q --
Foo tnotes
Manager Line o f Business Program Federal Trade Commission The author depended heavily on several staf f members of the LB Pro gram for statistical and clerical support in particular Joe Cholka George Pascoe and Haro lene Jenkins Helpful comments were given b y Richard Caves Dennis Mueller Michael Porter and F M Scherer When the paper was in its early stages o f development both S teve Garber and Jon Rasmussen former economists in the LB Program provided eery useful cri tiques The views expressed here are my own o f course
1 The basic exp licit model of the role of advertising is Dorfman and S teiner ( 1954 ) and the models in this paper may be seen as ano ther extens ion of their work
2 For a review o f models of oligopoly which treat prices and quantities as relevant variables see Shubik ( 1 959) Extens ions to take advertis ing int o account are fairly s traightforward
3 Dorfman and S teiner dealt with a price-setting firm For a monopolis t inverting the demand relation to show Pbull P (Q A) _instead o f Q- (P A) has no effect on the results o f the analysis For oligopolists who face a sys tem of demand relations inversion does have some impac t on the results The dif ferences are not trivial nonetheless they will no t be explored in this study
4 Shubik ( 1 959) p
5 From a formal mathematical point o f view the function is similar to a1 i Legrangian funct ion of the form - r rr
L bull 11 + A ( j - (i i i+j
where the indicates some fixed level of the profits industry Following the same logic which resul ts in and income constraint as the marginal utility of the of the indus tr in the obj ective of firm i
6 Hall amp Weiss ( 1 967) p 323
7 See Schmallensee ( 1 97 2 ) Ch 2 pp 1 6-47
8 On this poin t see Imel amp Helmberger ( 1 9 7 1 )
9 Long ( 1 970) p 130 - 1 39
1 0 See Schmallensee ( 1 972) Ch 3 pp 48-8 7
rrj i+j ) of the rest of the
the identification profits of the rest
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980
Figure
Page 12b
4
10
H 10 H
l
(27)
Cl - o) o
[ (1 -
[ (1
t y ij z
j
are given in Figure 5
Page 13
The equation in (24) was constructed as a hybrid of two results derived
from first-order equations for the non-cooperative and cooperative models
It may also be constructed directly by using l bull (1 - o) + on as thei i
th 2objective function for the i-- firm Treating o as a constant the first
order equation is
0 an aa
i middot
n i
Using equations (3) and (7) and setting --- bull 0 givesaai
(28) - 0
After collecting terms and converting to matrix notation we get
(29) - Cl - o) + 0 z-1vi yii i
N
j bulll
+ 0
+ o v1
d-1 r z z
-(30) v Cl - o) rd
-
o) r +d
d r J _ -1- d z
and middot
d) r + 0 rjd
l r z
v - z --(31) zv
- (1 0- +d) t r z- d-
If we now impose the simplified demand structure by substituting (11)
we get
- (1 - o) a +( 32) + oy and
Figure
l
-y
v
Y
H
r
I I
v
a
y
H 10 H
v
a
bulla
HH 10
bullbull
(33)
i
Page 14
v - (1 - o) [ a l + (y - a) ] The corresponding equation for Vis given in (24)
If the relation between industry advertising intensity (V) and
concentration (H) is non - linear via the impact of concentration
on the degree of cooperation does the relation take the form of a
auadratic as some other investigators have proposed (Greer ( 1971 ) Martin (1979) Strickland amp Weiss (1976) ) That question can
be approached from two perspectives - conceptual and empirical In
the next section I will try to apply some statistical material to it
here I want to explore the question in the context of the models
developed above
Referring to equation (24) taking the deriviative with respect
to H and letting bull (H) gives
bull( 3 3a ) 3V (y - a) (1 - o) + (1 - H) 3H dH
Given that o and H each are bounded by 0 and 1 and that o is an in-
creasing function of H the term in the brackets in non-negative The
sign of the deriviative then is the same as the sign of (y - a) If the industry effect is dominant (y gt a) then advertising in-
tensity will increase with concentration If the two effects are equal
then advertising intensity will not vbullry with concentration and if the
brand-switching e ffect is larger advertising intensity will decrease
with concentration
For given advertising elasticity parameters then it is not pos-
sible to show an increase in A S up to some H and the n a decrease
he relation may be non - linear but it is monotonic over the permis-
sible range of H
Page 15
One possibility is to introduce a relation between (y - a) and H
If industries characterized as dominated by the brand-switching effect
also were more concentrated then a cross industry comparison would
show an inverted U shape relation between AS and H I know of no
support for such an association between ( y - a) and H
V ERROR SPECUCATION
So far I have implicitly assumed that there are no errors in the
model I now assume that the error
thvariable in the equation for the i- firm is additive that its mean is zero
that its variance is inversely proportionate to its sales and that all error
term covarlances are zero hat is
34) v bull (1 - o) [ a + (y - a) z ] + oy + u i i i
- - -
In vector notation
(35) bull (1 - o) [ at + lt - a) J +
E() bull 0 Cov (u) -
0
If we now ca lculate V bull bull we get
oy + U bull
(36) V bull (1 - o) la + (y - a)R J +
U bull E(U) bull 0 Var (U) bull
P age 16
The error term specifi c ation which follows Hall
wi th the ratio of profits
6 heir analysis seems
as wellbull particularly given
and Weiss was
applied by them to equa tions to equity or
asse ts as dependen t variables to carry the same
weight in the present context the substantial
evidence presented by marketing analysis on the tendency of company de-
cision makers t o use the advertising to sales ratio as the decision
7variable
The zero covariance assumption on the other hand1 is particularly
troublesome Within a probably not true since
many even ts outside the
single industry it is
industry would have
industries
similar effec ts on all the
8firms in the industry Between it is probably not true either
since many companies produce in more than one indus try and even ts in the
firm would tend to effect all of its activities The extension of the
model to allow for non-zero covariances is certainly called for
--- -
Page Ll
Athird characteristic of this specification also deserves some comment
only on equation is shown for the firm or industry Several haveinvestigators
included an advertising equation ina larger system of equations (Comanor and
Wilson (1974) (1979) and Weiss (1976) so in myMartin Strickland Idid
dissertation1 Ihave ewo defenses for presenting only single equation results
here One is the suggestion by Comanor Wilson that simultaneousand equation
verybias may not be important in this context he second is that Iam will-
ing to see the model extended to include a profitability equation and I intend
to move that direction in the near futurein
Onthe inclusion of aconcentration equation on the other hand I amskep-
tical he problem isthat in some fundamental sense equations in models of
industrial organization should be oriented to the decision-making contexts of firms
Ifeach firm in anindustry with Nfirms has only one decision variable then
there can be only Nindependent equations which are based on the first-order
optimization equations It not show large ofis difficult of course to a number
functional relations which follow from the first-order conditions point here he
isthat only Nof them can be independent
Consider for illustrative purposes only a situation in wich teh N
itsquantities say q are set exogeneously and each firm determines advertising1
of and level say ai where a is the true optimal value a a bulla
1 1 i i + i
Ifp a) then the Nq
s and the Na s will determine N p s say pii bullp
i(l i 1 i
Now given Eand 9 we can determine both the vector s andthe vector v since
- bull til
ands We can also determine S s z bull s-1 s-ii
v and we
industry concentration isHbull (alternatively
V and Hare functions of the same
pendent Furthermore if there are
VM) and (H1 H2 bull bull bull )contain
VI STATISTI CAL APPLICATIONS
Page 18
Given the observed values of might be tempted to conclude that we
have a two equation system in which v and z are jointly determined That would i i
be inappropriate however since the 2N random variables (v ) are determined by
onlt N basic random variables t The variance-covariance matrix for ( ) must
be singular bull
The extension of this observation to industry level variables is straightshy
forward Since observed industry advertising intensity is V bull l v and observed 4
C4 bull z ) it follows thati l i
set of random variables t and cannot be indeshy-
M industries then the variables (V1 v2 at most M independent random variables
Given the assumptions made above y a and o are constant within a
given industry That being the case (34) can be written as
(37)
s1 bull (1 - o) (y - a)
If and
So + +
(1 - o) a + oy
81 are esti mated for an industry using a suitable statisticalSo procedure giving and 81 an estimate of y is then available The sumSo
Intra-industry analysis
1974 of Business Re ort -
categories are identified
Page 19
of So and S1 is y so y bull So + S1 is the estimate of y
If o is assumed to be bounded by zero and one the sign of 61 is also
the sign of y - a The estimate of So is of cou se an estimate of
(1 - o ) a + oy Moreove if y gt a then a is less than 6o- (1 - B) a+ 0(
Cor esponding results would
hold if y bull a or y lt a
If 61 gt 0 then we would estimate that a lt 6o
bull
For purposes of estimating this model 32 manufacturing industry categories
were selected The data are fram the Line of Business forms filed with
the Federal Trade Commission they are discussed in detail in the Annual Line
which is available from the FTC The 32 industry
in Table 1
Simple regressions for equ ation (37) were run for each of the 32 industries
with data for each of the firms which filed in the industry constituting an ob-
servation The results are given in Table 2 lines lA 2A
1974
the intercept
For all the A
equations generalized least quares was used th both and market
share bein scaled bv the square root of sales
The 32 industries were taken from the larger set of all manufacturing industry
categories in the 1974 data files Two selection cr iteria were used thereLB
had to be at least ten reporting companies and the industry had to be a consumer
goods industry or a producer goods industry which is strongly associated with a
related consumer goods industry so that adve rtising by the firms in the producer
goods industry impacts on demand in the related industry (eg soft drink syrup
and soft drinks) There were 135 industries with at least ten companies - 103
producer goods and 32 consumer goods
273
274
TABLE 1 INDUSTRY CATEGORIES USED
FTC
COOE OESCRI P T I ON _
2001 MEAT PACKINGt SAUSAGE S ANO OTHER PREPARED
MEAT PROOUCTS
2004 DAIRY PRODUCTS EXC F LUID MILK
2005 CANNED SPECIALTIES
2001 FROZEN SPECIALTIES
OEHYORATFOt AND
I NC LU 0 ING PRE SERVE S
SOUP MIXESbull
SEASO lNGSt ANO
OTHER PET FOOJ
RAIN MILL PROOUCTSt
ANO PREPARED
RELATED PRODUCTS
PRODUCTS
Z008 CANNEOt ORIEOt PICKLED FRUITS
ANO VEGET ABLES JAMS JELLIES DEHYDRATED I G TA3t E
S40CES ANO SALAD ORESSINGS
2010 DOG CAT A O
2012 FlOUR t OTHER RICE
MILLING fiLENOE) FLOUR
2014 B EADt CAKE ANO
20113 CO FECTIONERY
2026 BOTTLED A O CANNED SOFT DRINKS 2021 FLAVORING EXTRACTS AND SYRUPS EC
2029 MISC FOOOS AND KINDRED PROJUCTS EXC
ROASTED COFFEE
2301 MENS AND BOYS SUITS AND COATS
2302 MENS ANO BOYS FURNISHINGS
230 3 kOMENS AND MISSES OUTER EAR
2551 HOUSEHOLD FURNITURE
2702 PERIODICALS
Pqe 19a 1
RELATED 1972
src cooe
2011 f 3
202tX2026
2032
203pound
2033t4t5
2047 2041t4t5
2051 2065 2086
2081
209XZ05
231
232
233
251 272
2703 BOOKS
2704 ISC PVeLISHING
2806 ORGANIC FlEERS
2807 DRUG$ ETHICAL
23234
PT 2 S 3
2808 ORUCSt PROPRIETARY
2809 PERFU ES t COSMETICS ANO OTHER TaiLET
PREPARATIONS
2810 SOAP AND OTHER CLEANING PREPARATIONS 2815
2901
PEST1CIDES AND AGR ICULTURAL
PE OLEUM REF I N ING CHEMICALS EC
3606 HOUSEHOLD COOKING EQUIPMENT
3612 HOUSEHOLD APPLIANC ES NEC INCLUJING ELECshy
rqlC HOUSE ARFS AND FANS ANO SE I G MACHINES
3617 RADIO AND T V RECE I VING SETS 3808 PHOTOG APHIC E UIP amp SUP LfESt EXC
3qo3
PHO TOCOPYING
SPOR TING A D E QU I P M ENT f SUP P L I E S ATHLETTC GOODS NEC
3904 DOLLS GAMESt TOYS AND CHLOqENtS VE ICLES
PT283
2844
234tX2844
287 9
291 3631
3634t6t9
36H
PT386l
3949
15
943
middot
l A6l f 2 R(GRESSION RESULTS
EQ I NO
NR CODE cu 121
lA 2001 A (
2A 2004 r (
3A 2005 a
-R
ClS amp31
16
17
1 z
I TER-CEPT bull t
ooo9t21At 00022311 oooeos uu
OOl49tCAI 0010 2 middot 000960CCI
o049tl1At o04t55C61
MARKET SH ARE bull 5 I
-010670Ctl
-017600181
-001050
-0211AO
-0077 0
AOVER-TIS INti
lbl
000()39 000072
000018 0 0024liAI
-000079
SHARf
4ampgtVf Q bull
111
000140
-000170
RSQ
COl
66 485 862
607 6 3 t ()2
blti6 633
flASTI-
CI T Y PH
-007271(1
000441
-u0288l
clASTSHU (t)Ef
bull 10
tto
-234
210
( OOlltOCCI -126520 000922 181 001790
4A 200 7 18 00590liAI -038670 4e16 -032110 lell u c
OOOS74e 000700 025940
00067018) 0021001AI -02l100CAI
5 26
920
5A 6 c
200 8 26 0 03Sl9CA I 0012 26181 o 024511(
-005540
-0 81420 0001961AI O()Ol90CAI ooll240
441 604 891
-0020 19 274
6A 2010 16 Oe04224CAI o359lOIAI 9oq ll4015dCAI 061 6 -middot 005435CAI 00011 8 1AI 917 c 00535ltAI -009850((1 OOOA51J CAI -001610161 971
7A 2012 001862 000010 351 00187) oo1 a oooq45 ono1o6 439 c 001109 -027050 Oe0067SIAI -OOll60C81 897
- middotmiddot- -
8A 201t ll 000757 Ol4140CCI 814 ott9oocet 095 8 oooqo6 C C J 000069(81 811 c ootl2lampBI -080910CCI Oe009lUBI -OOtlOOCCI 960I
I DIIJIT pound11 1V OTIUTVTfAll-1pound1t- A _ 00 R - qq r - Q
1jl)OQCD
0
-middot middot--
l
_ __ _
I
i
0 00141 c c
91)7
I
T ABLE 2 REG ESSIOH RESULTS CCONTI
EO NR
IND CODE
iR bull
cos
I TEq-
CEPT
t1ARKE T
SHA E
AOVER-
Tl SING
SHARpound-cent
AuVER
RSO ILAST1-
C I T Y ELAST
SH CPtf (lt Cl) bull 11 I 41 I 5I (61 171 I 8 I f9) c 10)
9A 2 0 1 8 19 002397181 -007640 367 -005 246 le46 6 000605 000550CAI 597
-000029 G14180 OOl654CAJ -024li01At
lOA 2026 12 0028901AI -OlZl90 -0114 0 1 1 1 5 759
Ll (
llA 2021
f3 c
12A lOZ9
B
(
13A 2301
B
c
14A 2302
B
c
154 201
n (
16A 2551
8
10
35
10
20
19
24
001897(()
00141 6CCI
Oll9lHCI
010604 middot middot
Oll083
OOJ11JCAt
002059141
001411CIH
-000412
000293(81
000187
001159 (3)
ooo6a4cct
o00792CCI
OOl40dCAt
000006
-000076
OOlltl6fA)
o o0 5 64CCt
-034570
-o1middot o3o
-1659 90
Ol51lO
-046190CAt
Ol5450CA)
-005l60
002940
-030060(()
-0 43l50CCt
0 33650
-019080
00004
bullJ00741CAI
-000069
000466
0001 51CAt
000947CAt
O OOI)70CAt OOl750CA J
000175CCI
000671CAI
OOll5) f A t 002207CBI
0004b7CA I
- 007 1 70
001760
-001410CAt
- 0 16200CAt
-004210
- 1 207 70
705
96 0
542
466 805
460
614
921
832
9 61
996
424
550
9 15
56 1 651
884
-006114
019 037
03504lCA)
oo409Q
-141644((
491 -0 17664
607
871
lell
084
101
072
103
108 11IllOQfl)-- l
-037880fCJ OOl6l4CAJ -040l70fCI 0()
c
I
I
bull 9
9t
middot middot middotbullmiddot bull lraquo
TABLE 2 REGRE SION RESUlTS ICONTt
bull
EO I NO tHl I filER- f4A_K T AOVER- sbulluRE ttSQ flA Jl- iLAS T NR CODE cos CEPT SHARE JISING AOVfR bull (JTY SHR CiJCF
110Cl I f amp Itt 15J tbt bull 1c u caa
llA 2702 ll o079 6 81 Bt -019110 S6s -OlltO 111
a 004679 ooooo1 t69
( oOl26o -015160 001560 -022690 ltJr()I
ldA 270 3 19 0065lt -OIt2l60 220 -Ol57a2 11 8 e o 0 71t6IA t ol08 75tAt ( OOl983CiU -060llOC() OOl7191A t -O lltdOC A) middot 89
lqA 2704 10 005082 -Oil060 22b -l)i1977 lelO 6 -000080 oo2772C4 t 910 c 000907 -011 200 Oe04105CA t -046960 995
20A 2d06 1 II
O OlltStiAt -001120 837 00014 -))th O01380amp At -oooo 11 821 ( OOl526CAt -026700181 0002 5UAt 000530 969
2lA 2807 28 OOt9lt9CAt -00891 0 516 -0 03962 225I8 002095CCt OOOl lliAt 690 ( o02965CA t -056650((1 OOOt50CAJ -003l70ttl 911I
I 22A 2808 15 ot3l06 119730 655 1130) 090
0 - 007526CC t Oe003961At 835- middot-
c I oo7523 -067210 001l 2tA t -0091t80CCt 951
2lA 2809 21 Oe2067lCAt -1218101At 71 2 -10095((1 121 ttlIaOQ8 00797418) OOOI8lltt 686 Ill( 0092lU A t -Olltl70Ct t 000605CAt -OOit8701At 951 _
---- -
32 009506CAt
0 bull 06 981 I A J 006901 cA bull
00 24A 2810 -010960 5t 5 -OOltSO 7 56
8 -000011 s11 c -5811701At oooaoatAJ oot520IAI 888
-- 37 --- middot - -- -1 ) l lbull t5JJ r-l bullhmiddot1 --- - --
l5A
26A
27A
c
2RA
c
2 9A
31A
llA
24
13
597
Ci)55
14
l 6h
RSQ ELASTI- (lASTCOE f-
TABLE 2
EQ I NO NR CODE IU
REGRESSION RESULTS CCONTa
NR INTER- MARKET
(10S CEPT SHARE AOVER- SHARfcent
TIS lNG AOVER
) CITY StiR
c It) bull 5 co a I 1) bull 0 bull 9 c l 0 bull( 2 1
2815 16 003019 -009990 275 -o069bmiddotJ
6 001446 000224 282 c 002178 -054630 o02221CCI -0075 00 646
2901 29 00021t2 CAI 0)0520 592 0007 6 1 068
0 00105C8 t 000012141 60)g( OOOlJ51AJ -00126018) ooou271At -000u80CA) 974
36oa 1 3 003296(4) -oca29o -oo-q94
001755 000203 lt91t 002499((1 -04l820CCI OOl374CAI 0068)0 946
3612 0033281CI 011520 195 01684) oso
00 30 42C OI 00012 2 424 oo579ltAt -2 7l7801At 002 100IAI -003220 819
8 3617 btel 011789 081002259 009530
001448 oOOl OitCCI 732 OOl7 22CAI -0 46160CAI 0004961AI -OOll40181 989c
30A 3808 14 0021t 3 6fAI -000110 628 o023291C) -005 6 middot- _ middot- 0bull 02 04 1 IC I ooooo9 637 ( 001584((1 -0161)1)0 00021 91AI -0 00060
390) 14 002790CCt 006590 h8 t 0 0938 4 070
B OOl89bCCJ 000144 74
0021961(1 -OItlOlO OOl323CAI -004800 915
B
middot--- - middot 4 - middot- - dfll3904 12 004563181 0)8300 879 0428651(1 089 OQIll004 277CA I ooozstcea 919
c 0041)3181 -0-11250 0011881(1 -006270 972 0Ill
- J - bull bull bull bullbullbull middot - - middot - middot gt middotbull L)ofbullflbull middot middotfJltl -tn- middotmiddotmiddotmiddoto +f middot t middot bull
c
is
positive
2
Page 0 determineThe consumer goodproducer good split is sometimes difficult to
Sollle support for the split used in this paper is a comparison of goodness - of - fit
distributions of a
industries
set
advertising are
that industry
a sales increase of
the
highest aD10U3 those
measures for the two groups of industries Table 3 contains
measures for the 32- consumer goods industries and the 103 producer goods
2 Median R is 5 7 for the consumer goods set and 40 for the producer goods
Estimates - of y the elasticity of demand with respect to
given in column of Table 2 The hypothesis that y bull 0 with
alternative was applied for each ittdustry As shown in the
could not be rtjected for 24 of the 32 industries Of the
and 10 are positive For the remalnifg eight three are
and five are significantly positive
Of the threertegative elasetcities the largest (in absolute value)
cosmetics (code 2809) With all elasticity of -1 01 a one
industry advertising would gwerate a decrease in industry sales
more than one percetu
At the other extrema is toys and games (code 39 04) For
a one percent ittcrease in industry advertising would lead to
four-tenths of a pampreentmiddot Though not significantly different from zero
elasticity estimatbull for pt-bpuetary drugs (code 2808)
which are
Both drug industries have
to sales rafios (211 pe1 cent for 28 08 and 13 4 per cent
other hand flaIJo ing ext1ac s (mainly soft-drink SYUlS) also
is for
percent increase in
of slightly
a middottwo - tailed
table that hypothesis
24 14 are negative
significantly negative
9
the codetics and p-ropriacuy very high advertising
for 28 09) bull On the
has a high ind UY
advenisiltg t() sales ratio ( 8 0 per cent) bull but its estimated elasticity is very
small and not signifieantly different uom zero
middot It doe$ not seem unusual to find many elasticities near zero with a few
negative and a few positive If aggregate consumption is insensitive to aggregate
advertising as seems likely ld then what one industry gains must be lost by
others
Page 21
elasticity of demand does
(10) we can
if t 0 then
Finally a determination concerning the industry
not fix the individual firms elasticity Going back to equation
write y bull (1 - ) a + z y Since a gt 0 it follows thatii i i
Y gt 0 If Ylt 0 however y s sign is indeterminateij ij
Turning next to the coefficient of market share 20 of them are negative and
12 are positive with three of each sign being significantly different from zero
The largest positive value which is significant is for pet food (code 2010) and
a close second is mens amp boys suits and coats (code 23 01) The largest signifi
cant negative value is for cosmetics (code 28 09)
In the model some relations among y a 0 and 61 are implied If middota is
negative (1- y) is greater than one Since 61Y bull (1- a) (1 - ) 61Y is
pCsitive but it will be less than one if 0 is large enough Seventeen of the A
ys are negative in each of the seventeen industries 61 is also negative and les
1 (algebraicly) than y so 61Y is greater than one his is not evidence that the
degree of cooperation is low or zero in any industry of course A high o togethegt
with a negative y which is small (in absolute terms) relative to a can give a
value of 61Y greater than one On the other hand if 61Y were less than one
o would have to be greater than zero
For y gt 0 two conditions may hold If y lt a it follows that 61 lt 0 since
(y - a ) lt 0 It also follows that 61Y lt 0 regardless of the value of o bull
In addition the relation is an if-and-only-if one ie if 61Y is negative
then y lt a Three industries have negative values for 61y dairy products
except milk (code 20 04) synthetic fibers (code 2806) and photographic equipment
(code 3808) In all three cases y is quite small so a does not have to be very
large to give a negative value to al
If y gt a the model shows that 61-- gt 0 whatever the value of 0 bull There are
12 cases where both y and 61 are positive with 61 ranging from near zero
(grain milling code 20 12) to 1 20 ( proprietary drugs code 2808) Several
coeffiei-en_rs
--
15
o f to zero
Tab le 3 - Consumer Goods
Indus tr ies
2Distributions of R and Producer Goods
Page 20a
Consumer Producer
Goods Goods
0 - 9 9 0 3
10 - 19 9 0 1 1
20 - 29 9 3 20
30 - 39 9 3 17
40 - 49 9 5 16
50 - 59 9 8
69 9 6 Ii 1060
--f 670 79 9 2
80 - 89 9 4 4
90 - 100 1 1
Totals 32 10 3
Notes
2is used R i s cal-
culated as F F + ((N - K- 1 ) K ] 1 b_ F is the F
statistic Lor the hypothesis that all o f the non - intercept terms are simul taneously equal
1
2
--
-- Since generalized least squ-area regress ion
Page 12
o f these cases have values o f 6 1 Y rea ter than 0 8 (pet food (20 10) bread amp cakes
(20 14 ) miscellaneous processed foods (20 29 ) men s and boys suit s and coats
(23 01 ) p rop riet ary drugs (28 08 ) elect ric housewares (36 1 2) radio and
IV sets 17 and games and toys (39 For be than(36 ) 04)) 6 1 Y to greater 0 8
both o and cry mus t b e less than 0 2 That is in these industries the results
shown in Tab le 2 would hold only with very strong dominance of marke t demand
effects over brand swit ching e ffects and a very low degree o f cooperation
Using the concentration rat ios given in Table 4 the average concentration
ratio for the eight indust ries is 38 1 For the remaining 24 industries it is
43 2 This result would lend some support t o the proposition that concentrat ion
and cooperation are related if the group of eight industries are identi fied
as having low degrees of cooperation
One f inal note on model predict ions S ince both cr and o are assumed to
be non-negative 60 should also be non-negat ive In only one o f the 32 cas es
is 60 negative and it is not signif icant bull
cor pJ
7 1
7 0 4
1 4 7 5 1
)
7 3 z a o a t s 1 1 3 3 5 ( i V Z 8 tJ1 2 1 - 1 1 0 7 3 5 c middot v 2 8 1 0 3 2 -0 0 1 4 3 0 C 1 V 9 8 3 2 8 1 5 1 6 - 0 0 6 t1 6 9 C _ V 1 11 0 2 9 0 1 29 0 00 7 6 1 C l k V 7 3 1 3 6 08 1 3 - o n 4 9 9 4 3 3 3 6 1 2 2 4 0 1 6 4 3
3 1 7 1 3 0 1 1 7 d 7 8 3 3 8 0 8 0 0 2 3 i 2 3 0 6
P 4
3 9 0 3 1 4
1 2
0 09 3 8 4 39 1)1 0 4 l 8 6 5
T A LE 4 e I NCU S T R Y C A T E G O R Y
F TC NOM e tOOE ass
C 1 I c 2 )
2 0 0 1 1 6 2 0 0r 1 7 zo o l Z 2 0 0 7 1 8
z o o 2 6
z o t a 1 6
2 1) 1 2 1 5 2 0 1 4 1 1 2 0 1 a 1 9 2 0 26 1 2 2 0 2 7 1 0 2 0 29 - 3 5
2 3 0 1 10
2 3 02 z o
2 3 03 1 9
2 5 5 1 2 4 Z 1 0 Z 1 1 2 0 3 1 9 2 7 4 1 0 z a o 1 3
2 A 0 7 z s
E LA S T I -C I T Y
c 3 )
- 0 09 7 2 7
0 00 449
- 0 0 2 8 9 3
- 0 3 2 7 70 - 0 OZ il I q
o 4 u t s a 0 0 1 3 7 3
o 1 4 o o-0 0 5 2 4o -0 1 94 0 1
-C l 6 l l 4 0 1 JJ 3 7 0 3 t 0 4 2 0 0 4 0 9
- 0 4 1 8 44 - O l 7 t ) 4 -ll 7 1 4 0 b -0 3 5 7 8 Z
- 0 1 9 7 7 0 1)0 3 3 4
-o n 9 2
IJ A T A
CONV
ON-C ONV
( 4 )
t 1NV
C 1t-V C NV c ) v C - JNV C rlNV c n v c a v C ON V ( i) l V
C CtV
C t1tV C IJNV
C 1 V
c 1 v
Pa ge 2La
HEQ F t Noe x
c 5 )
6 2q
6 89 1 48
6 1 4 ) q 3
1 9 7 1 9
q z t s 5 a e z
l l 9 1
4 o 7 1 2 0
c oqt 1 Z
4 4 )-q 1 z
5 5 9 1 0 64 1 3 3
n 3
Inter industry analysi
Page 23
Equation ( 3 6 ) may be used to fo rmulate a rough test o f the cooperat ion
vs e f ficiency controversy I f we rest ricted ourselves to some subset o f
indus tries where the relat ion be tween y and a i s the same we would be lookingbull
at a sample o f indus tries for which the dependence o f V on H and o is a
quadratic wi th no squared terms
I f we now sub s titute (26) into ( 36 ) we get
(38) v bull a + (y - a) R + (y - a) o (H) - (y - a) H o (H) + u
If it is only d if ferences in marke t shares that affect advertising inten-
s i ty levels i f concentration does not af fect cooperation and i f o bull 0
( 38 ) reduces t o
( 39 ) V bull a + (y - a) H + U
Let this b e the null hyp o thes is and note that V is a linear func-
tion o f H If y gt a and the null hyp othesis holds the expected relat ion
between V and H is as shown by the pos itively sloped straight line in Figure
6 b etween points A and C
In this context the alterna ive hypothesis the degree o f
cooperation ( o) is a positive function o f concentrat ion
is that
(H) I f the func-
tion is as described in (26) and Figure 4 the relation between V and H
would be as shown by the curved line connec t ing po ints A and C in Figure 6 One way to test the null hypothe s is against this al ternative is to test fo r
the hypothesized curvature in the function relating V and H
Fi gure
R
6
v
ID c
H 1 0
I
i I I
sfigned
clasJ s
in def ini g
effec t is the
I
Page middot2lf
One charac teristic of produc ts which may be useful
classes wi th respec t to significanc e of the brand-switching
conveniencenon-convenienc e distinc tion developed by Porter ( 1 97 )
Using his identif ication of IRS indus tries as a s tar t ing point
each of the 32 industries used in this s tudy to one of the two
The ass ignments are not ed in Tab le 4
Using the conveniencenon-convenienc e dis t inc tion the
divided into two sub-samples
sample was
Four regressions were then run bull ene for e ch
sub-sample one for the to tal sample and one for the total sample with
a dummy variable for convenience goods Each regression was run wi th LB s
and wi th indus tries as ob servations The results are given in Tab le 5
For the LB level equations ( 1A - 1D) equation(37) was used ome
rearranging of terms gives v bull y - ( 1-o ) (y - a) ( 1-z ) + ui i i
bull y - 6 1 ( 1 - z ) + ui When observations are pooled across indus trif s yi
is an industry variable In these regressions the values o f y es t imated
in the first stage were used Since industries have been grouped so as
to reduce differences in (y - a) that term is t reated as a cons tant
within each group Finally H the Herfindahl index has been subst ituted
for o the degree of cooperation The resultant equation is linear
y ( 1 - z ) and H x ( l - z ) This is the equation for whi ch results arei i
reported in Table 5 In all regress ions all variables were weighted by the
square roo t of sales
in
raquo
- 34
3 1
i l T 4 8LE S R E G E S S I ON R E SUL T S i X T f N S I ON S T O T Y PE Of
AND L E V E L OF C ONC E kT R A T I O
E O O tt S
tP bull T Y P E NR C ONV O B S tO N-CONV
E L A S T I shy
T I C J T Y
O E - t-41( T SliA R E
H( fl F
1 1 - K O x middot K O X E R F
- H S a 1 1 - H S t 1 8 1 1 9 1
R S O
1 1 0 1
q z
t - bull 9 f 2 1
Z t P D l 6 fC ONV - 1 9 1 s s c 2 1
e - l O I e bull u q a
2 3 1 0 1 1 Nl N- t O V 5 7 1 ( 1 2 19 1 6 1 - J l S 2 5 4
l l l l t z q o a 1 - J o v z I
l ( I NO l l A l l ol u l t o o c bull J u l l
d S
c 3 4 t t r a l
bull Z D 1 0 l l 4 l l 2 3 -gt 2 l l 2 0 - 4 lS 5 4 l C 4 1 o e 6 t l b b 5 1 1 0 1 1 1 c - 1 9 5 1 t d S 2 1 tV
OQ10 T-R A T I O S A E G I VE N l P A RE N T H E S E S l E V E L S f S I G 1 F l C ANC E A - 99 5 B - q 9 t - s NS J UC E NO I Nl E R C ( P T I S U (j f middot) I N f Hf t OUA TIOI S 6 E f 3R E A O J U S T H E N T F OR 1-t E T E R O SIC E O A S T f C I T V pIraquo
1
middot
I
R SO I S C A L C UL A T E D A S f I f bull 1 1 bull 1lt 1 K t t- E E N I S THf U A1 B E R O f fJe S E fi V A T I O S F t S HtE F - S T A T I S
I N DE P E ND E N T y II
T I C F O T H E U Y P O HIE S I S f A l l C JE F F I C H N T A R E QU A L HI l f RO A f40 K I S T H f UUM 8 E R OF Q ( A fL f S
I l l
1 A
1 6
t C
f 1 r R I l l
1 2 1
L Jt
L B
l q
L 1
on s T Y P E bull 2 )
P OUC T
X
H S I l l
l O l A I bull l 0 7 1
O O b O J
1 2 00 1
1 b 2 1 C I l l Z U
l l S b 1 -1 1 0 1
rlt F X l l - fi C f l
bull 1 I
l l
- ()4 2 3 l f H
1 - 2 7 9 1
K O X
1 1 - ti E R F I
l d l
S4 l i A I 2 l b
1 3 00
K O X -t f R F t S J
I 1 - HER F t
)
60 9
z o bull
0
t l l
3 8 8
1 7 6
5 64
64
N f bull
tl S bull
I l l
I 4 1
t ONV
NON-t ONV
A l l
A ll
c o v NO- ( ONV
1 4 1
1 5 1
- l S b
c - 1 1 2 1
5 8 2 C A I 1 6 7 0 1
middot I l 9 H
l 1 il
1 4 1 6 1
E L A C) T J shy
l t C I T Y
I S I
l b l
- 0 2 0
1 - 1 4 9 1
O Z 3 2 1 A I C b 6 b l
0 0 9 2 -J
1 1 b 4 1
o z t t t l 1 2 2 1 1
oru - Ht tt F
1 gt X bull b I
Page 25 bull
Mean val -as ior relevan t variables for the two sub-samples are
Conv Non-Conv
Adv Sales 4 7 2 3
iit Share 2 7 3 2
Ela cicity - 05 3 02 2
Herf indahl 7 3 1 860
That the dist nct ion between coft9enienee goods and non - convenience
goods mat ters is supported not only by these means but also by the regression
results Equat ions lA and lB are different in almost every respec t Only the
concentrat ionmarket share int eract ion term seems to mat ter for convenience
good s Jus t he opposite holds for non - convenience goods both elasticity1
and market sl e are highly significant but the concentrat ionmarket share L
int erac t ion t=rm is not
Given that one minus market share is used these equ3 tions the coeffici t
of market share is the negative of what is sh in col (6) of Table 5 For the
regress ion for non - convenience LB s reported in equation lB then LB s with larger
market shares have lower advertising to sales ratios
On the question of the impact of the degree of cooperation the story is mixed
For convenienc good s the coefficient is pos itive as equation (31) predicts when
there is a po sitive degree of cooperat ion Fo r the non-convenience goods LB s middot
concentration has no effec t
Aggregat ion t o the indus try level and then rer unning the equations $ives the
results shown in (2a-2d) of Tab le 5 As expected R2 goes up and levels of significanc e
fmiddotJr individual coefficients gJ down In add ition the coeffic ient of H ( l -H) inmiddot
quat ion 2A is no t significant though it is po s i tive Since 2A is the indus try level
I
--
all but one of them and that one is not significant For 18 of the 31
wo uld have on its adve rtising to sal es ratio
cognate of LA the variable H ( l-H) pl ays the same role in 2A as H ( l -z 1) p lays
in LA
Vconomies of scale in advert ising
The literature on advert ing contains a number of definitionbull o f economies
of scale and of findings eone rning its presence and magnitude Before I loo k
a t the data with the hope o f mmenting about scale effect s then I wan t to
define the term as I use it
By economies of scale I mean a decrease in the number o f units of adver-
tising which are needed to generate a unit of sales as advert ising is increased
Given the definition of the firm s own elast icity of demand with respect to
advertising (see the text foll wing equation (1) ) the technical definition
is that rii gt 1 Diseconomies of scale are def ined symmetrically i e
Yu lt 1
ith a cro s s - section oi observations on advertising intensity
(v bull a s1) and onmiddot advertising (a ) we may regress the former on the latter 1 i i
The results of such regress ions for the 32 industries are given in Table 2
lines lB ZB bull bull bull 32B For the 32 eases the regression coefficient is positive it
positive coefficients the coefficient i s signif icant
The que tion now is whether these results ar e evidence of pervasive dis-
economies of scale in advertislng I think no t Given my as sumption that the
data I observe are equilibrium results I may ri ghtly conclud e that almost every-
where high levels of advert ising are as sociated in equilibrium with high ratios
of advertising to sales I may no t conclud e anything however about the
impact that a move by some firm away from its equilibrium level o f advertising
middot- Page 27
Given the assumed relation in equation ( 37 ) there is a corre spondshy
ing equilibrium equation relating v and ai The equation is non - linear i
and I have not found a simp le way to characterize it It is po ssib le to
det ermi ne its deriviative with respect to a b y takin$ tne de riviative of ( 3 7 ) i
When that is done the result is
v - 1(40) i bull sl a a z ( 1-z )i i ira-l
Sinc e all the other terms in (40) are non-negative the sign dep ends only
on the sign of 6 1 which is in turn de pend ent only on the sign of (y - a)
That the signs of the coefficients of a in equations lB 2B bull bull bull 32B arei
not always equal to the signs of the coefficients of z in LA 2A bull bull bull 32A i
is still a mystery to me and something to be explored
bull
Summary
My purpose in writing this pa per wa s t o attempt to integrate explicit micro-
economic mode lling reasoned econome tric specification of error terms and high
quality data to exp lore some questions about adver tising by large firms
al quite important improvements could be made in the first two of those areas
and much data ma ssag ing is yet to be done in the third
Concerning the subst an ce of what I have done I think six things are im-
Page 2 8
Severshy
V and Conclusions
portant
1 Explicit modeling is worth the ef for t if for no other reas on than
that it provides a basis for choosing fr om a varie ty of funct ional
forms
2 The same is true for error specif ica t ion I did in fact look at scatshy
ter diagrams for all 32 industry categor ies and they vir tually all show
the het eroscedast icity which I as sumed and for which I correc ted
3 The predicted relation be tween advertis ing in tensity and market share
shows up clearly in on ly 25 of the case s examined I have ye t to
explore why that may b e the cas e
4 Some evidence concerning the presence and impact of coopera tion was
produced but it is not clear ly pervasive
5 The dis tinction between convenience goods and non-c onvenie nce goods is
unambiguously a good one
6 Virtually no evidence concerning econ omies of scale in adver tis ing can
be gleaned fr om this study given its assumpt ions
More work is called for on many of these issues my study seems to have
raised more questions than it has answer ed This is pr obab lly due to the richshy
ne ss of the data source since I had the oppor tunity to address several issue s
at once
- -- - Q --
Foo tnotes
Manager Line o f Business Program Federal Trade Commission The author depended heavily on several staf f members of the LB Pro gram for statistical and clerical support in particular Joe Cholka George Pascoe and Haro lene Jenkins Helpful comments were given b y Richard Caves Dennis Mueller Michael Porter and F M Scherer When the paper was in its early stages o f development both S teve Garber and Jon Rasmussen former economists in the LB Program provided eery useful cri tiques The views expressed here are my own o f course
1 The basic exp licit model of the role of advertising is Dorfman and S teiner ( 1954 ) and the models in this paper may be seen as ano ther extens ion of their work
2 For a review o f models of oligopoly which treat prices and quantities as relevant variables see Shubik ( 1 959) Extens ions to take advertis ing int o account are fairly s traightforward
3 Dorfman and S teiner dealt with a price-setting firm For a monopolis t inverting the demand relation to show Pbull P (Q A) _instead o f Q- (P A) has no effect on the results o f the analysis For oligopolists who face a sys tem of demand relations inversion does have some impac t on the results The dif ferences are not trivial nonetheless they will no t be explored in this study
4 Shubik ( 1 959) p
5 From a formal mathematical point o f view the function is similar to a1 i Legrangian funct ion of the form - r rr
L bull 11 + A ( j - (i i i+j
where the indicates some fixed level of the profits industry Following the same logic which resul ts in and income constraint as the marginal utility of the of the indus tr in the obj ective of firm i
6 Hall amp Weiss ( 1 967) p 323
7 See Schmallensee ( 1 97 2 ) Ch 2 pp 1 6-47
8 On this poin t see Imel amp Helmberger ( 1 9 7 1 )
9 Long ( 1 970) p 130 - 1 39
1 0 See Schmallensee ( 1 972) Ch 3 pp 48-8 7
rrj i+j ) of the rest of the
the identification profits of the rest
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980
l
(27)
Cl - o) o
[ (1 -
[ (1
t y ij z
j
are given in Figure 5
Page 13
The equation in (24) was constructed as a hybrid of two results derived
from first-order equations for the non-cooperative and cooperative models
It may also be constructed directly by using l bull (1 - o) + on as thei i
th 2objective function for the i-- firm Treating o as a constant the first
order equation is
0 an aa
i middot
n i
Using equations (3) and (7) and setting --- bull 0 givesaai
(28) - 0
After collecting terms and converting to matrix notation we get
(29) - Cl - o) + 0 z-1vi yii i
N
j bulll
+ 0
+ o v1
d-1 r z z
-(30) v Cl - o) rd
-
o) r +d
d r J _ -1- d z
and middot
d) r + 0 rjd
l r z
v - z --(31) zv
- (1 0- +d) t r z- d-
If we now impose the simplified demand structure by substituting (11)
we get
- (1 - o) a +( 32) + oy and
Figure
l
-y
v
Y
H
r
I I
v
a
y
H 10 H
v
a
bulla
HH 10
bullbull
(33)
i
Page 14
v - (1 - o) [ a l + (y - a) ] The corresponding equation for Vis given in (24)
If the relation between industry advertising intensity (V) and
concentration (H) is non - linear via the impact of concentration
on the degree of cooperation does the relation take the form of a
auadratic as some other investigators have proposed (Greer ( 1971 ) Martin (1979) Strickland amp Weiss (1976) ) That question can
be approached from two perspectives - conceptual and empirical In
the next section I will try to apply some statistical material to it
here I want to explore the question in the context of the models
developed above
Referring to equation (24) taking the deriviative with respect
to H and letting bull (H) gives
bull( 3 3a ) 3V (y - a) (1 - o) + (1 - H) 3H dH
Given that o and H each are bounded by 0 and 1 and that o is an in-
creasing function of H the term in the brackets in non-negative The
sign of the deriviative then is the same as the sign of (y - a) If the industry effect is dominant (y gt a) then advertising in-
tensity will increase with concentration If the two effects are equal
then advertising intensity will not vbullry with concentration and if the
brand-switching e ffect is larger advertising intensity will decrease
with concentration
For given advertising elasticity parameters then it is not pos-
sible to show an increase in A S up to some H and the n a decrease
he relation may be non - linear but it is monotonic over the permis-
sible range of H
Page 15
One possibility is to introduce a relation between (y - a) and H
If industries characterized as dominated by the brand-switching effect
also were more concentrated then a cross industry comparison would
show an inverted U shape relation between AS and H I know of no
support for such an association between ( y - a) and H
V ERROR SPECUCATION
So far I have implicitly assumed that there are no errors in the
model I now assume that the error
thvariable in the equation for the i- firm is additive that its mean is zero
that its variance is inversely proportionate to its sales and that all error
term covarlances are zero hat is
34) v bull (1 - o) [ a + (y - a) z ] + oy + u i i i
- - -
In vector notation
(35) bull (1 - o) [ at + lt - a) J +
E() bull 0 Cov (u) -
0
If we now ca lculate V bull bull we get
oy + U bull
(36) V bull (1 - o) la + (y - a)R J +
U bull E(U) bull 0 Var (U) bull
P age 16
The error term specifi c ation which follows Hall
wi th the ratio of profits
6 heir analysis seems
as wellbull particularly given
and Weiss was
applied by them to equa tions to equity or
asse ts as dependen t variables to carry the same
weight in the present context the substantial
evidence presented by marketing analysis on the tendency of company de-
cision makers t o use the advertising to sales ratio as the decision
7variable
The zero covariance assumption on the other hand1 is particularly
troublesome Within a probably not true since
many even ts outside the
single industry it is
industry would have
industries
similar effec ts on all the
8firms in the industry Between it is probably not true either
since many companies produce in more than one indus try and even ts in the
firm would tend to effect all of its activities The extension of the
model to allow for non-zero covariances is certainly called for
--- -
Page Ll
Athird characteristic of this specification also deserves some comment
only on equation is shown for the firm or industry Several haveinvestigators
included an advertising equation ina larger system of equations (Comanor and
Wilson (1974) (1979) and Weiss (1976) so in myMartin Strickland Idid
dissertation1 Ihave ewo defenses for presenting only single equation results
here One is the suggestion by Comanor Wilson that simultaneousand equation
verybias may not be important in this context he second is that Iam will-
ing to see the model extended to include a profitability equation and I intend
to move that direction in the near futurein
Onthe inclusion of aconcentration equation on the other hand I amskep-
tical he problem isthat in some fundamental sense equations in models of
industrial organization should be oriented to the decision-making contexts of firms
Ifeach firm in anindustry with Nfirms has only one decision variable then
there can be only Nindependent equations which are based on the first-order
optimization equations It not show large ofis difficult of course to a number
functional relations which follow from the first-order conditions point here he
isthat only Nof them can be independent
Consider for illustrative purposes only a situation in wich teh N
itsquantities say q are set exogeneously and each firm determines advertising1
of and level say ai where a is the true optimal value a a bulla
1 1 i i + i
Ifp a) then the Nq
s and the Na s will determine N p s say pii bullp
i(l i 1 i
Now given Eand 9 we can determine both the vector s andthe vector v since
- bull til
ands We can also determine S s z bull s-1 s-ii
v and we
industry concentration isHbull (alternatively
V and Hare functions of the same
pendent Furthermore if there are
VM) and (H1 H2 bull bull bull )contain
VI STATISTI CAL APPLICATIONS
Page 18
Given the observed values of might be tempted to conclude that we
have a two equation system in which v and z are jointly determined That would i i
be inappropriate however since the 2N random variables (v ) are determined by
onlt N basic random variables t The variance-covariance matrix for ( ) must
be singular bull
The extension of this observation to industry level variables is straightshy
forward Since observed industry advertising intensity is V bull l v and observed 4
C4 bull z ) it follows thati l i
set of random variables t and cannot be indeshy-
M industries then the variables (V1 v2 at most M independent random variables
Given the assumptions made above y a and o are constant within a
given industry That being the case (34) can be written as
(37)
s1 bull (1 - o) (y - a)
If and
So + +
(1 - o) a + oy
81 are esti mated for an industry using a suitable statisticalSo procedure giving and 81 an estimate of y is then available The sumSo
Intra-industry analysis
1974 of Business Re ort -
categories are identified
Page 19
of So and S1 is y so y bull So + S1 is the estimate of y
If o is assumed to be bounded by zero and one the sign of 61 is also
the sign of y - a The estimate of So is of cou se an estimate of
(1 - o ) a + oy Moreove if y gt a then a is less than 6o- (1 - B) a+ 0(
Cor esponding results would
hold if y bull a or y lt a
If 61 gt 0 then we would estimate that a lt 6o
bull
For purposes of estimating this model 32 manufacturing industry categories
were selected The data are fram the Line of Business forms filed with
the Federal Trade Commission they are discussed in detail in the Annual Line
which is available from the FTC The 32 industry
in Table 1
Simple regressions for equ ation (37) were run for each of the 32 industries
with data for each of the firms which filed in the industry constituting an ob-
servation The results are given in Table 2 lines lA 2A
1974
the intercept
For all the A
equations generalized least quares was used th both and market
share bein scaled bv the square root of sales
The 32 industries were taken from the larger set of all manufacturing industry
categories in the 1974 data files Two selection cr iteria were used thereLB
had to be at least ten reporting companies and the industry had to be a consumer
goods industry or a producer goods industry which is strongly associated with a
related consumer goods industry so that adve rtising by the firms in the producer
goods industry impacts on demand in the related industry (eg soft drink syrup
and soft drinks) There were 135 industries with at least ten companies - 103
producer goods and 32 consumer goods
273
274
TABLE 1 INDUSTRY CATEGORIES USED
FTC
COOE OESCRI P T I ON _
2001 MEAT PACKINGt SAUSAGE S ANO OTHER PREPARED
MEAT PROOUCTS
2004 DAIRY PRODUCTS EXC F LUID MILK
2005 CANNED SPECIALTIES
2001 FROZEN SPECIALTIES
OEHYORATFOt AND
I NC LU 0 ING PRE SERVE S
SOUP MIXESbull
SEASO lNGSt ANO
OTHER PET FOOJ
RAIN MILL PROOUCTSt
ANO PREPARED
RELATED PRODUCTS
PRODUCTS
Z008 CANNEOt ORIEOt PICKLED FRUITS
ANO VEGET ABLES JAMS JELLIES DEHYDRATED I G TA3t E
S40CES ANO SALAD ORESSINGS
2010 DOG CAT A O
2012 FlOUR t OTHER RICE
MILLING fiLENOE) FLOUR
2014 B EADt CAKE ANO
20113 CO FECTIONERY
2026 BOTTLED A O CANNED SOFT DRINKS 2021 FLAVORING EXTRACTS AND SYRUPS EC
2029 MISC FOOOS AND KINDRED PROJUCTS EXC
ROASTED COFFEE
2301 MENS AND BOYS SUITS AND COATS
2302 MENS ANO BOYS FURNISHINGS
230 3 kOMENS AND MISSES OUTER EAR
2551 HOUSEHOLD FURNITURE
2702 PERIODICALS
Pqe 19a 1
RELATED 1972
src cooe
2011 f 3
202tX2026
2032
203pound
2033t4t5
2047 2041t4t5
2051 2065 2086
2081
209XZ05
231
232
233
251 272
2703 BOOKS
2704 ISC PVeLISHING
2806 ORGANIC FlEERS
2807 DRUG$ ETHICAL
23234
PT 2 S 3
2808 ORUCSt PROPRIETARY
2809 PERFU ES t COSMETICS ANO OTHER TaiLET
PREPARATIONS
2810 SOAP AND OTHER CLEANING PREPARATIONS 2815
2901
PEST1CIDES AND AGR ICULTURAL
PE OLEUM REF I N ING CHEMICALS EC
3606 HOUSEHOLD COOKING EQUIPMENT
3612 HOUSEHOLD APPLIANC ES NEC INCLUJING ELECshy
rqlC HOUSE ARFS AND FANS ANO SE I G MACHINES
3617 RADIO AND T V RECE I VING SETS 3808 PHOTOG APHIC E UIP amp SUP LfESt EXC
3qo3
PHO TOCOPYING
SPOR TING A D E QU I P M ENT f SUP P L I E S ATHLETTC GOODS NEC
3904 DOLLS GAMESt TOYS AND CHLOqENtS VE ICLES
PT283
2844
234tX2844
287 9
291 3631
3634t6t9
36H
PT386l
3949
15
943
middot
l A6l f 2 R(GRESSION RESULTS
EQ I NO
NR CODE cu 121
lA 2001 A (
2A 2004 r (
3A 2005 a
-R
ClS amp31
16
17
1 z
I TER-CEPT bull t
ooo9t21At 00022311 oooeos uu
OOl49tCAI 0010 2 middot 000960CCI
o049tl1At o04t55C61
MARKET SH ARE bull 5 I
-010670Ctl
-017600181
-001050
-0211AO
-0077 0
AOVER-TIS INti
lbl
000()39 000072
000018 0 0024liAI
-000079
SHARf
4ampgtVf Q bull
111
000140
-000170
RSQ
COl
66 485 862
607 6 3 t ()2
blti6 633
flASTI-
CI T Y PH
-007271(1
000441
-u0288l
clASTSHU (t)Ef
bull 10
tto
-234
210
( OOlltOCCI -126520 000922 181 001790
4A 200 7 18 00590liAI -038670 4e16 -032110 lell u c
OOOS74e 000700 025940
00067018) 0021001AI -02l100CAI
5 26
920
5A 6 c
200 8 26 0 03Sl9CA I 0012 26181 o 024511(
-005540
-0 81420 0001961AI O()Ol90CAI ooll240
441 604 891
-0020 19 274
6A 2010 16 Oe04224CAI o359lOIAI 9oq ll4015dCAI 061 6 -middot 005435CAI 00011 8 1AI 917 c 00535ltAI -009850((1 OOOA51J CAI -001610161 971
7A 2012 001862 000010 351 00187) oo1 a oooq45 ono1o6 439 c 001109 -027050 Oe0067SIAI -OOll60C81 897
- middotmiddot- -
8A 201t ll 000757 Ol4140CCI 814 ott9oocet 095 8 oooqo6 C C J 000069(81 811 c ootl2lampBI -080910CCI Oe009lUBI -OOtlOOCCI 960I
I DIIJIT pound11 1V OTIUTVTfAll-1pound1t- A _ 00 R - qq r - Q
1jl)OQCD
0
-middot middot--
l
_ __ _
I
i
0 00141 c c
91)7
I
T ABLE 2 REG ESSIOH RESULTS CCONTI
EO NR
IND CODE
iR bull
cos
I TEq-
CEPT
t1ARKE T
SHA E
AOVER-
Tl SING
SHARpound-cent
AuVER
RSO ILAST1-
C I T Y ELAST
SH CPtf (lt Cl) bull 11 I 41 I 5I (61 171 I 8 I f9) c 10)
9A 2 0 1 8 19 002397181 -007640 367 -005 246 le46 6 000605 000550CAI 597
-000029 G14180 OOl654CAJ -024li01At
lOA 2026 12 0028901AI -OlZl90 -0114 0 1 1 1 5 759
Ll (
llA 2021
f3 c
12A lOZ9
B
(
13A 2301
B
c
14A 2302
B
c
154 201
n (
16A 2551
8
10
35
10
20
19
24
001897(()
00141 6CCI
Oll9lHCI
010604 middot middot
Oll083
OOJ11JCAt
002059141
001411CIH
-000412
000293(81
000187
001159 (3)
ooo6a4cct
o00792CCI
OOl40dCAt
000006
-000076
OOlltl6fA)
o o0 5 64CCt
-034570
-o1middot o3o
-1659 90
Ol51lO
-046190CAt
Ol5450CA)
-005l60
002940
-030060(()
-0 43l50CCt
0 33650
-019080
00004
bullJ00741CAI
-000069
000466
0001 51CAt
000947CAt
O OOI)70CAt OOl750CA J
000175CCI
000671CAI
OOll5) f A t 002207CBI
0004b7CA I
- 007 1 70
001760
-001410CAt
- 0 16200CAt
-004210
- 1 207 70
705
96 0
542
466 805
460
614
921
832
9 61
996
424
550
9 15
56 1 651
884
-006114
019 037
03504lCA)
oo409Q
-141644((
491 -0 17664
607
871
lell
084
101
072
103
108 11IllOQfl)-- l
-037880fCJ OOl6l4CAJ -040l70fCI 0()
c
I
I
bull 9
9t
middot middot middotbullmiddot bull lraquo
TABLE 2 REGRE SION RESUlTS ICONTt
bull
EO I NO tHl I filER- f4A_K T AOVER- sbulluRE ttSQ flA Jl- iLAS T NR CODE cos CEPT SHARE JISING AOVfR bull (JTY SHR CiJCF
110Cl I f amp Itt 15J tbt bull 1c u caa
llA 2702 ll o079 6 81 Bt -019110 S6s -OlltO 111
a 004679 ooooo1 t69
( oOl26o -015160 001560 -022690 ltJr()I
ldA 270 3 19 0065lt -OIt2l60 220 -Ol57a2 11 8 e o 0 71t6IA t ol08 75tAt ( OOl983CiU -060llOC() OOl7191A t -O lltdOC A) middot 89
lqA 2704 10 005082 -Oil060 22b -l)i1977 lelO 6 -000080 oo2772C4 t 910 c 000907 -011 200 Oe04105CA t -046960 995
20A 2d06 1 II
O OlltStiAt -001120 837 00014 -))th O01380amp At -oooo 11 821 ( OOl526CAt -026700181 0002 5UAt 000530 969
2lA 2807 28 OOt9lt9CAt -00891 0 516 -0 03962 225I8 002095CCt OOOl lliAt 690 ( o02965CA t -056650((1 OOOt50CAJ -003l70ttl 911I
I 22A 2808 15 ot3l06 119730 655 1130) 090
0 - 007526CC t Oe003961At 835- middot-
c I oo7523 -067210 001l 2tA t -0091t80CCt 951
2lA 2809 21 Oe2067lCAt -1218101At 71 2 -10095((1 121 ttlIaOQ8 00797418) OOOI8lltt 686 Ill( 0092lU A t -Olltl70Ct t 000605CAt -OOit8701At 951 _
---- -
32 009506CAt
0 bull 06 981 I A J 006901 cA bull
00 24A 2810 -010960 5t 5 -OOltSO 7 56
8 -000011 s11 c -5811701At oooaoatAJ oot520IAI 888
-- 37 --- middot - -- -1 ) l lbull t5JJ r-l bullhmiddot1 --- - --
l5A
26A
27A
c
2RA
c
2 9A
31A
llA
24
13
597
Ci)55
14
l 6h
RSQ ELASTI- (lASTCOE f-
TABLE 2
EQ I NO NR CODE IU
REGRESSION RESULTS CCONTa
NR INTER- MARKET
(10S CEPT SHARE AOVER- SHARfcent
TIS lNG AOVER
) CITY StiR
c It) bull 5 co a I 1) bull 0 bull 9 c l 0 bull( 2 1
2815 16 003019 -009990 275 -o069bmiddotJ
6 001446 000224 282 c 002178 -054630 o02221CCI -0075 00 646
2901 29 00021t2 CAI 0)0520 592 0007 6 1 068
0 00105C8 t 000012141 60)g( OOOlJ51AJ -00126018) ooou271At -000u80CA) 974
36oa 1 3 003296(4) -oca29o -oo-q94
001755 000203 lt91t 002499((1 -04l820CCI OOl374CAI 0068)0 946
3612 0033281CI 011520 195 01684) oso
00 30 42C OI 00012 2 424 oo579ltAt -2 7l7801At 002 100IAI -003220 819
8 3617 btel 011789 081002259 009530
001448 oOOl OitCCI 732 OOl7 22CAI -0 46160CAI 0004961AI -OOll40181 989c
30A 3808 14 0021t 3 6fAI -000110 628 o023291C) -005 6 middot- _ middot- 0bull 02 04 1 IC I ooooo9 637 ( 001584((1 -0161)1)0 00021 91AI -0 00060
390) 14 002790CCt 006590 h8 t 0 0938 4 070
B OOl89bCCJ 000144 74
0021961(1 -OItlOlO OOl323CAI -004800 915
B
middot--- - middot 4 - middot- - dfll3904 12 004563181 0)8300 879 0428651(1 089 OQIll004 277CA I ooozstcea 919
c 0041)3181 -0-11250 0011881(1 -006270 972 0Ill
- J - bull bull bull bullbullbull middot - - middot - middot gt middotbull L)ofbullflbull middot middotfJltl -tn- middotmiddotmiddotmiddoto +f middot t middot bull
c
is
positive
2
Page 0 determineThe consumer goodproducer good split is sometimes difficult to
Sollle support for the split used in this paper is a comparison of goodness - of - fit
distributions of a
industries
set
advertising are
that industry
a sales increase of
the
highest aD10U3 those
measures for the two groups of industries Table 3 contains
measures for the 32- consumer goods industries and the 103 producer goods
2 Median R is 5 7 for the consumer goods set and 40 for the producer goods
Estimates - of y the elasticity of demand with respect to
given in column of Table 2 The hypothesis that y bull 0 with
alternative was applied for each ittdustry As shown in the
could not be rtjected for 24 of the 32 industries Of the
and 10 are positive For the remalnifg eight three are
and five are significantly positive
Of the threertegative elasetcities the largest (in absolute value)
cosmetics (code 2809) With all elasticity of -1 01 a one
industry advertising would gwerate a decrease in industry sales
more than one percetu
At the other extrema is toys and games (code 39 04) For
a one percent ittcrease in industry advertising would lead to
four-tenths of a pampreentmiddot Though not significantly different from zero
elasticity estimatbull for pt-bpuetary drugs (code 2808)
which are
Both drug industries have
to sales rafios (211 pe1 cent for 28 08 and 13 4 per cent
other hand flaIJo ing ext1ac s (mainly soft-drink SYUlS) also
is for
percent increase in
of slightly
a middottwo - tailed
table that hypothesis
24 14 are negative
significantly negative
9
the codetics and p-ropriacuy very high advertising
for 28 09) bull On the
has a high ind UY
advenisiltg t() sales ratio ( 8 0 per cent) bull but its estimated elasticity is very
small and not signifieantly different uom zero
middot It doe$ not seem unusual to find many elasticities near zero with a few
negative and a few positive If aggregate consumption is insensitive to aggregate
advertising as seems likely ld then what one industry gains must be lost by
others
Page 21
elasticity of demand does
(10) we can
if t 0 then
Finally a determination concerning the industry
not fix the individual firms elasticity Going back to equation
write y bull (1 - ) a + z y Since a gt 0 it follows thatii i i
Y gt 0 If Ylt 0 however y s sign is indeterminateij ij
Turning next to the coefficient of market share 20 of them are negative and
12 are positive with three of each sign being significantly different from zero
The largest positive value which is significant is for pet food (code 2010) and
a close second is mens amp boys suits and coats (code 23 01) The largest signifi
cant negative value is for cosmetics (code 28 09)
In the model some relations among y a 0 and 61 are implied If middota is
negative (1- y) is greater than one Since 61Y bull (1- a) (1 - ) 61Y is
pCsitive but it will be less than one if 0 is large enough Seventeen of the A
ys are negative in each of the seventeen industries 61 is also negative and les
1 (algebraicly) than y so 61Y is greater than one his is not evidence that the
degree of cooperation is low or zero in any industry of course A high o togethegt
with a negative y which is small (in absolute terms) relative to a can give a
value of 61Y greater than one On the other hand if 61Y were less than one
o would have to be greater than zero
For y gt 0 two conditions may hold If y lt a it follows that 61 lt 0 since
(y - a ) lt 0 It also follows that 61Y lt 0 regardless of the value of o bull
In addition the relation is an if-and-only-if one ie if 61Y is negative
then y lt a Three industries have negative values for 61y dairy products
except milk (code 20 04) synthetic fibers (code 2806) and photographic equipment
(code 3808) In all three cases y is quite small so a does not have to be very
large to give a negative value to al
If y gt a the model shows that 61-- gt 0 whatever the value of 0 bull There are
12 cases where both y and 61 are positive with 61 ranging from near zero
(grain milling code 20 12) to 1 20 ( proprietary drugs code 2808) Several
coeffiei-en_rs
--
15
o f to zero
Tab le 3 - Consumer Goods
Indus tr ies
2Distributions of R and Producer Goods
Page 20a
Consumer Producer
Goods Goods
0 - 9 9 0 3
10 - 19 9 0 1 1
20 - 29 9 3 20
30 - 39 9 3 17
40 - 49 9 5 16
50 - 59 9 8
69 9 6 Ii 1060
--f 670 79 9 2
80 - 89 9 4 4
90 - 100 1 1
Totals 32 10 3
Notes
2is used R i s cal-
culated as F F + ((N - K- 1 ) K ] 1 b_ F is the F
statistic Lor the hypothesis that all o f the non - intercept terms are simul taneously equal
1
2
--
-- Since generalized least squ-area regress ion
Page 12
o f these cases have values o f 6 1 Y rea ter than 0 8 (pet food (20 10) bread amp cakes
(20 14 ) miscellaneous processed foods (20 29 ) men s and boys suit s and coats
(23 01 ) p rop riet ary drugs (28 08 ) elect ric housewares (36 1 2) radio and
IV sets 17 and games and toys (39 For be than(36 ) 04)) 6 1 Y to greater 0 8
both o and cry mus t b e less than 0 2 That is in these industries the results
shown in Tab le 2 would hold only with very strong dominance of marke t demand
effects over brand swit ching e ffects and a very low degree o f cooperation
Using the concentration rat ios given in Table 4 the average concentration
ratio for the eight indust ries is 38 1 For the remaining 24 industries it is
43 2 This result would lend some support t o the proposition that concentrat ion
and cooperation are related if the group of eight industries are identi fied
as having low degrees of cooperation
One f inal note on model predict ions S ince both cr and o are assumed to
be non-negative 60 should also be non-negat ive In only one o f the 32 cas es
is 60 negative and it is not signif icant bull
cor pJ
7 1
7 0 4
1 4 7 5 1
)
7 3 z a o a t s 1 1 3 3 5 ( i V Z 8 tJ1 2 1 - 1 1 0 7 3 5 c middot v 2 8 1 0 3 2 -0 0 1 4 3 0 C 1 V 9 8 3 2 8 1 5 1 6 - 0 0 6 t1 6 9 C _ V 1 11 0 2 9 0 1 29 0 00 7 6 1 C l k V 7 3 1 3 6 08 1 3 - o n 4 9 9 4 3 3 3 6 1 2 2 4 0 1 6 4 3
3 1 7 1 3 0 1 1 7 d 7 8 3 3 8 0 8 0 0 2 3 i 2 3 0 6
P 4
3 9 0 3 1 4
1 2
0 09 3 8 4 39 1)1 0 4 l 8 6 5
T A LE 4 e I NCU S T R Y C A T E G O R Y
F TC NOM e tOOE ass
C 1 I c 2 )
2 0 0 1 1 6 2 0 0r 1 7 zo o l Z 2 0 0 7 1 8
z o o 2 6
z o t a 1 6
2 1) 1 2 1 5 2 0 1 4 1 1 2 0 1 a 1 9 2 0 26 1 2 2 0 2 7 1 0 2 0 29 - 3 5
2 3 0 1 10
2 3 02 z o
2 3 03 1 9
2 5 5 1 2 4 Z 1 0 Z 1 1 2 0 3 1 9 2 7 4 1 0 z a o 1 3
2 A 0 7 z s
E LA S T I -C I T Y
c 3 )
- 0 09 7 2 7
0 00 449
- 0 0 2 8 9 3
- 0 3 2 7 70 - 0 OZ il I q
o 4 u t s a 0 0 1 3 7 3
o 1 4 o o-0 0 5 2 4o -0 1 94 0 1
-C l 6 l l 4 0 1 JJ 3 7 0 3 t 0 4 2 0 0 4 0 9
- 0 4 1 8 44 - O l 7 t ) 4 -ll 7 1 4 0 b -0 3 5 7 8 Z
- 0 1 9 7 7 0 1)0 3 3 4
-o n 9 2
IJ A T A
CONV
ON-C ONV
( 4 )
t 1NV
C 1t-V C NV c ) v C - JNV C rlNV c n v c a v C ON V ( i) l V
C CtV
C t1tV C IJNV
C 1 V
c 1 v
Pa ge 2La
HEQ F t Noe x
c 5 )
6 2q
6 89 1 48
6 1 4 ) q 3
1 9 7 1 9
q z t s 5 a e z
l l 9 1
4 o 7 1 2 0
c oqt 1 Z
4 4 )-q 1 z
5 5 9 1 0 64 1 3 3
n 3
Inter industry analysi
Page 23
Equation ( 3 6 ) may be used to fo rmulate a rough test o f the cooperat ion
vs e f ficiency controversy I f we rest ricted ourselves to some subset o f
indus tries where the relat ion be tween y and a i s the same we would be lookingbull
at a sample o f indus tries for which the dependence o f V on H and o is a
quadratic wi th no squared terms
I f we now sub s titute (26) into ( 36 ) we get
(38) v bull a + (y - a) R + (y - a) o (H) - (y - a) H o (H) + u
If it is only d if ferences in marke t shares that affect advertising inten-
s i ty levels i f concentration does not af fect cooperation and i f o bull 0
( 38 ) reduces t o
( 39 ) V bull a + (y - a) H + U
Let this b e the null hyp o thes is and note that V is a linear func-
tion o f H If y gt a and the null hyp othesis holds the expected relat ion
between V and H is as shown by the pos itively sloped straight line in Figure
6 b etween points A and C
In this context the alterna ive hypothesis the degree o f
cooperation ( o) is a positive function o f concentrat ion
is that
(H) I f the func-
tion is as described in (26) and Figure 4 the relation between V and H
would be as shown by the curved line connec t ing po ints A and C in Figure 6 One way to test the null hypothe s is against this al ternative is to test fo r
the hypothesized curvature in the function relating V and H
Fi gure
R
6
v
ID c
H 1 0
I
i I I
sfigned
clasJ s
in def ini g
effec t is the
I
Page middot2lf
One charac teristic of produc ts which may be useful
classes wi th respec t to significanc e of the brand-switching
conveniencenon-convenienc e distinc tion developed by Porter ( 1 97 )
Using his identif ication of IRS indus tries as a s tar t ing point
each of the 32 industries used in this s tudy to one of the two
The ass ignments are not ed in Tab le 4
Using the conveniencenon-convenienc e dis t inc tion the
divided into two sub-samples
sample was
Four regressions were then run bull ene for e ch
sub-sample one for the to tal sample and one for the total sample with
a dummy variable for convenience goods Each regression was run wi th LB s
and wi th indus tries as ob servations The results are given in Tab le 5
For the LB level equations ( 1A - 1D) equation(37) was used ome
rearranging of terms gives v bull y - ( 1-o ) (y - a) ( 1-z ) + ui i i
bull y - 6 1 ( 1 - z ) + ui When observations are pooled across indus trif s yi
is an industry variable In these regressions the values o f y es t imated
in the first stage were used Since industries have been grouped so as
to reduce differences in (y - a) that term is t reated as a cons tant
within each group Finally H the Herfindahl index has been subst ituted
for o the degree of cooperation The resultant equation is linear
y ( 1 - z ) and H x ( l - z ) This is the equation for whi ch results arei i
reported in Table 5 In all regress ions all variables were weighted by the
square roo t of sales
in
raquo
- 34
3 1
i l T 4 8LE S R E G E S S I ON R E SUL T S i X T f N S I ON S T O T Y PE Of
AND L E V E L OF C ONC E kT R A T I O
E O O tt S
tP bull T Y P E NR C ONV O B S tO N-CONV
E L A S T I shy
T I C J T Y
O E - t-41( T SliA R E
H( fl F
1 1 - K O x middot K O X E R F
- H S a 1 1 - H S t 1 8 1 1 9 1
R S O
1 1 0 1
q z
t - bull 9 f 2 1
Z t P D l 6 fC ONV - 1 9 1 s s c 2 1
e - l O I e bull u q a
2 3 1 0 1 1 Nl N- t O V 5 7 1 ( 1 2 19 1 6 1 - J l S 2 5 4
l l l l t z q o a 1 - J o v z I
l ( I NO l l A l l ol u l t o o c bull J u l l
d S
c 3 4 t t r a l
bull Z D 1 0 l l 4 l l 2 3 -gt 2 l l 2 0 - 4 lS 5 4 l C 4 1 o e 6 t l b b 5 1 1 0 1 1 1 c - 1 9 5 1 t d S 2 1 tV
OQ10 T-R A T I O S A E G I VE N l P A RE N T H E S E S l E V E L S f S I G 1 F l C ANC E A - 99 5 B - q 9 t - s NS J UC E NO I Nl E R C ( P T I S U (j f middot) I N f Hf t OUA TIOI S 6 E f 3R E A O J U S T H E N T F OR 1-t E T E R O SIC E O A S T f C I T V pIraquo
1
middot
I
R SO I S C A L C UL A T E D A S f I f bull 1 1 bull 1lt 1 K t t- E E N I S THf U A1 B E R O f fJe S E fi V A T I O S F t S HtE F - S T A T I S
I N DE P E ND E N T y II
T I C F O T H E U Y P O HIE S I S f A l l C JE F F I C H N T A R E QU A L HI l f RO A f40 K I S T H f UUM 8 E R OF Q ( A fL f S
I l l
1 A
1 6
t C
f 1 r R I l l
1 2 1
L Jt
L B
l q
L 1
on s T Y P E bull 2 )
P OUC T
X
H S I l l
l O l A I bull l 0 7 1
O O b O J
1 2 00 1
1 b 2 1 C I l l Z U
l l S b 1 -1 1 0 1
rlt F X l l - fi C f l
bull 1 I
l l
- ()4 2 3 l f H
1 - 2 7 9 1
K O X
1 1 - ti E R F I
l d l
S4 l i A I 2 l b
1 3 00
K O X -t f R F t S J
I 1 - HER F t
)
60 9
z o bull
0
t l l
3 8 8
1 7 6
5 64
64
N f bull
tl S bull
I l l
I 4 1
t ONV
NON-t ONV
A l l
A ll
c o v NO- ( ONV
1 4 1
1 5 1
- l S b
c - 1 1 2 1
5 8 2 C A I 1 6 7 0 1
middot I l 9 H
l 1 il
1 4 1 6 1
E L A C) T J shy
l t C I T Y
I S I
l b l
- 0 2 0
1 - 1 4 9 1
O Z 3 2 1 A I C b 6 b l
0 0 9 2 -J
1 1 b 4 1
o z t t t l 1 2 2 1 1
oru - Ht tt F
1 gt X bull b I
Page 25 bull
Mean val -as ior relevan t variables for the two sub-samples are
Conv Non-Conv
Adv Sales 4 7 2 3
iit Share 2 7 3 2
Ela cicity - 05 3 02 2
Herf indahl 7 3 1 860
That the dist nct ion between coft9enienee goods and non - convenience
goods mat ters is supported not only by these means but also by the regression
results Equat ions lA and lB are different in almost every respec t Only the
concentrat ionmarket share int eract ion term seems to mat ter for convenience
good s Jus t he opposite holds for non - convenience goods both elasticity1
and market sl e are highly significant but the concentrat ionmarket share L
int erac t ion t=rm is not
Given that one minus market share is used these equ3 tions the coeffici t
of market share is the negative of what is sh in col (6) of Table 5 For the
regress ion for non - convenience LB s reported in equation lB then LB s with larger
market shares have lower advertising to sales ratios
On the question of the impact of the degree of cooperation the story is mixed
For convenienc good s the coefficient is pos itive as equation (31) predicts when
there is a po sitive degree of cooperat ion Fo r the non-convenience goods LB s middot
concentration has no effec t
Aggregat ion t o the indus try level and then rer unning the equations $ives the
results shown in (2a-2d) of Tab le 5 As expected R2 goes up and levels of significanc e
fmiddotJr individual coefficients gJ down In add ition the coeffic ient of H ( l -H) inmiddot
quat ion 2A is no t significant though it is po s i tive Since 2A is the indus try level
I
--
all but one of them and that one is not significant For 18 of the 31
wo uld have on its adve rtising to sal es ratio
cognate of LA the variable H ( l-H) pl ays the same role in 2A as H ( l -z 1) p lays
in LA
Vconomies of scale in advert ising
The literature on advert ing contains a number of definitionbull o f economies
of scale and of findings eone rning its presence and magnitude Before I loo k
a t the data with the hope o f mmenting about scale effect s then I wan t to
define the term as I use it
By economies of scale I mean a decrease in the number o f units of adver-
tising which are needed to generate a unit of sales as advert ising is increased
Given the definition of the firm s own elast icity of demand with respect to
advertising (see the text foll wing equation (1) ) the technical definition
is that rii gt 1 Diseconomies of scale are def ined symmetrically i e
Yu lt 1
ith a cro s s - section oi observations on advertising intensity
(v bull a s1) and onmiddot advertising (a ) we may regress the former on the latter 1 i i
The results of such regress ions for the 32 industries are given in Table 2
lines lB ZB bull bull bull 32B For the 32 eases the regression coefficient is positive it
positive coefficients the coefficient i s signif icant
The que tion now is whether these results ar e evidence of pervasive dis-
economies of scale in advertislng I think no t Given my as sumption that the
data I observe are equilibrium results I may ri ghtly conclud e that almost every-
where high levels of advert ising are as sociated in equilibrium with high ratios
of advertising to sales I may no t conclud e anything however about the
impact that a move by some firm away from its equilibrium level o f advertising
middot- Page 27
Given the assumed relation in equation ( 37 ) there is a corre spondshy
ing equilibrium equation relating v and ai The equation is non - linear i
and I have not found a simp le way to characterize it It is po ssib le to
det ermi ne its deriviative with respect to a b y takin$ tne de riviative of ( 3 7 ) i
When that is done the result is
v - 1(40) i bull sl a a z ( 1-z )i i ira-l
Sinc e all the other terms in (40) are non-negative the sign dep ends only
on the sign of 6 1 which is in turn de pend ent only on the sign of (y - a)
That the signs of the coefficients of a in equations lB 2B bull bull bull 32B arei
not always equal to the signs of the coefficients of z in LA 2A bull bull bull 32A i
is still a mystery to me and something to be explored
bull
Summary
My purpose in writing this pa per wa s t o attempt to integrate explicit micro-
economic mode lling reasoned econome tric specification of error terms and high
quality data to exp lore some questions about adver tising by large firms
al quite important improvements could be made in the first two of those areas
and much data ma ssag ing is yet to be done in the third
Concerning the subst an ce of what I have done I think six things are im-
Page 2 8
Severshy
V and Conclusions
portant
1 Explicit modeling is worth the ef for t if for no other reas on than
that it provides a basis for choosing fr om a varie ty of funct ional
forms
2 The same is true for error specif ica t ion I did in fact look at scatshy
ter diagrams for all 32 industry categor ies and they vir tually all show
the het eroscedast icity which I as sumed and for which I correc ted
3 The predicted relation be tween advertis ing in tensity and market share
shows up clearly in on ly 25 of the case s examined I have ye t to
explore why that may b e the cas e
4 Some evidence concerning the presence and impact of coopera tion was
produced but it is not clear ly pervasive
5 The dis tinction between convenience goods and non-c onvenie nce goods is
unambiguously a good one
6 Virtually no evidence concerning econ omies of scale in adver tis ing can
be gleaned fr om this study given its assumpt ions
More work is called for on many of these issues my study seems to have
raised more questions than it has answer ed This is pr obab lly due to the richshy
ne ss of the data source since I had the oppor tunity to address several issue s
at once
- -- - Q --
Foo tnotes
Manager Line o f Business Program Federal Trade Commission The author depended heavily on several staf f members of the LB Pro gram for statistical and clerical support in particular Joe Cholka George Pascoe and Haro lene Jenkins Helpful comments were given b y Richard Caves Dennis Mueller Michael Porter and F M Scherer When the paper was in its early stages o f development both S teve Garber and Jon Rasmussen former economists in the LB Program provided eery useful cri tiques The views expressed here are my own o f course
1 The basic exp licit model of the role of advertising is Dorfman and S teiner ( 1954 ) and the models in this paper may be seen as ano ther extens ion of their work
2 For a review o f models of oligopoly which treat prices and quantities as relevant variables see Shubik ( 1 959) Extens ions to take advertis ing int o account are fairly s traightforward
3 Dorfman and S teiner dealt with a price-setting firm For a monopolis t inverting the demand relation to show Pbull P (Q A) _instead o f Q- (P A) has no effect on the results o f the analysis For oligopolists who face a sys tem of demand relations inversion does have some impac t on the results The dif ferences are not trivial nonetheless they will no t be explored in this study
4 Shubik ( 1 959) p
5 From a formal mathematical point o f view the function is similar to a1 i Legrangian funct ion of the form - r rr
L bull 11 + A ( j - (i i i+j
where the indicates some fixed level of the profits industry Following the same logic which resul ts in and income constraint as the marginal utility of the of the indus tr in the obj ective of firm i
6 Hall amp Weiss ( 1 967) p 323
7 See Schmallensee ( 1 97 2 ) Ch 2 pp 1 6-47
8 On this poin t see Imel amp Helmberger ( 1 9 7 1 )
9 Long ( 1 970) p 130 - 1 39
1 0 See Schmallensee ( 1 972) Ch 3 pp 48-8 7
rrj i+j ) of the rest of the
the identification profits of the rest
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980
Figure
l
-y
v
Y
H
r
I I
v
a
y
H 10 H
v
a
bulla
HH 10
bullbull
(33)
i
Page 14
v - (1 - o) [ a l + (y - a) ] The corresponding equation for Vis given in (24)
If the relation between industry advertising intensity (V) and
concentration (H) is non - linear via the impact of concentration
on the degree of cooperation does the relation take the form of a
auadratic as some other investigators have proposed (Greer ( 1971 ) Martin (1979) Strickland amp Weiss (1976) ) That question can
be approached from two perspectives - conceptual and empirical In
the next section I will try to apply some statistical material to it
here I want to explore the question in the context of the models
developed above
Referring to equation (24) taking the deriviative with respect
to H and letting bull (H) gives
bull( 3 3a ) 3V (y - a) (1 - o) + (1 - H) 3H dH
Given that o and H each are bounded by 0 and 1 and that o is an in-
creasing function of H the term in the brackets in non-negative The
sign of the deriviative then is the same as the sign of (y - a) If the industry effect is dominant (y gt a) then advertising in-
tensity will increase with concentration If the two effects are equal
then advertising intensity will not vbullry with concentration and if the
brand-switching e ffect is larger advertising intensity will decrease
with concentration
For given advertising elasticity parameters then it is not pos-
sible to show an increase in A S up to some H and the n a decrease
he relation may be non - linear but it is monotonic over the permis-
sible range of H
Page 15
One possibility is to introduce a relation between (y - a) and H
If industries characterized as dominated by the brand-switching effect
also were more concentrated then a cross industry comparison would
show an inverted U shape relation between AS and H I know of no
support for such an association between ( y - a) and H
V ERROR SPECUCATION
So far I have implicitly assumed that there are no errors in the
model I now assume that the error
thvariable in the equation for the i- firm is additive that its mean is zero
that its variance is inversely proportionate to its sales and that all error
term covarlances are zero hat is
34) v bull (1 - o) [ a + (y - a) z ] + oy + u i i i
- - -
In vector notation
(35) bull (1 - o) [ at + lt - a) J +
E() bull 0 Cov (u) -
0
If we now ca lculate V bull bull we get
oy + U bull
(36) V bull (1 - o) la + (y - a)R J +
U bull E(U) bull 0 Var (U) bull
P age 16
The error term specifi c ation which follows Hall
wi th the ratio of profits
6 heir analysis seems
as wellbull particularly given
and Weiss was
applied by them to equa tions to equity or
asse ts as dependen t variables to carry the same
weight in the present context the substantial
evidence presented by marketing analysis on the tendency of company de-
cision makers t o use the advertising to sales ratio as the decision
7variable
The zero covariance assumption on the other hand1 is particularly
troublesome Within a probably not true since
many even ts outside the
single industry it is
industry would have
industries
similar effec ts on all the
8firms in the industry Between it is probably not true either
since many companies produce in more than one indus try and even ts in the
firm would tend to effect all of its activities The extension of the
model to allow for non-zero covariances is certainly called for
--- -
Page Ll
Athird characteristic of this specification also deserves some comment
only on equation is shown for the firm or industry Several haveinvestigators
included an advertising equation ina larger system of equations (Comanor and
Wilson (1974) (1979) and Weiss (1976) so in myMartin Strickland Idid
dissertation1 Ihave ewo defenses for presenting only single equation results
here One is the suggestion by Comanor Wilson that simultaneousand equation
verybias may not be important in this context he second is that Iam will-
ing to see the model extended to include a profitability equation and I intend
to move that direction in the near futurein
Onthe inclusion of aconcentration equation on the other hand I amskep-
tical he problem isthat in some fundamental sense equations in models of
industrial organization should be oriented to the decision-making contexts of firms
Ifeach firm in anindustry with Nfirms has only one decision variable then
there can be only Nindependent equations which are based on the first-order
optimization equations It not show large ofis difficult of course to a number
functional relations which follow from the first-order conditions point here he
isthat only Nof them can be independent
Consider for illustrative purposes only a situation in wich teh N
itsquantities say q are set exogeneously and each firm determines advertising1
of and level say ai where a is the true optimal value a a bulla
1 1 i i + i
Ifp a) then the Nq
s and the Na s will determine N p s say pii bullp
i(l i 1 i
Now given Eand 9 we can determine both the vector s andthe vector v since
- bull til
ands We can also determine S s z bull s-1 s-ii
v and we
industry concentration isHbull (alternatively
V and Hare functions of the same
pendent Furthermore if there are
VM) and (H1 H2 bull bull bull )contain
VI STATISTI CAL APPLICATIONS
Page 18
Given the observed values of might be tempted to conclude that we
have a two equation system in which v and z are jointly determined That would i i
be inappropriate however since the 2N random variables (v ) are determined by
onlt N basic random variables t The variance-covariance matrix for ( ) must
be singular bull
The extension of this observation to industry level variables is straightshy
forward Since observed industry advertising intensity is V bull l v and observed 4
C4 bull z ) it follows thati l i
set of random variables t and cannot be indeshy-
M industries then the variables (V1 v2 at most M independent random variables
Given the assumptions made above y a and o are constant within a
given industry That being the case (34) can be written as
(37)
s1 bull (1 - o) (y - a)
If and
So + +
(1 - o) a + oy
81 are esti mated for an industry using a suitable statisticalSo procedure giving and 81 an estimate of y is then available The sumSo
Intra-industry analysis
1974 of Business Re ort -
categories are identified
Page 19
of So and S1 is y so y bull So + S1 is the estimate of y
If o is assumed to be bounded by zero and one the sign of 61 is also
the sign of y - a The estimate of So is of cou se an estimate of
(1 - o ) a + oy Moreove if y gt a then a is less than 6o- (1 - B) a+ 0(
Cor esponding results would
hold if y bull a or y lt a
If 61 gt 0 then we would estimate that a lt 6o
bull
For purposes of estimating this model 32 manufacturing industry categories
were selected The data are fram the Line of Business forms filed with
the Federal Trade Commission they are discussed in detail in the Annual Line
which is available from the FTC The 32 industry
in Table 1
Simple regressions for equ ation (37) were run for each of the 32 industries
with data for each of the firms which filed in the industry constituting an ob-
servation The results are given in Table 2 lines lA 2A
1974
the intercept
For all the A
equations generalized least quares was used th both and market
share bein scaled bv the square root of sales
The 32 industries were taken from the larger set of all manufacturing industry
categories in the 1974 data files Two selection cr iteria were used thereLB
had to be at least ten reporting companies and the industry had to be a consumer
goods industry or a producer goods industry which is strongly associated with a
related consumer goods industry so that adve rtising by the firms in the producer
goods industry impacts on demand in the related industry (eg soft drink syrup
and soft drinks) There were 135 industries with at least ten companies - 103
producer goods and 32 consumer goods
273
274
TABLE 1 INDUSTRY CATEGORIES USED
FTC
COOE OESCRI P T I ON _
2001 MEAT PACKINGt SAUSAGE S ANO OTHER PREPARED
MEAT PROOUCTS
2004 DAIRY PRODUCTS EXC F LUID MILK
2005 CANNED SPECIALTIES
2001 FROZEN SPECIALTIES
OEHYORATFOt AND
I NC LU 0 ING PRE SERVE S
SOUP MIXESbull
SEASO lNGSt ANO
OTHER PET FOOJ
RAIN MILL PROOUCTSt
ANO PREPARED
RELATED PRODUCTS
PRODUCTS
Z008 CANNEOt ORIEOt PICKLED FRUITS
ANO VEGET ABLES JAMS JELLIES DEHYDRATED I G TA3t E
S40CES ANO SALAD ORESSINGS
2010 DOG CAT A O
2012 FlOUR t OTHER RICE
MILLING fiLENOE) FLOUR
2014 B EADt CAKE ANO
20113 CO FECTIONERY
2026 BOTTLED A O CANNED SOFT DRINKS 2021 FLAVORING EXTRACTS AND SYRUPS EC
2029 MISC FOOOS AND KINDRED PROJUCTS EXC
ROASTED COFFEE
2301 MENS AND BOYS SUITS AND COATS
2302 MENS ANO BOYS FURNISHINGS
230 3 kOMENS AND MISSES OUTER EAR
2551 HOUSEHOLD FURNITURE
2702 PERIODICALS
Pqe 19a 1
RELATED 1972
src cooe
2011 f 3
202tX2026
2032
203pound
2033t4t5
2047 2041t4t5
2051 2065 2086
2081
209XZ05
231
232
233
251 272
2703 BOOKS
2704 ISC PVeLISHING
2806 ORGANIC FlEERS
2807 DRUG$ ETHICAL
23234
PT 2 S 3
2808 ORUCSt PROPRIETARY
2809 PERFU ES t COSMETICS ANO OTHER TaiLET
PREPARATIONS
2810 SOAP AND OTHER CLEANING PREPARATIONS 2815
2901
PEST1CIDES AND AGR ICULTURAL
PE OLEUM REF I N ING CHEMICALS EC
3606 HOUSEHOLD COOKING EQUIPMENT
3612 HOUSEHOLD APPLIANC ES NEC INCLUJING ELECshy
rqlC HOUSE ARFS AND FANS ANO SE I G MACHINES
3617 RADIO AND T V RECE I VING SETS 3808 PHOTOG APHIC E UIP amp SUP LfESt EXC
3qo3
PHO TOCOPYING
SPOR TING A D E QU I P M ENT f SUP P L I E S ATHLETTC GOODS NEC
3904 DOLLS GAMESt TOYS AND CHLOqENtS VE ICLES
PT283
2844
234tX2844
287 9
291 3631
3634t6t9
36H
PT386l
3949
15
943
middot
l A6l f 2 R(GRESSION RESULTS
EQ I NO
NR CODE cu 121
lA 2001 A (
2A 2004 r (
3A 2005 a
-R
ClS amp31
16
17
1 z
I TER-CEPT bull t
ooo9t21At 00022311 oooeos uu
OOl49tCAI 0010 2 middot 000960CCI
o049tl1At o04t55C61
MARKET SH ARE bull 5 I
-010670Ctl
-017600181
-001050
-0211AO
-0077 0
AOVER-TIS INti
lbl
000()39 000072
000018 0 0024liAI
-000079
SHARf
4ampgtVf Q bull
111
000140
-000170
RSQ
COl
66 485 862
607 6 3 t ()2
blti6 633
flASTI-
CI T Y PH
-007271(1
000441
-u0288l
clASTSHU (t)Ef
bull 10
tto
-234
210
( OOlltOCCI -126520 000922 181 001790
4A 200 7 18 00590liAI -038670 4e16 -032110 lell u c
OOOS74e 000700 025940
00067018) 0021001AI -02l100CAI
5 26
920
5A 6 c
200 8 26 0 03Sl9CA I 0012 26181 o 024511(
-005540
-0 81420 0001961AI O()Ol90CAI ooll240
441 604 891
-0020 19 274
6A 2010 16 Oe04224CAI o359lOIAI 9oq ll4015dCAI 061 6 -middot 005435CAI 00011 8 1AI 917 c 00535ltAI -009850((1 OOOA51J CAI -001610161 971
7A 2012 001862 000010 351 00187) oo1 a oooq45 ono1o6 439 c 001109 -027050 Oe0067SIAI -OOll60C81 897
- middotmiddot- -
8A 201t ll 000757 Ol4140CCI 814 ott9oocet 095 8 oooqo6 C C J 000069(81 811 c ootl2lampBI -080910CCI Oe009lUBI -OOtlOOCCI 960I
I DIIJIT pound11 1V OTIUTVTfAll-1pound1t- A _ 00 R - qq r - Q
1jl)OQCD
0
-middot middot--
l
_ __ _
I
i
0 00141 c c
91)7
I
T ABLE 2 REG ESSIOH RESULTS CCONTI
EO NR
IND CODE
iR bull
cos
I TEq-
CEPT
t1ARKE T
SHA E
AOVER-
Tl SING
SHARpound-cent
AuVER
RSO ILAST1-
C I T Y ELAST
SH CPtf (lt Cl) bull 11 I 41 I 5I (61 171 I 8 I f9) c 10)
9A 2 0 1 8 19 002397181 -007640 367 -005 246 le46 6 000605 000550CAI 597
-000029 G14180 OOl654CAJ -024li01At
lOA 2026 12 0028901AI -OlZl90 -0114 0 1 1 1 5 759
Ll (
llA 2021
f3 c
12A lOZ9
B
(
13A 2301
B
c
14A 2302
B
c
154 201
n (
16A 2551
8
10
35
10
20
19
24
001897(()
00141 6CCI
Oll9lHCI
010604 middot middot
Oll083
OOJ11JCAt
002059141
001411CIH
-000412
000293(81
000187
001159 (3)
ooo6a4cct
o00792CCI
OOl40dCAt
000006
-000076
OOlltl6fA)
o o0 5 64CCt
-034570
-o1middot o3o
-1659 90
Ol51lO
-046190CAt
Ol5450CA)
-005l60
002940
-030060(()
-0 43l50CCt
0 33650
-019080
00004
bullJ00741CAI
-000069
000466
0001 51CAt
000947CAt
O OOI)70CAt OOl750CA J
000175CCI
000671CAI
OOll5) f A t 002207CBI
0004b7CA I
- 007 1 70
001760
-001410CAt
- 0 16200CAt
-004210
- 1 207 70
705
96 0
542
466 805
460
614
921
832
9 61
996
424
550
9 15
56 1 651
884
-006114
019 037
03504lCA)
oo409Q
-141644((
491 -0 17664
607
871
lell
084
101
072
103
108 11IllOQfl)-- l
-037880fCJ OOl6l4CAJ -040l70fCI 0()
c
I
I
bull 9
9t
middot middot middotbullmiddot bull lraquo
TABLE 2 REGRE SION RESUlTS ICONTt
bull
EO I NO tHl I filER- f4A_K T AOVER- sbulluRE ttSQ flA Jl- iLAS T NR CODE cos CEPT SHARE JISING AOVfR bull (JTY SHR CiJCF
110Cl I f amp Itt 15J tbt bull 1c u caa
llA 2702 ll o079 6 81 Bt -019110 S6s -OlltO 111
a 004679 ooooo1 t69
( oOl26o -015160 001560 -022690 ltJr()I
ldA 270 3 19 0065lt -OIt2l60 220 -Ol57a2 11 8 e o 0 71t6IA t ol08 75tAt ( OOl983CiU -060llOC() OOl7191A t -O lltdOC A) middot 89
lqA 2704 10 005082 -Oil060 22b -l)i1977 lelO 6 -000080 oo2772C4 t 910 c 000907 -011 200 Oe04105CA t -046960 995
20A 2d06 1 II
O OlltStiAt -001120 837 00014 -))th O01380amp At -oooo 11 821 ( OOl526CAt -026700181 0002 5UAt 000530 969
2lA 2807 28 OOt9lt9CAt -00891 0 516 -0 03962 225I8 002095CCt OOOl lliAt 690 ( o02965CA t -056650((1 OOOt50CAJ -003l70ttl 911I
I 22A 2808 15 ot3l06 119730 655 1130) 090
0 - 007526CC t Oe003961At 835- middot-
c I oo7523 -067210 001l 2tA t -0091t80CCt 951
2lA 2809 21 Oe2067lCAt -1218101At 71 2 -10095((1 121 ttlIaOQ8 00797418) OOOI8lltt 686 Ill( 0092lU A t -Olltl70Ct t 000605CAt -OOit8701At 951 _
---- -
32 009506CAt
0 bull 06 981 I A J 006901 cA bull
00 24A 2810 -010960 5t 5 -OOltSO 7 56
8 -000011 s11 c -5811701At oooaoatAJ oot520IAI 888
-- 37 --- middot - -- -1 ) l lbull t5JJ r-l bullhmiddot1 --- - --
l5A
26A
27A
c
2RA
c
2 9A
31A
llA
24
13
597
Ci)55
14
l 6h
RSQ ELASTI- (lASTCOE f-
TABLE 2
EQ I NO NR CODE IU
REGRESSION RESULTS CCONTa
NR INTER- MARKET
(10S CEPT SHARE AOVER- SHARfcent
TIS lNG AOVER
) CITY StiR
c It) bull 5 co a I 1) bull 0 bull 9 c l 0 bull( 2 1
2815 16 003019 -009990 275 -o069bmiddotJ
6 001446 000224 282 c 002178 -054630 o02221CCI -0075 00 646
2901 29 00021t2 CAI 0)0520 592 0007 6 1 068
0 00105C8 t 000012141 60)g( OOOlJ51AJ -00126018) ooou271At -000u80CA) 974
36oa 1 3 003296(4) -oca29o -oo-q94
001755 000203 lt91t 002499((1 -04l820CCI OOl374CAI 0068)0 946
3612 0033281CI 011520 195 01684) oso
00 30 42C OI 00012 2 424 oo579ltAt -2 7l7801At 002 100IAI -003220 819
8 3617 btel 011789 081002259 009530
001448 oOOl OitCCI 732 OOl7 22CAI -0 46160CAI 0004961AI -OOll40181 989c
30A 3808 14 0021t 3 6fAI -000110 628 o023291C) -005 6 middot- _ middot- 0bull 02 04 1 IC I ooooo9 637 ( 001584((1 -0161)1)0 00021 91AI -0 00060
390) 14 002790CCt 006590 h8 t 0 0938 4 070
B OOl89bCCJ 000144 74
0021961(1 -OItlOlO OOl323CAI -004800 915
B
middot--- - middot 4 - middot- - dfll3904 12 004563181 0)8300 879 0428651(1 089 OQIll004 277CA I ooozstcea 919
c 0041)3181 -0-11250 0011881(1 -006270 972 0Ill
- J - bull bull bull bullbullbull middot - - middot - middot gt middotbull L)ofbullflbull middot middotfJltl -tn- middotmiddotmiddotmiddoto +f middot t middot bull
c
is
positive
2
Page 0 determineThe consumer goodproducer good split is sometimes difficult to
Sollle support for the split used in this paper is a comparison of goodness - of - fit
distributions of a
industries
set
advertising are
that industry
a sales increase of
the
highest aD10U3 those
measures for the two groups of industries Table 3 contains
measures for the 32- consumer goods industries and the 103 producer goods
2 Median R is 5 7 for the consumer goods set and 40 for the producer goods
Estimates - of y the elasticity of demand with respect to
given in column of Table 2 The hypothesis that y bull 0 with
alternative was applied for each ittdustry As shown in the
could not be rtjected for 24 of the 32 industries Of the
and 10 are positive For the remalnifg eight three are
and five are significantly positive
Of the threertegative elasetcities the largest (in absolute value)
cosmetics (code 2809) With all elasticity of -1 01 a one
industry advertising would gwerate a decrease in industry sales
more than one percetu
At the other extrema is toys and games (code 39 04) For
a one percent ittcrease in industry advertising would lead to
four-tenths of a pampreentmiddot Though not significantly different from zero
elasticity estimatbull for pt-bpuetary drugs (code 2808)
which are
Both drug industries have
to sales rafios (211 pe1 cent for 28 08 and 13 4 per cent
other hand flaIJo ing ext1ac s (mainly soft-drink SYUlS) also
is for
percent increase in
of slightly
a middottwo - tailed
table that hypothesis
24 14 are negative
significantly negative
9
the codetics and p-ropriacuy very high advertising
for 28 09) bull On the
has a high ind UY
advenisiltg t() sales ratio ( 8 0 per cent) bull but its estimated elasticity is very
small and not signifieantly different uom zero
middot It doe$ not seem unusual to find many elasticities near zero with a few
negative and a few positive If aggregate consumption is insensitive to aggregate
advertising as seems likely ld then what one industry gains must be lost by
others
Page 21
elasticity of demand does
(10) we can
if t 0 then
Finally a determination concerning the industry
not fix the individual firms elasticity Going back to equation
write y bull (1 - ) a + z y Since a gt 0 it follows thatii i i
Y gt 0 If Ylt 0 however y s sign is indeterminateij ij
Turning next to the coefficient of market share 20 of them are negative and
12 are positive with three of each sign being significantly different from zero
The largest positive value which is significant is for pet food (code 2010) and
a close second is mens amp boys suits and coats (code 23 01) The largest signifi
cant negative value is for cosmetics (code 28 09)
In the model some relations among y a 0 and 61 are implied If middota is
negative (1- y) is greater than one Since 61Y bull (1- a) (1 - ) 61Y is
pCsitive but it will be less than one if 0 is large enough Seventeen of the A
ys are negative in each of the seventeen industries 61 is also negative and les
1 (algebraicly) than y so 61Y is greater than one his is not evidence that the
degree of cooperation is low or zero in any industry of course A high o togethegt
with a negative y which is small (in absolute terms) relative to a can give a
value of 61Y greater than one On the other hand if 61Y were less than one
o would have to be greater than zero
For y gt 0 two conditions may hold If y lt a it follows that 61 lt 0 since
(y - a ) lt 0 It also follows that 61Y lt 0 regardless of the value of o bull
In addition the relation is an if-and-only-if one ie if 61Y is negative
then y lt a Three industries have negative values for 61y dairy products
except milk (code 20 04) synthetic fibers (code 2806) and photographic equipment
(code 3808) In all three cases y is quite small so a does not have to be very
large to give a negative value to al
If y gt a the model shows that 61-- gt 0 whatever the value of 0 bull There are
12 cases where both y and 61 are positive with 61 ranging from near zero
(grain milling code 20 12) to 1 20 ( proprietary drugs code 2808) Several
coeffiei-en_rs
--
15
o f to zero
Tab le 3 - Consumer Goods
Indus tr ies
2Distributions of R and Producer Goods
Page 20a
Consumer Producer
Goods Goods
0 - 9 9 0 3
10 - 19 9 0 1 1
20 - 29 9 3 20
30 - 39 9 3 17
40 - 49 9 5 16
50 - 59 9 8
69 9 6 Ii 1060
--f 670 79 9 2
80 - 89 9 4 4
90 - 100 1 1
Totals 32 10 3
Notes
2is used R i s cal-
culated as F F + ((N - K- 1 ) K ] 1 b_ F is the F
statistic Lor the hypothesis that all o f the non - intercept terms are simul taneously equal
1
2
--
-- Since generalized least squ-area regress ion
Page 12
o f these cases have values o f 6 1 Y rea ter than 0 8 (pet food (20 10) bread amp cakes
(20 14 ) miscellaneous processed foods (20 29 ) men s and boys suit s and coats
(23 01 ) p rop riet ary drugs (28 08 ) elect ric housewares (36 1 2) radio and
IV sets 17 and games and toys (39 For be than(36 ) 04)) 6 1 Y to greater 0 8
both o and cry mus t b e less than 0 2 That is in these industries the results
shown in Tab le 2 would hold only with very strong dominance of marke t demand
effects over brand swit ching e ffects and a very low degree o f cooperation
Using the concentration rat ios given in Table 4 the average concentration
ratio for the eight indust ries is 38 1 For the remaining 24 industries it is
43 2 This result would lend some support t o the proposition that concentrat ion
and cooperation are related if the group of eight industries are identi fied
as having low degrees of cooperation
One f inal note on model predict ions S ince both cr and o are assumed to
be non-negative 60 should also be non-negat ive In only one o f the 32 cas es
is 60 negative and it is not signif icant bull
cor pJ
7 1
7 0 4
1 4 7 5 1
)
7 3 z a o a t s 1 1 3 3 5 ( i V Z 8 tJ1 2 1 - 1 1 0 7 3 5 c middot v 2 8 1 0 3 2 -0 0 1 4 3 0 C 1 V 9 8 3 2 8 1 5 1 6 - 0 0 6 t1 6 9 C _ V 1 11 0 2 9 0 1 29 0 00 7 6 1 C l k V 7 3 1 3 6 08 1 3 - o n 4 9 9 4 3 3 3 6 1 2 2 4 0 1 6 4 3
3 1 7 1 3 0 1 1 7 d 7 8 3 3 8 0 8 0 0 2 3 i 2 3 0 6
P 4
3 9 0 3 1 4
1 2
0 09 3 8 4 39 1)1 0 4 l 8 6 5
T A LE 4 e I NCU S T R Y C A T E G O R Y
F TC NOM e tOOE ass
C 1 I c 2 )
2 0 0 1 1 6 2 0 0r 1 7 zo o l Z 2 0 0 7 1 8
z o o 2 6
z o t a 1 6
2 1) 1 2 1 5 2 0 1 4 1 1 2 0 1 a 1 9 2 0 26 1 2 2 0 2 7 1 0 2 0 29 - 3 5
2 3 0 1 10
2 3 02 z o
2 3 03 1 9
2 5 5 1 2 4 Z 1 0 Z 1 1 2 0 3 1 9 2 7 4 1 0 z a o 1 3
2 A 0 7 z s
E LA S T I -C I T Y
c 3 )
- 0 09 7 2 7
0 00 449
- 0 0 2 8 9 3
- 0 3 2 7 70 - 0 OZ il I q
o 4 u t s a 0 0 1 3 7 3
o 1 4 o o-0 0 5 2 4o -0 1 94 0 1
-C l 6 l l 4 0 1 JJ 3 7 0 3 t 0 4 2 0 0 4 0 9
- 0 4 1 8 44 - O l 7 t ) 4 -ll 7 1 4 0 b -0 3 5 7 8 Z
- 0 1 9 7 7 0 1)0 3 3 4
-o n 9 2
IJ A T A
CONV
ON-C ONV
( 4 )
t 1NV
C 1t-V C NV c ) v C - JNV C rlNV c n v c a v C ON V ( i) l V
C CtV
C t1tV C IJNV
C 1 V
c 1 v
Pa ge 2La
HEQ F t Noe x
c 5 )
6 2q
6 89 1 48
6 1 4 ) q 3
1 9 7 1 9
q z t s 5 a e z
l l 9 1
4 o 7 1 2 0
c oqt 1 Z
4 4 )-q 1 z
5 5 9 1 0 64 1 3 3
n 3
Inter industry analysi
Page 23
Equation ( 3 6 ) may be used to fo rmulate a rough test o f the cooperat ion
vs e f ficiency controversy I f we rest ricted ourselves to some subset o f
indus tries where the relat ion be tween y and a i s the same we would be lookingbull
at a sample o f indus tries for which the dependence o f V on H and o is a
quadratic wi th no squared terms
I f we now sub s titute (26) into ( 36 ) we get
(38) v bull a + (y - a) R + (y - a) o (H) - (y - a) H o (H) + u
If it is only d if ferences in marke t shares that affect advertising inten-
s i ty levels i f concentration does not af fect cooperation and i f o bull 0
( 38 ) reduces t o
( 39 ) V bull a + (y - a) H + U
Let this b e the null hyp o thes is and note that V is a linear func-
tion o f H If y gt a and the null hyp othesis holds the expected relat ion
between V and H is as shown by the pos itively sloped straight line in Figure
6 b etween points A and C
In this context the alterna ive hypothesis the degree o f
cooperation ( o) is a positive function o f concentrat ion
is that
(H) I f the func-
tion is as described in (26) and Figure 4 the relation between V and H
would be as shown by the curved line connec t ing po ints A and C in Figure 6 One way to test the null hypothe s is against this al ternative is to test fo r
the hypothesized curvature in the function relating V and H
Fi gure
R
6
v
ID c
H 1 0
I
i I I
sfigned
clasJ s
in def ini g
effec t is the
I
Page middot2lf
One charac teristic of produc ts which may be useful
classes wi th respec t to significanc e of the brand-switching
conveniencenon-convenienc e distinc tion developed by Porter ( 1 97 )
Using his identif ication of IRS indus tries as a s tar t ing point
each of the 32 industries used in this s tudy to one of the two
The ass ignments are not ed in Tab le 4
Using the conveniencenon-convenienc e dis t inc tion the
divided into two sub-samples
sample was
Four regressions were then run bull ene for e ch
sub-sample one for the to tal sample and one for the total sample with
a dummy variable for convenience goods Each regression was run wi th LB s
and wi th indus tries as ob servations The results are given in Tab le 5
For the LB level equations ( 1A - 1D) equation(37) was used ome
rearranging of terms gives v bull y - ( 1-o ) (y - a) ( 1-z ) + ui i i
bull y - 6 1 ( 1 - z ) + ui When observations are pooled across indus trif s yi
is an industry variable In these regressions the values o f y es t imated
in the first stage were used Since industries have been grouped so as
to reduce differences in (y - a) that term is t reated as a cons tant
within each group Finally H the Herfindahl index has been subst ituted
for o the degree of cooperation The resultant equation is linear
y ( 1 - z ) and H x ( l - z ) This is the equation for whi ch results arei i
reported in Table 5 In all regress ions all variables were weighted by the
square roo t of sales
in
raquo
- 34
3 1
i l T 4 8LE S R E G E S S I ON R E SUL T S i X T f N S I ON S T O T Y PE Of
AND L E V E L OF C ONC E kT R A T I O
E O O tt S
tP bull T Y P E NR C ONV O B S tO N-CONV
E L A S T I shy
T I C J T Y
O E - t-41( T SliA R E
H( fl F
1 1 - K O x middot K O X E R F
- H S a 1 1 - H S t 1 8 1 1 9 1
R S O
1 1 0 1
q z
t - bull 9 f 2 1
Z t P D l 6 fC ONV - 1 9 1 s s c 2 1
e - l O I e bull u q a
2 3 1 0 1 1 Nl N- t O V 5 7 1 ( 1 2 19 1 6 1 - J l S 2 5 4
l l l l t z q o a 1 - J o v z I
l ( I NO l l A l l ol u l t o o c bull J u l l
d S
c 3 4 t t r a l
bull Z D 1 0 l l 4 l l 2 3 -gt 2 l l 2 0 - 4 lS 5 4 l C 4 1 o e 6 t l b b 5 1 1 0 1 1 1 c - 1 9 5 1 t d S 2 1 tV
OQ10 T-R A T I O S A E G I VE N l P A RE N T H E S E S l E V E L S f S I G 1 F l C ANC E A - 99 5 B - q 9 t - s NS J UC E NO I Nl E R C ( P T I S U (j f middot) I N f Hf t OUA TIOI S 6 E f 3R E A O J U S T H E N T F OR 1-t E T E R O SIC E O A S T f C I T V pIraquo
1
middot
I
R SO I S C A L C UL A T E D A S f I f bull 1 1 bull 1lt 1 K t t- E E N I S THf U A1 B E R O f fJe S E fi V A T I O S F t S HtE F - S T A T I S
I N DE P E ND E N T y II
T I C F O T H E U Y P O HIE S I S f A l l C JE F F I C H N T A R E QU A L HI l f RO A f40 K I S T H f UUM 8 E R OF Q ( A fL f S
I l l
1 A
1 6
t C
f 1 r R I l l
1 2 1
L Jt
L B
l q
L 1
on s T Y P E bull 2 )
P OUC T
X
H S I l l
l O l A I bull l 0 7 1
O O b O J
1 2 00 1
1 b 2 1 C I l l Z U
l l S b 1 -1 1 0 1
rlt F X l l - fi C f l
bull 1 I
l l
- ()4 2 3 l f H
1 - 2 7 9 1
K O X
1 1 - ti E R F I
l d l
S4 l i A I 2 l b
1 3 00
K O X -t f R F t S J
I 1 - HER F t
)
60 9
z o bull
0
t l l
3 8 8
1 7 6
5 64
64
N f bull
tl S bull
I l l
I 4 1
t ONV
NON-t ONV
A l l
A ll
c o v NO- ( ONV
1 4 1
1 5 1
- l S b
c - 1 1 2 1
5 8 2 C A I 1 6 7 0 1
middot I l 9 H
l 1 il
1 4 1 6 1
E L A C) T J shy
l t C I T Y
I S I
l b l
- 0 2 0
1 - 1 4 9 1
O Z 3 2 1 A I C b 6 b l
0 0 9 2 -J
1 1 b 4 1
o z t t t l 1 2 2 1 1
oru - Ht tt F
1 gt X bull b I
Page 25 bull
Mean val -as ior relevan t variables for the two sub-samples are
Conv Non-Conv
Adv Sales 4 7 2 3
iit Share 2 7 3 2
Ela cicity - 05 3 02 2
Herf indahl 7 3 1 860
That the dist nct ion between coft9enienee goods and non - convenience
goods mat ters is supported not only by these means but also by the regression
results Equat ions lA and lB are different in almost every respec t Only the
concentrat ionmarket share int eract ion term seems to mat ter for convenience
good s Jus t he opposite holds for non - convenience goods both elasticity1
and market sl e are highly significant but the concentrat ionmarket share L
int erac t ion t=rm is not
Given that one minus market share is used these equ3 tions the coeffici t
of market share is the negative of what is sh in col (6) of Table 5 For the
regress ion for non - convenience LB s reported in equation lB then LB s with larger
market shares have lower advertising to sales ratios
On the question of the impact of the degree of cooperation the story is mixed
For convenienc good s the coefficient is pos itive as equation (31) predicts when
there is a po sitive degree of cooperat ion Fo r the non-convenience goods LB s middot
concentration has no effec t
Aggregat ion t o the indus try level and then rer unning the equations $ives the
results shown in (2a-2d) of Tab le 5 As expected R2 goes up and levels of significanc e
fmiddotJr individual coefficients gJ down In add ition the coeffic ient of H ( l -H) inmiddot
quat ion 2A is no t significant though it is po s i tive Since 2A is the indus try level
I
--
all but one of them and that one is not significant For 18 of the 31
wo uld have on its adve rtising to sal es ratio
cognate of LA the variable H ( l-H) pl ays the same role in 2A as H ( l -z 1) p lays
in LA
Vconomies of scale in advert ising
The literature on advert ing contains a number of definitionbull o f economies
of scale and of findings eone rning its presence and magnitude Before I loo k
a t the data with the hope o f mmenting about scale effect s then I wan t to
define the term as I use it
By economies of scale I mean a decrease in the number o f units of adver-
tising which are needed to generate a unit of sales as advert ising is increased
Given the definition of the firm s own elast icity of demand with respect to
advertising (see the text foll wing equation (1) ) the technical definition
is that rii gt 1 Diseconomies of scale are def ined symmetrically i e
Yu lt 1
ith a cro s s - section oi observations on advertising intensity
(v bull a s1) and onmiddot advertising (a ) we may regress the former on the latter 1 i i
The results of such regress ions for the 32 industries are given in Table 2
lines lB ZB bull bull bull 32B For the 32 eases the regression coefficient is positive it
positive coefficients the coefficient i s signif icant
The que tion now is whether these results ar e evidence of pervasive dis-
economies of scale in advertislng I think no t Given my as sumption that the
data I observe are equilibrium results I may ri ghtly conclud e that almost every-
where high levels of advert ising are as sociated in equilibrium with high ratios
of advertising to sales I may no t conclud e anything however about the
impact that a move by some firm away from its equilibrium level o f advertising
middot- Page 27
Given the assumed relation in equation ( 37 ) there is a corre spondshy
ing equilibrium equation relating v and ai The equation is non - linear i
and I have not found a simp le way to characterize it It is po ssib le to
det ermi ne its deriviative with respect to a b y takin$ tne de riviative of ( 3 7 ) i
When that is done the result is
v - 1(40) i bull sl a a z ( 1-z )i i ira-l
Sinc e all the other terms in (40) are non-negative the sign dep ends only
on the sign of 6 1 which is in turn de pend ent only on the sign of (y - a)
That the signs of the coefficients of a in equations lB 2B bull bull bull 32B arei
not always equal to the signs of the coefficients of z in LA 2A bull bull bull 32A i
is still a mystery to me and something to be explored
bull
Summary
My purpose in writing this pa per wa s t o attempt to integrate explicit micro-
economic mode lling reasoned econome tric specification of error terms and high
quality data to exp lore some questions about adver tising by large firms
al quite important improvements could be made in the first two of those areas
and much data ma ssag ing is yet to be done in the third
Concerning the subst an ce of what I have done I think six things are im-
Page 2 8
Severshy
V and Conclusions
portant
1 Explicit modeling is worth the ef for t if for no other reas on than
that it provides a basis for choosing fr om a varie ty of funct ional
forms
2 The same is true for error specif ica t ion I did in fact look at scatshy
ter diagrams for all 32 industry categor ies and they vir tually all show
the het eroscedast icity which I as sumed and for which I correc ted
3 The predicted relation be tween advertis ing in tensity and market share
shows up clearly in on ly 25 of the case s examined I have ye t to
explore why that may b e the cas e
4 Some evidence concerning the presence and impact of coopera tion was
produced but it is not clear ly pervasive
5 The dis tinction between convenience goods and non-c onvenie nce goods is
unambiguously a good one
6 Virtually no evidence concerning econ omies of scale in adver tis ing can
be gleaned fr om this study given its assumpt ions
More work is called for on many of these issues my study seems to have
raised more questions than it has answer ed This is pr obab lly due to the richshy
ne ss of the data source since I had the oppor tunity to address several issue s
at once
- -- - Q --
Foo tnotes
Manager Line o f Business Program Federal Trade Commission The author depended heavily on several staf f members of the LB Pro gram for statistical and clerical support in particular Joe Cholka George Pascoe and Haro lene Jenkins Helpful comments were given b y Richard Caves Dennis Mueller Michael Porter and F M Scherer When the paper was in its early stages o f development both S teve Garber and Jon Rasmussen former economists in the LB Program provided eery useful cri tiques The views expressed here are my own o f course
1 The basic exp licit model of the role of advertising is Dorfman and S teiner ( 1954 ) and the models in this paper may be seen as ano ther extens ion of their work
2 For a review o f models of oligopoly which treat prices and quantities as relevant variables see Shubik ( 1 959) Extens ions to take advertis ing int o account are fairly s traightforward
3 Dorfman and S teiner dealt with a price-setting firm For a monopolis t inverting the demand relation to show Pbull P (Q A) _instead o f Q- (P A) has no effect on the results o f the analysis For oligopolists who face a sys tem of demand relations inversion does have some impac t on the results The dif ferences are not trivial nonetheless they will no t be explored in this study
4 Shubik ( 1 959) p
5 From a formal mathematical point o f view the function is similar to a1 i Legrangian funct ion of the form - r rr
L bull 11 + A ( j - (i i i+j
where the indicates some fixed level of the profits industry Following the same logic which resul ts in and income constraint as the marginal utility of the of the indus tr in the obj ective of firm i
6 Hall amp Weiss ( 1 967) p 323
7 See Schmallensee ( 1 97 2 ) Ch 2 pp 1 6-47
8 On this poin t see Imel amp Helmberger ( 1 9 7 1 )
9 Long ( 1 970) p 130 - 1 39
1 0 See Schmallensee ( 1 972) Ch 3 pp 48-8 7
rrj i+j ) of the rest of the
the identification profits of the rest
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980
(33)
i
Page 14
v - (1 - o) [ a l + (y - a) ] The corresponding equation for Vis given in (24)
If the relation between industry advertising intensity (V) and
concentration (H) is non - linear via the impact of concentration
on the degree of cooperation does the relation take the form of a
auadratic as some other investigators have proposed (Greer ( 1971 ) Martin (1979) Strickland amp Weiss (1976) ) That question can
be approached from two perspectives - conceptual and empirical In
the next section I will try to apply some statistical material to it
here I want to explore the question in the context of the models
developed above
Referring to equation (24) taking the deriviative with respect
to H and letting bull (H) gives
bull( 3 3a ) 3V (y - a) (1 - o) + (1 - H) 3H dH
Given that o and H each are bounded by 0 and 1 and that o is an in-
creasing function of H the term in the brackets in non-negative The
sign of the deriviative then is the same as the sign of (y - a) If the industry effect is dominant (y gt a) then advertising in-
tensity will increase with concentration If the two effects are equal
then advertising intensity will not vbullry with concentration and if the
brand-switching e ffect is larger advertising intensity will decrease
with concentration
For given advertising elasticity parameters then it is not pos-
sible to show an increase in A S up to some H and the n a decrease
he relation may be non - linear but it is monotonic over the permis-
sible range of H
Page 15
One possibility is to introduce a relation between (y - a) and H
If industries characterized as dominated by the brand-switching effect
also were more concentrated then a cross industry comparison would
show an inverted U shape relation between AS and H I know of no
support for such an association between ( y - a) and H
V ERROR SPECUCATION
So far I have implicitly assumed that there are no errors in the
model I now assume that the error
thvariable in the equation for the i- firm is additive that its mean is zero
that its variance is inversely proportionate to its sales and that all error
term covarlances are zero hat is
34) v bull (1 - o) [ a + (y - a) z ] + oy + u i i i
- - -
In vector notation
(35) bull (1 - o) [ at + lt - a) J +
E() bull 0 Cov (u) -
0
If we now ca lculate V bull bull we get
oy + U bull
(36) V bull (1 - o) la + (y - a)R J +
U bull E(U) bull 0 Var (U) bull
P age 16
The error term specifi c ation which follows Hall
wi th the ratio of profits
6 heir analysis seems
as wellbull particularly given
and Weiss was
applied by them to equa tions to equity or
asse ts as dependen t variables to carry the same
weight in the present context the substantial
evidence presented by marketing analysis on the tendency of company de-
cision makers t o use the advertising to sales ratio as the decision
7variable
The zero covariance assumption on the other hand1 is particularly
troublesome Within a probably not true since
many even ts outside the
single industry it is
industry would have
industries
similar effec ts on all the
8firms in the industry Between it is probably not true either
since many companies produce in more than one indus try and even ts in the
firm would tend to effect all of its activities The extension of the
model to allow for non-zero covariances is certainly called for
--- -
Page Ll
Athird characteristic of this specification also deserves some comment
only on equation is shown for the firm or industry Several haveinvestigators
included an advertising equation ina larger system of equations (Comanor and
Wilson (1974) (1979) and Weiss (1976) so in myMartin Strickland Idid
dissertation1 Ihave ewo defenses for presenting only single equation results
here One is the suggestion by Comanor Wilson that simultaneousand equation
verybias may not be important in this context he second is that Iam will-
ing to see the model extended to include a profitability equation and I intend
to move that direction in the near futurein
Onthe inclusion of aconcentration equation on the other hand I amskep-
tical he problem isthat in some fundamental sense equations in models of
industrial organization should be oriented to the decision-making contexts of firms
Ifeach firm in anindustry with Nfirms has only one decision variable then
there can be only Nindependent equations which are based on the first-order
optimization equations It not show large ofis difficult of course to a number
functional relations which follow from the first-order conditions point here he
isthat only Nof them can be independent
Consider for illustrative purposes only a situation in wich teh N
itsquantities say q are set exogeneously and each firm determines advertising1
of and level say ai where a is the true optimal value a a bulla
1 1 i i + i
Ifp a) then the Nq
s and the Na s will determine N p s say pii bullp
i(l i 1 i
Now given Eand 9 we can determine both the vector s andthe vector v since
- bull til
ands We can also determine S s z bull s-1 s-ii
v and we
industry concentration isHbull (alternatively
V and Hare functions of the same
pendent Furthermore if there are
VM) and (H1 H2 bull bull bull )contain
VI STATISTI CAL APPLICATIONS
Page 18
Given the observed values of might be tempted to conclude that we
have a two equation system in which v and z are jointly determined That would i i
be inappropriate however since the 2N random variables (v ) are determined by
onlt N basic random variables t The variance-covariance matrix for ( ) must
be singular bull
The extension of this observation to industry level variables is straightshy
forward Since observed industry advertising intensity is V bull l v and observed 4
C4 bull z ) it follows thati l i
set of random variables t and cannot be indeshy-
M industries then the variables (V1 v2 at most M independent random variables
Given the assumptions made above y a and o are constant within a
given industry That being the case (34) can be written as
(37)
s1 bull (1 - o) (y - a)
If and
So + +
(1 - o) a + oy
81 are esti mated for an industry using a suitable statisticalSo procedure giving and 81 an estimate of y is then available The sumSo
Intra-industry analysis
1974 of Business Re ort -
categories are identified
Page 19
of So and S1 is y so y bull So + S1 is the estimate of y
If o is assumed to be bounded by zero and one the sign of 61 is also
the sign of y - a The estimate of So is of cou se an estimate of
(1 - o ) a + oy Moreove if y gt a then a is less than 6o- (1 - B) a+ 0(
Cor esponding results would
hold if y bull a or y lt a
If 61 gt 0 then we would estimate that a lt 6o
bull
For purposes of estimating this model 32 manufacturing industry categories
were selected The data are fram the Line of Business forms filed with
the Federal Trade Commission they are discussed in detail in the Annual Line
which is available from the FTC The 32 industry
in Table 1
Simple regressions for equ ation (37) were run for each of the 32 industries
with data for each of the firms which filed in the industry constituting an ob-
servation The results are given in Table 2 lines lA 2A
1974
the intercept
For all the A
equations generalized least quares was used th both and market
share bein scaled bv the square root of sales
The 32 industries were taken from the larger set of all manufacturing industry
categories in the 1974 data files Two selection cr iteria were used thereLB
had to be at least ten reporting companies and the industry had to be a consumer
goods industry or a producer goods industry which is strongly associated with a
related consumer goods industry so that adve rtising by the firms in the producer
goods industry impacts on demand in the related industry (eg soft drink syrup
and soft drinks) There were 135 industries with at least ten companies - 103
producer goods and 32 consumer goods
273
274
TABLE 1 INDUSTRY CATEGORIES USED
FTC
COOE OESCRI P T I ON _
2001 MEAT PACKINGt SAUSAGE S ANO OTHER PREPARED
MEAT PROOUCTS
2004 DAIRY PRODUCTS EXC F LUID MILK
2005 CANNED SPECIALTIES
2001 FROZEN SPECIALTIES
OEHYORATFOt AND
I NC LU 0 ING PRE SERVE S
SOUP MIXESbull
SEASO lNGSt ANO
OTHER PET FOOJ
RAIN MILL PROOUCTSt
ANO PREPARED
RELATED PRODUCTS
PRODUCTS
Z008 CANNEOt ORIEOt PICKLED FRUITS
ANO VEGET ABLES JAMS JELLIES DEHYDRATED I G TA3t E
S40CES ANO SALAD ORESSINGS
2010 DOG CAT A O
2012 FlOUR t OTHER RICE
MILLING fiLENOE) FLOUR
2014 B EADt CAKE ANO
20113 CO FECTIONERY
2026 BOTTLED A O CANNED SOFT DRINKS 2021 FLAVORING EXTRACTS AND SYRUPS EC
2029 MISC FOOOS AND KINDRED PROJUCTS EXC
ROASTED COFFEE
2301 MENS AND BOYS SUITS AND COATS
2302 MENS ANO BOYS FURNISHINGS
230 3 kOMENS AND MISSES OUTER EAR
2551 HOUSEHOLD FURNITURE
2702 PERIODICALS
Pqe 19a 1
RELATED 1972
src cooe
2011 f 3
202tX2026
2032
203pound
2033t4t5
2047 2041t4t5
2051 2065 2086
2081
209XZ05
231
232
233
251 272
2703 BOOKS
2704 ISC PVeLISHING
2806 ORGANIC FlEERS
2807 DRUG$ ETHICAL
23234
PT 2 S 3
2808 ORUCSt PROPRIETARY
2809 PERFU ES t COSMETICS ANO OTHER TaiLET
PREPARATIONS
2810 SOAP AND OTHER CLEANING PREPARATIONS 2815
2901
PEST1CIDES AND AGR ICULTURAL
PE OLEUM REF I N ING CHEMICALS EC
3606 HOUSEHOLD COOKING EQUIPMENT
3612 HOUSEHOLD APPLIANC ES NEC INCLUJING ELECshy
rqlC HOUSE ARFS AND FANS ANO SE I G MACHINES
3617 RADIO AND T V RECE I VING SETS 3808 PHOTOG APHIC E UIP amp SUP LfESt EXC
3qo3
PHO TOCOPYING
SPOR TING A D E QU I P M ENT f SUP P L I E S ATHLETTC GOODS NEC
3904 DOLLS GAMESt TOYS AND CHLOqENtS VE ICLES
PT283
2844
234tX2844
287 9
291 3631
3634t6t9
36H
PT386l
3949
15
943
middot
l A6l f 2 R(GRESSION RESULTS
EQ I NO
NR CODE cu 121
lA 2001 A (
2A 2004 r (
3A 2005 a
-R
ClS amp31
16
17
1 z
I TER-CEPT bull t
ooo9t21At 00022311 oooeos uu
OOl49tCAI 0010 2 middot 000960CCI
o049tl1At o04t55C61
MARKET SH ARE bull 5 I
-010670Ctl
-017600181
-001050
-0211AO
-0077 0
AOVER-TIS INti
lbl
000()39 000072
000018 0 0024liAI
-000079
SHARf
4ampgtVf Q bull
111
000140
-000170
RSQ
COl
66 485 862
607 6 3 t ()2
blti6 633
flASTI-
CI T Y PH
-007271(1
000441
-u0288l
clASTSHU (t)Ef
bull 10
tto
-234
210
( OOlltOCCI -126520 000922 181 001790
4A 200 7 18 00590liAI -038670 4e16 -032110 lell u c
OOOS74e 000700 025940
00067018) 0021001AI -02l100CAI
5 26
920
5A 6 c
200 8 26 0 03Sl9CA I 0012 26181 o 024511(
-005540
-0 81420 0001961AI O()Ol90CAI ooll240
441 604 891
-0020 19 274
6A 2010 16 Oe04224CAI o359lOIAI 9oq ll4015dCAI 061 6 -middot 005435CAI 00011 8 1AI 917 c 00535ltAI -009850((1 OOOA51J CAI -001610161 971
7A 2012 001862 000010 351 00187) oo1 a oooq45 ono1o6 439 c 001109 -027050 Oe0067SIAI -OOll60C81 897
- middotmiddot- -
8A 201t ll 000757 Ol4140CCI 814 ott9oocet 095 8 oooqo6 C C J 000069(81 811 c ootl2lampBI -080910CCI Oe009lUBI -OOtlOOCCI 960I
I DIIJIT pound11 1V OTIUTVTfAll-1pound1t- A _ 00 R - qq r - Q
1jl)OQCD
0
-middot middot--
l
_ __ _
I
i
0 00141 c c
91)7
I
T ABLE 2 REG ESSIOH RESULTS CCONTI
EO NR
IND CODE
iR bull
cos
I TEq-
CEPT
t1ARKE T
SHA E
AOVER-
Tl SING
SHARpound-cent
AuVER
RSO ILAST1-
C I T Y ELAST
SH CPtf (lt Cl) bull 11 I 41 I 5I (61 171 I 8 I f9) c 10)
9A 2 0 1 8 19 002397181 -007640 367 -005 246 le46 6 000605 000550CAI 597
-000029 G14180 OOl654CAJ -024li01At
lOA 2026 12 0028901AI -OlZl90 -0114 0 1 1 1 5 759
Ll (
llA 2021
f3 c
12A lOZ9
B
(
13A 2301
B
c
14A 2302
B
c
154 201
n (
16A 2551
8
10
35
10
20
19
24
001897(()
00141 6CCI
Oll9lHCI
010604 middot middot
Oll083
OOJ11JCAt
002059141
001411CIH
-000412
000293(81
000187
001159 (3)
ooo6a4cct
o00792CCI
OOl40dCAt
000006
-000076
OOlltl6fA)
o o0 5 64CCt
-034570
-o1middot o3o
-1659 90
Ol51lO
-046190CAt
Ol5450CA)
-005l60
002940
-030060(()
-0 43l50CCt
0 33650
-019080
00004
bullJ00741CAI
-000069
000466
0001 51CAt
000947CAt
O OOI)70CAt OOl750CA J
000175CCI
000671CAI
OOll5) f A t 002207CBI
0004b7CA I
- 007 1 70
001760
-001410CAt
- 0 16200CAt
-004210
- 1 207 70
705
96 0
542
466 805
460
614
921
832
9 61
996
424
550
9 15
56 1 651
884
-006114
019 037
03504lCA)
oo409Q
-141644((
491 -0 17664
607
871
lell
084
101
072
103
108 11IllOQfl)-- l
-037880fCJ OOl6l4CAJ -040l70fCI 0()
c
I
I
bull 9
9t
middot middot middotbullmiddot bull lraquo
TABLE 2 REGRE SION RESUlTS ICONTt
bull
EO I NO tHl I filER- f4A_K T AOVER- sbulluRE ttSQ flA Jl- iLAS T NR CODE cos CEPT SHARE JISING AOVfR bull (JTY SHR CiJCF
110Cl I f amp Itt 15J tbt bull 1c u caa
llA 2702 ll o079 6 81 Bt -019110 S6s -OlltO 111
a 004679 ooooo1 t69
( oOl26o -015160 001560 -022690 ltJr()I
ldA 270 3 19 0065lt -OIt2l60 220 -Ol57a2 11 8 e o 0 71t6IA t ol08 75tAt ( OOl983CiU -060llOC() OOl7191A t -O lltdOC A) middot 89
lqA 2704 10 005082 -Oil060 22b -l)i1977 lelO 6 -000080 oo2772C4 t 910 c 000907 -011 200 Oe04105CA t -046960 995
20A 2d06 1 II
O OlltStiAt -001120 837 00014 -))th O01380amp At -oooo 11 821 ( OOl526CAt -026700181 0002 5UAt 000530 969
2lA 2807 28 OOt9lt9CAt -00891 0 516 -0 03962 225I8 002095CCt OOOl lliAt 690 ( o02965CA t -056650((1 OOOt50CAJ -003l70ttl 911I
I 22A 2808 15 ot3l06 119730 655 1130) 090
0 - 007526CC t Oe003961At 835- middot-
c I oo7523 -067210 001l 2tA t -0091t80CCt 951
2lA 2809 21 Oe2067lCAt -1218101At 71 2 -10095((1 121 ttlIaOQ8 00797418) OOOI8lltt 686 Ill( 0092lU A t -Olltl70Ct t 000605CAt -OOit8701At 951 _
---- -
32 009506CAt
0 bull 06 981 I A J 006901 cA bull
00 24A 2810 -010960 5t 5 -OOltSO 7 56
8 -000011 s11 c -5811701At oooaoatAJ oot520IAI 888
-- 37 --- middot - -- -1 ) l lbull t5JJ r-l bullhmiddot1 --- - --
l5A
26A
27A
c
2RA
c
2 9A
31A
llA
24
13
597
Ci)55
14
l 6h
RSQ ELASTI- (lASTCOE f-
TABLE 2
EQ I NO NR CODE IU
REGRESSION RESULTS CCONTa
NR INTER- MARKET
(10S CEPT SHARE AOVER- SHARfcent
TIS lNG AOVER
) CITY StiR
c It) bull 5 co a I 1) bull 0 bull 9 c l 0 bull( 2 1
2815 16 003019 -009990 275 -o069bmiddotJ
6 001446 000224 282 c 002178 -054630 o02221CCI -0075 00 646
2901 29 00021t2 CAI 0)0520 592 0007 6 1 068
0 00105C8 t 000012141 60)g( OOOlJ51AJ -00126018) ooou271At -000u80CA) 974
36oa 1 3 003296(4) -oca29o -oo-q94
001755 000203 lt91t 002499((1 -04l820CCI OOl374CAI 0068)0 946
3612 0033281CI 011520 195 01684) oso
00 30 42C OI 00012 2 424 oo579ltAt -2 7l7801At 002 100IAI -003220 819
8 3617 btel 011789 081002259 009530
001448 oOOl OitCCI 732 OOl7 22CAI -0 46160CAI 0004961AI -OOll40181 989c
30A 3808 14 0021t 3 6fAI -000110 628 o023291C) -005 6 middot- _ middot- 0bull 02 04 1 IC I ooooo9 637 ( 001584((1 -0161)1)0 00021 91AI -0 00060
390) 14 002790CCt 006590 h8 t 0 0938 4 070
B OOl89bCCJ 000144 74
0021961(1 -OItlOlO OOl323CAI -004800 915
B
middot--- - middot 4 - middot- - dfll3904 12 004563181 0)8300 879 0428651(1 089 OQIll004 277CA I ooozstcea 919
c 0041)3181 -0-11250 0011881(1 -006270 972 0Ill
- J - bull bull bull bullbullbull middot - - middot - middot gt middotbull L)ofbullflbull middot middotfJltl -tn- middotmiddotmiddotmiddoto +f middot t middot bull
c
is
positive
2
Page 0 determineThe consumer goodproducer good split is sometimes difficult to
Sollle support for the split used in this paper is a comparison of goodness - of - fit
distributions of a
industries
set
advertising are
that industry
a sales increase of
the
highest aD10U3 those
measures for the two groups of industries Table 3 contains
measures for the 32- consumer goods industries and the 103 producer goods
2 Median R is 5 7 for the consumer goods set and 40 for the producer goods
Estimates - of y the elasticity of demand with respect to
given in column of Table 2 The hypothesis that y bull 0 with
alternative was applied for each ittdustry As shown in the
could not be rtjected for 24 of the 32 industries Of the
and 10 are positive For the remalnifg eight three are
and five are significantly positive
Of the threertegative elasetcities the largest (in absolute value)
cosmetics (code 2809) With all elasticity of -1 01 a one
industry advertising would gwerate a decrease in industry sales
more than one percetu
At the other extrema is toys and games (code 39 04) For
a one percent ittcrease in industry advertising would lead to
four-tenths of a pampreentmiddot Though not significantly different from zero
elasticity estimatbull for pt-bpuetary drugs (code 2808)
which are
Both drug industries have
to sales rafios (211 pe1 cent for 28 08 and 13 4 per cent
other hand flaIJo ing ext1ac s (mainly soft-drink SYUlS) also
is for
percent increase in
of slightly
a middottwo - tailed
table that hypothesis
24 14 are negative
significantly negative
9
the codetics and p-ropriacuy very high advertising
for 28 09) bull On the
has a high ind UY
advenisiltg t() sales ratio ( 8 0 per cent) bull but its estimated elasticity is very
small and not signifieantly different uom zero
middot It doe$ not seem unusual to find many elasticities near zero with a few
negative and a few positive If aggregate consumption is insensitive to aggregate
advertising as seems likely ld then what one industry gains must be lost by
others
Page 21
elasticity of demand does
(10) we can
if t 0 then
Finally a determination concerning the industry
not fix the individual firms elasticity Going back to equation
write y bull (1 - ) a + z y Since a gt 0 it follows thatii i i
Y gt 0 If Ylt 0 however y s sign is indeterminateij ij
Turning next to the coefficient of market share 20 of them are negative and
12 are positive with three of each sign being significantly different from zero
The largest positive value which is significant is for pet food (code 2010) and
a close second is mens amp boys suits and coats (code 23 01) The largest signifi
cant negative value is for cosmetics (code 28 09)
In the model some relations among y a 0 and 61 are implied If middota is
negative (1- y) is greater than one Since 61Y bull (1- a) (1 - ) 61Y is
pCsitive but it will be less than one if 0 is large enough Seventeen of the A
ys are negative in each of the seventeen industries 61 is also negative and les
1 (algebraicly) than y so 61Y is greater than one his is not evidence that the
degree of cooperation is low or zero in any industry of course A high o togethegt
with a negative y which is small (in absolute terms) relative to a can give a
value of 61Y greater than one On the other hand if 61Y were less than one
o would have to be greater than zero
For y gt 0 two conditions may hold If y lt a it follows that 61 lt 0 since
(y - a ) lt 0 It also follows that 61Y lt 0 regardless of the value of o bull
In addition the relation is an if-and-only-if one ie if 61Y is negative
then y lt a Three industries have negative values for 61y dairy products
except milk (code 20 04) synthetic fibers (code 2806) and photographic equipment
(code 3808) In all three cases y is quite small so a does not have to be very
large to give a negative value to al
If y gt a the model shows that 61-- gt 0 whatever the value of 0 bull There are
12 cases where both y and 61 are positive with 61 ranging from near zero
(grain milling code 20 12) to 1 20 ( proprietary drugs code 2808) Several
coeffiei-en_rs
--
15
o f to zero
Tab le 3 - Consumer Goods
Indus tr ies
2Distributions of R and Producer Goods
Page 20a
Consumer Producer
Goods Goods
0 - 9 9 0 3
10 - 19 9 0 1 1
20 - 29 9 3 20
30 - 39 9 3 17
40 - 49 9 5 16
50 - 59 9 8
69 9 6 Ii 1060
--f 670 79 9 2
80 - 89 9 4 4
90 - 100 1 1
Totals 32 10 3
Notes
2is used R i s cal-
culated as F F + ((N - K- 1 ) K ] 1 b_ F is the F
statistic Lor the hypothesis that all o f the non - intercept terms are simul taneously equal
1
2
--
-- Since generalized least squ-area regress ion
Page 12
o f these cases have values o f 6 1 Y rea ter than 0 8 (pet food (20 10) bread amp cakes
(20 14 ) miscellaneous processed foods (20 29 ) men s and boys suit s and coats
(23 01 ) p rop riet ary drugs (28 08 ) elect ric housewares (36 1 2) radio and
IV sets 17 and games and toys (39 For be than(36 ) 04)) 6 1 Y to greater 0 8
both o and cry mus t b e less than 0 2 That is in these industries the results
shown in Tab le 2 would hold only with very strong dominance of marke t demand
effects over brand swit ching e ffects and a very low degree o f cooperation
Using the concentration rat ios given in Table 4 the average concentration
ratio for the eight indust ries is 38 1 For the remaining 24 industries it is
43 2 This result would lend some support t o the proposition that concentrat ion
and cooperation are related if the group of eight industries are identi fied
as having low degrees of cooperation
One f inal note on model predict ions S ince both cr and o are assumed to
be non-negative 60 should also be non-negat ive In only one o f the 32 cas es
is 60 negative and it is not signif icant bull
cor pJ
7 1
7 0 4
1 4 7 5 1
)
7 3 z a o a t s 1 1 3 3 5 ( i V Z 8 tJ1 2 1 - 1 1 0 7 3 5 c middot v 2 8 1 0 3 2 -0 0 1 4 3 0 C 1 V 9 8 3 2 8 1 5 1 6 - 0 0 6 t1 6 9 C _ V 1 11 0 2 9 0 1 29 0 00 7 6 1 C l k V 7 3 1 3 6 08 1 3 - o n 4 9 9 4 3 3 3 6 1 2 2 4 0 1 6 4 3
3 1 7 1 3 0 1 1 7 d 7 8 3 3 8 0 8 0 0 2 3 i 2 3 0 6
P 4
3 9 0 3 1 4
1 2
0 09 3 8 4 39 1)1 0 4 l 8 6 5
T A LE 4 e I NCU S T R Y C A T E G O R Y
F TC NOM e tOOE ass
C 1 I c 2 )
2 0 0 1 1 6 2 0 0r 1 7 zo o l Z 2 0 0 7 1 8
z o o 2 6
z o t a 1 6
2 1) 1 2 1 5 2 0 1 4 1 1 2 0 1 a 1 9 2 0 26 1 2 2 0 2 7 1 0 2 0 29 - 3 5
2 3 0 1 10
2 3 02 z o
2 3 03 1 9
2 5 5 1 2 4 Z 1 0 Z 1 1 2 0 3 1 9 2 7 4 1 0 z a o 1 3
2 A 0 7 z s
E LA S T I -C I T Y
c 3 )
- 0 09 7 2 7
0 00 449
- 0 0 2 8 9 3
- 0 3 2 7 70 - 0 OZ il I q
o 4 u t s a 0 0 1 3 7 3
o 1 4 o o-0 0 5 2 4o -0 1 94 0 1
-C l 6 l l 4 0 1 JJ 3 7 0 3 t 0 4 2 0 0 4 0 9
- 0 4 1 8 44 - O l 7 t ) 4 -ll 7 1 4 0 b -0 3 5 7 8 Z
- 0 1 9 7 7 0 1)0 3 3 4
-o n 9 2
IJ A T A
CONV
ON-C ONV
( 4 )
t 1NV
C 1t-V C NV c ) v C - JNV C rlNV c n v c a v C ON V ( i) l V
C CtV
C t1tV C IJNV
C 1 V
c 1 v
Pa ge 2La
HEQ F t Noe x
c 5 )
6 2q
6 89 1 48
6 1 4 ) q 3
1 9 7 1 9
q z t s 5 a e z
l l 9 1
4 o 7 1 2 0
c oqt 1 Z
4 4 )-q 1 z
5 5 9 1 0 64 1 3 3
n 3
Inter industry analysi
Page 23
Equation ( 3 6 ) may be used to fo rmulate a rough test o f the cooperat ion
vs e f ficiency controversy I f we rest ricted ourselves to some subset o f
indus tries where the relat ion be tween y and a i s the same we would be lookingbull
at a sample o f indus tries for which the dependence o f V on H and o is a
quadratic wi th no squared terms
I f we now sub s titute (26) into ( 36 ) we get
(38) v bull a + (y - a) R + (y - a) o (H) - (y - a) H o (H) + u
If it is only d if ferences in marke t shares that affect advertising inten-
s i ty levels i f concentration does not af fect cooperation and i f o bull 0
( 38 ) reduces t o
( 39 ) V bull a + (y - a) H + U
Let this b e the null hyp o thes is and note that V is a linear func-
tion o f H If y gt a and the null hyp othesis holds the expected relat ion
between V and H is as shown by the pos itively sloped straight line in Figure
6 b etween points A and C
In this context the alterna ive hypothesis the degree o f
cooperation ( o) is a positive function o f concentrat ion
is that
(H) I f the func-
tion is as described in (26) and Figure 4 the relation between V and H
would be as shown by the curved line connec t ing po ints A and C in Figure 6 One way to test the null hypothe s is against this al ternative is to test fo r
the hypothesized curvature in the function relating V and H
Fi gure
R
6
v
ID c
H 1 0
I
i I I
sfigned
clasJ s
in def ini g
effec t is the
I
Page middot2lf
One charac teristic of produc ts which may be useful
classes wi th respec t to significanc e of the brand-switching
conveniencenon-convenienc e distinc tion developed by Porter ( 1 97 )
Using his identif ication of IRS indus tries as a s tar t ing point
each of the 32 industries used in this s tudy to one of the two
The ass ignments are not ed in Tab le 4
Using the conveniencenon-convenienc e dis t inc tion the
divided into two sub-samples
sample was
Four regressions were then run bull ene for e ch
sub-sample one for the to tal sample and one for the total sample with
a dummy variable for convenience goods Each regression was run wi th LB s
and wi th indus tries as ob servations The results are given in Tab le 5
For the LB level equations ( 1A - 1D) equation(37) was used ome
rearranging of terms gives v bull y - ( 1-o ) (y - a) ( 1-z ) + ui i i
bull y - 6 1 ( 1 - z ) + ui When observations are pooled across indus trif s yi
is an industry variable In these regressions the values o f y es t imated
in the first stage were used Since industries have been grouped so as
to reduce differences in (y - a) that term is t reated as a cons tant
within each group Finally H the Herfindahl index has been subst ituted
for o the degree of cooperation The resultant equation is linear
y ( 1 - z ) and H x ( l - z ) This is the equation for whi ch results arei i
reported in Table 5 In all regress ions all variables were weighted by the
square roo t of sales
in
raquo
- 34
3 1
i l T 4 8LE S R E G E S S I ON R E SUL T S i X T f N S I ON S T O T Y PE Of
AND L E V E L OF C ONC E kT R A T I O
E O O tt S
tP bull T Y P E NR C ONV O B S tO N-CONV
E L A S T I shy
T I C J T Y
O E - t-41( T SliA R E
H( fl F
1 1 - K O x middot K O X E R F
- H S a 1 1 - H S t 1 8 1 1 9 1
R S O
1 1 0 1
q z
t - bull 9 f 2 1
Z t P D l 6 fC ONV - 1 9 1 s s c 2 1
e - l O I e bull u q a
2 3 1 0 1 1 Nl N- t O V 5 7 1 ( 1 2 19 1 6 1 - J l S 2 5 4
l l l l t z q o a 1 - J o v z I
l ( I NO l l A l l ol u l t o o c bull J u l l
d S
c 3 4 t t r a l
bull Z D 1 0 l l 4 l l 2 3 -gt 2 l l 2 0 - 4 lS 5 4 l C 4 1 o e 6 t l b b 5 1 1 0 1 1 1 c - 1 9 5 1 t d S 2 1 tV
OQ10 T-R A T I O S A E G I VE N l P A RE N T H E S E S l E V E L S f S I G 1 F l C ANC E A - 99 5 B - q 9 t - s NS J UC E NO I Nl E R C ( P T I S U (j f middot) I N f Hf t OUA TIOI S 6 E f 3R E A O J U S T H E N T F OR 1-t E T E R O SIC E O A S T f C I T V pIraquo
1
middot
I
R SO I S C A L C UL A T E D A S f I f bull 1 1 bull 1lt 1 K t t- E E N I S THf U A1 B E R O f fJe S E fi V A T I O S F t S HtE F - S T A T I S
I N DE P E ND E N T y II
T I C F O T H E U Y P O HIE S I S f A l l C JE F F I C H N T A R E QU A L HI l f RO A f40 K I S T H f UUM 8 E R OF Q ( A fL f S
I l l
1 A
1 6
t C
f 1 r R I l l
1 2 1
L Jt
L B
l q
L 1
on s T Y P E bull 2 )
P OUC T
X
H S I l l
l O l A I bull l 0 7 1
O O b O J
1 2 00 1
1 b 2 1 C I l l Z U
l l S b 1 -1 1 0 1
rlt F X l l - fi C f l
bull 1 I
l l
- ()4 2 3 l f H
1 - 2 7 9 1
K O X
1 1 - ti E R F I
l d l
S4 l i A I 2 l b
1 3 00
K O X -t f R F t S J
I 1 - HER F t
)
60 9
z o bull
0
t l l
3 8 8
1 7 6
5 64
64
N f bull
tl S bull
I l l
I 4 1
t ONV
NON-t ONV
A l l
A ll
c o v NO- ( ONV
1 4 1
1 5 1
- l S b
c - 1 1 2 1
5 8 2 C A I 1 6 7 0 1
middot I l 9 H
l 1 il
1 4 1 6 1
E L A C) T J shy
l t C I T Y
I S I
l b l
- 0 2 0
1 - 1 4 9 1
O Z 3 2 1 A I C b 6 b l
0 0 9 2 -J
1 1 b 4 1
o z t t t l 1 2 2 1 1
oru - Ht tt F
1 gt X bull b I
Page 25 bull
Mean val -as ior relevan t variables for the two sub-samples are
Conv Non-Conv
Adv Sales 4 7 2 3
iit Share 2 7 3 2
Ela cicity - 05 3 02 2
Herf indahl 7 3 1 860
That the dist nct ion between coft9enienee goods and non - convenience
goods mat ters is supported not only by these means but also by the regression
results Equat ions lA and lB are different in almost every respec t Only the
concentrat ionmarket share int eract ion term seems to mat ter for convenience
good s Jus t he opposite holds for non - convenience goods both elasticity1
and market sl e are highly significant but the concentrat ionmarket share L
int erac t ion t=rm is not
Given that one minus market share is used these equ3 tions the coeffici t
of market share is the negative of what is sh in col (6) of Table 5 For the
regress ion for non - convenience LB s reported in equation lB then LB s with larger
market shares have lower advertising to sales ratios
On the question of the impact of the degree of cooperation the story is mixed
For convenienc good s the coefficient is pos itive as equation (31) predicts when
there is a po sitive degree of cooperat ion Fo r the non-convenience goods LB s middot
concentration has no effec t
Aggregat ion t o the indus try level and then rer unning the equations $ives the
results shown in (2a-2d) of Tab le 5 As expected R2 goes up and levels of significanc e
fmiddotJr individual coefficients gJ down In add ition the coeffic ient of H ( l -H) inmiddot
quat ion 2A is no t significant though it is po s i tive Since 2A is the indus try level
I
--
all but one of them and that one is not significant For 18 of the 31
wo uld have on its adve rtising to sal es ratio
cognate of LA the variable H ( l-H) pl ays the same role in 2A as H ( l -z 1) p lays
in LA
Vconomies of scale in advert ising
The literature on advert ing contains a number of definitionbull o f economies
of scale and of findings eone rning its presence and magnitude Before I loo k
a t the data with the hope o f mmenting about scale effect s then I wan t to
define the term as I use it
By economies of scale I mean a decrease in the number o f units of adver-
tising which are needed to generate a unit of sales as advert ising is increased
Given the definition of the firm s own elast icity of demand with respect to
advertising (see the text foll wing equation (1) ) the technical definition
is that rii gt 1 Diseconomies of scale are def ined symmetrically i e
Yu lt 1
ith a cro s s - section oi observations on advertising intensity
(v bull a s1) and onmiddot advertising (a ) we may regress the former on the latter 1 i i
The results of such regress ions for the 32 industries are given in Table 2
lines lB ZB bull bull bull 32B For the 32 eases the regression coefficient is positive it
positive coefficients the coefficient i s signif icant
The que tion now is whether these results ar e evidence of pervasive dis-
economies of scale in advertislng I think no t Given my as sumption that the
data I observe are equilibrium results I may ri ghtly conclud e that almost every-
where high levels of advert ising are as sociated in equilibrium with high ratios
of advertising to sales I may no t conclud e anything however about the
impact that a move by some firm away from its equilibrium level o f advertising
middot- Page 27
Given the assumed relation in equation ( 37 ) there is a corre spondshy
ing equilibrium equation relating v and ai The equation is non - linear i
and I have not found a simp le way to characterize it It is po ssib le to
det ermi ne its deriviative with respect to a b y takin$ tne de riviative of ( 3 7 ) i
When that is done the result is
v - 1(40) i bull sl a a z ( 1-z )i i ira-l
Sinc e all the other terms in (40) are non-negative the sign dep ends only
on the sign of 6 1 which is in turn de pend ent only on the sign of (y - a)
That the signs of the coefficients of a in equations lB 2B bull bull bull 32B arei
not always equal to the signs of the coefficients of z in LA 2A bull bull bull 32A i
is still a mystery to me and something to be explored
bull
Summary
My purpose in writing this pa per wa s t o attempt to integrate explicit micro-
economic mode lling reasoned econome tric specification of error terms and high
quality data to exp lore some questions about adver tising by large firms
al quite important improvements could be made in the first two of those areas
and much data ma ssag ing is yet to be done in the third
Concerning the subst an ce of what I have done I think six things are im-
Page 2 8
Severshy
V and Conclusions
portant
1 Explicit modeling is worth the ef for t if for no other reas on than
that it provides a basis for choosing fr om a varie ty of funct ional
forms
2 The same is true for error specif ica t ion I did in fact look at scatshy
ter diagrams for all 32 industry categor ies and they vir tually all show
the het eroscedast icity which I as sumed and for which I correc ted
3 The predicted relation be tween advertis ing in tensity and market share
shows up clearly in on ly 25 of the case s examined I have ye t to
explore why that may b e the cas e
4 Some evidence concerning the presence and impact of coopera tion was
produced but it is not clear ly pervasive
5 The dis tinction between convenience goods and non-c onvenie nce goods is
unambiguously a good one
6 Virtually no evidence concerning econ omies of scale in adver tis ing can
be gleaned fr om this study given its assumpt ions
More work is called for on many of these issues my study seems to have
raised more questions than it has answer ed This is pr obab lly due to the richshy
ne ss of the data source since I had the oppor tunity to address several issue s
at once
- -- - Q --
Foo tnotes
Manager Line o f Business Program Federal Trade Commission The author depended heavily on several staf f members of the LB Pro gram for statistical and clerical support in particular Joe Cholka George Pascoe and Haro lene Jenkins Helpful comments were given b y Richard Caves Dennis Mueller Michael Porter and F M Scherer When the paper was in its early stages o f development both S teve Garber and Jon Rasmussen former economists in the LB Program provided eery useful cri tiques The views expressed here are my own o f course
1 The basic exp licit model of the role of advertising is Dorfman and S teiner ( 1954 ) and the models in this paper may be seen as ano ther extens ion of their work
2 For a review o f models of oligopoly which treat prices and quantities as relevant variables see Shubik ( 1 959) Extens ions to take advertis ing int o account are fairly s traightforward
3 Dorfman and S teiner dealt with a price-setting firm For a monopolis t inverting the demand relation to show Pbull P (Q A) _instead o f Q- (P A) has no effect on the results o f the analysis For oligopolists who face a sys tem of demand relations inversion does have some impac t on the results The dif ferences are not trivial nonetheless they will no t be explored in this study
4 Shubik ( 1 959) p
5 From a formal mathematical point o f view the function is similar to a1 i Legrangian funct ion of the form - r rr
L bull 11 + A ( j - (i i i+j
where the indicates some fixed level of the profits industry Following the same logic which resul ts in and income constraint as the marginal utility of the of the indus tr in the obj ective of firm i
6 Hall amp Weiss ( 1 967) p 323
7 See Schmallensee ( 1 97 2 ) Ch 2 pp 1 6-47
8 On this poin t see Imel amp Helmberger ( 1 9 7 1 )
9 Long ( 1 970) p 130 - 1 39
1 0 See Schmallensee ( 1 972) Ch 3 pp 48-8 7
rrj i+j ) of the rest of the
the identification profits of the rest
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980
Page 15
One possibility is to introduce a relation between (y - a) and H
If industries characterized as dominated by the brand-switching effect
also were more concentrated then a cross industry comparison would
show an inverted U shape relation between AS and H I know of no
support for such an association between ( y - a) and H
V ERROR SPECUCATION
So far I have implicitly assumed that there are no errors in the
model I now assume that the error
thvariable in the equation for the i- firm is additive that its mean is zero
that its variance is inversely proportionate to its sales and that all error
term covarlances are zero hat is
34) v bull (1 - o) [ a + (y - a) z ] + oy + u i i i
- - -
In vector notation
(35) bull (1 - o) [ at + lt - a) J +
E() bull 0 Cov (u) -
0
If we now ca lculate V bull bull we get
oy + U bull
(36) V bull (1 - o) la + (y - a)R J +
U bull E(U) bull 0 Var (U) bull
P age 16
The error term specifi c ation which follows Hall
wi th the ratio of profits
6 heir analysis seems
as wellbull particularly given
and Weiss was
applied by them to equa tions to equity or
asse ts as dependen t variables to carry the same
weight in the present context the substantial
evidence presented by marketing analysis on the tendency of company de-
cision makers t o use the advertising to sales ratio as the decision
7variable
The zero covariance assumption on the other hand1 is particularly
troublesome Within a probably not true since
many even ts outside the
single industry it is
industry would have
industries
similar effec ts on all the
8firms in the industry Between it is probably not true either
since many companies produce in more than one indus try and even ts in the
firm would tend to effect all of its activities The extension of the
model to allow for non-zero covariances is certainly called for
--- -
Page Ll
Athird characteristic of this specification also deserves some comment
only on equation is shown for the firm or industry Several haveinvestigators
included an advertising equation ina larger system of equations (Comanor and
Wilson (1974) (1979) and Weiss (1976) so in myMartin Strickland Idid
dissertation1 Ihave ewo defenses for presenting only single equation results
here One is the suggestion by Comanor Wilson that simultaneousand equation
verybias may not be important in this context he second is that Iam will-
ing to see the model extended to include a profitability equation and I intend
to move that direction in the near futurein
Onthe inclusion of aconcentration equation on the other hand I amskep-
tical he problem isthat in some fundamental sense equations in models of
industrial organization should be oriented to the decision-making contexts of firms
Ifeach firm in anindustry with Nfirms has only one decision variable then
there can be only Nindependent equations which are based on the first-order
optimization equations It not show large ofis difficult of course to a number
functional relations which follow from the first-order conditions point here he
isthat only Nof them can be independent
Consider for illustrative purposes only a situation in wich teh N
itsquantities say q are set exogeneously and each firm determines advertising1
of and level say ai where a is the true optimal value a a bulla
1 1 i i + i
Ifp a) then the Nq
s and the Na s will determine N p s say pii bullp
i(l i 1 i
Now given Eand 9 we can determine both the vector s andthe vector v since
- bull til
ands We can also determine S s z bull s-1 s-ii
v and we
industry concentration isHbull (alternatively
V and Hare functions of the same
pendent Furthermore if there are
VM) and (H1 H2 bull bull bull )contain
VI STATISTI CAL APPLICATIONS
Page 18
Given the observed values of might be tempted to conclude that we
have a two equation system in which v and z are jointly determined That would i i
be inappropriate however since the 2N random variables (v ) are determined by
onlt N basic random variables t The variance-covariance matrix for ( ) must
be singular bull
The extension of this observation to industry level variables is straightshy
forward Since observed industry advertising intensity is V bull l v and observed 4
C4 bull z ) it follows thati l i
set of random variables t and cannot be indeshy-
M industries then the variables (V1 v2 at most M independent random variables
Given the assumptions made above y a and o are constant within a
given industry That being the case (34) can be written as
(37)
s1 bull (1 - o) (y - a)
If and
So + +
(1 - o) a + oy
81 are esti mated for an industry using a suitable statisticalSo procedure giving and 81 an estimate of y is then available The sumSo
Intra-industry analysis
1974 of Business Re ort -
categories are identified
Page 19
of So and S1 is y so y bull So + S1 is the estimate of y
If o is assumed to be bounded by zero and one the sign of 61 is also
the sign of y - a The estimate of So is of cou se an estimate of
(1 - o ) a + oy Moreove if y gt a then a is less than 6o- (1 - B) a+ 0(
Cor esponding results would
hold if y bull a or y lt a
If 61 gt 0 then we would estimate that a lt 6o
bull
For purposes of estimating this model 32 manufacturing industry categories
were selected The data are fram the Line of Business forms filed with
the Federal Trade Commission they are discussed in detail in the Annual Line
which is available from the FTC The 32 industry
in Table 1
Simple regressions for equ ation (37) were run for each of the 32 industries
with data for each of the firms which filed in the industry constituting an ob-
servation The results are given in Table 2 lines lA 2A
1974
the intercept
For all the A
equations generalized least quares was used th both and market
share bein scaled bv the square root of sales
The 32 industries were taken from the larger set of all manufacturing industry
categories in the 1974 data files Two selection cr iteria were used thereLB
had to be at least ten reporting companies and the industry had to be a consumer
goods industry or a producer goods industry which is strongly associated with a
related consumer goods industry so that adve rtising by the firms in the producer
goods industry impacts on demand in the related industry (eg soft drink syrup
and soft drinks) There were 135 industries with at least ten companies - 103
producer goods and 32 consumer goods
273
274
TABLE 1 INDUSTRY CATEGORIES USED
FTC
COOE OESCRI P T I ON _
2001 MEAT PACKINGt SAUSAGE S ANO OTHER PREPARED
MEAT PROOUCTS
2004 DAIRY PRODUCTS EXC F LUID MILK
2005 CANNED SPECIALTIES
2001 FROZEN SPECIALTIES
OEHYORATFOt AND
I NC LU 0 ING PRE SERVE S
SOUP MIXESbull
SEASO lNGSt ANO
OTHER PET FOOJ
RAIN MILL PROOUCTSt
ANO PREPARED
RELATED PRODUCTS
PRODUCTS
Z008 CANNEOt ORIEOt PICKLED FRUITS
ANO VEGET ABLES JAMS JELLIES DEHYDRATED I G TA3t E
S40CES ANO SALAD ORESSINGS
2010 DOG CAT A O
2012 FlOUR t OTHER RICE
MILLING fiLENOE) FLOUR
2014 B EADt CAKE ANO
20113 CO FECTIONERY
2026 BOTTLED A O CANNED SOFT DRINKS 2021 FLAVORING EXTRACTS AND SYRUPS EC
2029 MISC FOOOS AND KINDRED PROJUCTS EXC
ROASTED COFFEE
2301 MENS AND BOYS SUITS AND COATS
2302 MENS ANO BOYS FURNISHINGS
230 3 kOMENS AND MISSES OUTER EAR
2551 HOUSEHOLD FURNITURE
2702 PERIODICALS
Pqe 19a 1
RELATED 1972
src cooe
2011 f 3
202tX2026
2032
203pound
2033t4t5
2047 2041t4t5
2051 2065 2086
2081
209XZ05
231
232
233
251 272
2703 BOOKS
2704 ISC PVeLISHING
2806 ORGANIC FlEERS
2807 DRUG$ ETHICAL
23234
PT 2 S 3
2808 ORUCSt PROPRIETARY
2809 PERFU ES t COSMETICS ANO OTHER TaiLET
PREPARATIONS
2810 SOAP AND OTHER CLEANING PREPARATIONS 2815
2901
PEST1CIDES AND AGR ICULTURAL
PE OLEUM REF I N ING CHEMICALS EC
3606 HOUSEHOLD COOKING EQUIPMENT
3612 HOUSEHOLD APPLIANC ES NEC INCLUJING ELECshy
rqlC HOUSE ARFS AND FANS ANO SE I G MACHINES
3617 RADIO AND T V RECE I VING SETS 3808 PHOTOG APHIC E UIP amp SUP LfESt EXC
3qo3
PHO TOCOPYING
SPOR TING A D E QU I P M ENT f SUP P L I E S ATHLETTC GOODS NEC
3904 DOLLS GAMESt TOYS AND CHLOqENtS VE ICLES
PT283
2844
234tX2844
287 9
291 3631
3634t6t9
36H
PT386l
3949
15
943
middot
l A6l f 2 R(GRESSION RESULTS
EQ I NO
NR CODE cu 121
lA 2001 A (
2A 2004 r (
3A 2005 a
-R
ClS amp31
16
17
1 z
I TER-CEPT bull t
ooo9t21At 00022311 oooeos uu
OOl49tCAI 0010 2 middot 000960CCI
o049tl1At o04t55C61
MARKET SH ARE bull 5 I
-010670Ctl
-017600181
-001050
-0211AO
-0077 0
AOVER-TIS INti
lbl
000()39 000072
000018 0 0024liAI
-000079
SHARf
4ampgtVf Q bull
111
000140
-000170
RSQ
COl
66 485 862
607 6 3 t ()2
blti6 633
flASTI-
CI T Y PH
-007271(1
000441
-u0288l
clASTSHU (t)Ef
bull 10
tto
-234
210
( OOlltOCCI -126520 000922 181 001790
4A 200 7 18 00590liAI -038670 4e16 -032110 lell u c
OOOS74e 000700 025940
00067018) 0021001AI -02l100CAI
5 26
920
5A 6 c
200 8 26 0 03Sl9CA I 0012 26181 o 024511(
-005540
-0 81420 0001961AI O()Ol90CAI ooll240
441 604 891
-0020 19 274
6A 2010 16 Oe04224CAI o359lOIAI 9oq ll4015dCAI 061 6 -middot 005435CAI 00011 8 1AI 917 c 00535ltAI -009850((1 OOOA51J CAI -001610161 971
7A 2012 001862 000010 351 00187) oo1 a oooq45 ono1o6 439 c 001109 -027050 Oe0067SIAI -OOll60C81 897
- middotmiddot- -
8A 201t ll 000757 Ol4140CCI 814 ott9oocet 095 8 oooqo6 C C J 000069(81 811 c ootl2lampBI -080910CCI Oe009lUBI -OOtlOOCCI 960I
I DIIJIT pound11 1V OTIUTVTfAll-1pound1t- A _ 00 R - qq r - Q
1jl)OQCD
0
-middot middot--
l
_ __ _
I
i
0 00141 c c
91)7
I
T ABLE 2 REG ESSIOH RESULTS CCONTI
EO NR
IND CODE
iR bull
cos
I TEq-
CEPT
t1ARKE T
SHA E
AOVER-
Tl SING
SHARpound-cent
AuVER
RSO ILAST1-
C I T Y ELAST
SH CPtf (lt Cl) bull 11 I 41 I 5I (61 171 I 8 I f9) c 10)
9A 2 0 1 8 19 002397181 -007640 367 -005 246 le46 6 000605 000550CAI 597
-000029 G14180 OOl654CAJ -024li01At
lOA 2026 12 0028901AI -OlZl90 -0114 0 1 1 1 5 759
Ll (
llA 2021
f3 c
12A lOZ9
B
(
13A 2301
B
c
14A 2302
B
c
154 201
n (
16A 2551
8
10
35
10
20
19
24
001897(()
00141 6CCI
Oll9lHCI
010604 middot middot
Oll083
OOJ11JCAt
002059141
001411CIH
-000412
000293(81
000187
001159 (3)
ooo6a4cct
o00792CCI
OOl40dCAt
000006
-000076
OOlltl6fA)
o o0 5 64CCt
-034570
-o1middot o3o
-1659 90
Ol51lO
-046190CAt
Ol5450CA)
-005l60
002940
-030060(()
-0 43l50CCt
0 33650
-019080
00004
bullJ00741CAI
-000069
000466
0001 51CAt
000947CAt
O OOI)70CAt OOl750CA J
000175CCI
000671CAI
OOll5) f A t 002207CBI
0004b7CA I
- 007 1 70
001760
-001410CAt
- 0 16200CAt
-004210
- 1 207 70
705
96 0
542
466 805
460
614
921
832
9 61
996
424
550
9 15
56 1 651
884
-006114
019 037
03504lCA)
oo409Q
-141644((
491 -0 17664
607
871
lell
084
101
072
103
108 11IllOQfl)-- l
-037880fCJ OOl6l4CAJ -040l70fCI 0()
c
I
I
bull 9
9t
middot middot middotbullmiddot bull lraquo
TABLE 2 REGRE SION RESUlTS ICONTt
bull
EO I NO tHl I filER- f4A_K T AOVER- sbulluRE ttSQ flA Jl- iLAS T NR CODE cos CEPT SHARE JISING AOVfR bull (JTY SHR CiJCF
110Cl I f amp Itt 15J tbt bull 1c u caa
llA 2702 ll o079 6 81 Bt -019110 S6s -OlltO 111
a 004679 ooooo1 t69
( oOl26o -015160 001560 -022690 ltJr()I
ldA 270 3 19 0065lt -OIt2l60 220 -Ol57a2 11 8 e o 0 71t6IA t ol08 75tAt ( OOl983CiU -060llOC() OOl7191A t -O lltdOC A) middot 89
lqA 2704 10 005082 -Oil060 22b -l)i1977 lelO 6 -000080 oo2772C4 t 910 c 000907 -011 200 Oe04105CA t -046960 995
20A 2d06 1 II
O OlltStiAt -001120 837 00014 -))th O01380amp At -oooo 11 821 ( OOl526CAt -026700181 0002 5UAt 000530 969
2lA 2807 28 OOt9lt9CAt -00891 0 516 -0 03962 225I8 002095CCt OOOl lliAt 690 ( o02965CA t -056650((1 OOOt50CAJ -003l70ttl 911I
I 22A 2808 15 ot3l06 119730 655 1130) 090
0 - 007526CC t Oe003961At 835- middot-
c I oo7523 -067210 001l 2tA t -0091t80CCt 951
2lA 2809 21 Oe2067lCAt -1218101At 71 2 -10095((1 121 ttlIaOQ8 00797418) OOOI8lltt 686 Ill( 0092lU A t -Olltl70Ct t 000605CAt -OOit8701At 951 _
---- -
32 009506CAt
0 bull 06 981 I A J 006901 cA bull
00 24A 2810 -010960 5t 5 -OOltSO 7 56
8 -000011 s11 c -5811701At oooaoatAJ oot520IAI 888
-- 37 --- middot - -- -1 ) l lbull t5JJ r-l bullhmiddot1 --- - --
l5A
26A
27A
c
2RA
c
2 9A
31A
llA
24
13
597
Ci)55
14
l 6h
RSQ ELASTI- (lASTCOE f-
TABLE 2
EQ I NO NR CODE IU
REGRESSION RESULTS CCONTa
NR INTER- MARKET
(10S CEPT SHARE AOVER- SHARfcent
TIS lNG AOVER
) CITY StiR
c It) bull 5 co a I 1) bull 0 bull 9 c l 0 bull( 2 1
2815 16 003019 -009990 275 -o069bmiddotJ
6 001446 000224 282 c 002178 -054630 o02221CCI -0075 00 646
2901 29 00021t2 CAI 0)0520 592 0007 6 1 068
0 00105C8 t 000012141 60)g( OOOlJ51AJ -00126018) ooou271At -000u80CA) 974
36oa 1 3 003296(4) -oca29o -oo-q94
001755 000203 lt91t 002499((1 -04l820CCI OOl374CAI 0068)0 946
3612 0033281CI 011520 195 01684) oso
00 30 42C OI 00012 2 424 oo579ltAt -2 7l7801At 002 100IAI -003220 819
8 3617 btel 011789 081002259 009530
001448 oOOl OitCCI 732 OOl7 22CAI -0 46160CAI 0004961AI -OOll40181 989c
30A 3808 14 0021t 3 6fAI -000110 628 o023291C) -005 6 middot- _ middot- 0bull 02 04 1 IC I ooooo9 637 ( 001584((1 -0161)1)0 00021 91AI -0 00060
390) 14 002790CCt 006590 h8 t 0 0938 4 070
B OOl89bCCJ 000144 74
0021961(1 -OItlOlO OOl323CAI -004800 915
B
middot--- - middot 4 - middot- - dfll3904 12 004563181 0)8300 879 0428651(1 089 OQIll004 277CA I ooozstcea 919
c 0041)3181 -0-11250 0011881(1 -006270 972 0Ill
- J - bull bull bull bullbullbull middot - - middot - middot gt middotbull L)ofbullflbull middot middotfJltl -tn- middotmiddotmiddotmiddoto +f middot t middot bull
c
is
positive
2
Page 0 determineThe consumer goodproducer good split is sometimes difficult to
Sollle support for the split used in this paper is a comparison of goodness - of - fit
distributions of a
industries
set
advertising are
that industry
a sales increase of
the
highest aD10U3 those
measures for the two groups of industries Table 3 contains
measures for the 32- consumer goods industries and the 103 producer goods
2 Median R is 5 7 for the consumer goods set and 40 for the producer goods
Estimates - of y the elasticity of demand with respect to
given in column of Table 2 The hypothesis that y bull 0 with
alternative was applied for each ittdustry As shown in the
could not be rtjected for 24 of the 32 industries Of the
and 10 are positive For the remalnifg eight three are
and five are significantly positive
Of the threertegative elasetcities the largest (in absolute value)
cosmetics (code 2809) With all elasticity of -1 01 a one
industry advertising would gwerate a decrease in industry sales
more than one percetu
At the other extrema is toys and games (code 39 04) For
a one percent ittcrease in industry advertising would lead to
four-tenths of a pampreentmiddot Though not significantly different from zero
elasticity estimatbull for pt-bpuetary drugs (code 2808)
which are
Both drug industries have
to sales rafios (211 pe1 cent for 28 08 and 13 4 per cent
other hand flaIJo ing ext1ac s (mainly soft-drink SYUlS) also
is for
percent increase in
of slightly
a middottwo - tailed
table that hypothesis
24 14 are negative
significantly negative
9
the codetics and p-ropriacuy very high advertising
for 28 09) bull On the
has a high ind UY
advenisiltg t() sales ratio ( 8 0 per cent) bull but its estimated elasticity is very
small and not signifieantly different uom zero
middot It doe$ not seem unusual to find many elasticities near zero with a few
negative and a few positive If aggregate consumption is insensitive to aggregate
advertising as seems likely ld then what one industry gains must be lost by
others
Page 21
elasticity of demand does
(10) we can
if t 0 then
Finally a determination concerning the industry
not fix the individual firms elasticity Going back to equation
write y bull (1 - ) a + z y Since a gt 0 it follows thatii i i
Y gt 0 If Ylt 0 however y s sign is indeterminateij ij
Turning next to the coefficient of market share 20 of them are negative and
12 are positive with three of each sign being significantly different from zero
The largest positive value which is significant is for pet food (code 2010) and
a close second is mens amp boys suits and coats (code 23 01) The largest signifi
cant negative value is for cosmetics (code 28 09)
In the model some relations among y a 0 and 61 are implied If middota is
negative (1- y) is greater than one Since 61Y bull (1- a) (1 - ) 61Y is
pCsitive but it will be less than one if 0 is large enough Seventeen of the A
ys are negative in each of the seventeen industries 61 is also negative and les
1 (algebraicly) than y so 61Y is greater than one his is not evidence that the
degree of cooperation is low or zero in any industry of course A high o togethegt
with a negative y which is small (in absolute terms) relative to a can give a
value of 61Y greater than one On the other hand if 61Y were less than one
o would have to be greater than zero
For y gt 0 two conditions may hold If y lt a it follows that 61 lt 0 since
(y - a ) lt 0 It also follows that 61Y lt 0 regardless of the value of o bull
In addition the relation is an if-and-only-if one ie if 61Y is negative
then y lt a Three industries have negative values for 61y dairy products
except milk (code 20 04) synthetic fibers (code 2806) and photographic equipment
(code 3808) In all three cases y is quite small so a does not have to be very
large to give a negative value to al
If y gt a the model shows that 61-- gt 0 whatever the value of 0 bull There are
12 cases where both y and 61 are positive with 61 ranging from near zero
(grain milling code 20 12) to 1 20 ( proprietary drugs code 2808) Several
coeffiei-en_rs
--
15
o f to zero
Tab le 3 - Consumer Goods
Indus tr ies
2Distributions of R and Producer Goods
Page 20a
Consumer Producer
Goods Goods
0 - 9 9 0 3
10 - 19 9 0 1 1
20 - 29 9 3 20
30 - 39 9 3 17
40 - 49 9 5 16
50 - 59 9 8
69 9 6 Ii 1060
--f 670 79 9 2
80 - 89 9 4 4
90 - 100 1 1
Totals 32 10 3
Notes
2is used R i s cal-
culated as F F + ((N - K- 1 ) K ] 1 b_ F is the F
statistic Lor the hypothesis that all o f the non - intercept terms are simul taneously equal
1
2
--
-- Since generalized least squ-area regress ion
Page 12
o f these cases have values o f 6 1 Y rea ter than 0 8 (pet food (20 10) bread amp cakes
(20 14 ) miscellaneous processed foods (20 29 ) men s and boys suit s and coats
(23 01 ) p rop riet ary drugs (28 08 ) elect ric housewares (36 1 2) radio and
IV sets 17 and games and toys (39 For be than(36 ) 04)) 6 1 Y to greater 0 8
both o and cry mus t b e less than 0 2 That is in these industries the results
shown in Tab le 2 would hold only with very strong dominance of marke t demand
effects over brand swit ching e ffects and a very low degree o f cooperation
Using the concentration rat ios given in Table 4 the average concentration
ratio for the eight indust ries is 38 1 For the remaining 24 industries it is
43 2 This result would lend some support t o the proposition that concentrat ion
and cooperation are related if the group of eight industries are identi fied
as having low degrees of cooperation
One f inal note on model predict ions S ince both cr and o are assumed to
be non-negative 60 should also be non-negat ive In only one o f the 32 cas es
is 60 negative and it is not signif icant bull
cor pJ
7 1
7 0 4
1 4 7 5 1
)
7 3 z a o a t s 1 1 3 3 5 ( i V Z 8 tJ1 2 1 - 1 1 0 7 3 5 c middot v 2 8 1 0 3 2 -0 0 1 4 3 0 C 1 V 9 8 3 2 8 1 5 1 6 - 0 0 6 t1 6 9 C _ V 1 11 0 2 9 0 1 29 0 00 7 6 1 C l k V 7 3 1 3 6 08 1 3 - o n 4 9 9 4 3 3 3 6 1 2 2 4 0 1 6 4 3
3 1 7 1 3 0 1 1 7 d 7 8 3 3 8 0 8 0 0 2 3 i 2 3 0 6
P 4
3 9 0 3 1 4
1 2
0 09 3 8 4 39 1)1 0 4 l 8 6 5
T A LE 4 e I NCU S T R Y C A T E G O R Y
F TC NOM e tOOE ass
C 1 I c 2 )
2 0 0 1 1 6 2 0 0r 1 7 zo o l Z 2 0 0 7 1 8
z o o 2 6
z o t a 1 6
2 1) 1 2 1 5 2 0 1 4 1 1 2 0 1 a 1 9 2 0 26 1 2 2 0 2 7 1 0 2 0 29 - 3 5
2 3 0 1 10
2 3 02 z o
2 3 03 1 9
2 5 5 1 2 4 Z 1 0 Z 1 1 2 0 3 1 9 2 7 4 1 0 z a o 1 3
2 A 0 7 z s
E LA S T I -C I T Y
c 3 )
- 0 09 7 2 7
0 00 449
- 0 0 2 8 9 3
- 0 3 2 7 70 - 0 OZ il I q
o 4 u t s a 0 0 1 3 7 3
o 1 4 o o-0 0 5 2 4o -0 1 94 0 1
-C l 6 l l 4 0 1 JJ 3 7 0 3 t 0 4 2 0 0 4 0 9
- 0 4 1 8 44 - O l 7 t ) 4 -ll 7 1 4 0 b -0 3 5 7 8 Z
- 0 1 9 7 7 0 1)0 3 3 4
-o n 9 2
IJ A T A
CONV
ON-C ONV
( 4 )
t 1NV
C 1t-V C NV c ) v C - JNV C rlNV c n v c a v C ON V ( i) l V
C CtV
C t1tV C IJNV
C 1 V
c 1 v
Pa ge 2La
HEQ F t Noe x
c 5 )
6 2q
6 89 1 48
6 1 4 ) q 3
1 9 7 1 9
q z t s 5 a e z
l l 9 1
4 o 7 1 2 0
c oqt 1 Z
4 4 )-q 1 z
5 5 9 1 0 64 1 3 3
n 3
Inter industry analysi
Page 23
Equation ( 3 6 ) may be used to fo rmulate a rough test o f the cooperat ion
vs e f ficiency controversy I f we rest ricted ourselves to some subset o f
indus tries where the relat ion be tween y and a i s the same we would be lookingbull
at a sample o f indus tries for which the dependence o f V on H and o is a
quadratic wi th no squared terms
I f we now sub s titute (26) into ( 36 ) we get
(38) v bull a + (y - a) R + (y - a) o (H) - (y - a) H o (H) + u
If it is only d if ferences in marke t shares that affect advertising inten-
s i ty levels i f concentration does not af fect cooperation and i f o bull 0
( 38 ) reduces t o
( 39 ) V bull a + (y - a) H + U
Let this b e the null hyp o thes is and note that V is a linear func-
tion o f H If y gt a and the null hyp othesis holds the expected relat ion
between V and H is as shown by the pos itively sloped straight line in Figure
6 b etween points A and C
In this context the alterna ive hypothesis the degree o f
cooperation ( o) is a positive function o f concentrat ion
is that
(H) I f the func-
tion is as described in (26) and Figure 4 the relation between V and H
would be as shown by the curved line connec t ing po ints A and C in Figure 6 One way to test the null hypothe s is against this al ternative is to test fo r
the hypothesized curvature in the function relating V and H
Fi gure
R
6
v
ID c
H 1 0
I
i I I
sfigned
clasJ s
in def ini g
effec t is the
I
Page middot2lf
One charac teristic of produc ts which may be useful
classes wi th respec t to significanc e of the brand-switching
conveniencenon-convenienc e distinc tion developed by Porter ( 1 97 )
Using his identif ication of IRS indus tries as a s tar t ing point
each of the 32 industries used in this s tudy to one of the two
The ass ignments are not ed in Tab le 4
Using the conveniencenon-convenienc e dis t inc tion the
divided into two sub-samples
sample was
Four regressions were then run bull ene for e ch
sub-sample one for the to tal sample and one for the total sample with
a dummy variable for convenience goods Each regression was run wi th LB s
and wi th indus tries as ob servations The results are given in Tab le 5
For the LB level equations ( 1A - 1D) equation(37) was used ome
rearranging of terms gives v bull y - ( 1-o ) (y - a) ( 1-z ) + ui i i
bull y - 6 1 ( 1 - z ) + ui When observations are pooled across indus trif s yi
is an industry variable In these regressions the values o f y es t imated
in the first stage were used Since industries have been grouped so as
to reduce differences in (y - a) that term is t reated as a cons tant
within each group Finally H the Herfindahl index has been subst ituted
for o the degree of cooperation The resultant equation is linear
y ( 1 - z ) and H x ( l - z ) This is the equation for whi ch results arei i
reported in Table 5 In all regress ions all variables were weighted by the
square roo t of sales
in
raquo
- 34
3 1
i l T 4 8LE S R E G E S S I ON R E SUL T S i X T f N S I ON S T O T Y PE Of
AND L E V E L OF C ONC E kT R A T I O
E O O tt S
tP bull T Y P E NR C ONV O B S tO N-CONV
E L A S T I shy
T I C J T Y
O E - t-41( T SliA R E
H( fl F
1 1 - K O x middot K O X E R F
- H S a 1 1 - H S t 1 8 1 1 9 1
R S O
1 1 0 1
q z
t - bull 9 f 2 1
Z t P D l 6 fC ONV - 1 9 1 s s c 2 1
e - l O I e bull u q a
2 3 1 0 1 1 Nl N- t O V 5 7 1 ( 1 2 19 1 6 1 - J l S 2 5 4
l l l l t z q o a 1 - J o v z I
l ( I NO l l A l l ol u l t o o c bull J u l l
d S
c 3 4 t t r a l
bull Z D 1 0 l l 4 l l 2 3 -gt 2 l l 2 0 - 4 lS 5 4 l C 4 1 o e 6 t l b b 5 1 1 0 1 1 1 c - 1 9 5 1 t d S 2 1 tV
OQ10 T-R A T I O S A E G I VE N l P A RE N T H E S E S l E V E L S f S I G 1 F l C ANC E A - 99 5 B - q 9 t - s NS J UC E NO I Nl E R C ( P T I S U (j f middot) I N f Hf t OUA TIOI S 6 E f 3R E A O J U S T H E N T F OR 1-t E T E R O SIC E O A S T f C I T V pIraquo
1
middot
I
R SO I S C A L C UL A T E D A S f I f bull 1 1 bull 1lt 1 K t t- E E N I S THf U A1 B E R O f fJe S E fi V A T I O S F t S HtE F - S T A T I S
I N DE P E ND E N T y II
T I C F O T H E U Y P O HIE S I S f A l l C JE F F I C H N T A R E QU A L HI l f RO A f40 K I S T H f UUM 8 E R OF Q ( A fL f S
I l l
1 A
1 6
t C
f 1 r R I l l
1 2 1
L Jt
L B
l q
L 1
on s T Y P E bull 2 )
P OUC T
X
H S I l l
l O l A I bull l 0 7 1
O O b O J
1 2 00 1
1 b 2 1 C I l l Z U
l l S b 1 -1 1 0 1
rlt F X l l - fi C f l
bull 1 I
l l
- ()4 2 3 l f H
1 - 2 7 9 1
K O X
1 1 - ti E R F I
l d l
S4 l i A I 2 l b
1 3 00
K O X -t f R F t S J
I 1 - HER F t
)
60 9
z o bull
0
t l l
3 8 8
1 7 6
5 64
64
N f bull
tl S bull
I l l
I 4 1
t ONV
NON-t ONV
A l l
A ll
c o v NO- ( ONV
1 4 1
1 5 1
- l S b
c - 1 1 2 1
5 8 2 C A I 1 6 7 0 1
middot I l 9 H
l 1 il
1 4 1 6 1
E L A C) T J shy
l t C I T Y
I S I
l b l
- 0 2 0
1 - 1 4 9 1
O Z 3 2 1 A I C b 6 b l
0 0 9 2 -J
1 1 b 4 1
o z t t t l 1 2 2 1 1
oru - Ht tt F
1 gt X bull b I
Page 25 bull
Mean val -as ior relevan t variables for the two sub-samples are
Conv Non-Conv
Adv Sales 4 7 2 3
iit Share 2 7 3 2
Ela cicity - 05 3 02 2
Herf indahl 7 3 1 860
That the dist nct ion between coft9enienee goods and non - convenience
goods mat ters is supported not only by these means but also by the regression
results Equat ions lA and lB are different in almost every respec t Only the
concentrat ionmarket share int eract ion term seems to mat ter for convenience
good s Jus t he opposite holds for non - convenience goods both elasticity1
and market sl e are highly significant but the concentrat ionmarket share L
int erac t ion t=rm is not
Given that one minus market share is used these equ3 tions the coeffici t
of market share is the negative of what is sh in col (6) of Table 5 For the
regress ion for non - convenience LB s reported in equation lB then LB s with larger
market shares have lower advertising to sales ratios
On the question of the impact of the degree of cooperation the story is mixed
For convenienc good s the coefficient is pos itive as equation (31) predicts when
there is a po sitive degree of cooperat ion Fo r the non-convenience goods LB s middot
concentration has no effec t
Aggregat ion t o the indus try level and then rer unning the equations $ives the
results shown in (2a-2d) of Tab le 5 As expected R2 goes up and levels of significanc e
fmiddotJr individual coefficients gJ down In add ition the coeffic ient of H ( l -H) inmiddot
quat ion 2A is no t significant though it is po s i tive Since 2A is the indus try level
I
--
all but one of them and that one is not significant For 18 of the 31
wo uld have on its adve rtising to sal es ratio
cognate of LA the variable H ( l-H) pl ays the same role in 2A as H ( l -z 1) p lays
in LA
Vconomies of scale in advert ising
The literature on advert ing contains a number of definitionbull o f economies
of scale and of findings eone rning its presence and magnitude Before I loo k
a t the data with the hope o f mmenting about scale effect s then I wan t to
define the term as I use it
By economies of scale I mean a decrease in the number o f units of adver-
tising which are needed to generate a unit of sales as advert ising is increased
Given the definition of the firm s own elast icity of demand with respect to
advertising (see the text foll wing equation (1) ) the technical definition
is that rii gt 1 Diseconomies of scale are def ined symmetrically i e
Yu lt 1
ith a cro s s - section oi observations on advertising intensity
(v bull a s1) and onmiddot advertising (a ) we may regress the former on the latter 1 i i
The results of such regress ions for the 32 industries are given in Table 2
lines lB ZB bull bull bull 32B For the 32 eases the regression coefficient is positive it
positive coefficients the coefficient i s signif icant
The que tion now is whether these results ar e evidence of pervasive dis-
economies of scale in advertislng I think no t Given my as sumption that the
data I observe are equilibrium results I may ri ghtly conclud e that almost every-
where high levels of advert ising are as sociated in equilibrium with high ratios
of advertising to sales I may no t conclud e anything however about the
impact that a move by some firm away from its equilibrium level o f advertising
middot- Page 27
Given the assumed relation in equation ( 37 ) there is a corre spondshy
ing equilibrium equation relating v and ai The equation is non - linear i
and I have not found a simp le way to characterize it It is po ssib le to
det ermi ne its deriviative with respect to a b y takin$ tne de riviative of ( 3 7 ) i
When that is done the result is
v - 1(40) i bull sl a a z ( 1-z )i i ira-l
Sinc e all the other terms in (40) are non-negative the sign dep ends only
on the sign of 6 1 which is in turn de pend ent only on the sign of (y - a)
That the signs of the coefficients of a in equations lB 2B bull bull bull 32B arei
not always equal to the signs of the coefficients of z in LA 2A bull bull bull 32A i
is still a mystery to me and something to be explored
bull
Summary
My purpose in writing this pa per wa s t o attempt to integrate explicit micro-
economic mode lling reasoned econome tric specification of error terms and high
quality data to exp lore some questions about adver tising by large firms
al quite important improvements could be made in the first two of those areas
and much data ma ssag ing is yet to be done in the third
Concerning the subst an ce of what I have done I think six things are im-
Page 2 8
Severshy
V and Conclusions
portant
1 Explicit modeling is worth the ef for t if for no other reas on than
that it provides a basis for choosing fr om a varie ty of funct ional
forms
2 The same is true for error specif ica t ion I did in fact look at scatshy
ter diagrams for all 32 industry categor ies and they vir tually all show
the het eroscedast icity which I as sumed and for which I correc ted
3 The predicted relation be tween advertis ing in tensity and market share
shows up clearly in on ly 25 of the case s examined I have ye t to
explore why that may b e the cas e
4 Some evidence concerning the presence and impact of coopera tion was
produced but it is not clear ly pervasive
5 The dis tinction between convenience goods and non-c onvenie nce goods is
unambiguously a good one
6 Virtually no evidence concerning econ omies of scale in adver tis ing can
be gleaned fr om this study given its assumpt ions
More work is called for on many of these issues my study seems to have
raised more questions than it has answer ed This is pr obab lly due to the richshy
ne ss of the data source since I had the oppor tunity to address several issue s
at once
- -- - Q --
Foo tnotes
Manager Line o f Business Program Federal Trade Commission The author depended heavily on several staf f members of the LB Pro gram for statistical and clerical support in particular Joe Cholka George Pascoe and Haro lene Jenkins Helpful comments were given b y Richard Caves Dennis Mueller Michael Porter and F M Scherer When the paper was in its early stages o f development both S teve Garber and Jon Rasmussen former economists in the LB Program provided eery useful cri tiques The views expressed here are my own o f course
1 The basic exp licit model of the role of advertising is Dorfman and S teiner ( 1954 ) and the models in this paper may be seen as ano ther extens ion of their work
2 For a review o f models of oligopoly which treat prices and quantities as relevant variables see Shubik ( 1 959) Extens ions to take advertis ing int o account are fairly s traightforward
3 Dorfman and S teiner dealt with a price-setting firm For a monopolis t inverting the demand relation to show Pbull P (Q A) _instead o f Q- (P A) has no effect on the results o f the analysis For oligopolists who face a sys tem of demand relations inversion does have some impac t on the results The dif ferences are not trivial nonetheless they will no t be explored in this study
4 Shubik ( 1 959) p
5 From a formal mathematical point o f view the function is similar to a1 i Legrangian funct ion of the form - r rr
L bull 11 + A ( j - (i i i+j
where the indicates some fixed level of the profits industry Following the same logic which resul ts in and income constraint as the marginal utility of the of the indus tr in the obj ective of firm i
6 Hall amp Weiss ( 1 967) p 323
7 See Schmallensee ( 1 97 2 ) Ch 2 pp 1 6-47
8 On this poin t see Imel amp Helmberger ( 1 9 7 1 )
9 Long ( 1 970) p 130 - 1 39
1 0 See Schmallensee ( 1 972) Ch 3 pp 48-8 7
rrj i+j ) of the rest of the
the identification profits of the rest
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980
P age 16
The error term specifi c ation which follows Hall
wi th the ratio of profits
6 heir analysis seems
as wellbull particularly given
and Weiss was
applied by them to equa tions to equity or
asse ts as dependen t variables to carry the same
weight in the present context the substantial
evidence presented by marketing analysis on the tendency of company de-
cision makers t o use the advertising to sales ratio as the decision
7variable
The zero covariance assumption on the other hand1 is particularly
troublesome Within a probably not true since
many even ts outside the
single industry it is
industry would have
industries
similar effec ts on all the
8firms in the industry Between it is probably not true either
since many companies produce in more than one indus try and even ts in the
firm would tend to effect all of its activities The extension of the
model to allow for non-zero covariances is certainly called for
--- -
Page Ll
Athird characteristic of this specification also deserves some comment
only on equation is shown for the firm or industry Several haveinvestigators
included an advertising equation ina larger system of equations (Comanor and
Wilson (1974) (1979) and Weiss (1976) so in myMartin Strickland Idid
dissertation1 Ihave ewo defenses for presenting only single equation results
here One is the suggestion by Comanor Wilson that simultaneousand equation
verybias may not be important in this context he second is that Iam will-
ing to see the model extended to include a profitability equation and I intend
to move that direction in the near futurein
Onthe inclusion of aconcentration equation on the other hand I amskep-
tical he problem isthat in some fundamental sense equations in models of
industrial organization should be oriented to the decision-making contexts of firms
Ifeach firm in anindustry with Nfirms has only one decision variable then
there can be only Nindependent equations which are based on the first-order
optimization equations It not show large ofis difficult of course to a number
functional relations which follow from the first-order conditions point here he
isthat only Nof them can be independent
Consider for illustrative purposes only a situation in wich teh N
itsquantities say q are set exogeneously and each firm determines advertising1
of and level say ai where a is the true optimal value a a bulla
1 1 i i + i
Ifp a) then the Nq
s and the Na s will determine N p s say pii bullp
i(l i 1 i
Now given Eand 9 we can determine both the vector s andthe vector v since
- bull til
ands We can also determine S s z bull s-1 s-ii
v and we
industry concentration isHbull (alternatively
V and Hare functions of the same
pendent Furthermore if there are
VM) and (H1 H2 bull bull bull )contain
VI STATISTI CAL APPLICATIONS
Page 18
Given the observed values of might be tempted to conclude that we
have a two equation system in which v and z are jointly determined That would i i
be inappropriate however since the 2N random variables (v ) are determined by
onlt N basic random variables t The variance-covariance matrix for ( ) must
be singular bull
The extension of this observation to industry level variables is straightshy
forward Since observed industry advertising intensity is V bull l v and observed 4
C4 bull z ) it follows thati l i
set of random variables t and cannot be indeshy-
M industries then the variables (V1 v2 at most M independent random variables
Given the assumptions made above y a and o are constant within a
given industry That being the case (34) can be written as
(37)
s1 bull (1 - o) (y - a)
If and
So + +
(1 - o) a + oy
81 are esti mated for an industry using a suitable statisticalSo procedure giving and 81 an estimate of y is then available The sumSo
Intra-industry analysis
1974 of Business Re ort -
categories are identified
Page 19
of So and S1 is y so y bull So + S1 is the estimate of y
If o is assumed to be bounded by zero and one the sign of 61 is also
the sign of y - a The estimate of So is of cou se an estimate of
(1 - o ) a + oy Moreove if y gt a then a is less than 6o- (1 - B) a+ 0(
Cor esponding results would
hold if y bull a or y lt a
If 61 gt 0 then we would estimate that a lt 6o
bull
For purposes of estimating this model 32 manufacturing industry categories
were selected The data are fram the Line of Business forms filed with
the Federal Trade Commission they are discussed in detail in the Annual Line
which is available from the FTC The 32 industry
in Table 1
Simple regressions for equ ation (37) were run for each of the 32 industries
with data for each of the firms which filed in the industry constituting an ob-
servation The results are given in Table 2 lines lA 2A
1974
the intercept
For all the A
equations generalized least quares was used th both and market
share bein scaled bv the square root of sales
The 32 industries were taken from the larger set of all manufacturing industry
categories in the 1974 data files Two selection cr iteria were used thereLB
had to be at least ten reporting companies and the industry had to be a consumer
goods industry or a producer goods industry which is strongly associated with a
related consumer goods industry so that adve rtising by the firms in the producer
goods industry impacts on demand in the related industry (eg soft drink syrup
and soft drinks) There were 135 industries with at least ten companies - 103
producer goods and 32 consumer goods
273
274
TABLE 1 INDUSTRY CATEGORIES USED
FTC
COOE OESCRI P T I ON _
2001 MEAT PACKINGt SAUSAGE S ANO OTHER PREPARED
MEAT PROOUCTS
2004 DAIRY PRODUCTS EXC F LUID MILK
2005 CANNED SPECIALTIES
2001 FROZEN SPECIALTIES
OEHYORATFOt AND
I NC LU 0 ING PRE SERVE S
SOUP MIXESbull
SEASO lNGSt ANO
OTHER PET FOOJ
RAIN MILL PROOUCTSt
ANO PREPARED
RELATED PRODUCTS
PRODUCTS
Z008 CANNEOt ORIEOt PICKLED FRUITS
ANO VEGET ABLES JAMS JELLIES DEHYDRATED I G TA3t E
S40CES ANO SALAD ORESSINGS
2010 DOG CAT A O
2012 FlOUR t OTHER RICE
MILLING fiLENOE) FLOUR
2014 B EADt CAKE ANO
20113 CO FECTIONERY
2026 BOTTLED A O CANNED SOFT DRINKS 2021 FLAVORING EXTRACTS AND SYRUPS EC
2029 MISC FOOOS AND KINDRED PROJUCTS EXC
ROASTED COFFEE
2301 MENS AND BOYS SUITS AND COATS
2302 MENS ANO BOYS FURNISHINGS
230 3 kOMENS AND MISSES OUTER EAR
2551 HOUSEHOLD FURNITURE
2702 PERIODICALS
Pqe 19a 1
RELATED 1972
src cooe
2011 f 3
202tX2026
2032
203pound
2033t4t5
2047 2041t4t5
2051 2065 2086
2081
209XZ05
231
232
233
251 272
2703 BOOKS
2704 ISC PVeLISHING
2806 ORGANIC FlEERS
2807 DRUG$ ETHICAL
23234
PT 2 S 3
2808 ORUCSt PROPRIETARY
2809 PERFU ES t COSMETICS ANO OTHER TaiLET
PREPARATIONS
2810 SOAP AND OTHER CLEANING PREPARATIONS 2815
2901
PEST1CIDES AND AGR ICULTURAL
PE OLEUM REF I N ING CHEMICALS EC
3606 HOUSEHOLD COOKING EQUIPMENT
3612 HOUSEHOLD APPLIANC ES NEC INCLUJING ELECshy
rqlC HOUSE ARFS AND FANS ANO SE I G MACHINES
3617 RADIO AND T V RECE I VING SETS 3808 PHOTOG APHIC E UIP amp SUP LfESt EXC
3qo3
PHO TOCOPYING
SPOR TING A D E QU I P M ENT f SUP P L I E S ATHLETTC GOODS NEC
3904 DOLLS GAMESt TOYS AND CHLOqENtS VE ICLES
PT283
2844
234tX2844
287 9
291 3631
3634t6t9
36H
PT386l
3949
15
943
middot
l A6l f 2 R(GRESSION RESULTS
EQ I NO
NR CODE cu 121
lA 2001 A (
2A 2004 r (
3A 2005 a
-R
ClS amp31
16
17
1 z
I TER-CEPT bull t
ooo9t21At 00022311 oooeos uu
OOl49tCAI 0010 2 middot 000960CCI
o049tl1At o04t55C61
MARKET SH ARE bull 5 I
-010670Ctl
-017600181
-001050
-0211AO
-0077 0
AOVER-TIS INti
lbl
000()39 000072
000018 0 0024liAI
-000079
SHARf
4ampgtVf Q bull
111
000140
-000170
RSQ
COl
66 485 862
607 6 3 t ()2
blti6 633
flASTI-
CI T Y PH
-007271(1
000441
-u0288l
clASTSHU (t)Ef
bull 10
tto
-234
210
( OOlltOCCI -126520 000922 181 001790
4A 200 7 18 00590liAI -038670 4e16 -032110 lell u c
OOOS74e 000700 025940
00067018) 0021001AI -02l100CAI
5 26
920
5A 6 c
200 8 26 0 03Sl9CA I 0012 26181 o 024511(
-005540
-0 81420 0001961AI O()Ol90CAI ooll240
441 604 891
-0020 19 274
6A 2010 16 Oe04224CAI o359lOIAI 9oq ll4015dCAI 061 6 -middot 005435CAI 00011 8 1AI 917 c 00535ltAI -009850((1 OOOA51J CAI -001610161 971
7A 2012 001862 000010 351 00187) oo1 a oooq45 ono1o6 439 c 001109 -027050 Oe0067SIAI -OOll60C81 897
- middotmiddot- -
8A 201t ll 000757 Ol4140CCI 814 ott9oocet 095 8 oooqo6 C C J 000069(81 811 c ootl2lampBI -080910CCI Oe009lUBI -OOtlOOCCI 960I
I DIIJIT pound11 1V OTIUTVTfAll-1pound1t- A _ 00 R - qq r - Q
1jl)OQCD
0
-middot middot--
l
_ __ _
I
i
0 00141 c c
91)7
I
T ABLE 2 REG ESSIOH RESULTS CCONTI
EO NR
IND CODE
iR bull
cos
I TEq-
CEPT
t1ARKE T
SHA E
AOVER-
Tl SING
SHARpound-cent
AuVER
RSO ILAST1-
C I T Y ELAST
SH CPtf (lt Cl) bull 11 I 41 I 5I (61 171 I 8 I f9) c 10)
9A 2 0 1 8 19 002397181 -007640 367 -005 246 le46 6 000605 000550CAI 597
-000029 G14180 OOl654CAJ -024li01At
lOA 2026 12 0028901AI -OlZl90 -0114 0 1 1 1 5 759
Ll (
llA 2021
f3 c
12A lOZ9
B
(
13A 2301
B
c
14A 2302
B
c
154 201
n (
16A 2551
8
10
35
10
20
19
24
001897(()
00141 6CCI
Oll9lHCI
010604 middot middot
Oll083
OOJ11JCAt
002059141
001411CIH
-000412
000293(81
000187
001159 (3)
ooo6a4cct
o00792CCI
OOl40dCAt
000006
-000076
OOlltl6fA)
o o0 5 64CCt
-034570
-o1middot o3o
-1659 90
Ol51lO
-046190CAt
Ol5450CA)
-005l60
002940
-030060(()
-0 43l50CCt
0 33650
-019080
00004
bullJ00741CAI
-000069
000466
0001 51CAt
000947CAt
O OOI)70CAt OOl750CA J
000175CCI
000671CAI
OOll5) f A t 002207CBI
0004b7CA I
- 007 1 70
001760
-001410CAt
- 0 16200CAt
-004210
- 1 207 70
705
96 0
542
466 805
460
614
921
832
9 61
996
424
550
9 15
56 1 651
884
-006114
019 037
03504lCA)
oo409Q
-141644((
491 -0 17664
607
871
lell
084
101
072
103
108 11IllOQfl)-- l
-037880fCJ OOl6l4CAJ -040l70fCI 0()
c
I
I
bull 9
9t
middot middot middotbullmiddot bull lraquo
TABLE 2 REGRE SION RESUlTS ICONTt
bull
EO I NO tHl I filER- f4A_K T AOVER- sbulluRE ttSQ flA Jl- iLAS T NR CODE cos CEPT SHARE JISING AOVfR bull (JTY SHR CiJCF
110Cl I f amp Itt 15J tbt bull 1c u caa
llA 2702 ll o079 6 81 Bt -019110 S6s -OlltO 111
a 004679 ooooo1 t69
( oOl26o -015160 001560 -022690 ltJr()I
ldA 270 3 19 0065lt -OIt2l60 220 -Ol57a2 11 8 e o 0 71t6IA t ol08 75tAt ( OOl983CiU -060llOC() OOl7191A t -O lltdOC A) middot 89
lqA 2704 10 005082 -Oil060 22b -l)i1977 lelO 6 -000080 oo2772C4 t 910 c 000907 -011 200 Oe04105CA t -046960 995
20A 2d06 1 II
O OlltStiAt -001120 837 00014 -))th O01380amp At -oooo 11 821 ( OOl526CAt -026700181 0002 5UAt 000530 969
2lA 2807 28 OOt9lt9CAt -00891 0 516 -0 03962 225I8 002095CCt OOOl lliAt 690 ( o02965CA t -056650((1 OOOt50CAJ -003l70ttl 911I
I 22A 2808 15 ot3l06 119730 655 1130) 090
0 - 007526CC t Oe003961At 835- middot-
c I oo7523 -067210 001l 2tA t -0091t80CCt 951
2lA 2809 21 Oe2067lCAt -1218101At 71 2 -10095((1 121 ttlIaOQ8 00797418) OOOI8lltt 686 Ill( 0092lU A t -Olltl70Ct t 000605CAt -OOit8701At 951 _
---- -
32 009506CAt
0 bull 06 981 I A J 006901 cA bull
00 24A 2810 -010960 5t 5 -OOltSO 7 56
8 -000011 s11 c -5811701At oooaoatAJ oot520IAI 888
-- 37 --- middot - -- -1 ) l lbull t5JJ r-l bullhmiddot1 --- - --
l5A
26A
27A
c
2RA
c
2 9A
31A
llA
24
13
597
Ci)55
14
l 6h
RSQ ELASTI- (lASTCOE f-
TABLE 2
EQ I NO NR CODE IU
REGRESSION RESULTS CCONTa
NR INTER- MARKET
(10S CEPT SHARE AOVER- SHARfcent
TIS lNG AOVER
) CITY StiR
c It) bull 5 co a I 1) bull 0 bull 9 c l 0 bull( 2 1
2815 16 003019 -009990 275 -o069bmiddotJ
6 001446 000224 282 c 002178 -054630 o02221CCI -0075 00 646
2901 29 00021t2 CAI 0)0520 592 0007 6 1 068
0 00105C8 t 000012141 60)g( OOOlJ51AJ -00126018) ooou271At -000u80CA) 974
36oa 1 3 003296(4) -oca29o -oo-q94
001755 000203 lt91t 002499((1 -04l820CCI OOl374CAI 0068)0 946
3612 0033281CI 011520 195 01684) oso
00 30 42C OI 00012 2 424 oo579ltAt -2 7l7801At 002 100IAI -003220 819
8 3617 btel 011789 081002259 009530
001448 oOOl OitCCI 732 OOl7 22CAI -0 46160CAI 0004961AI -OOll40181 989c
30A 3808 14 0021t 3 6fAI -000110 628 o023291C) -005 6 middot- _ middot- 0bull 02 04 1 IC I ooooo9 637 ( 001584((1 -0161)1)0 00021 91AI -0 00060
390) 14 002790CCt 006590 h8 t 0 0938 4 070
B OOl89bCCJ 000144 74
0021961(1 -OItlOlO OOl323CAI -004800 915
B
middot--- - middot 4 - middot- - dfll3904 12 004563181 0)8300 879 0428651(1 089 OQIll004 277CA I ooozstcea 919
c 0041)3181 -0-11250 0011881(1 -006270 972 0Ill
- J - bull bull bull bullbullbull middot - - middot - middot gt middotbull L)ofbullflbull middot middotfJltl -tn- middotmiddotmiddotmiddoto +f middot t middot bull
c
is
positive
2
Page 0 determineThe consumer goodproducer good split is sometimes difficult to
Sollle support for the split used in this paper is a comparison of goodness - of - fit
distributions of a
industries
set
advertising are
that industry
a sales increase of
the
highest aD10U3 those
measures for the two groups of industries Table 3 contains
measures for the 32- consumer goods industries and the 103 producer goods
2 Median R is 5 7 for the consumer goods set and 40 for the producer goods
Estimates - of y the elasticity of demand with respect to
given in column of Table 2 The hypothesis that y bull 0 with
alternative was applied for each ittdustry As shown in the
could not be rtjected for 24 of the 32 industries Of the
and 10 are positive For the remalnifg eight three are
and five are significantly positive
Of the threertegative elasetcities the largest (in absolute value)
cosmetics (code 2809) With all elasticity of -1 01 a one
industry advertising would gwerate a decrease in industry sales
more than one percetu
At the other extrema is toys and games (code 39 04) For
a one percent ittcrease in industry advertising would lead to
four-tenths of a pampreentmiddot Though not significantly different from zero
elasticity estimatbull for pt-bpuetary drugs (code 2808)
which are
Both drug industries have
to sales rafios (211 pe1 cent for 28 08 and 13 4 per cent
other hand flaIJo ing ext1ac s (mainly soft-drink SYUlS) also
is for
percent increase in
of slightly
a middottwo - tailed
table that hypothesis
24 14 are negative
significantly negative
9
the codetics and p-ropriacuy very high advertising
for 28 09) bull On the
has a high ind UY
advenisiltg t() sales ratio ( 8 0 per cent) bull but its estimated elasticity is very
small and not signifieantly different uom zero
middot It doe$ not seem unusual to find many elasticities near zero with a few
negative and a few positive If aggregate consumption is insensitive to aggregate
advertising as seems likely ld then what one industry gains must be lost by
others
Page 21
elasticity of demand does
(10) we can
if t 0 then
Finally a determination concerning the industry
not fix the individual firms elasticity Going back to equation
write y bull (1 - ) a + z y Since a gt 0 it follows thatii i i
Y gt 0 If Ylt 0 however y s sign is indeterminateij ij
Turning next to the coefficient of market share 20 of them are negative and
12 are positive with three of each sign being significantly different from zero
The largest positive value which is significant is for pet food (code 2010) and
a close second is mens amp boys suits and coats (code 23 01) The largest signifi
cant negative value is for cosmetics (code 28 09)
In the model some relations among y a 0 and 61 are implied If middota is
negative (1- y) is greater than one Since 61Y bull (1- a) (1 - ) 61Y is
pCsitive but it will be less than one if 0 is large enough Seventeen of the A
ys are negative in each of the seventeen industries 61 is also negative and les
1 (algebraicly) than y so 61Y is greater than one his is not evidence that the
degree of cooperation is low or zero in any industry of course A high o togethegt
with a negative y which is small (in absolute terms) relative to a can give a
value of 61Y greater than one On the other hand if 61Y were less than one
o would have to be greater than zero
For y gt 0 two conditions may hold If y lt a it follows that 61 lt 0 since
(y - a ) lt 0 It also follows that 61Y lt 0 regardless of the value of o bull
In addition the relation is an if-and-only-if one ie if 61Y is negative
then y lt a Three industries have negative values for 61y dairy products
except milk (code 20 04) synthetic fibers (code 2806) and photographic equipment
(code 3808) In all three cases y is quite small so a does not have to be very
large to give a negative value to al
If y gt a the model shows that 61-- gt 0 whatever the value of 0 bull There are
12 cases where both y and 61 are positive with 61 ranging from near zero
(grain milling code 20 12) to 1 20 ( proprietary drugs code 2808) Several
coeffiei-en_rs
--
15
o f to zero
Tab le 3 - Consumer Goods
Indus tr ies
2Distributions of R and Producer Goods
Page 20a
Consumer Producer
Goods Goods
0 - 9 9 0 3
10 - 19 9 0 1 1
20 - 29 9 3 20
30 - 39 9 3 17
40 - 49 9 5 16
50 - 59 9 8
69 9 6 Ii 1060
--f 670 79 9 2
80 - 89 9 4 4
90 - 100 1 1
Totals 32 10 3
Notes
2is used R i s cal-
culated as F F + ((N - K- 1 ) K ] 1 b_ F is the F
statistic Lor the hypothesis that all o f the non - intercept terms are simul taneously equal
1
2
--
-- Since generalized least squ-area regress ion
Page 12
o f these cases have values o f 6 1 Y rea ter than 0 8 (pet food (20 10) bread amp cakes
(20 14 ) miscellaneous processed foods (20 29 ) men s and boys suit s and coats
(23 01 ) p rop riet ary drugs (28 08 ) elect ric housewares (36 1 2) radio and
IV sets 17 and games and toys (39 For be than(36 ) 04)) 6 1 Y to greater 0 8
both o and cry mus t b e less than 0 2 That is in these industries the results
shown in Tab le 2 would hold only with very strong dominance of marke t demand
effects over brand swit ching e ffects and a very low degree o f cooperation
Using the concentration rat ios given in Table 4 the average concentration
ratio for the eight indust ries is 38 1 For the remaining 24 industries it is
43 2 This result would lend some support t o the proposition that concentrat ion
and cooperation are related if the group of eight industries are identi fied
as having low degrees of cooperation
One f inal note on model predict ions S ince both cr and o are assumed to
be non-negative 60 should also be non-negat ive In only one o f the 32 cas es
is 60 negative and it is not signif icant bull
cor pJ
7 1
7 0 4
1 4 7 5 1
)
7 3 z a o a t s 1 1 3 3 5 ( i V Z 8 tJ1 2 1 - 1 1 0 7 3 5 c middot v 2 8 1 0 3 2 -0 0 1 4 3 0 C 1 V 9 8 3 2 8 1 5 1 6 - 0 0 6 t1 6 9 C _ V 1 11 0 2 9 0 1 29 0 00 7 6 1 C l k V 7 3 1 3 6 08 1 3 - o n 4 9 9 4 3 3 3 6 1 2 2 4 0 1 6 4 3
3 1 7 1 3 0 1 1 7 d 7 8 3 3 8 0 8 0 0 2 3 i 2 3 0 6
P 4
3 9 0 3 1 4
1 2
0 09 3 8 4 39 1)1 0 4 l 8 6 5
T A LE 4 e I NCU S T R Y C A T E G O R Y
F TC NOM e tOOE ass
C 1 I c 2 )
2 0 0 1 1 6 2 0 0r 1 7 zo o l Z 2 0 0 7 1 8
z o o 2 6
z o t a 1 6
2 1) 1 2 1 5 2 0 1 4 1 1 2 0 1 a 1 9 2 0 26 1 2 2 0 2 7 1 0 2 0 29 - 3 5
2 3 0 1 10
2 3 02 z o
2 3 03 1 9
2 5 5 1 2 4 Z 1 0 Z 1 1 2 0 3 1 9 2 7 4 1 0 z a o 1 3
2 A 0 7 z s
E LA S T I -C I T Y
c 3 )
- 0 09 7 2 7
0 00 449
- 0 0 2 8 9 3
- 0 3 2 7 70 - 0 OZ il I q
o 4 u t s a 0 0 1 3 7 3
o 1 4 o o-0 0 5 2 4o -0 1 94 0 1
-C l 6 l l 4 0 1 JJ 3 7 0 3 t 0 4 2 0 0 4 0 9
- 0 4 1 8 44 - O l 7 t ) 4 -ll 7 1 4 0 b -0 3 5 7 8 Z
- 0 1 9 7 7 0 1)0 3 3 4
-o n 9 2
IJ A T A
CONV
ON-C ONV
( 4 )
t 1NV
C 1t-V C NV c ) v C - JNV C rlNV c n v c a v C ON V ( i) l V
C CtV
C t1tV C IJNV
C 1 V
c 1 v
Pa ge 2La
HEQ F t Noe x
c 5 )
6 2q
6 89 1 48
6 1 4 ) q 3
1 9 7 1 9
q z t s 5 a e z
l l 9 1
4 o 7 1 2 0
c oqt 1 Z
4 4 )-q 1 z
5 5 9 1 0 64 1 3 3
n 3
Inter industry analysi
Page 23
Equation ( 3 6 ) may be used to fo rmulate a rough test o f the cooperat ion
vs e f ficiency controversy I f we rest ricted ourselves to some subset o f
indus tries where the relat ion be tween y and a i s the same we would be lookingbull
at a sample o f indus tries for which the dependence o f V on H and o is a
quadratic wi th no squared terms
I f we now sub s titute (26) into ( 36 ) we get
(38) v bull a + (y - a) R + (y - a) o (H) - (y - a) H o (H) + u
If it is only d if ferences in marke t shares that affect advertising inten-
s i ty levels i f concentration does not af fect cooperation and i f o bull 0
( 38 ) reduces t o
( 39 ) V bull a + (y - a) H + U
Let this b e the null hyp o thes is and note that V is a linear func-
tion o f H If y gt a and the null hyp othesis holds the expected relat ion
between V and H is as shown by the pos itively sloped straight line in Figure
6 b etween points A and C
In this context the alterna ive hypothesis the degree o f
cooperation ( o) is a positive function o f concentrat ion
is that
(H) I f the func-
tion is as described in (26) and Figure 4 the relation between V and H
would be as shown by the curved line connec t ing po ints A and C in Figure 6 One way to test the null hypothe s is against this al ternative is to test fo r
the hypothesized curvature in the function relating V and H
Fi gure
R
6
v
ID c
H 1 0
I
i I I
sfigned
clasJ s
in def ini g
effec t is the
I
Page middot2lf
One charac teristic of produc ts which may be useful
classes wi th respec t to significanc e of the brand-switching
conveniencenon-convenienc e distinc tion developed by Porter ( 1 97 )
Using his identif ication of IRS indus tries as a s tar t ing point
each of the 32 industries used in this s tudy to one of the two
The ass ignments are not ed in Tab le 4
Using the conveniencenon-convenienc e dis t inc tion the
divided into two sub-samples
sample was
Four regressions were then run bull ene for e ch
sub-sample one for the to tal sample and one for the total sample with
a dummy variable for convenience goods Each regression was run wi th LB s
and wi th indus tries as ob servations The results are given in Tab le 5
For the LB level equations ( 1A - 1D) equation(37) was used ome
rearranging of terms gives v bull y - ( 1-o ) (y - a) ( 1-z ) + ui i i
bull y - 6 1 ( 1 - z ) + ui When observations are pooled across indus trif s yi
is an industry variable In these regressions the values o f y es t imated
in the first stage were used Since industries have been grouped so as
to reduce differences in (y - a) that term is t reated as a cons tant
within each group Finally H the Herfindahl index has been subst ituted
for o the degree of cooperation The resultant equation is linear
y ( 1 - z ) and H x ( l - z ) This is the equation for whi ch results arei i
reported in Table 5 In all regress ions all variables were weighted by the
square roo t of sales
in
raquo
- 34
3 1
i l T 4 8LE S R E G E S S I ON R E SUL T S i X T f N S I ON S T O T Y PE Of
AND L E V E L OF C ONC E kT R A T I O
E O O tt S
tP bull T Y P E NR C ONV O B S tO N-CONV
E L A S T I shy
T I C J T Y
O E - t-41( T SliA R E
H( fl F
1 1 - K O x middot K O X E R F
- H S a 1 1 - H S t 1 8 1 1 9 1
R S O
1 1 0 1
q z
t - bull 9 f 2 1
Z t P D l 6 fC ONV - 1 9 1 s s c 2 1
e - l O I e bull u q a
2 3 1 0 1 1 Nl N- t O V 5 7 1 ( 1 2 19 1 6 1 - J l S 2 5 4
l l l l t z q o a 1 - J o v z I
l ( I NO l l A l l ol u l t o o c bull J u l l
d S
c 3 4 t t r a l
bull Z D 1 0 l l 4 l l 2 3 -gt 2 l l 2 0 - 4 lS 5 4 l C 4 1 o e 6 t l b b 5 1 1 0 1 1 1 c - 1 9 5 1 t d S 2 1 tV
OQ10 T-R A T I O S A E G I VE N l P A RE N T H E S E S l E V E L S f S I G 1 F l C ANC E A - 99 5 B - q 9 t - s NS J UC E NO I Nl E R C ( P T I S U (j f middot) I N f Hf t OUA TIOI S 6 E f 3R E A O J U S T H E N T F OR 1-t E T E R O SIC E O A S T f C I T V pIraquo
1
middot
I
R SO I S C A L C UL A T E D A S f I f bull 1 1 bull 1lt 1 K t t- E E N I S THf U A1 B E R O f fJe S E fi V A T I O S F t S HtE F - S T A T I S
I N DE P E ND E N T y II
T I C F O T H E U Y P O HIE S I S f A l l C JE F F I C H N T A R E QU A L HI l f RO A f40 K I S T H f UUM 8 E R OF Q ( A fL f S
I l l
1 A
1 6
t C
f 1 r R I l l
1 2 1
L Jt
L B
l q
L 1
on s T Y P E bull 2 )
P OUC T
X
H S I l l
l O l A I bull l 0 7 1
O O b O J
1 2 00 1
1 b 2 1 C I l l Z U
l l S b 1 -1 1 0 1
rlt F X l l - fi C f l
bull 1 I
l l
- ()4 2 3 l f H
1 - 2 7 9 1
K O X
1 1 - ti E R F I
l d l
S4 l i A I 2 l b
1 3 00
K O X -t f R F t S J
I 1 - HER F t
)
60 9
z o bull
0
t l l
3 8 8
1 7 6
5 64
64
N f bull
tl S bull
I l l
I 4 1
t ONV
NON-t ONV
A l l
A ll
c o v NO- ( ONV
1 4 1
1 5 1
- l S b
c - 1 1 2 1
5 8 2 C A I 1 6 7 0 1
middot I l 9 H
l 1 il
1 4 1 6 1
E L A C) T J shy
l t C I T Y
I S I
l b l
- 0 2 0
1 - 1 4 9 1
O Z 3 2 1 A I C b 6 b l
0 0 9 2 -J
1 1 b 4 1
o z t t t l 1 2 2 1 1
oru - Ht tt F
1 gt X bull b I
Page 25 bull
Mean val -as ior relevan t variables for the two sub-samples are
Conv Non-Conv
Adv Sales 4 7 2 3
iit Share 2 7 3 2
Ela cicity - 05 3 02 2
Herf indahl 7 3 1 860
That the dist nct ion between coft9enienee goods and non - convenience
goods mat ters is supported not only by these means but also by the regression
results Equat ions lA and lB are different in almost every respec t Only the
concentrat ionmarket share int eract ion term seems to mat ter for convenience
good s Jus t he opposite holds for non - convenience goods both elasticity1
and market sl e are highly significant but the concentrat ionmarket share L
int erac t ion t=rm is not
Given that one minus market share is used these equ3 tions the coeffici t
of market share is the negative of what is sh in col (6) of Table 5 For the
regress ion for non - convenience LB s reported in equation lB then LB s with larger
market shares have lower advertising to sales ratios
On the question of the impact of the degree of cooperation the story is mixed
For convenienc good s the coefficient is pos itive as equation (31) predicts when
there is a po sitive degree of cooperat ion Fo r the non-convenience goods LB s middot
concentration has no effec t
Aggregat ion t o the indus try level and then rer unning the equations $ives the
results shown in (2a-2d) of Tab le 5 As expected R2 goes up and levels of significanc e
fmiddotJr individual coefficients gJ down In add ition the coeffic ient of H ( l -H) inmiddot
quat ion 2A is no t significant though it is po s i tive Since 2A is the indus try level
I
--
all but one of them and that one is not significant For 18 of the 31
wo uld have on its adve rtising to sal es ratio
cognate of LA the variable H ( l-H) pl ays the same role in 2A as H ( l -z 1) p lays
in LA
Vconomies of scale in advert ising
The literature on advert ing contains a number of definitionbull o f economies
of scale and of findings eone rning its presence and magnitude Before I loo k
a t the data with the hope o f mmenting about scale effect s then I wan t to
define the term as I use it
By economies of scale I mean a decrease in the number o f units of adver-
tising which are needed to generate a unit of sales as advert ising is increased
Given the definition of the firm s own elast icity of demand with respect to
advertising (see the text foll wing equation (1) ) the technical definition
is that rii gt 1 Diseconomies of scale are def ined symmetrically i e
Yu lt 1
ith a cro s s - section oi observations on advertising intensity
(v bull a s1) and onmiddot advertising (a ) we may regress the former on the latter 1 i i
The results of such regress ions for the 32 industries are given in Table 2
lines lB ZB bull bull bull 32B For the 32 eases the regression coefficient is positive it
positive coefficients the coefficient i s signif icant
The que tion now is whether these results ar e evidence of pervasive dis-
economies of scale in advertislng I think no t Given my as sumption that the
data I observe are equilibrium results I may ri ghtly conclud e that almost every-
where high levels of advert ising are as sociated in equilibrium with high ratios
of advertising to sales I may no t conclud e anything however about the
impact that a move by some firm away from its equilibrium level o f advertising
middot- Page 27
Given the assumed relation in equation ( 37 ) there is a corre spondshy
ing equilibrium equation relating v and ai The equation is non - linear i
and I have not found a simp le way to characterize it It is po ssib le to
det ermi ne its deriviative with respect to a b y takin$ tne de riviative of ( 3 7 ) i
When that is done the result is
v - 1(40) i bull sl a a z ( 1-z )i i ira-l
Sinc e all the other terms in (40) are non-negative the sign dep ends only
on the sign of 6 1 which is in turn de pend ent only on the sign of (y - a)
That the signs of the coefficients of a in equations lB 2B bull bull bull 32B arei
not always equal to the signs of the coefficients of z in LA 2A bull bull bull 32A i
is still a mystery to me and something to be explored
bull
Summary
My purpose in writing this pa per wa s t o attempt to integrate explicit micro-
economic mode lling reasoned econome tric specification of error terms and high
quality data to exp lore some questions about adver tising by large firms
al quite important improvements could be made in the first two of those areas
and much data ma ssag ing is yet to be done in the third
Concerning the subst an ce of what I have done I think six things are im-
Page 2 8
Severshy
V and Conclusions
portant
1 Explicit modeling is worth the ef for t if for no other reas on than
that it provides a basis for choosing fr om a varie ty of funct ional
forms
2 The same is true for error specif ica t ion I did in fact look at scatshy
ter diagrams for all 32 industry categor ies and they vir tually all show
the het eroscedast icity which I as sumed and for which I correc ted
3 The predicted relation be tween advertis ing in tensity and market share
shows up clearly in on ly 25 of the case s examined I have ye t to
explore why that may b e the cas e
4 Some evidence concerning the presence and impact of coopera tion was
produced but it is not clear ly pervasive
5 The dis tinction between convenience goods and non-c onvenie nce goods is
unambiguously a good one
6 Virtually no evidence concerning econ omies of scale in adver tis ing can
be gleaned fr om this study given its assumpt ions
More work is called for on many of these issues my study seems to have
raised more questions than it has answer ed This is pr obab lly due to the richshy
ne ss of the data source since I had the oppor tunity to address several issue s
at once
- -- - Q --
Foo tnotes
Manager Line o f Business Program Federal Trade Commission The author depended heavily on several staf f members of the LB Pro gram for statistical and clerical support in particular Joe Cholka George Pascoe and Haro lene Jenkins Helpful comments were given b y Richard Caves Dennis Mueller Michael Porter and F M Scherer When the paper was in its early stages o f development both S teve Garber and Jon Rasmussen former economists in the LB Program provided eery useful cri tiques The views expressed here are my own o f course
1 The basic exp licit model of the role of advertising is Dorfman and S teiner ( 1954 ) and the models in this paper may be seen as ano ther extens ion of their work
2 For a review o f models of oligopoly which treat prices and quantities as relevant variables see Shubik ( 1 959) Extens ions to take advertis ing int o account are fairly s traightforward
3 Dorfman and S teiner dealt with a price-setting firm For a monopolis t inverting the demand relation to show Pbull P (Q A) _instead o f Q- (P A) has no effect on the results o f the analysis For oligopolists who face a sys tem of demand relations inversion does have some impac t on the results The dif ferences are not trivial nonetheless they will no t be explored in this study
4 Shubik ( 1 959) p
5 From a formal mathematical point o f view the function is similar to a1 i Legrangian funct ion of the form - r rr
L bull 11 + A ( j - (i i i+j
where the indicates some fixed level of the profits industry Following the same logic which resul ts in and income constraint as the marginal utility of the of the indus tr in the obj ective of firm i
6 Hall amp Weiss ( 1 967) p 323
7 See Schmallensee ( 1 97 2 ) Ch 2 pp 1 6-47
8 On this poin t see Imel amp Helmberger ( 1 9 7 1 )
9 Long ( 1 970) p 130 - 1 39
1 0 See Schmallensee ( 1 972) Ch 3 pp 48-8 7
rrj i+j ) of the rest of the
the identification profits of the rest
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980
--- -
Page Ll
Athird characteristic of this specification also deserves some comment
only on equation is shown for the firm or industry Several haveinvestigators
included an advertising equation ina larger system of equations (Comanor and
Wilson (1974) (1979) and Weiss (1976) so in myMartin Strickland Idid
dissertation1 Ihave ewo defenses for presenting only single equation results
here One is the suggestion by Comanor Wilson that simultaneousand equation
verybias may not be important in this context he second is that Iam will-
ing to see the model extended to include a profitability equation and I intend
to move that direction in the near futurein
Onthe inclusion of aconcentration equation on the other hand I amskep-
tical he problem isthat in some fundamental sense equations in models of
industrial organization should be oriented to the decision-making contexts of firms
Ifeach firm in anindustry with Nfirms has only one decision variable then
there can be only Nindependent equations which are based on the first-order
optimization equations It not show large ofis difficult of course to a number
functional relations which follow from the first-order conditions point here he
isthat only Nof them can be independent
Consider for illustrative purposes only a situation in wich teh N
itsquantities say q are set exogeneously and each firm determines advertising1
of and level say ai where a is the true optimal value a a bulla
1 1 i i + i
Ifp a) then the Nq
s and the Na s will determine N p s say pii bullp
i(l i 1 i
Now given Eand 9 we can determine both the vector s andthe vector v since
- bull til
ands We can also determine S s z bull s-1 s-ii
v and we
industry concentration isHbull (alternatively
V and Hare functions of the same
pendent Furthermore if there are
VM) and (H1 H2 bull bull bull )contain
VI STATISTI CAL APPLICATIONS
Page 18
Given the observed values of might be tempted to conclude that we
have a two equation system in which v and z are jointly determined That would i i
be inappropriate however since the 2N random variables (v ) are determined by
onlt N basic random variables t The variance-covariance matrix for ( ) must
be singular bull
The extension of this observation to industry level variables is straightshy
forward Since observed industry advertising intensity is V bull l v and observed 4
C4 bull z ) it follows thati l i
set of random variables t and cannot be indeshy-
M industries then the variables (V1 v2 at most M independent random variables
Given the assumptions made above y a and o are constant within a
given industry That being the case (34) can be written as
(37)
s1 bull (1 - o) (y - a)
If and
So + +
(1 - o) a + oy
81 are esti mated for an industry using a suitable statisticalSo procedure giving and 81 an estimate of y is then available The sumSo
Intra-industry analysis
1974 of Business Re ort -
categories are identified
Page 19
of So and S1 is y so y bull So + S1 is the estimate of y
If o is assumed to be bounded by zero and one the sign of 61 is also
the sign of y - a The estimate of So is of cou se an estimate of
(1 - o ) a + oy Moreove if y gt a then a is less than 6o- (1 - B) a+ 0(
Cor esponding results would
hold if y bull a or y lt a
If 61 gt 0 then we would estimate that a lt 6o
bull
For purposes of estimating this model 32 manufacturing industry categories
were selected The data are fram the Line of Business forms filed with
the Federal Trade Commission they are discussed in detail in the Annual Line
which is available from the FTC The 32 industry
in Table 1
Simple regressions for equ ation (37) were run for each of the 32 industries
with data for each of the firms which filed in the industry constituting an ob-
servation The results are given in Table 2 lines lA 2A
1974
the intercept
For all the A
equations generalized least quares was used th both and market
share bein scaled bv the square root of sales
The 32 industries were taken from the larger set of all manufacturing industry
categories in the 1974 data files Two selection cr iteria were used thereLB
had to be at least ten reporting companies and the industry had to be a consumer
goods industry or a producer goods industry which is strongly associated with a
related consumer goods industry so that adve rtising by the firms in the producer
goods industry impacts on demand in the related industry (eg soft drink syrup
and soft drinks) There were 135 industries with at least ten companies - 103
producer goods and 32 consumer goods
273
274
TABLE 1 INDUSTRY CATEGORIES USED
FTC
COOE OESCRI P T I ON _
2001 MEAT PACKINGt SAUSAGE S ANO OTHER PREPARED
MEAT PROOUCTS
2004 DAIRY PRODUCTS EXC F LUID MILK
2005 CANNED SPECIALTIES
2001 FROZEN SPECIALTIES
OEHYORATFOt AND
I NC LU 0 ING PRE SERVE S
SOUP MIXESbull
SEASO lNGSt ANO
OTHER PET FOOJ
RAIN MILL PROOUCTSt
ANO PREPARED
RELATED PRODUCTS
PRODUCTS
Z008 CANNEOt ORIEOt PICKLED FRUITS
ANO VEGET ABLES JAMS JELLIES DEHYDRATED I G TA3t E
S40CES ANO SALAD ORESSINGS
2010 DOG CAT A O
2012 FlOUR t OTHER RICE
MILLING fiLENOE) FLOUR
2014 B EADt CAKE ANO
20113 CO FECTIONERY
2026 BOTTLED A O CANNED SOFT DRINKS 2021 FLAVORING EXTRACTS AND SYRUPS EC
2029 MISC FOOOS AND KINDRED PROJUCTS EXC
ROASTED COFFEE
2301 MENS AND BOYS SUITS AND COATS
2302 MENS ANO BOYS FURNISHINGS
230 3 kOMENS AND MISSES OUTER EAR
2551 HOUSEHOLD FURNITURE
2702 PERIODICALS
Pqe 19a 1
RELATED 1972
src cooe
2011 f 3
202tX2026
2032
203pound
2033t4t5
2047 2041t4t5
2051 2065 2086
2081
209XZ05
231
232
233
251 272
2703 BOOKS
2704 ISC PVeLISHING
2806 ORGANIC FlEERS
2807 DRUG$ ETHICAL
23234
PT 2 S 3
2808 ORUCSt PROPRIETARY
2809 PERFU ES t COSMETICS ANO OTHER TaiLET
PREPARATIONS
2810 SOAP AND OTHER CLEANING PREPARATIONS 2815
2901
PEST1CIDES AND AGR ICULTURAL
PE OLEUM REF I N ING CHEMICALS EC
3606 HOUSEHOLD COOKING EQUIPMENT
3612 HOUSEHOLD APPLIANC ES NEC INCLUJING ELECshy
rqlC HOUSE ARFS AND FANS ANO SE I G MACHINES
3617 RADIO AND T V RECE I VING SETS 3808 PHOTOG APHIC E UIP amp SUP LfESt EXC
3qo3
PHO TOCOPYING
SPOR TING A D E QU I P M ENT f SUP P L I E S ATHLETTC GOODS NEC
3904 DOLLS GAMESt TOYS AND CHLOqENtS VE ICLES
PT283
2844
234tX2844
287 9
291 3631
3634t6t9
36H
PT386l
3949
15
943
middot
l A6l f 2 R(GRESSION RESULTS
EQ I NO
NR CODE cu 121
lA 2001 A (
2A 2004 r (
3A 2005 a
-R
ClS amp31
16
17
1 z
I TER-CEPT bull t
ooo9t21At 00022311 oooeos uu
OOl49tCAI 0010 2 middot 000960CCI
o049tl1At o04t55C61
MARKET SH ARE bull 5 I
-010670Ctl
-017600181
-001050
-0211AO
-0077 0
AOVER-TIS INti
lbl
000()39 000072
000018 0 0024liAI
-000079
SHARf
4ampgtVf Q bull
111
000140
-000170
RSQ
COl
66 485 862
607 6 3 t ()2
blti6 633
flASTI-
CI T Y PH
-007271(1
000441
-u0288l
clASTSHU (t)Ef
bull 10
tto
-234
210
( OOlltOCCI -126520 000922 181 001790
4A 200 7 18 00590liAI -038670 4e16 -032110 lell u c
OOOS74e 000700 025940
00067018) 0021001AI -02l100CAI
5 26
920
5A 6 c
200 8 26 0 03Sl9CA I 0012 26181 o 024511(
-005540
-0 81420 0001961AI O()Ol90CAI ooll240
441 604 891
-0020 19 274
6A 2010 16 Oe04224CAI o359lOIAI 9oq ll4015dCAI 061 6 -middot 005435CAI 00011 8 1AI 917 c 00535ltAI -009850((1 OOOA51J CAI -001610161 971
7A 2012 001862 000010 351 00187) oo1 a oooq45 ono1o6 439 c 001109 -027050 Oe0067SIAI -OOll60C81 897
- middotmiddot- -
8A 201t ll 000757 Ol4140CCI 814 ott9oocet 095 8 oooqo6 C C J 000069(81 811 c ootl2lampBI -080910CCI Oe009lUBI -OOtlOOCCI 960I
I DIIJIT pound11 1V OTIUTVTfAll-1pound1t- A _ 00 R - qq r - Q
1jl)OQCD
0
-middot middot--
l
_ __ _
I
i
0 00141 c c
91)7
I
T ABLE 2 REG ESSIOH RESULTS CCONTI
EO NR
IND CODE
iR bull
cos
I TEq-
CEPT
t1ARKE T
SHA E
AOVER-
Tl SING
SHARpound-cent
AuVER
RSO ILAST1-
C I T Y ELAST
SH CPtf (lt Cl) bull 11 I 41 I 5I (61 171 I 8 I f9) c 10)
9A 2 0 1 8 19 002397181 -007640 367 -005 246 le46 6 000605 000550CAI 597
-000029 G14180 OOl654CAJ -024li01At
lOA 2026 12 0028901AI -OlZl90 -0114 0 1 1 1 5 759
Ll (
llA 2021
f3 c
12A lOZ9
B
(
13A 2301
B
c
14A 2302
B
c
154 201
n (
16A 2551
8
10
35
10
20
19
24
001897(()
00141 6CCI
Oll9lHCI
010604 middot middot
Oll083
OOJ11JCAt
002059141
001411CIH
-000412
000293(81
000187
001159 (3)
ooo6a4cct
o00792CCI
OOl40dCAt
000006
-000076
OOlltl6fA)
o o0 5 64CCt
-034570
-o1middot o3o
-1659 90
Ol51lO
-046190CAt
Ol5450CA)
-005l60
002940
-030060(()
-0 43l50CCt
0 33650
-019080
00004
bullJ00741CAI
-000069
000466
0001 51CAt
000947CAt
O OOI)70CAt OOl750CA J
000175CCI
000671CAI
OOll5) f A t 002207CBI
0004b7CA I
- 007 1 70
001760
-001410CAt
- 0 16200CAt
-004210
- 1 207 70
705
96 0
542
466 805
460
614
921
832
9 61
996
424
550
9 15
56 1 651
884
-006114
019 037
03504lCA)
oo409Q
-141644((
491 -0 17664
607
871
lell
084
101
072
103
108 11IllOQfl)-- l
-037880fCJ OOl6l4CAJ -040l70fCI 0()
c
I
I
bull 9
9t
middot middot middotbullmiddot bull lraquo
TABLE 2 REGRE SION RESUlTS ICONTt
bull
EO I NO tHl I filER- f4A_K T AOVER- sbulluRE ttSQ flA Jl- iLAS T NR CODE cos CEPT SHARE JISING AOVfR bull (JTY SHR CiJCF
110Cl I f amp Itt 15J tbt bull 1c u caa
llA 2702 ll o079 6 81 Bt -019110 S6s -OlltO 111
a 004679 ooooo1 t69
( oOl26o -015160 001560 -022690 ltJr()I
ldA 270 3 19 0065lt -OIt2l60 220 -Ol57a2 11 8 e o 0 71t6IA t ol08 75tAt ( OOl983CiU -060llOC() OOl7191A t -O lltdOC A) middot 89
lqA 2704 10 005082 -Oil060 22b -l)i1977 lelO 6 -000080 oo2772C4 t 910 c 000907 -011 200 Oe04105CA t -046960 995
20A 2d06 1 II
O OlltStiAt -001120 837 00014 -))th O01380amp At -oooo 11 821 ( OOl526CAt -026700181 0002 5UAt 000530 969
2lA 2807 28 OOt9lt9CAt -00891 0 516 -0 03962 225I8 002095CCt OOOl lliAt 690 ( o02965CA t -056650((1 OOOt50CAJ -003l70ttl 911I
I 22A 2808 15 ot3l06 119730 655 1130) 090
0 - 007526CC t Oe003961At 835- middot-
c I oo7523 -067210 001l 2tA t -0091t80CCt 951
2lA 2809 21 Oe2067lCAt -1218101At 71 2 -10095((1 121 ttlIaOQ8 00797418) OOOI8lltt 686 Ill( 0092lU A t -Olltl70Ct t 000605CAt -OOit8701At 951 _
---- -
32 009506CAt
0 bull 06 981 I A J 006901 cA bull
00 24A 2810 -010960 5t 5 -OOltSO 7 56
8 -000011 s11 c -5811701At oooaoatAJ oot520IAI 888
-- 37 --- middot - -- -1 ) l lbull t5JJ r-l bullhmiddot1 --- - --
l5A
26A
27A
c
2RA
c
2 9A
31A
llA
24
13
597
Ci)55
14
l 6h
RSQ ELASTI- (lASTCOE f-
TABLE 2
EQ I NO NR CODE IU
REGRESSION RESULTS CCONTa
NR INTER- MARKET
(10S CEPT SHARE AOVER- SHARfcent
TIS lNG AOVER
) CITY StiR
c It) bull 5 co a I 1) bull 0 bull 9 c l 0 bull( 2 1
2815 16 003019 -009990 275 -o069bmiddotJ
6 001446 000224 282 c 002178 -054630 o02221CCI -0075 00 646
2901 29 00021t2 CAI 0)0520 592 0007 6 1 068
0 00105C8 t 000012141 60)g( OOOlJ51AJ -00126018) ooou271At -000u80CA) 974
36oa 1 3 003296(4) -oca29o -oo-q94
001755 000203 lt91t 002499((1 -04l820CCI OOl374CAI 0068)0 946
3612 0033281CI 011520 195 01684) oso
00 30 42C OI 00012 2 424 oo579ltAt -2 7l7801At 002 100IAI -003220 819
8 3617 btel 011789 081002259 009530
001448 oOOl OitCCI 732 OOl7 22CAI -0 46160CAI 0004961AI -OOll40181 989c
30A 3808 14 0021t 3 6fAI -000110 628 o023291C) -005 6 middot- _ middot- 0bull 02 04 1 IC I ooooo9 637 ( 001584((1 -0161)1)0 00021 91AI -0 00060
390) 14 002790CCt 006590 h8 t 0 0938 4 070
B OOl89bCCJ 000144 74
0021961(1 -OItlOlO OOl323CAI -004800 915
B
middot--- - middot 4 - middot- - dfll3904 12 004563181 0)8300 879 0428651(1 089 OQIll004 277CA I ooozstcea 919
c 0041)3181 -0-11250 0011881(1 -006270 972 0Ill
- J - bull bull bull bullbullbull middot - - middot - middot gt middotbull L)ofbullflbull middot middotfJltl -tn- middotmiddotmiddotmiddoto +f middot t middot bull
c
is
positive
2
Page 0 determineThe consumer goodproducer good split is sometimes difficult to
Sollle support for the split used in this paper is a comparison of goodness - of - fit
distributions of a
industries
set
advertising are
that industry
a sales increase of
the
highest aD10U3 those
measures for the two groups of industries Table 3 contains
measures for the 32- consumer goods industries and the 103 producer goods
2 Median R is 5 7 for the consumer goods set and 40 for the producer goods
Estimates - of y the elasticity of demand with respect to
given in column of Table 2 The hypothesis that y bull 0 with
alternative was applied for each ittdustry As shown in the
could not be rtjected for 24 of the 32 industries Of the
and 10 are positive For the remalnifg eight three are
and five are significantly positive
Of the threertegative elasetcities the largest (in absolute value)
cosmetics (code 2809) With all elasticity of -1 01 a one
industry advertising would gwerate a decrease in industry sales
more than one percetu
At the other extrema is toys and games (code 39 04) For
a one percent ittcrease in industry advertising would lead to
four-tenths of a pampreentmiddot Though not significantly different from zero
elasticity estimatbull for pt-bpuetary drugs (code 2808)
which are
Both drug industries have
to sales rafios (211 pe1 cent for 28 08 and 13 4 per cent
other hand flaIJo ing ext1ac s (mainly soft-drink SYUlS) also
is for
percent increase in
of slightly
a middottwo - tailed
table that hypothesis
24 14 are negative
significantly negative
9
the codetics and p-ropriacuy very high advertising
for 28 09) bull On the
has a high ind UY
advenisiltg t() sales ratio ( 8 0 per cent) bull but its estimated elasticity is very
small and not signifieantly different uom zero
middot It doe$ not seem unusual to find many elasticities near zero with a few
negative and a few positive If aggregate consumption is insensitive to aggregate
advertising as seems likely ld then what one industry gains must be lost by
others
Page 21
elasticity of demand does
(10) we can
if t 0 then
Finally a determination concerning the industry
not fix the individual firms elasticity Going back to equation
write y bull (1 - ) a + z y Since a gt 0 it follows thatii i i
Y gt 0 If Ylt 0 however y s sign is indeterminateij ij
Turning next to the coefficient of market share 20 of them are negative and
12 are positive with three of each sign being significantly different from zero
The largest positive value which is significant is for pet food (code 2010) and
a close second is mens amp boys suits and coats (code 23 01) The largest signifi
cant negative value is for cosmetics (code 28 09)
In the model some relations among y a 0 and 61 are implied If middota is
negative (1- y) is greater than one Since 61Y bull (1- a) (1 - ) 61Y is
pCsitive but it will be less than one if 0 is large enough Seventeen of the A
ys are negative in each of the seventeen industries 61 is also negative and les
1 (algebraicly) than y so 61Y is greater than one his is not evidence that the
degree of cooperation is low or zero in any industry of course A high o togethegt
with a negative y which is small (in absolute terms) relative to a can give a
value of 61Y greater than one On the other hand if 61Y were less than one
o would have to be greater than zero
For y gt 0 two conditions may hold If y lt a it follows that 61 lt 0 since
(y - a ) lt 0 It also follows that 61Y lt 0 regardless of the value of o bull
In addition the relation is an if-and-only-if one ie if 61Y is negative
then y lt a Three industries have negative values for 61y dairy products
except milk (code 20 04) synthetic fibers (code 2806) and photographic equipment
(code 3808) In all three cases y is quite small so a does not have to be very
large to give a negative value to al
If y gt a the model shows that 61-- gt 0 whatever the value of 0 bull There are
12 cases where both y and 61 are positive with 61 ranging from near zero
(grain milling code 20 12) to 1 20 ( proprietary drugs code 2808) Several
coeffiei-en_rs
--
15
o f to zero
Tab le 3 - Consumer Goods
Indus tr ies
2Distributions of R and Producer Goods
Page 20a
Consumer Producer
Goods Goods
0 - 9 9 0 3
10 - 19 9 0 1 1
20 - 29 9 3 20
30 - 39 9 3 17
40 - 49 9 5 16
50 - 59 9 8
69 9 6 Ii 1060
--f 670 79 9 2
80 - 89 9 4 4
90 - 100 1 1
Totals 32 10 3
Notes
2is used R i s cal-
culated as F F + ((N - K- 1 ) K ] 1 b_ F is the F
statistic Lor the hypothesis that all o f the non - intercept terms are simul taneously equal
1
2
--
-- Since generalized least squ-area regress ion
Page 12
o f these cases have values o f 6 1 Y rea ter than 0 8 (pet food (20 10) bread amp cakes
(20 14 ) miscellaneous processed foods (20 29 ) men s and boys suit s and coats
(23 01 ) p rop riet ary drugs (28 08 ) elect ric housewares (36 1 2) radio and
IV sets 17 and games and toys (39 For be than(36 ) 04)) 6 1 Y to greater 0 8
both o and cry mus t b e less than 0 2 That is in these industries the results
shown in Tab le 2 would hold only with very strong dominance of marke t demand
effects over brand swit ching e ffects and a very low degree o f cooperation
Using the concentration rat ios given in Table 4 the average concentration
ratio for the eight indust ries is 38 1 For the remaining 24 industries it is
43 2 This result would lend some support t o the proposition that concentrat ion
and cooperation are related if the group of eight industries are identi fied
as having low degrees of cooperation
One f inal note on model predict ions S ince both cr and o are assumed to
be non-negative 60 should also be non-negat ive In only one o f the 32 cas es
is 60 negative and it is not signif icant bull
cor pJ
7 1
7 0 4
1 4 7 5 1
)
7 3 z a o a t s 1 1 3 3 5 ( i V Z 8 tJ1 2 1 - 1 1 0 7 3 5 c middot v 2 8 1 0 3 2 -0 0 1 4 3 0 C 1 V 9 8 3 2 8 1 5 1 6 - 0 0 6 t1 6 9 C _ V 1 11 0 2 9 0 1 29 0 00 7 6 1 C l k V 7 3 1 3 6 08 1 3 - o n 4 9 9 4 3 3 3 6 1 2 2 4 0 1 6 4 3
3 1 7 1 3 0 1 1 7 d 7 8 3 3 8 0 8 0 0 2 3 i 2 3 0 6
P 4
3 9 0 3 1 4
1 2
0 09 3 8 4 39 1)1 0 4 l 8 6 5
T A LE 4 e I NCU S T R Y C A T E G O R Y
F TC NOM e tOOE ass
C 1 I c 2 )
2 0 0 1 1 6 2 0 0r 1 7 zo o l Z 2 0 0 7 1 8
z o o 2 6
z o t a 1 6
2 1) 1 2 1 5 2 0 1 4 1 1 2 0 1 a 1 9 2 0 26 1 2 2 0 2 7 1 0 2 0 29 - 3 5
2 3 0 1 10
2 3 02 z o
2 3 03 1 9
2 5 5 1 2 4 Z 1 0 Z 1 1 2 0 3 1 9 2 7 4 1 0 z a o 1 3
2 A 0 7 z s
E LA S T I -C I T Y
c 3 )
- 0 09 7 2 7
0 00 449
- 0 0 2 8 9 3
- 0 3 2 7 70 - 0 OZ il I q
o 4 u t s a 0 0 1 3 7 3
o 1 4 o o-0 0 5 2 4o -0 1 94 0 1
-C l 6 l l 4 0 1 JJ 3 7 0 3 t 0 4 2 0 0 4 0 9
- 0 4 1 8 44 - O l 7 t ) 4 -ll 7 1 4 0 b -0 3 5 7 8 Z
- 0 1 9 7 7 0 1)0 3 3 4
-o n 9 2
IJ A T A
CONV
ON-C ONV
( 4 )
t 1NV
C 1t-V C NV c ) v C - JNV C rlNV c n v c a v C ON V ( i) l V
C CtV
C t1tV C IJNV
C 1 V
c 1 v
Pa ge 2La
HEQ F t Noe x
c 5 )
6 2q
6 89 1 48
6 1 4 ) q 3
1 9 7 1 9
q z t s 5 a e z
l l 9 1
4 o 7 1 2 0
c oqt 1 Z
4 4 )-q 1 z
5 5 9 1 0 64 1 3 3
n 3
Inter industry analysi
Page 23
Equation ( 3 6 ) may be used to fo rmulate a rough test o f the cooperat ion
vs e f ficiency controversy I f we rest ricted ourselves to some subset o f
indus tries where the relat ion be tween y and a i s the same we would be lookingbull
at a sample o f indus tries for which the dependence o f V on H and o is a
quadratic wi th no squared terms
I f we now sub s titute (26) into ( 36 ) we get
(38) v bull a + (y - a) R + (y - a) o (H) - (y - a) H o (H) + u
If it is only d if ferences in marke t shares that affect advertising inten-
s i ty levels i f concentration does not af fect cooperation and i f o bull 0
( 38 ) reduces t o
( 39 ) V bull a + (y - a) H + U
Let this b e the null hyp o thes is and note that V is a linear func-
tion o f H If y gt a and the null hyp othesis holds the expected relat ion
between V and H is as shown by the pos itively sloped straight line in Figure
6 b etween points A and C
In this context the alterna ive hypothesis the degree o f
cooperation ( o) is a positive function o f concentrat ion
is that
(H) I f the func-
tion is as described in (26) and Figure 4 the relation between V and H
would be as shown by the curved line connec t ing po ints A and C in Figure 6 One way to test the null hypothe s is against this al ternative is to test fo r
the hypothesized curvature in the function relating V and H
Fi gure
R
6
v
ID c
H 1 0
I
i I I
sfigned
clasJ s
in def ini g
effec t is the
I
Page middot2lf
One charac teristic of produc ts which may be useful
classes wi th respec t to significanc e of the brand-switching
conveniencenon-convenienc e distinc tion developed by Porter ( 1 97 )
Using his identif ication of IRS indus tries as a s tar t ing point
each of the 32 industries used in this s tudy to one of the two
The ass ignments are not ed in Tab le 4
Using the conveniencenon-convenienc e dis t inc tion the
divided into two sub-samples
sample was
Four regressions were then run bull ene for e ch
sub-sample one for the to tal sample and one for the total sample with
a dummy variable for convenience goods Each regression was run wi th LB s
and wi th indus tries as ob servations The results are given in Tab le 5
For the LB level equations ( 1A - 1D) equation(37) was used ome
rearranging of terms gives v bull y - ( 1-o ) (y - a) ( 1-z ) + ui i i
bull y - 6 1 ( 1 - z ) + ui When observations are pooled across indus trif s yi
is an industry variable In these regressions the values o f y es t imated
in the first stage were used Since industries have been grouped so as
to reduce differences in (y - a) that term is t reated as a cons tant
within each group Finally H the Herfindahl index has been subst ituted
for o the degree of cooperation The resultant equation is linear
y ( 1 - z ) and H x ( l - z ) This is the equation for whi ch results arei i
reported in Table 5 In all regress ions all variables were weighted by the
square roo t of sales
in
raquo
- 34
3 1
i l T 4 8LE S R E G E S S I ON R E SUL T S i X T f N S I ON S T O T Y PE Of
AND L E V E L OF C ONC E kT R A T I O
E O O tt S
tP bull T Y P E NR C ONV O B S tO N-CONV
E L A S T I shy
T I C J T Y
O E - t-41( T SliA R E
H( fl F
1 1 - K O x middot K O X E R F
- H S a 1 1 - H S t 1 8 1 1 9 1
R S O
1 1 0 1
q z
t - bull 9 f 2 1
Z t P D l 6 fC ONV - 1 9 1 s s c 2 1
e - l O I e bull u q a
2 3 1 0 1 1 Nl N- t O V 5 7 1 ( 1 2 19 1 6 1 - J l S 2 5 4
l l l l t z q o a 1 - J o v z I
l ( I NO l l A l l ol u l t o o c bull J u l l
d S
c 3 4 t t r a l
bull Z D 1 0 l l 4 l l 2 3 -gt 2 l l 2 0 - 4 lS 5 4 l C 4 1 o e 6 t l b b 5 1 1 0 1 1 1 c - 1 9 5 1 t d S 2 1 tV
OQ10 T-R A T I O S A E G I VE N l P A RE N T H E S E S l E V E L S f S I G 1 F l C ANC E A - 99 5 B - q 9 t - s NS J UC E NO I Nl E R C ( P T I S U (j f middot) I N f Hf t OUA TIOI S 6 E f 3R E A O J U S T H E N T F OR 1-t E T E R O SIC E O A S T f C I T V pIraquo
1
middot
I
R SO I S C A L C UL A T E D A S f I f bull 1 1 bull 1lt 1 K t t- E E N I S THf U A1 B E R O f fJe S E fi V A T I O S F t S HtE F - S T A T I S
I N DE P E ND E N T y II
T I C F O T H E U Y P O HIE S I S f A l l C JE F F I C H N T A R E QU A L HI l f RO A f40 K I S T H f UUM 8 E R OF Q ( A fL f S
I l l
1 A
1 6
t C
f 1 r R I l l
1 2 1
L Jt
L B
l q
L 1
on s T Y P E bull 2 )
P OUC T
X
H S I l l
l O l A I bull l 0 7 1
O O b O J
1 2 00 1
1 b 2 1 C I l l Z U
l l S b 1 -1 1 0 1
rlt F X l l - fi C f l
bull 1 I
l l
- ()4 2 3 l f H
1 - 2 7 9 1
K O X
1 1 - ti E R F I
l d l
S4 l i A I 2 l b
1 3 00
K O X -t f R F t S J
I 1 - HER F t
)
60 9
z o bull
0
t l l
3 8 8
1 7 6
5 64
64
N f bull
tl S bull
I l l
I 4 1
t ONV
NON-t ONV
A l l
A ll
c o v NO- ( ONV
1 4 1
1 5 1
- l S b
c - 1 1 2 1
5 8 2 C A I 1 6 7 0 1
middot I l 9 H
l 1 il
1 4 1 6 1
E L A C) T J shy
l t C I T Y
I S I
l b l
- 0 2 0
1 - 1 4 9 1
O Z 3 2 1 A I C b 6 b l
0 0 9 2 -J
1 1 b 4 1
o z t t t l 1 2 2 1 1
oru - Ht tt F
1 gt X bull b I
Page 25 bull
Mean val -as ior relevan t variables for the two sub-samples are
Conv Non-Conv
Adv Sales 4 7 2 3
iit Share 2 7 3 2
Ela cicity - 05 3 02 2
Herf indahl 7 3 1 860
That the dist nct ion between coft9enienee goods and non - convenience
goods mat ters is supported not only by these means but also by the regression
results Equat ions lA and lB are different in almost every respec t Only the
concentrat ionmarket share int eract ion term seems to mat ter for convenience
good s Jus t he opposite holds for non - convenience goods both elasticity1
and market sl e are highly significant but the concentrat ionmarket share L
int erac t ion t=rm is not
Given that one minus market share is used these equ3 tions the coeffici t
of market share is the negative of what is sh in col (6) of Table 5 For the
regress ion for non - convenience LB s reported in equation lB then LB s with larger
market shares have lower advertising to sales ratios
On the question of the impact of the degree of cooperation the story is mixed
For convenienc good s the coefficient is pos itive as equation (31) predicts when
there is a po sitive degree of cooperat ion Fo r the non-convenience goods LB s middot
concentration has no effec t
Aggregat ion t o the indus try level and then rer unning the equations $ives the
results shown in (2a-2d) of Tab le 5 As expected R2 goes up and levels of significanc e
fmiddotJr individual coefficients gJ down In add ition the coeffic ient of H ( l -H) inmiddot
quat ion 2A is no t significant though it is po s i tive Since 2A is the indus try level
I
--
all but one of them and that one is not significant For 18 of the 31
wo uld have on its adve rtising to sal es ratio
cognate of LA the variable H ( l-H) pl ays the same role in 2A as H ( l -z 1) p lays
in LA
Vconomies of scale in advert ising
The literature on advert ing contains a number of definitionbull o f economies
of scale and of findings eone rning its presence and magnitude Before I loo k
a t the data with the hope o f mmenting about scale effect s then I wan t to
define the term as I use it
By economies of scale I mean a decrease in the number o f units of adver-
tising which are needed to generate a unit of sales as advert ising is increased
Given the definition of the firm s own elast icity of demand with respect to
advertising (see the text foll wing equation (1) ) the technical definition
is that rii gt 1 Diseconomies of scale are def ined symmetrically i e
Yu lt 1
ith a cro s s - section oi observations on advertising intensity
(v bull a s1) and onmiddot advertising (a ) we may regress the former on the latter 1 i i
The results of such regress ions for the 32 industries are given in Table 2
lines lB ZB bull bull bull 32B For the 32 eases the regression coefficient is positive it
positive coefficients the coefficient i s signif icant
The que tion now is whether these results ar e evidence of pervasive dis-
economies of scale in advertislng I think no t Given my as sumption that the
data I observe are equilibrium results I may ri ghtly conclud e that almost every-
where high levels of advert ising are as sociated in equilibrium with high ratios
of advertising to sales I may no t conclud e anything however about the
impact that a move by some firm away from its equilibrium level o f advertising
middot- Page 27
Given the assumed relation in equation ( 37 ) there is a corre spondshy
ing equilibrium equation relating v and ai The equation is non - linear i
and I have not found a simp le way to characterize it It is po ssib le to
det ermi ne its deriviative with respect to a b y takin$ tne de riviative of ( 3 7 ) i
When that is done the result is
v - 1(40) i bull sl a a z ( 1-z )i i ira-l
Sinc e all the other terms in (40) are non-negative the sign dep ends only
on the sign of 6 1 which is in turn de pend ent only on the sign of (y - a)
That the signs of the coefficients of a in equations lB 2B bull bull bull 32B arei
not always equal to the signs of the coefficients of z in LA 2A bull bull bull 32A i
is still a mystery to me and something to be explored
bull
Summary
My purpose in writing this pa per wa s t o attempt to integrate explicit micro-
economic mode lling reasoned econome tric specification of error terms and high
quality data to exp lore some questions about adver tising by large firms
al quite important improvements could be made in the first two of those areas
and much data ma ssag ing is yet to be done in the third
Concerning the subst an ce of what I have done I think six things are im-
Page 2 8
Severshy
V and Conclusions
portant
1 Explicit modeling is worth the ef for t if for no other reas on than
that it provides a basis for choosing fr om a varie ty of funct ional
forms
2 The same is true for error specif ica t ion I did in fact look at scatshy
ter diagrams for all 32 industry categor ies and they vir tually all show
the het eroscedast icity which I as sumed and for which I correc ted
3 The predicted relation be tween advertis ing in tensity and market share
shows up clearly in on ly 25 of the case s examined I have ye t to
explore why that may b e the cas e
4 Some evidence concerning the presence and impact of coopera tion was
produced but it is not clear ly pervasive
5 The dis tinction between convenience goods and non-c onvenie nce goods is
unambiguously a good one
6 Virtually no evidence concerning econ omies of scale in adver tis ing can
be gleaned fr om this study given its assumpt ions
More work is called for on many of these issues my study seems to have
raised more questions than it has answer ed This is pr obab lly due to the richshy
ne ss of the data source since I had the oppor tunity to address several issue s
at once
- -- - Q --
Foo tnotes
Manager Line o f Business Program Federal Trade Commission The author depended heavily on several staf f members of the LB Pro gram for statistical and clerical support in particular Joe Cholka George Pascoe and Haro lene Jenkins Helpful comments were given b y Richard Caves Dennis Mueller Michael Porter and F M Scherer When the paper was in its early stages o f development both S teve Garber and Jon Rasmussen former economists in the LB Program provided eery useful cri tiques The views expressed here are my own o f course
1 The basic exp licit model of the role of advertising is Dorfman and S teiner ( 1954 ) and the models in this paper may be seen as ano ther extens ion of their work
2 For a review o f models of oligopoly which treat prices and quantities as relevant variables see Shubik ( 1 959) Extens ions to take advertis ing int o account are fairly s traightforward
3 Dorfman and S teiner dealt with a price-setting firm For a monopolis t inverting the demand relation to show Pbull P (Q A) _instead o f Q- (P A) has no effect on the results o f the analysis For oligopolists who face a sys tem of demand relations inversion does have some impac t on the results The dif ferences are not trivial nonetheless they will no t be explored in this study
4 Shubik ( 1 959) p
5 From a formal mathematical point o f view the function is similar to a1 i Legrangian funct ion of the form - r rr
L bull 11 + A ( j - (i i i+j
where the indicates some fixed level of the profits industry Following the same logic which resul ts in and income constraint as the marginal utility of the of the indus tr in the obj ective of firm i
6 Hall amp Weiss ( 1 967) p 323
7 See Schmallensee ( 1 97 2 ) Ch 2 pp 1 6-47
8 On this poin t see Imel amp Helmberger ( 1 9 7 1 )
9 Long ( 1 970) p 130 - 1 39
1 0 See Schmallensee ( 1 972) Ch 3 pp 48-8 7
rrj i+j ) of the rest of the
the identification profits of the rest
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980
v and we
industry concentration isHbull (alternatively
V and Hare functions of the same
pendent Furthermore if there are
VM) and (H1 H2 bull bull bull )contain
VI STATISTI CAL APPLICATIONS
Page 18
Given the observed values of might be tempted to conclude that we
have a two equation system in which v and z are jointly determined That would i i
be inappropriate however since the 2N random variables (v ) are determined by
onlt N basic random variables t The variance-covariance matrix for ( ) must
be singular bull
The extension of this observation to industry level variables is straightshy
forward Since observed industry advertising intensity is V bull l v and observed 4
C4 bull z ) it follows thati l i
set of random variables t and cannot be indeshy-
M industries then the variables (V1 v2 at most M independent random variables
Given the assumptions made above y a and o are constant within a
given industry That being the case (34) can be written as
(37)
s1 bull (1 - o) (y - a)
If and
So + +
(1 - o) a + oy
81 are esti mated for an industry using a suitable statisticalSo procedure giving and 81 an estimate of y is then available The sumSo
Intra-industry analysis
1974 of Business Re ort -
categories are identified
Page 19
of So and S1 is y so y bull So + S1 is the estimate of y
If o is assumed to be bounded by zero and one the sign of 61 is also
the sign of y - a The estimate of So is of cou se an estimate of
(1 - o ) a + oy Moreove if y gt a then a is less than 6o- (1 - B) a+ 0(
Cor esponding results would
hold if y bull a or y lt a
If 61 gt 0 then we would estimate that a lt 6o
bull
For purposes of estimating this model 32 manufacturing industry categories
were selected The data are fram the Line of Business forms filed with
the Federal Trade Commission they are discussed in detail in the Annual Line
which is available from the FTC The 32 industry
in Table 1
Simple regressions for equ ation (37) were run for each of the 32 industries
with data for each of the firms which filed in the industry constituting an ob-
servation The results are given in Table 2 lines lA 2A
1974
the intercept
For all the A
equations generalized least quares was used th both and market
share bein scaled bv the square root of sales
The 32 industries were taken from the larger set of all manufacturing industry
categories in the 1974 data files Two selection cr iteria were used thereLB
had to be at least ten reporting companies and the industry had to be a consumer
goods industry or a producer goods industry which is strongly associated with a
related consumer goods industry so that adve rtising by the firms in the producer
goods industry impacts on demand in the related industry (eg soft drink syrup
and soft drinks) There were 135 industries with at least ten companies - 103
producer goods and 32 consumer goods
273
274
TABLE 1 INDUSTRY CATEGORIES USED
FTC
COOE OESCRI P T I ON _
2001 MEAT PACKINGt SAUSAGE S ANO OTHER PREPARED
MEAT PROOUCTS
2004 DAIRY PRODUCTS EXC F LUID MILK
2005 CANNED SPECIALTIES
2001 FROZEN SPECIALTIES
OEHYORATFOt AND
I NC LU 0 ING PRE SERVE S
SOUP MIXESbull
SEASO lNGSt ANO
OTHER PET FOOJ
RAIN MILL PROOUCTSt
ANO PREPARED
RELATED PRODUCTS
PRODUCTS
Z008 CANNEOt ORIEOt PICKLED FRUITS
ANO VEGET ABLES JAMS JELLIES DEHYDRATED I G TA3t E
S40CES ANO SALAD ORESSINGS
2010 DOG CAT A O
2012 FlOUR t OTHER RICE
MILLING fiLENOE) FLOUR
2014 B EADt CAKE ANO
20113 CO FECTIONERY
2026 BOTTLED A O CANNED SOFT DRINKS 2021 FLAVORING EXTRACTS AND SYRUPS EC
2029 MISC FOOOS AND KINDRED PROJUCTS EXC
ROASTED COFFEE
2301 MENS AND BOYS SUITS AND COATS
2302 MENS ANO BOYS FURNISHINGS
230 3 kOMENS AND MISSES OUTER EAR
2551 HOUSEHOLD FURNITURE
2702 PERIODICALS
Pqe 19a 1
RELATED 1972
src cooe
2011 f 3
202tX2026
2032
203pound
2033t4t5
2047 2041t4t5
2051 2065 2086
2081
209XZ05
231
232
233
251 272
2703 BOOKS
2704 ISC PVeLISHING
2806 ORGANIC FlEERS
2807 DRUG$ ETHICAL
23234
PT 2 S 3
2808 ORUCSt PROPRIETARY
2809 PERFU ES t COSMETICS ANO OTHER TaiLET
PREPARATIONS
2810 SOAP AND OTHER CLEANING PREPARATIONS 2815
2901
PEST1CIDES AND AGR ICULTURAL
PE OLEUM REF I N ING CHEMICALS EC
3606 HOUSEHOLD COOKING EQUIPMENT
3612 HOUSEHOLD APPLIANC ES NEC INCLUJING ELECshy
rqlC HOUSE ARFS AND FANS ANO SE I G MACHINES
3617 RADIO AND T V RECE I VING SETS 3808 PHOTOG APHIC E UIP amp SUP LfESt EXC
3qo3
PHO TOCOPYING
SPOR TING A D E QU I P M ENT f SUP P L I E S ATHLETTC GOODS NEC
3904 DOLLS GAMESt TOYS AND CHLOqENtS VE ICLES
PT283
2844
234tX2844
287 9
291 3631
3634t6t9
36H
PT386l
3949
15
943
middot
l A6l f 2 R(GRESSION RESULTS
EQ I NO
NR CODE cu 121
lA 2001 A (
2A 2004 r (
3A 2005 a
-R
ClS amp31
16
17
1 z
I TER-CEPT bull t
ooo9t21At 00022311 oooeos uu
OOl49tCAI 0010 2 middot 000960CCI
o049tl1At o04t55C61
MARKET SH ARE bull 5 I
-010670Ctl
-017600181
-001050
-0211AO
-0077 0
AOVER-TIS INti
lbl
000()39 000072
000018 0 0024liAI
-000079
SHARf
4ampgtVf Q bull
111
000140
-000170
RSQ
COl
66 485 862
607 6 3 t ()2
blti6 633
flASTI-
CI T Y PH
-007271(1
000441
-u0288l
clASTSHU (t)Ef
bull 10
tto
-234
210
( OOlltOCCI -126520 000922 181 001790
4A 200 7 18 00590liAI -038670 4e16 -032110 lell u c
OOOS74e 000700 025940
00067018) 0021001AI -02l100CAI
5 26
920
5A 6 c
200 8 26 0 03Sl9CA I 0012 26181 o 024511(
-005540
-0 81420 0001961AI O()Ol90CAI ooll240
441 604 891
-0020 19 274
6A 2010 16 Oe04224CAI o359lOIAI 9oq ll4015dCAI 061 6 -middot 005435CAI 00011 8 1AI 917 c 00535ltAI -009850((1 OOOA51J CAI -001610161 971
7A 2012 001862 000010 351 00187) oo1 a oooq45 ono1o6 439 c 001109 -027050 Oe0067SIAI -OOll60C81 897
- middotmiddot- -
8A 201t ll 000757 Ol4140CCI 814 ott9oocet 095 8 oooqo6 C C J 000069(81 811 c ootl2lampBI -080910CCI Oe009lUBI -OOtlOOCCI 960I
I DIIJIT pound11 1V OTIUTVTfAll-1pound1t- A _ 00 R - qq r - Q
1jl)OQCD
0
-middot middot--
l
_ __ _
I
i
0 00141 c c
91)7
I
T ABLE 2 REG ESSIOH RESULTS CCONTI
EO NR
IND CODE
iR bull
cos
I TEq-
CEPT
t1ARKE T
SHA E
AOVER-
Tl SING
SHARpound-cent
AuVER
RSO ILAST1-
C I T Y ELAST
SH CPtf (lt Cl) bull 11 I 41 I 5I (61 171 I 8 I f9) c 10)
9A 2 0 1 8 19 002397181 -007640 367 -005 246 le46 6 000605 000550CAI 597
-000029 G14180 OOl654CAJ -024li01At
lOA 2026 12 0028901AI -OlZl90 -0114 0 1 1 1 5 759
Ll (
llA 2021
f3 c
12A lOZ9
B
(
13A 2301
B
c
14A 2302
B
c
154 201
n (
16A 2551
8
10
35
10
20
19
24
001897(()
00141 6CCI
Oll9lHCI
010604 middot middot
Oll083
OOJ11JCAt
002059141
001411CIH
-000412
000293(81
000187
001159 (3)
ooo6a4cct
o00792CCI
OOl40dCAt
000006
-000076
OOlltl6fA)
o o0 5 64CCt
-034570
-o1middot o3o
-1659 90
Ol51lO
-046190CAt
Ol5450CA)
-005l60
002940
-030060(()
-0 43l50CCt
0 33650
-019080
00004
bullJ00741CAI
-000069
000466
0001 51CAt
000947CAt
O OOI)70CAt OOl750CA J
000175CCI
000671CAI
OOll5) f A t 002207CBI
0004b7CA I
- 007 1 70
001760
-001410CAt
- 0 16200CAt
-004210
- 1 207 70
705
96 0
542
466 805
460
614
921
832
9 61
996
424
550
9 15
56 1 651
884
-006114
019 037
03504lCA)
oo409Q
-141644((
491 -0 17664
607
871
lell
084
101
072
103
108 11IllOQfl)-- l
-037880fCJ OOl6l4CAJ -040l70fCI 0()
c
I
I
bull 9
9t
middot middot middotbullmiddot bull lraquo
TABLE 2 REGRE SION RESUlTS ICONTt
bull
EO I NO tHl I filER- f4A_K T AOVER- sbulluRE ttSQ flA Jl- iLAS T NR CODE cos CEPT SHARE JISING AOVfR bull (JTY SHR CiJCF
110Cl I f amp Itt 15J tbt bull 1c u caa
llA 2702 ll o079 6 81 Bt -019110 S6s -OlltO 111
a 004679 ooooo1 t69
( oOl26o -015160 001560 -022690 ltJr()I
ldA 270 3 19 0065lt -OIt2l60 220 -Ol57a2 11 8 e o 0 71t6IA t ol08 75tAt ( OOl983CiU -060llOC() OOl7191A t -O lltdOC A) middot 89
lqA 2704 10 005082 -Oil060 22b -l)i1977 lelO 6 -000080 oo2772C4 t 910 c 000907 -011 200 Oe04105CA t -046960 995
20A 2d06 1 II
O OlltStiAt -001120 837 00014 -))th O01380amp At -oooo 11 821 ( OOl526CAt -026700181 0002 5UAt 000530 969
2lA 2807 28 OOt9lt9CAt -00891 0 516 -0 03962 225I8 002095CCt OOOl lliAt 690 ( o02965CA t -056650((1 OOOt50CAJ -003l70ttl 911I
I 22A 2808 15 ot3l06 119730 655 1130) 090
0 - 007526CC t Oe003961At 835- middot-
c I oo7523 -067210 001l 2tA t -0091t80CCt 951
2lA 2809 21 Oe2067lCAt -1218101At 71 2 -10095((1 121 ttlIaOQ8 00797418) OOOI8lltt 686 Ill( 0092lU A t -Olltl70Ct t 000605CAt -OOit8701At 951 _
---- -
32 009506CAt
0 bull 06 981 I A J 006901 cA bull
00 24A 2810 -010960 5t 5 -OOltSO 7 56
8 -000011 s11 c -5811701At oooaoatAJ oot520IAI 888
-- 37 --- middot - -- -1 ) l lbull t5JJ r-l bullhmiddot1 --- - --
l5A
26A
27A
c
2RA
c
2 9A
31A
llA
24
13
597
Ci)55
14
l 6h
RSQ ELASTI- (lASTCOE f-
TABLE 2
EQ I NO NR CODE IU
REGRESSION RESULTS CCONTa
NR INTER- MARKET
(10S CEPT SHARE AOVER- SHARfcent
TIS lNG AOVER
) CITY StiR
c It) bull 5 co a I 1) bull 0 bull 9 c l 0 bull( 2 1
2815 16 003019 -009990 275 -o069bmiddotJ
6 001446 000224 282 c 002178 -054630 o02221CCI -0075 00 646
2901 29 00021t2 CAI 0)0520 592 0007 6 1 068
0 00105C8 t 000012141 60)g( OOOlJ51AJ -00126018) ooou271At -000u80CA) 974
36oa 1 3 003296(4) -oca29o -oo-q94
001755 000203 lt91t 002499((1 -04l820CCI OOl374CAI 0068)0 946
3612 0033281CI 011520 195 01684) oso
00 30 42C OI 00012 2 424 oo579ltAt -2 7l7801At 002 100IAI -003220 819
8 3617 btel 011789 081002259 009530
001448 oOOl OitCCI 732 OOl7 22CAI -0 46160CAI 0004961AI -OOll40181 989c
30A 3808 14 0021t 3 6fAI -000110 628 o023291C) -005 6 middot- _ middot- 0bull 02 04 1 IC I ooooo9 637 ( 001584((1 -0161)1)0 00021 91AI -0 00060
390) 14 002790CCt 006590 h8 t 0 0938 4 070
B OOl89bCCJ 000144 74
0021961(1 -OItlOlO OOl323CAI -004800 915
B
middot--- - middot 4 - middot- - dfll3904 12 004563181 0)8300 879 0428651(1 089 OQIll004 277CA I ooozstcea 919
c 0041)3181 -0-11250 0011881(1 -006270 972 0Ill
- J - bull bull bull bullbullbull middot - - middot - middot gt middotbull L)ofbullflbull middot middotfJltl -tn- middotmiddotmiddotmiddoto +f middot t middot bull
c
is
positive
2
Page 0 determineThe consumer goodproducer good split is sometimes difficult to
Sollle support for the split used in this paper is a comparison of goodness - of - fit
distributions of a
industries
set
advertising are
that industry
a sales increase of
the
highest aD10U3 those
measures for the two groups of industries Table 3 contains
measures for the 32- consumer goods industries and the 103 producer goods
2 Median R is 5 7 for the consumer goods set and 40 for the producer goods
Estimates - of y the elasticity of demand with respect to
given in column of Table 2 The hypothesis that y bull 0 with
alternative was applied for each ittdustry As shown in the
could not be rtjected for 24 of the 32 industries Of the
and 10 are positive For the remalnifg eight three are
and five are significantly positive
Of the threertegative elasetcities the largest (in absolute value)
cosmetics (code 2809) With all elasticity of -1 01 a one
industry advertising would gwerate a decrease in industry sales
more than one percetu
At the other extrema is toys and games (code 39 04) For
a one percent ittcrease in industry advertising would lead to
four-tenths of a pampreentmiddot Though not significantly different from zero
elasticity estimatbull for pt-bpuetary drugs (code 2808)
which are
Both drug industries have
to sales rafios (211 pe1 cent for 28 08 and 13 4 per cent
other hand flaIJo ing ext1ac s (mainly soft-drink SYUlS) also
is for
percent increase in
of slightly
a middottwo - tailed
table that hypothesis
24 14 are negative
significantly negative
9
the codetics and p-ropriacuy very high advertising
for 28 09) bull On the
has a high ind UY
advenisiltg t() sales ratio ( 8 0 per cent) bull but its estimated elasticity is very
small and not signifieantly different uom zero
middot It doe$ not seem unusual to find many elasticities near zero with a few
negative and a few positive If aggregate consumption is insensitive to aggregate
advertising as seems likely ld then what one industry gains must be lost by
others
Page 21
elasticity of demand does
(10) we can
if t 0 then
Finally a determination concerning the industry
not fix the individual firms elasticity Going back to equation
write y bull (1 - ) a + z y Since a gt 0 it follows thatii i i
Y gt 0 If Ylt 0 however y s sign is indeterminateij ij
Turning next to the coefficient of market share 20 of them are negative and
12 are positive with three of each sign being significantly different from zero
The largest positive value which is significant is for pet food (code 2010) and
a close second is mens amp boys suits and coats (code 23 01) The largest signifi
cant negative value is for cosmetics (code 28 09)
In the model some relations among y a 0 and 61 are implied If middota is
negative (1- y) is greater than one Since 61Y bull (1- a) (1 - ) 61Y is
pCsitive but it will be less than one if 0 is large enough Seventeen of the A
ys are negative in each of the seventeen industries 61 is also negative and les
1 (algebraicly) than y so 61Y is greater than one his is not evidence that the
degree of cooperation is low or zero in any industry of course A high o togethegt
with a negative y which is small (in absolute terms) relative to a can give a
value of 61Y greater than one On the other hand if 61Y were less than one
o would have to be greater than zero
For y gt 0 two conditions may hold If y lt a it follows that 61 lt 0 since
(y - a ) lt 0 It also follows that 61Y lt 0 regardless of the value of o bull
In addition the relation is an if-and-only-if one ie if 61Y is negative
then y lt a Three industries have negative values for 61y dairy products
except milk (code 20 04) synthetic fibers (code 2806) and photographic equipment
(code 3808) In all three cases y is quite small so a does not have to be very
large to give a negative value to al
If y gt a the model shows that 61-- gt 0 whatever the value of 0 bull There are
12 cases where both y and 61 are positive with 61 ranging from near zero
(grain milling code 20 12) to 1 20 ( proprietary drugs code 2808) Several
coeffiei-en_rs
--
15
o f to zero
Tab le 3 - Consumer Goods
Indus tr ies
2Distributions of R and Producer Goods
Page 20a
Consumer Producer
Goods Goods
0 - 9 9 0 3
10 - 19 9 0 1 1
20 - 29 9 3 20
30 - 39 9 3 17
40 - 49 9 5 16
50 - 59 9 8
69 9 6 Ii 1060
--f 670 79 9 2
80 - 89 9 4 4
90 - 100 1 1
Totals 32 10 3
Notes
2is used R i s cal-
culated as F F + ((N - K- 1 ) K ] 1 b_ F is the F
statistic Lor the hypothesis that all o f the non - intercept terms are simul taneously equal
1
2
--
-- Since generalized least squ-area regress ion
Page 12
o f these cases have values o f 6 1 Y rea ter than 0 8 (pet food (20 10) bread amp cakes
(20 14 ) miscellaneous processed foods (20 29 ) men s and boys suit s and coats
(23 01 ) p rop riet ary drugs (28 08 ) elect ric housewares (36 1 2) radio and
IV sets 17 and games and toys (39 For be than(36 ) 04)) 6 1 Y to greater 0 8
both o and cry mus t b e less than 0 2 That is in these industries the results
shown in Tab le 2 would hold only with very strong dominance of marke t demand
effects over brand swit ching e ffects and a very low degree o f cooperation
Using the concentration rat ios given in Table 4 the average concentration
ratio for the eight indust ries is 38 1 For the remaining 24 industries it is
43 2 This result would lend some support t o the proposition that concentrat ion
and cooperation are related if the group of eight industries are identi fied
as having low degrees of cooperation
One f inal note on model predict ions S ince both cr and o are assumed to
be non-negative 60 should also be non-negat ive In only one o f the 32 cas es
is 60 negative and it is not signif icant bull
cor pJ
7 1
7 0 4
1 4 7 5 1
)
7 3 z a o a t s 1 1 3 3 5 ( i V Z 8 tJ1 2 1 - 1 1 0 7 3 5 c middot v 2 8 1 0 3 2 -0 0 1 4 3 0 C 1 V 9 8 3 2 8 1 5 1 6 - 0 0 6 t1 6 9 C _ V 1 11 0 2 9 0 1 29 0 00 7 6 1 C l k V 7 3 1 3 6 08 1 3 - o n 4 9 9 4 3 3 3 6 1 2 2 4 0 1 6 4 3
3 1 7 1 3 0 1 1 7 d 7 8 3 3 8 0 8 0 0 2 3 i 2 3 0 6
P 4
3 9 0 3 1 4
1 2
0 09 3 8 4 39 1)1 0 4 l 8 6 5
T A LE 4 e I NCU S T R Y C A T E G O R Y
F TC NOM e tOOE ass
C 1 I c 2 )
2 0 0 1 1 6 2 0 0r 1 7 zo o l Z 2 0 0 7 1 8
z o o 2 6
z o t a 1 6
2 1) 1 2 1 5 2 0 1 4 1 1 2 0 1 a 1 9 2 0 26 1 2 2 0 2 7 1 0 2 0 29 - 3 5
2 3 0 1 10
2 3 02 z o
2 3 03 1 9
2 5 5 1 2 4 Z 1 0 Z 1 1 2 0 3 1 9 2 7 4 1 0 z a o 1 3
2 A 0 7 z s
E LA S T I -C I T Y
c 3 )
- 0 09 7 2 7
0 00 449
- 0 0 2 8 9 3
- 0 3 2 7 70 - 0 OZ il I q
o 4 u t s a 0 0 1 3 7 3
o 1 4 o o-0 0 5 2 4o -0 1 94 0 1
-C l 6 l l 4 0 1 JJ 3 7 0 3 t 0 4 2 0 0 4 0 9
- 0 4 1 8 44 - O l 7 t ) 4 -ll 7 1 4 0 b -0 3 5 7 8 Z
- 0 1 9 7 7 0 1)0 3 3 4
-o n 9 2
IJ A T A
CONV
ON-C ONV
( 4 )
t 1NV
C 1t-V C NV c ) v C - JNV C rlNV c n v c a v C ON V ( i) l V
C CtV
C t1tV C IJNV
C 1 V
c 1 v
Pa ge 2La
HEQ F t Noe x
c 5 )
6 2q
6 89 1 48
6 1 4 ) q 3
1 9 7 1 9
q z t s 5 a e z
l l 9 1
4 o 7 1 2 0
c oqt 1 Z
4 4 )-q 1 z
5 5 9 1 0 64 1 3 3
n 3
Inter industry analysi
Page 23
Equation ( 3 6 ) may be used to fo rmulate a rough test o f the cooperat ion
vs e f ficiency controversy I f we rest ricted ourselves to some subset o f
indus tries where the relat ion be tween y and a i s the same we would be lookingbull
at a sample o f indus tries for which the dependence o f V on H and o is a
quadratic wi th no squared terms
I f we now sub s titute (26) into ( 36 ) we get
(38) v bull a + (y - a) R + (y - a) o (H) - (y - a) H o (H) + u
If it is only d if ferences in marke t shares that affect advertising inten-
s i ty levels i f concentration does not af fect cooperation and i f o bull 0
( 38 ) reduces t o
( 39 ) V bull a + (y - a) H + U
Let this b e the null hyp o thes is and note that V is a linear func-
tion o f H If y gt a and the null hyp othesis holds the expected relat ion
between V and H is as shown by the pos itively sloped straight line in Figure
6 b etween points A and C
In this context the alterna ive hypothesis the degree o f
cooperation ( o) is a positive function o f concentrat ion
is that
(H) I f the func-
tion is as described in (26) and Figure 4 the relation between V and H
would be as shown by the curved line connec t ing po ints A and C in Figure 6 One way to test the null hypothe s is against this al ternative is to test fo r
the hypothesized curvature in the function relating V and H
Fi gure
R
6
v
ID c
H 1 0
I
i I I
sfigned
clasJ s
in def ini g
effec t is the
I
Page middot2lf
One charac teristic of produc ts which may be useful
classes wi th respec t to significanc e of the brand-switching
conveniencenon-convenienc e distinc tion developed by Porter ( 1 97 )
Using his identif ication of IRS indus tries as a s tar t ing point
each of the 32 industries used in this s tudy to one of the two
The ass ignments are not ed in Tab le 4
Using the conveniencenon-convenienc e dis t inc tion the
divided into two sub-samples
sample was
Four regressions were then run bull ene for e ch
sub-sample one for the to tal sample and one for the total sample with
a dummy variable for convenience goods Each regression was run wi th LB s
and wi th indus tries as ob servations The results are given in Tab le 5
For the LB level equations ( 1A - 1D) equation(37) was used ome
rearranging of terms gives v bull y - ( 1-o ) (y - a) ( 1-z ) + ui i i
bull y - 6 1 ( 1 - z ) + ui When observations are pooled across indus trif s yi
is an industry variable In these regressions the values o f y es t imated
in the first stage were used Since industries have been grouped so as
to reduce differences in (y - a) that term is t reated as a cons tant
within each group Finally H the Herfindahl index has been subst ituted
for o the degree of cooperation The resultant equation is linear
y ( 1 - z ) and H x ( l - z ) This is the equation for whi ch results arei i
reported in Table 5 In all regress ions all variables were weighted by the
square roo t of sales
in
raquo
- 34
3 1
i l T 4 8LE S R E G E S S I ON R E SUL T S i X T f N S I ON S T O T Y PE Of
AND L E V E L OF C ONC E kT R A T I O
E O O tt S
tP bull T Y P E NR C ONV O B S tO N-CONV
E L A S T I shy
T I C J T Y
O E - t-41( T SliA R E
H( fl F
1 1 - K O x middot K O X E R F
- H S a 1 1 - H S t 1 8 1 1 9 1
R S O
1 1 0 1
q z
t - bull 9 f 2 1
Z t P D l 6 fC ONV - 1 9 1 s s c 2 1
e - l O I e bull u q a
2 3 1 0 1 1 Nl N- t O V 5 7 1 ( 1 2 19 1 6 1 - J l S 2 5 4
l l l l t z q o a 1 - J o v z I
l ( I NO l l A l l ol u l t o o c bull J u l l
d S
c 3 4 t t r a l
bull Z D 1 0 l l 4 l l 2 3 -gt 2 l l 2 0 - 4 lS 5 4 l C 4 1 o e 6 t l b b 5 1 1 0 1 1 1 c - 1 9 5 1 t d S 2 1 tV
OQ10 T-R A T I O S A E G I VE N l P A RE N T H E S E S l E V E L S f S I G 1 F l C ANC E A - 99 5 B - q 9 t - s NS J UC E NO I Nl E R C ( P T I S U (j f middot) I N f Hf t OUA TIOI S 6 E f 3R E A O J U S T H E N T F OR 1-t E T E R O SIC E O A S T f C I T V pIraquo
1
middot
I
R SO I S C A L C UL A T E D A S f I f bull 1 1 bull 1lt 1 K t t- E E N I S THf U A1 B E R O f fJe S E fi V A T I O S F t S HtE F - S T A T I S
I N DE P E ND E N T y II
T I C F O T H E U Y P O HIE S I S f A l l C JE F F I C H N T A R E QU A L HI l f RO A f40 K I S T H f UUM 8 E R OF Q ( A fL f S
I l l
1 A
1 6
t C
f 1 r R I l l
1 2 1
L Jt
L B
l q
L 1
on s T Y P E bull 2 )
P OUC T
X
H S I l l
l O l A I bull l 0 7 1
O O b O J
1 2 00 1
1 b 2 1 C I l l Z U
l l S b 1 -1 1 0 1
rlt F X l l - fi C f l
bull 1 I
l l
- ()4 2 3 l f H
1 - 2 7 9 1
K O X
1 1 - ti E R F I
l d l
S4 l i A I 2 l b
1 3 00
K O X -t f R F t S J
I 1 - HER F t
)
60 9
z o bull
0
t l l
3 8 8
1 7 6
5 64
64
N f bull
tl S bull
I l l
I 4 1
t ONV
NON-t ONV
A l l
A ll
c o v NO- ( ONV
1 4 1
1 5 1
- l S b
c - 1 1 2 1
5 8 2 C A I 1 6 7 0 1
middot I l 9 H
l 1 il
1 4 1 6 1
E L A C) T J shy
l t C I T Y
I S I
l b l
- 0 2 0
1 - 1 4 9 1
O Z 3 2 1 A I C b 6 b l
0 0 9 2 -J
1 1 b 4 1
o z t t t l 1 2 2 1 1
oru - Ht tt F
1 gt X bull b I
Page 25 bull
Mean val -as ior relevan t variables for the two sub-samples are
Conv Non-Conv
Adv Sales 4 7 2 3
iit Share 2 7 3 2
Ela cicity - 05 3 02 2
Herf indahl 7 3 1 860
That the dist nct ion between coft9enienee goods and non - convenience
goods mat ters is supported not only by these means but also by the regression
results Equat ions lA and lB are different in almost every respec t Only the
concentrat ionmarket share int eract ion term seems to mat ter for convenience
good s Jus t he opposite holds for non - convenience goods both elasticity1
and market sl e are highly significant but the concentrat ionmarket share L
int erac t ion t=rm is not
Given that one minus market share is used these equ3 tions the coeffici t
of market share is the negative of what is sh in col (6) of Table 5 For the
regress ion for non - convenience LB s reported in equation lB then LB s with larger
market shares have lower advertising to sales ratios
On the question of the impact of the degree of cooperation the story is mixed
For convenienc good s the coefficient is pos itive as equation (31) predicts when
there is a po sitive degree of cooperat ion Fo r the non-convenience goods LB s middot
concentration has no effec t
Aggregat ion t o the indus try level and then rer unning the equations $ives the
results shown in (2a-2d) of Tab le 5 As expected R2 goes up and levels of significanc e
fmiddotJr individual coefficients gJ down In add ition the coeffic ient of H ( l -H) inmiddot
quat ion 2A is no t significant though it is po s i tive Since 2A is the indus try level
I
--
all but one of them and that one is not significant For 18 of the 31
wo uld have on its adve rtising to sal es ratio
cognate of LA the variable H ( l-H) pl ays the same role in 2A as H ( l -z 1) p lays
in LA
Vconomies of scale in advert ising
The literature on advert ing contains a number of definitionbull o f economies
of scale and of findings eone rning its presence and magnitude Before I loo k
a t the data with the hope o f mmenting about scale effect s then I wan t to
define the term as I use it
By economies of scale I mean a decrease in the number o f units of adver-
tising which are needed to generate a unit of sales as advert ising is increased
Given the definition of the firm s own elast icity of demand with respect to
advertising (see the text foll wing equation (1) ) the technical definition
is that rii gt 1 Diseconomies of scale are def ined symmetrically i e
Yu lt 1
ith a cro s s - section oi observations on advertising intensity
(v bull a s1) and onmiddot advertising (a ) we may regress the former on the latter 1 i i
The results of such regress ions for the 32 industries are given in Table 2
lines lB ZB bull bull bull 32B For the 32 eases the regression coefficient is positive it
positive coefficients the coefficient i s signif icant
The que tion now is whether these results ar e evidence of pervasive dis-
economies of scale in advertislng I think no t Given my as sumption that the
data I observe are equilibrium results I may ri ghtly conclud e that almost every-
where high levels of advert ising are as sociated in equilibrium with high ratios
of advertising to sales I may no t conclud e anything however about the
impact that a move by some firm away from its equilibrium level o f advertising
middot- Page 27
Given the assumed relation in equation ( 37 ) there is a corre spondshy
ing equilibrium equation relating v and ai The equation is non - linear i
and I have not found a simp le way to characterize it It is po ssib le to
det ermi ne its deriviative with respect to a b y takin$ tne de riviative of ( 3 7 ) i
When that is done the result is
v - 1(40) i bull sl a a z ( 1-z )i i ira-l
Sinc e all the other terms in (40) are non-negative the sign dep ends only
on the sign of 6 1 which is in turn de pend ent only on the sign of (y - a)
That the signs of the coefficients of a in equations lB 2B bull bull bull 32B arei
not always equal to the signs of the coefficients of z in LA 2A bull bull bull 32A i
is still a mystery to me and something to be explored
bull
Summary
My purpose in writing this pa per wa s t o attempt to integrate explicit micro-
economic mode lling reasoned econome tric specification of error terms and high
quality data to exp lore some questions about adver tising by large firms
al quite important improvements could be made in the first two of those areas
and much data ma ssag ing is yet to be done in the third
Concerning the subst an ce of what I have done I think six things are im-
Page 2 8
Severshy
V and Conclusions
portant
1 Explicit modeling is worth the ef for t if for no other reas on than
that it provides a basis for choosing fr om a varie ty of funct ional
forms
2 The same is true for error specif ica t ion I did in fact look at scatshy
ter diagrams for all 32 industry categor ies and they vir tually all show
the het eroscedast icity which I as sumed and for which I correc ted
3 The predicted relation be tween advertis ing in tensity and market share
shows up clearly in on ly 25 of the case s examined I have ye t to
explore why that may b e the cas e
4 Some evidence concerning the presence and impact of coopera tion was
produced but it is not clear ly pervasive
5 The dis tinction between convenience goods and non-c onvenie nce goods is
unambiguously a good one
6 Virtually no evidence concerning econ omies of scale in adver tis ing can
be gleaned fr om this study given its assumpt ions
More work is called for on many of these issues my study seems to have
raised more questions than it has answer ed This is pr obab lly due to the richshy
ne ss of the data source since I had the oppor tunity to address several issue s
at once
- -- - Q --
Foo tnotes
Manager Line o f Business Program Federal Trade Commission The author depended heavily on several staf f members of the LB Pro gram for statistical and clerical support in particular Joe Cholka George Pascoe and Haro lene Jenkins Helpful comments were given b y Richard Caves Dennis Mueller Michael Porter and F M Scherer When the paper was in its early stages o f development both S teve Garber and Jon Rasmussen former economists in the LB Program provided eery useful cri tiques The views expressed here are my own o f course
1 The basic exp licit model of the role of advertising is Dorfman and S teiner ( 1954 ) and the models in this paper may be seen as ano ther extens ion of their work
2 For a review o f models of oligopoly which treat prices and quantities as relevant variables see Shubik ( 1 959) Extens ions to take advertis ing int o account are fairly s traightforward
3 Dorfman and S teiner dealt with a price-setting firm For a monopolis t inverting the demand relation to show Pbull P (Q A) _instead o f Q- (P A) has no effect on the results o f the analysis For oligopolists who face a sys tem of demand relations inversion does have some impac t on the results The dif ferences are not trivial nonetheless they will no t be explored in this study
4 Shubik ( 1 959) p
5 From a formal mathematical point o f view the function is similar to a1 i Legrangian funct ion of the form - r rr
L bull 11 + A ( j - (i i i+j
where the indicates some fixed level of the profits industry Following the same logic which resul ts in and income constraint as the marginal utility of the of the indus tr in the obj ective of firm i
6 Hall amp Weiss ( 1 967) p 323
7 See Schmallensee ( 1 97 2 ) Ch 2 pp 1 6-47
8 On this poin t see Imel amp Helmberger ( 1 9 7 1 )
9 Long ( 1 970) p 130 - 1 39
1 0 See Schmallensee ( 1 972) Ch 3 pp 48-8 7
rrj i+j ) of the rest of the
the identification profits of the rest
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980
Intra-industry analysis
1974 of Business Re ort -
categories are identified
Page 19
of So and S1 is y so y bull So + S1 is the estimate of y
If o is assumed to be bounded by zero and one the sign of 61 is also
the sign of y - a The estimate of So is of cou se an estimate of
(1 - o ) a + oy Moreove if y gt a then a is less than 6o- (1 - B) a+ 0(
Cor esponding results would
hold if y bull a or y lt a
If 61 gt 0 then we would estimate that a lt 6o
bull
For purposes of estimating this model 32 manufacturing industry categories
were selected The data are fram the Line of Business forms filed with
the Federal Trade Commission they are discussed in detail in the Annual Line
which is available from the FTC The 32 industry
in Table 1
Simple regressions for equ ation (37) were run for each of the 32 industries
with data for each of the firms which filed in the industry constituting an ob-
servation The results are given in Table 2 lines lA 2A
1974
the intercept
For all the A
equations generalized least quares was used th both and market
share bein scaled bv the square root of sales
The 32 industries were taken from the larger set of all manufacturing industry
categories in the 1974 data files Two selection cr iteria were used thereLB
had to be at least ten reporting companies and the industry had to be a consumer
goods industry or a producer goods industry which is strongly associated with a
related consumer goods industry so that adve rtising by the firms in the producer
goods industry impacts on demand in the related industry (eg soft drink syrup
and soft drinks) There were 135 industries with at least ten companies - 103
producer goods and 32 consumer goods
273
274
TABLE 1 INDUSTRY CATEGORIES USED
FTC
COOE OESCRI P T I ON _
2001 MEAT PACKINGt SAUSAGE S ANO OTHER PREPARED
MEAT PROOUCTS
2004 DAIRY PRODUCTS EXC F LUID MILK
2005 CANNED SPECIALTIES
2001 FROZEN SPECIALTIES
OEHYORATFOt AND
I NC LU 0 ING PRE SERVE S
SOUP MIXESbull
SEASO lNGSt ANO
OTHER PET FOOJ
RAIN MILL PROOUCTSt
ANO PREPARED
RELATED PRODUCTS
PRODUCTS
Z008 CANNEOt ORIEOt PICKLED FRUITS
ANO VEGET ABLES JAMS JELLIES DEHYDRATED I G TA3t E
S40CES ANO SALAD ORESSINGS
2010 DOG CAT A O
2012 FlOUR t OTHER RICE
MILLING fiLENOE) FLOUR
2014 B EADt CAKE ANO
20113 CO FECTIONERY
2026 BOTTLED A O CANNED SOFT DRINKS 2021 FLAVORING EXTRACTS AND SYRUPS EC
2029 MISC FOOOS AND KINDRED PROJUCTS EXC
ROASTED COFFEE
2301 MENS AND BOYS SUITS AND COATS
2302 MENS ANO BOYS FURNISHINGS
230 3 kOMENS AND MISSES OUTER EAR
2551 HOUSEHOLD FURNITURE
2702 PERIODICALS
Pqe 19a 1
RELATED 1972
src cooe
2011 f 3
202tX2026
2032
203pound
2033t4t5
2047 2041t4t5
2051 2065 2086
2081
209XZ05
231
232
233
251 272
2703 BOOKS
2704 ISC PVeLISHING
2806 ORGANIC FlEERS
2807 DRUG$ ETHICAL
23234
PT 2 S 3
2808 ORUCSt PROPRIETARY
2809 PERFU ES t COSMETICS ANO OTHER TaiLET
PREPARATIONS
2810 SOAP AND OTHER CLEANING PREPARATIONS 2815
2901
PEST1CIDES AND AGR ICULTURAL
PE OLEUM REF I N ING CHEMICALS EC
3606 HOUSEHOLD COOKING EQUIPMENT
3612 HOUSEHOLD APPLIANC ES NEC INCLUJING ELECshy
rqlC HOUSE ARFS AND FANS ANO SE I G MACHINES
3617 RADIO AND T V RECE I VING SETS 3808 PHOTOG APHIC E UIP amp SUP LfESt EXC
3qo3
PHO TOCOPYING
SPOR TING A D E QU I P M ENT f SUP P L I E S ATHLETTC GOODS NEC
3904 DOLLS GAMESt TOYS AND CHLOqENtS VE ICLES
PT283
2844
234tX2844
287 9
291 3631
3634t6t9
36H
PT386l
3949
15
943
middot
l A6l f 2 R(GRESSION RESULTS
EQ I NO
NR CODE cu 121
lA 2001 A (
2A 2004 r (
3A 2005 a
-R
ClS amp31
16
17
1 z
I TER-CEPT bull t
ooo9t21At 00022311 oooeos uu
OOl49tCAI 0010 2 middot 000960CCI
o049tl1At o04t55C61
MARKET SH ARE bull 5 I
-010670Ctl
-017600181
-001050
-0211AO
-0077 0
AOVER-TIS INti
lbl
000()39 000072
000018 0 0024liAI
-000079
SHARf
4ampgtVf Q bull
111
000140
-000170
RSQ
COl
66 485 862
607 6 3 t ()2
blti6 633
flASTI-
CI T Y PH
-007271(1
000441
-u0288l
clASTSHU (t)Ef
bull 10
tto
-234
210
( OOlltOCCI -126520 000922 181 001790
4A 200 7 18 00590liAI -038670 4e16 -032110 lell u c
OOOS74e 000700 025940
00067018) 0021001AI -02l100CAI
5 26
920
5A 6 c
200 8 26 0 03Sl9CA I 0012 26181 o 024511(
-005540
-0 81420 0001961AI O()Ol90CAI ooll240
441 604 891
-0020 19 274
6A 2010 16 Oe04224CAI o359lOIAI 9oq ll4015dCAI 061 6 -middot 005435CAI 00011 8 1AI 917 c 00535ltAI -009850((1 OOOA51J CAI -001610161 971
7A 2012 001862 000010 351 00187) oo1 a oooq45 ono1o6 439 c 001109 -027050 Oe0067SIAI -OOll60C81 897
- middotmiddot- -
8A 201t ll 000757 Ol4140CCI 814 ott9oocet 095 8 oooqo6 C C J 000069(81 811 c ootl2lampBI -080910CCI Oe009lUBI -OOtlOOCCI 960I
I DIIJIT pound11 1V OTIUTVTfAll-1pound1t- A _ 00 R - qq r - Q
1jl)OQCD
0
-middot middot--
l
_ __ _
I
i
0 00141 c c
91)7
I
T ABLE 2 REG ESSIOH RESULTS CCONTI
EO NR
IND CODE
iR bull
cos
I TEq-
CEPT
t1ARKE T
SHA E
AOVER-
Tl SING
SHARpound-cent
AuVER
RSO ILAST1-
C I T Y ELAST
SH CPtf (lt Cl) bull 11 I 41 I 5I (61 171 I 8 I f9) c 10)
9A 2 0 1 8 19 002397181 -007640 367 -005 246 le46 6 000605 000550CAI 597
-000029 G14180 OOl654CAJ -024li01At
lOA 2026 12 0028901AI -OlZl90 -0114 0 1 1 1 5 759
Ll (
llA 2021
f3 c
12A lOZ9
B
(
13A 2301
B
c
14A 2302
B
c
154 201
n (
16A 2551
8
10
35
10
20
19
24
001897(()
00141 6CCI
Oll9lHCI
010604 middot middot
Oll083
OOJ11JCAt
002059141
001411CIH
-000412
000293(81
000187
001159 (3)
ooo6a4cct
o00792CCI
OOl40dCAt
000006
-000076
OOlltl6fA)
o o0 5 64CCt
-034570
-o1middot o3o
-1659 90
Ol51lO
-046190CAt
Ol5450CA)
-005l60
002940
-030060(()
-0 43l50CCt
0 33650
-019080
00004
bullJ00741CAI
-000069
000466
0001 51CAt
000947CAt
O OOI)70CAt OOl750CA J
000175CCI
000671CAI
OOll5) f A t 002207CBI
0004b7CA I
- 007 1 70
001760
-001410CAt
- 0 16200CAt
-004210
- 1 207 70
705
96 0
542
466 805
460
614
921
832
9 61
996
424
550
9 15
56 1 651
884
-006114
019 037
03504lCA)
oo409Q
-141644((
491 -0 17664
607
871
lell
084
101
072
103
108 11IllOQfl)-- l
-037880fCJ OOl6l4CAJ -040l70fCI 0()
c
I
I
bull 9
9t
middot middot middotbullmiddot bull lraquo
TABLE 2 REGRE SION RESUlTS ICONTt
bull
EO I NO tHl I filER- f4A_K T AOVER- sbulluRE ttSQ flA Jl- iLAS T NR CODE cos CEPT SHARE JISING AOVfR bull (JTY SHR CiJCF
110Cl I f amp Itt 15J tbt bull 1c u caa
llA 2702 ll o079 6 81 Bt -019110 S6s -OlltO 111
a 004679 ooooo1 t69
( oOl26o -015160 001560 -022690 ltJr()I
ldA 270 3 19 0065lt -OIt2l60 220 -Ol57a2 11 8 e o 0 71t6IA t ol08 75tAt ( OOl983CiU -060llOC() OOl7191A t -O lltdOC A) middot 89
lqA 2704 10 005082 -Oil060 22b -l)i1977 lelO 6 -000080 oo2772C4 t 910 c 000907 -011 200 Oe04105CA t -046960 995
20A 2d06 1 II
O OlltStiAt -001120 837 00014 -))th O01380amp At -oooo 11 821 ( OOl526CAt -026700181 0002 5UAt 000530 969
2lA 2807 28 OOt9lt9CAt -00891 0 516 -0 03962 225I8 002095CCt OOOl lliAt 690 ( o02965CA t -056650((1 OOOt50CAJ -003l70ttl 911I
I 22A 2808 15 ot3l06 119730 655 1130) 090
0 - 007526CC t Oe003961At 835- middot-
c I oo7523 -067210 001l 2tA t -0091t80CCt 951
2lA 2809 21 Oe2067lCAt -1218101At 71 2 -10095((1 121 ttlIaOQ8 00797418) OOOI8lltt 686 Ill( 0092lU A t -Olltl70Ct t 000605CAt -OOit8701At 951 _
---- -
32 009506CAt
0 bull 06 981 I A J 006901 cA bull
00 24A 2810 -010960 5t 5 -OOltSO 7 56
8 -000011 s11 c -5811701At oooaoatAJ oot520IAI 888
-- 37 --- middot - -- -1 ) l lbull t5JJ r-l bullhmiddot1 --- - --
l5A
26A
27A
c
2RA
c
2 9A
31A
llA
24
13
597
Ci)55
14
l 6h
RSQ ELASTI- (lASTCOE f-
TABLE 2
EQ I NO NR CODE IU
REGRESSION RESULTS CCONTa
NR INTER- MARKET
(10S CEPT SHARE AOVER- SHARfcent
TIS lNG AOVER
) CITY StiR
c It) bull 5 co a I 1) bull 0 bull 9 c l 0 bull( 2 1
2815 16 003019 -009990 275 -o069bmiddotJ
6 001446 000224 282 c 002178 -054630 o02221CCI -0075 00 646
2901 29 00021t2 CAI 0)0520 592 0007 6 1 068
0 00105C8 t 000012141 60)g( OOOlJ51AJ -00126018) ooou271At -000u80CA) 974
36oa 1 3 003296(4) -oca29o -oo-q94
001755 000203 lt91t 002499((1 -04l820CCI OOl374CAI 0068)0 946
3612 0033281CI 011520 195 01684) oso
00 30 42C OI 00012 2 424 oo579ltAt -2 7l7801At 002 100IAI -003220 819
8 3617 btel 011789 081002259 009530
001448 oOOl OitCCI 732 OOl7 22CAI -0 46160CAI 0004961AI -OOll40181 989c
30A 3808 14 0021t 3 6fAI -000110 628 o023291C) -005 6 middot- _ middot- 0bull 02 04 1 IC I ooooo9 637 ( 001584((1 -0161)1)0 00021 91AI -0 00060
390) 14 002790CCt 006590 h8 t 0 0938 4 070
B OOl89bCCJ 000144 74
0021961(1 -OItlOlO OOl323CAI -004800 915
B
middot--- - middot 4 - middot- - dfll3904 12 004563181 0)8300 879 0428651(1 089 OQIll004 277CA I ooozstcea 919
c 0041)3181 -0-11250 0011881(1 -006270 972 0Ill
- J - bull bull bull bullbullbull middot - - middot - middot gt middotbull L)ofbullflbull middot middotfJltl -tn- middotmiddotmiddotmiddoto +f middot t middot bull
c
is
positive
2
Page 0 determineThe consumer goodproducer good split is sometimes difficult to
Sollle support for the split used in this paper is a comparison of goodness - of - fit
distributions of a
industries
set
advertising are
that industry
a sales increase of
the
highest aD10U3 those
measures for the two groups of industries Table 3 contains
measures for the 32- consumer goods industries and the 103 producer goods
2 Median R is 5 7 for the consumer goods set and 40 for the producer goods
Estimates - of y the elasticity of demand with respect to
given in column of Table 2 The hypothesis that y bull 0 with
alternative was applied for each ittdustry As shown in the
could not be rtjected for 24 of the 32 industries Of the
and 10 are positive For the remalnifg eight three are
and five are significantly positive
Of the threertegative elasetcities the largest (in absolute value)
cosmetics (code 2809) With all elasticity of -1 01 a one
industry advertising would gwerate a decrease in industry sales
more than one percetu
At the other extrema is toys and games (code 39 04) For
a one percent ittcrease in industry advertising would lead to
four-tenths of a pampreentmiddot Though not significantly different from zero
elasticity estimatbull for pt-bpuetary drugs (code 2808)
which are
Both drug industries have
to sales rafios (211 pe1 cent for 28 08 and 13 4 per cent
other hand flaIJo ing ext1ac s (mainly soft-drink SYUlS) also
is for
percent increase in
of slightly
a middottwo - tailed
table that hypothesis
24 14 are negative
significantly negative
9
the codetics and p-ropriacuy very high advertising
for 28 09) bull On the
has a high ind UY
advenisiltg t() sales ratio ( 8 0 per cent) bull but its estimated elasticity is very
small and not signifieantly different uom zero
middot It doe$ not seem unusual to find many elasticities near zero with a few
negative and a few positive If aggregate consumption is insensitive to aggregate
advertising as seems likely ld then what one industry gains must be lost by
others
Page 21
elasticity of demand does
(10) we can
if t 0 then
Finally a determination concerning the industry
not fix the individual firms elasticity Going back to equation
write y bull (1 - ) a + z y Since a gt 0 it follows thatii i i
Y gt 0 If Ylt 0 however y s sign is indeterminateij ij
Turning next to the coefficient of market share 20 of them are negative and
12 are positive with three of each sign being significantly different from zero
The largest positive value which is significant is for pet food (code 2010) and
a close second is mens amp boys suits and coats (code 23 01) The largest signifi
cant negative value is for cosmetics (code 28 09)
In the model some relations among y a 0 and 61 are implied If middota is
negative (1- y) is greater than one Since 61Y bull (1- a) (1 - ) 61Y is
pCsitive but it will be less than one if 0 is large enough Seventeen of the A
ys are negative in each of the seventeen industries 61 is also negative and les
1 (algebraicly) than y so 61Y is greater than one his is not evidence that the
degree of cooperation is low or zero in any industry of course A high o togethegt
with a negative y which is small (in absolute terms) relative to a can give a
value of 61Y greater than one On the other hand if 61Y were less than one
o would have to be greater than zero
For y gt 0 two conditions may hold If y lt a it follows that 61 lt 0 since
(y - a ) lt 0 It also follows that 61Y lt 0 regardless of the value of o bull
In addition the relation is an if-and-only-if one ie if 61Y is negative
then y lt a Three industries have negative values for 61y dairy products
except milk (code 20 04) synthetic fibers (code 2806) and photographic equipment
(code 3808) In all three cases y is quite small so a does not have to be very
large to give a negative value to al
If y gt a the model shows that 61-- gt 0 whatever the value of 0 bull There are
12 cases where both y and 61 are positive with 61 ranging from near zero
(grain milling code 20 12) to 1 20 ( proprietary drugs code 2808) Several
coeffiei-en_rs
--
15
o f to zero
Tab le 3 - Consumer Goods
Indus tr ies
2Distributions of R and Producer Goods
Page 20a
Consumer Producer
Goods Goods
0 - 9 9 0 3
10 - 19 9 0 1 1
20 - 29 9 3 20
30 - 39 9 3 17
40 - 49 9 5 16
50 - 59 9 8
69 9 6 Ii 1060
--f 670 79 9 2
80 - 89 9 4 4
90 - 100 1 1
Totals 32 10 3
Notes
2is used R i s cal-
culated as F F + ((N - K- 1 ) K ] 1 b_ F is the F
statistic Lor the hypothesis that all o f the non - intercept terms are simul taneously equal
1
2
--
-- Since generalized least squ-area regress ion
Page 12
o f these cases have values o f 6 1 Y rea ter than 0 8 (pet food (20 10) bread amp cakes
(20 14 ) miscellaneous processed foods (20 29 ) men s and boys suit s and coats
(23 01 ) p rop riet ary drugs (28 08 ) elect ric housewares (36 1 2) radio and
IV sets 17 and games and toys (39 For be than(36 ) 04)) 6 1 Y to greater 0 8
both o and cry mus t b e less than 0 2 That is in these industries the results
shown in Tab le 2 would hold only with very strong dominance of marke t demand
effects over brand swit ching e ffects and a very low degree o f cooperation
Using the concentration rat ios given in Table 4 the average concentration
ratio for the eight indust ries is 38 1 For the remaining 24 industries it is
43 2 This result would lend some support t o the proposition that concentrat ion
and cooperation are related if the group of eight industries are identi fied
as having low degrees of cooperation
One f inal note on model predict ions S ince both cr and o are assumed to
be non-negative 60 should also be non-negat ive In only one o f the 32 cas es
is 60 negative and it is not signif icant bull
cor pJ
7 1
7 0 4
1 4 7 5 1
)
7 3 z a o a t s 1 1 3 3 5 ( i V Z 8 tJ1 2 1 - 1 1 0 7 3 5 c middot v 2 8 1 0 3 2 -0 0 1 4 3 0 C 1 V 9 8 3 2 8 1 5 1 6 - 0 0 6 t1 6 9 C _ V 1 11 0 2 9 0 1 29 0 00 7 6 1 C l k V 7 3 1 3 6 08 1 3 - o n 4 9 9 4 3 3 3 6 1 2 2 4 0 1 6 4 3
3 1 7 1 3 0 1 1 7 d 7 8 3 3 8 0 8 0 0 2 3 i 2 3 0 6
P 4
3 9 0 3 1 4
1 2
0 09 3 8 4 39 1)1 0 4 l 8 6 5
T A LE 4 e I NCU S T R Y C A T E G O R Y
F TC NOM e tOOE ass
C 1 I c 2 )
2 0 0 1 1 6 2 0 0r 1 7 zo o l Z 2 0 0 7 1 8
z o o 2 6
z o t a 1 6
2 1) 1 2 1 5 2 0 1 4 1 1 2 0 1 a 1 9 2 0 26 1 2 2 0 2 7 1 0 2 0 29 - 3 5
2 3 0 1 10
2 3 02 z o
2 3 03 1 9
2 5 5 1 2 4 Z 1 0 Z 1 1 2 0 3 1 9 2 7 4 1 0 z a o 1 3
2 A 0 7 z s
E LA S T I -C I T Y
c 3 )
- 0 09 7 2 7
0 00 449
- 0 0 2 8 9 3
- 0 3 2 7 70 - 0 OZ il I q
o 4 u t s a 0 0 1 3 7 3
o 1 4 o o-0 0 5 2 4o -0 1 94 0 1
-C l 6 l l 4 0 1 JJ 3 7 0 3 t 0 4 2 0 0 4 0 9
- 0 4 1 8 44 - O l 7 t ) 4 -ll 7 1 4 0 b -0 3 5 7 8 Z
- 0 1 9 7 7 0 1)0 3 3 4
-o n 9 2
IJ A T A
CONV
ON-C ONV
( 4 )
t 1NV
C 1t-V C NV c ) v C - JNV C rlNV c n v c a v C ON V ( i) l V
C CtV
C t1tV C IJNV
C 1 V
c 1 v
Pa ge 2La
HEQ F t Noe x
c 5 )
6 2q
6 89 1 48
6 1 4 ) q 3
1 9 7 1 9
q z t s 5 a e z
l l 9 1
4 o 7 1 2 0
c oqt 1 Z
4 4 )-q 1 z
5 5 9 1 0 64 1 3 3
n 3
Inter industry analysi
Page 23
Equation ( 3 6 ) may be used to fo rmulate a rough test o f the cooperat ion
vs e f ficiency controversy I f we rest ricted ourselves to some subset o f
indus tries where the relat ion be tween y and a i s the same we would be lookingbull
at a sample o f indus tries for which the dependence o f V on H and o is a
quadratic wi th no squared terms
I f we now sub s titute (26) into ( 36 ) we get
(38) v bull a + (y - a) R + (y - a) o (H) - (y - a) H o (H) + u
If it is only d if ferences in marke t shares that affect advertising inten-
s i ty levels i f concentration does not af fect cooperation and i f o bull 0
( 38 ) reduces t o
( 39 ) V bull a + (y - a) H + U
Let this b e the null hyp o thes is and note that V is a linear func-
tion o f H If y gt a and the null hyp othesis holds the expected relat ion
between V and H is as shown by the pos itively sloped straight line in Figure
6 b etween points A and C
In this context the alterna ive hypothesis the degree o f
cooperation ( o) is a positive function o f concentrat ion
is that
(H) I f the func-
tion is as described in (26) and Figure 4 the relation between V and H
would be as shown by the curved line connec t ing po ints A and C in Figure 6 One way to test the null hypothe s is against this al ternative is to test fo r
the hypothesized curvature in the function relating V and H
Fi gure
R
6
v
ID c
H 1 0
I
i I I
sfigned
clasJ s
in def ini g
effec t is the
I
Page middot2lf
One charac teristic of produc ts which may be useful
classes wi th respec t to significanc e of the brand-switching
conveniencenon-convenienc e distinc tion developed by Porter ( 1 97 )
Using his identif ication of IRS indus tries as a s tar t ing point
each of the 32 industries used in this s tudy to one of the two
The ass ignments are not ed in Tab le 4
Using the conveniencenon-convenienc e dis t inc tion the
divided into two sub-samples
sample was
Four regressions were then run bull ene for e ch
sub-sample one for the to tal sample and one for the total sample with
a dummy variable for convenience goods Each regression was run wi th LB s
and wi th indus tries as ob servations The results are given in Tab le 5
For the LB level equations ( 1A - 1D) equation(37) was used ome
rearranging of terms gives v bull y - ( 1-o ) (y - a) ( 1-z ) + ui i i
bull y - 6 1 ( 1 - z ) + ui When observations are pooled across indus trif s yi
is an industry variable In these regressions the values o f y es t imated
in the first stage were used Since industries have been grouped so as
to reduce differences in (y - a) that term is t reated as a cons tant
within each group Finally H the Herfindahl index has been subst ituted
for o the degree of cooperation The resultant equation is linear
y ( 1 - z ) and H x ( l - z ) This is the equation for whi ch results arei i
reported in Table 5 In all regress ions all variables were weighted by the
square roo t of sales
in
raquo
- 34
3 1
i l T 4 8LE S R E G E S S I ON R E SUL T S i X T f N S I ON S T O T Y PE Of
AND L E V E L OF C ONC E kT R A T I O
E O O tt S
tP bull T Y P E NR C ONV O B S tO N-CONV
E L A S T I shy
T I C J T Y
O E - t-41( T SliA R E
H( fl F
1 1 - K O x middot K O X E R F
- H S a 1 1 - H S t 1 8 1 1 9 1
R S O
1 1 0 1
q z
t - bull 9 f 2 1
Z t P D l 6 fC ONV - 1 9 1 s s c 2 1
e - l O I e bull u q a
2 3 1 0 1 1 Nl N- t O V 5 7 1 ( 1 2 19 1 6 1 - J l S 2 5 4
l l l l t z q o a 1 - J o v z I
l ( I NO l l A l l ol u l t o o c bull J u l l
d S
c 3 4 t t r a l
bull Z D 1 0 l l 4 l l 2 3 -gt 2 l l 2 0 - 4 lS 5 4 l C 4 1 o e 6 t l b b 5 1 1 0 1 1 1 c - 1 9 5 1 t d S 2 1 tV
OQ10 T-R A T I O S A E G I VE N l P A RE N T H E S E S l E V E L S f S I G 1 F l C ANC E A - 99 5 B - q 9 t - s NS J UC E NO I Nl E R C ( P T I S U (j f middot) I N f Hf t OUA TIOI S 6 E f 3R E A O J U S T H E N T F OR 1-t E T E R O SIC E O A S T f C I T V pIraquo
1
middot
I
R SO I S C A L C UL A T E D A S f I f bull 1 1 bull 1lt 1 K t t- E E N I S THf U A1 B E R O f fJe S E fi V A T I O S F t S HtE F - S T A T I S
I N DE P E ND E N T y II
T I C F O T H E U Y P O HIE S I S f A l l C JE F F I C H N T A R E QU A L HI l f RO A f40 K I S T H f UUM 8 E R OF Q ( A fL f S
I l l
1 A
1 6
t C
f 1 r R I l l
1 2 1
L Jt
L B
l q
L 1
on s T Y P E bull 2 )
P OUC T
X
H S I l l
l O l A I bull l 0 7 1
O O b O J
1 2 00 1
1 b 2 1 C I l l Z U
l l S b 1 -1 1 0 1
rlt F X l l - fi C f l
bull 1 I
l l
- ()4 2 3 l f H
1 - 2 7 9 1
K O X
1 1 - ti E R F I
l d l
S4 l i A I 2 l b
1 3 00
K O X -t f R F t S J
I 1 - HER F t
)
60 9
z o bull
0
t l l
3 8 8
1 7 6
5 64
64
N f bull
tl S bull
I l l
I 4 1
t ONV
NON-t ONV
A l l
A ll
c o v NO- ( ONV
1 4 1
1 5 1
- l S b
c - 1 1 2 1
5 8 2 C A I 1 6 7 0 1
middot I l 9 H
l 1 il
1 4 1 6 1
E L A C) T J shy
l t C I T Y
I S I
l b l
- 0 2 0
1 - 1 4 9 1
O Z 3 2 1 A I C b 6 b l
0 0 9 2 -J
1 1 b 4 1
o z t t t l 1 2 2 1 1
oru - Ht tt F
1 gt X bull b I
Page 25 bull
Mean val -as ior relevan t variables for the two sub-samples are
Conv Non-Conv
Adv Sales 4 7 2 3
iit Share 2 7 3 2
Ela cicity - 05 3 02 2
Herf indahl 7 3 1 860
That the dist nct ion between coft9enienee goods and non - convenience
goods mat ters is supported not only by these means but also by the regression
results Equat ions lA and lB are different in almost every respec t Only the
concentrat ionmarket share int eract ion term seems to mat ter for convenience
good s Jus t he opposite holds for non - convenience goods both elasticity1
and market sl e are highly significant but the concentrat ionmarket share L
int erac t ion t=rm is not
Given that one minus market share is used these equ3 tions the coeffici t
of market share is the negative of what is sh in col (6) of Table 5 For the
regress ion for non - convenience LB s reported in equation lB then LB s with larger
market shares have lower advertising to sales ratios
On the question of the impact of the degree of cooperation the story is mixed
For convenienc good s the coefficient is pos itive as equation (31) predicts when
there is a po sitive degree of cooperat ion Fo r the non-convenience goods LB s middot
concentration has no effec t
Aggregat ion t o the indus try level and then rer unning the equations $ives the
results shown in (2a-2d) of Tab le 5 As expected R2 goes up and levels of significanc e
fmiddotJr individual coefficients gJ down In add ition the coeffic ient of H ( l -H) inmiddot
quat ion 2A is no t significant though it is po s i tive Since 2A is the indus try level
I
--
all but one of them and that one is not significant For 18 of the 31
wo uld have on its adve rtising to sal es ratio
cognate of LA the variable H ( l-H) pl ays the same role in 2A as H ( l -z 1) p lays
in LA
Vconomies of scale in advert ising
The literature on advert ing contains a number of definitionbull o f economies
of scale and of findings eone rning its presence and magnitude Before I loo k
a t the data with the hope o f mmenting about scale effect s then I wan t to
define the term as I use it
By economies of scale I mean a decrease in the number o f units of adver-
tising which are needed to generate a unit of sales as advert ising is increased
Given the definition of the firm s own elast icity of demand with respect to
advertising (see the text foll wing equation (1) ) the technical definition
is that rii gt 1 Diseconomies of scale are def ined symmetrically i e
Yu lt 1
ith a cro s s - section oi observations on advertising intensity
(v bull a s1) and onmiddot advertising (a ) we may regress the former on the latter 1 i i
The results of such regress ions for the 32 industries are given in Table 2
lines lB ZB bull bull bull 32B For the 32 eases the regression coefficient is positive it
positive coefficients the coefficient i s signif icant
The que tion now is whether these results ar e evidence of pervasive dis-
economies of scale in advertislng I think no t Given my as sumption that the
data I observe are equilibrium results I may ri ghtly conclud e that almost every-
where high levels of advert ising are as sociated in equilibrium with high ratios
of advertising to sales I may no t conclud e anything however about the
impact that a move by some firm away from its equilibrium level o f advertising
middot- Page 27
Given the assumed relation in equation ( 37 ) there is a corre spondshy
ing equilibrium equation relating v and ai The equation is non - linear i
and I have not found a simp le way to characterize it It is po ssib le to
det ermi ne its deriviative with respect to a b y takin$ tne de riviative of ( 3 7 ) i
When that is done the result is
v - 1(40) i bull sl a a z ( 1-z )i i ira-l
Sinc e all the other terms in (40) are non-negative the sign dep ends only
on the sign of 6 1 which is in turn de pend ent only on the sign of (y - a)
That the signs of the coefficients of a in equations lB 2B bull bull bull 32B arei
not always equal to the signs of the coefficients of z in LA 2A bull bull bull 32A i
is still a mystery to me and something to be explored
bull
Summary
My purpose in writing this pa per wa s t o attempt to integrate explicit micro-
economic mode lling reasoned econome tric specification of error terms and high
quality data to exp lore some questions about adver tising by large firms
al quite important improvements could be made in the first two of those areas
and much data ma ssag ing is yet to be done in the third
Concerning the subst an ce of what I have done I think six things are im-
Page 2 8
Severshy
V and Conclusions
portant
1 Explicit modeling is worth the ef for t if for no other reas on than
that it provides a basis for choosing fr om a varie ty of funct ional
forms
2 The same is true for error specif ica t ion I did in fact look at scatshy
ter diagrams for all 32 industry categor ies and they vir tually all show
the het eroscedast icity which I as sumed and for which I correc ted
3 The predicted relation be tween advertis ing in tensity and market share
shows up clearly in on ly 25 of the case s examined I have ye t to
explore why that may b e the cas e
4 Some evidence concerning the presence and impact of coopera tion was
produced but it is not clear ly pervasive
5 The dis tinction between convenience goods and non-c onvenie nce goods is
unambiguously a good one
6 Virtually no evidence concerning econ omies of scale in adver tis ing can
be gleaned fr om this study given its assumpt ions
More work is called for on many of these issues my study seems to have
raised more questions than it has answer ed This is pr obab lly due to the richshy
ne ss of the data source since I had the oppor tunity to address several issue s
at once
- -- - Q --
Foo tnotes
Manager Line o f Business Program Federal Trade Commission The author depended heavily on several staf f members of the LB Pro gram for statistical and clerical support in particular Joe Cholka George Pascoe and Haro lene Jenkins Helpful comments were given b y Richard Caves Dennis Mueller Michael Porter and F M Scherer When the paper was in its early stages o f development both S teve Garber and Jon Rasmussen former economists in the LB Program provided eery useful cri tiques The views expressed here are my own o f course
1 The basic exp licit model of the role of advertising is Dorfman and S teiner ( 1954 ) and the models in this paper may be seen as ano ther extens ion of their work
2 For a review o f models of oligopoly which treat prices and quantities as relevant variables see Shubik ( 1 959) Extens ions to take advertis ing int o account are fairly s traightforward
3 Dorfman and S teiner dealt with a price-setting firm For a monopolis t inverting the demand relation to show Pbull P (Q A) _instead o f Q- (P A) has no effect on the results o f the analysis For oligopolists who face a sys tem of demand relations inversion does have some impac t on the results The dif ferences are not trivial nonetheless they will no t be explored in this study
4 Shubik ( 1 959) p
5 From a formal mathematical point o f view the function is similar to a1 i Legrangian funct ion of the form - r rr
L bull 11 + A ( j - (i i i+j
where the indicates some fixed level of the profits industry Following the same logic which resul ts in and income constraint as the marginal utility of the of the indus tr in the obj ective of firm i
6 Hall amp Weiss ( 1 967) p 323
7 See Schmallensee ( 1 97 2 ) Ch 2 pp 1 6-47
8 On this poin t see Imel amp Helmberger ( 1 9 7 1 )
9 Long ( 1 970) p 130 - 1 39
1 0 See Schmallensee ( 1 972) Ch 3 pp 48-8 7
rrj i+j ) of the rest of the
the identification profits of the rest
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980
273
274
TABLE 1 INDUSTRY CATEGORIES USED
FTC
COOE OESCRI P T I ON _
2001 MEAT PACKINGt SAUSAGE S ANO OTHER PREPARED
MEAT PROOUCTS
2004 DAIRY PRODUCTS EXC F LUID MILK
2005 CANNED SPECIALTIES
2001 FROZEN SPECIALTIES
OEHYORATFOt AND
I NC LU 0 ING PRE SERVE S
SOUP MIXESbull
SEASO lNGSt ANO
OTHER PET FOOJ
RAIN MILL PROOUCTSt
ANO PREPARED
RELATED PRODUCTS
PRODUCTS
Z008 CANNEOt ORIEOt PICKLED FRUITS
ANO VEGET ABLES JAMS JELLIES DEHYDRATED I G TA3t E
S40CES ANO SALAD ORESSINGS
2010 DOG CAT A O
2012 FlOUR t OTHER RICE
MILLING fiLENOE) FLOUR
2014 B EADt CAKE ANO
20113 CO FECTIONERY
2026 BOTTLED A O CANNED SOFT DRINKS 2021 FLAVORING EXTRACTS AND SYRUPS EC
2029 MISC FOOOS AND KINDRED PROJUCTS EXC
ROASTED COFFEE
2301 MENS AND BOYS SUITS AND COATS
2302 MENS ANO BOYS FURNISHINGS
230 3 kOMENS AND MISSES OUTER EAR
2551 HOUSEHOLD FURNITURE
2702 PERIODICALS
Pqe 19a 1
RELATED 1972
src cooe
2011 f 3
202tX2026
2032
203pound
2033t4t5
2047 2041t4t5
2051 2065 2086
2081
209XZ05
231
232
233
251 272
2703 BOOKS
2704 ISC PVeLISHING
2806 ORGANIC FlEERS
2807 DRUG$ ETHICAL
23234
PT 2 S 3
2808 ORUCSt PROPRIETARY
2809 PERFU ES t COSMETICS ANO OTHER TaiLET
PREPARATIONS
2810 SOAP AND OTHER CLEANING PREPARATIONS 2815
2901
PEST1CIDES AND AGR ICULTURAL
PE OLEUM REF I N ING CHEMICALS EC
3606 HOUSEHOLD COOKING EQUIPMENT
3612 HOUSEHOLD APPLIANC ES NEC INCLUJING ELECshy
rqlC HOUSE ARFS AND FANS ANO SE I G MACHINES
3617 RADIO AND T V RECE I VING SETS 3808 PHOTOG APHIC E UIP amp SUP LfESt EXC
3qo3
PHO TOCOPYING
SPOR TING A D E QU I P M ENT f SUP P L I E S ATHLETTC GOODS NEC
3904 DOLLS GAMESt TOYS AND CHLOqENtS VE ICLES
PT283
2844
234tX2844
287 9
291 3631
3634t6t9
36H
PT386l
3949
15
943
middot
l A6l f 2 R(GRESSION RESULTS
EQ I NO
NR CODE cu 121
lA 2001 A (
2A 2004 r (
3A 2005 a
-R
ClS amp31
16
17
1 z
I TER-CEPT bull t
ooo9t21At 00022311 oooeos uu
OOl49tCAI 0010 2 middot 000960CCI
o049tl1At o04t55C61
MARKET SH ARE bull 5 I
-010670Ctl
-017600181
-001050
-0211AO
-0077 0
AOVER-TIS INti
lbl
000()39 000072
000018 0 0024liAI
-000079
SHARf
4ampgtVf Q bull
111
000140
-000170
RSQ
COl
66 485 862
607 6 3 t ()2
blti6 633
flASTI-
CI T Y PH
-007271(1
000441
-u0288l
clASTSHU (t)Ef
bull 10
tto
-234
210
( OOlltOCCI -126520 000922 181 001790
4A 200 7 18 00590liAI -038670 4e16 -032110 lell u c
OOOS74e 000700 025940
00067018) 0021001AI -02l100CAI
5 26
920
5A 6 c
200 8 26 0 03Sl9CA I 0012 26181 o 024511(
-005540
-0 81420 0001961AI O()Ol90CAI ooll240
441 604 891
-0020 19 274
6A 2010 16 Oe04224CAI o359lOIAI 9oq ll4015dCAI 061 6 -middot 005435CAI 00011 8 1AI 917 c 00535ltAI -009850((1 OOOA51J CAI -001610161 971
7A 2012 001862 000010 351 00187) oo1 a oooq45 ono1o6 439 c 001109 -027050 Oe0067SIAI -OOll60C81 897
- middotmiddot- -
8A 201t ll 000757 Ol4140CCI 814 ott9oocet 095 8 oooqo6 C C J 000069(81 811 c ootl2lampBI -080910CCI Oe009lUBI -OOtlOOCCI 960I
I DIIJIT pound11 1V OTIUTVTfAll-1pound1t- A _ 00 R - qq r - Q
1jl)OQCD
0
-middot middot--
l
_ __ _
I
i
0 00141 c c
91)7
I
T ABLE 2 REG ESSIOH RESULTS CCONTI
EO NR
IND CODE
iR bull
cos
I TEq-
CEPT
t1ARKE T
SHA E
AOVER-
Tl SING
SHARpound-cent
AuVER
RSO ILAST1-
C I T Y ELAST
SH CPtf (lt Cl) bull 11 I 41 I 5I (61 171 I 8 I f9) c 10)
9A 2 0 1 8 19 002397181 -007640 367 -005 246 le46 6 000605 000550CAI 597
-000029 G14180 OOl654CAJ -024li01At
lOA 2026 12 0028901AI -OlZl90 -0114 0 1 1 1 5 759
Ll (
llA 2021
f3 c
12A lOZ9
B
(
13A 2301
B
c
14A 2302
B
c
154 201
n (
16A 2551
8
10
35
10
20
19
24
001897(()
00141 6CCI
Oll9lHCI
010604 middot middot
Oll083
OOJ11JCAt
002059141
001411CIH
-000412
000293(81
000187
001159 (3)
ooo6a4cct
o00792CCI
OOl40dCAt
000006
-000076
OOlltl6fA)
o o0 5 64CCt
-034570
-o1middot o3o
-1659 90
Ol51lO
-046190CAt
Ol5450CA)
-005l60
002940
-030060(()
-0 43l50CCt
0 33650
-019080
00004
bullJ00741CAI
-000069
000466
0001 51CAt
000947CAt
O OOI)70CAt OOl750CA J
000175CCI
000671CAI
OOll5) f A t 002207CBI
0004b7CA I
- 007 1 70
001760
-001410CAt
- 0 16200CAt
-004210
- 1 207 70
705
96 0
542
466 805
460
614
921
832
9 61
996
424
550
9 15
56 1 651
884
-006114
019 037
03504lCA)
oo409Q
-141644((
491 -0 17664
607
871
lell
084
101
072
103
108 11IllOQfl)-- l
-037880fCJ OOl6l4CAJ -040l70fCI 0()
c
I
I
bull 9
9t
middot middot middotbullmiddot bull lraquo
TABLE 2 REGRE SION RESUlTS ICONTt
bull
EO I NO tHl I filER- f4A_K T AOVER- sbulluRE ttSQ flA Jl- iLAS T NR CODE cos CEPT SHARE JISING AOVfR bull (JTY SHR CiJCF
110Cl I f amp Itt 15J tbt bull 1c u caa
llA 2702 ll o079 6 81 Bt -019110 S6s -OlltO 111
a 004679 ooooo1 t69
( oOl26o -015160 001560 -022690 ltJr()I
ldA 270 3 19 0065lt -OIt2l60 220 -Ol57a2 11 8 e o 0 71t6IA t ol08 75tAt ( OOl983CiU -060llOC() OOl7191A t -O lltdOC A) middot 89
lqA 2704 10 005082 -Oil060 22b -l)i1977 lelO 6 -000080 oo2772C4 t 910 c 000907 -011 200 Oe04105CA t -046960 995
20A 2d06 1 II
O OlltStiAt -001120 837 00014 -))th O01380amp At -oooo 11 821 ( OOl526CAt -026700181 0002 5UAt 000530 969
2lA 2807 28 OOt9lt9CAt -00891 0 516 -0 03962 225I8 002095CCt OOOl lliAt 690 ( o02965CA t -056650((1 OOOt50CAJ -003l70ttl 911I
I 22A 2808 15 ot3l06 119730 655 1130) 090
0 - 007526CC t Oe003961At 835- middot-
c I oo7523 -067210 001l 2tA t -0091t80CCt 951
2lA 2809 21 Oe2067lCAt -1218101At 71 2 -10095((1 121 ttlIaOQ8 00797418) OOOI8lltt 686 Ill( 0092lU A t -Olltl70Ct t 000605CAt -OOit8701At 951 _
---- -
32 009506CAt
0 bull 06 981 I A J 006901 cA bull
00 24A 2810 -010960 5t 5 -OOltSO 7 56
8 -000011 s11 c -5811701At oooaoatAJ oot520IAI 888
-- 37 --- middot - -- -1 ) l lbull t5JJ r-l bullhmiddot1 --- - --
l5A
26A
27A
c
2RA
c
2 9A
31A
llA
24
13
597
Ci)55
14
l 6h
RSQ ELASTI- (lASTCOE f-
TABLE 2
EQ I NO NR CODE IU
REGRESSION RESULTS CCONTa
NR INTER- MARKET
(10S CEPT SHARE AOVER- SHARfcent
TIS lNG AOVER
) CITY StiR
c It) bull 5 co a I 1) bull 0 bull 9 c l 0 bull( 2 1
2815 16 003019 -009990 275 -o069bmiddotJ
6 001446 000224 282 c 002178 -054630 o02221CCI -0075 00 646
2901 29 00021t2 CAI 0)0520 592 0007 6 1 068
0 00105C8 t 000012141 60)g( OOOlJ51AJ -00126018) ooou271At -000u80CA) 974
36oa 1 3 003296(4) -oca29o -oo-q94
001755 000203 lt91t 002499((1 -04l820CCI OOl374CAI 0068)0 946
3612 0033281CI 011520 195 01684) oso
00 30 42C OI 00012 2 424 oo579ltAt -2 7l7801At 002 100IAI -003220 819
8 3617 btel 011789 081002259 009530
001448 oOOl OitCCI 732 OOl7 22CAI -0 46160CAI 0004961AI -OOll40181 989c
30A 3808 14 0021t 3 6fAI -000110 628 o023291C) -005 6 middot- _ middot- 0bull 02 04 1 IC I ooooo9 637 ( 001584((1 -0161)1)0 00021 91AI -0 00060
390) 14 002790CCt 006590 h8 t 0 0938 4 070
B OOl89bCCJ 000144 74
0021961(1 -OItlOlO OOl323CAI -004800 915
B
middot--- - middot 4 - middot- - dfll3904 12 004563181 0)8300 879 0428651(1 089 OQIll004 277CA I ooozstcea 919
c 0041)3181 -0-11250 0011881(1 -006270 972 0Ill
- J - bull bull bull bullbullbull middot - - middot - middot gt middotbull L)ofbullflbull middot middotfJltl -tn- middotmiddotmiddotmiddoto +f middot t middot bull
c
is
positive
2
Page 0 determineThe consumer goodproducer good split is sometimes difficult to
Sollle support for the split used in this paper is a comparison of goodness - of - fit
distributions of a
industries
set
advertising are
that industry
a sales increase of
the
highest aD10U3 those
measures for the two groups of industries Table 3 contains
measures for the 32- consumer goods industries and the 103 producer goods
2 Median R is 5 7 for the consumer goods set and 40 for the producer goods
Estimates - of y the elasticity of demand with respect to
given in column of Table 2 The hypothesis that y bull 0 with
alternative was applied for each ittdustry As shown in the
could not be rtjected for 24 of the 32 industries Of the
and 10 are positive For the remalnifg eight three are
and five are significantly positive
Of the threertegative elasetcities the largest (in absolute value)
cosmetics (code 2809) With all elasticity of -1 01 a one
industry advertising would gwerate a decrease in industry sales
more than one percetu
At the other extrema is toys and games (code 39 04) For
a one percent ittcrease in industry advertising would lead to
four-tenths of a pampreentmiddot Though not significantly different from zero
elasticity estimatbull for pt-bpuetary drugs (code 2808)
which are
Both drug industries have
to sales rafios (211 pe1 cent for 28 08 and 13 4 per cent
other hand flaIJo ing ext1ac s (mainly soft-drink SYUlS) also
is for
percent increase in
of slightly
a middottwo - tailed
table that hypothesis
24 14 are negative
significantly negative
9
the codetics and p-ropriacuy very high advertising
for 28 09) bull On the
has a high ind UY
advenisiltg t() sales ratio ( 8 0 per cent) bull but its estimated elasticity is very
small and not signifieantly different uom zero
middot It doe$ not seem unusual to find many elasticities near zero with a few
negative and a few positive If aggregate consumption is insensitive to aggregate
advertising as seems likely ld then what one industry gains must be lost by
others
Page 21
elasticity of demand does
(10) we can
if t 0 then
Finally a determination concerning the industry
not fix the individual firms elasticity Going back to equation
write y bull (1 - ) a + z y Since a gt 0 it follows thatii i i
Y gt 0 If Ylt 0 however y s sign is indeterminateij ij
Turning next to the coefficient of market share 20 of them are negative and
12 are positive with three of each sign being significantly different from zero
The largest positive value which is significant is for pet food (code 2010) and
a close second is mens amp boys suits and coats (code 23 01) The largest signifi
cant negative value is for cosmetics (code 28 09)
In the model some relations among y a 0 and 61 are implied If middota is
negative (1- y) is greater than one Since 61Y bull (1- a) (1 - ) 61Y is
pCsitive but it will be less than one if 0 is large enough Seventeen of the A
ys are negative in each of the seventeen industries 61 is also negative and les
1 (algebraicly) than y so 61Y is greater than one his is not evidence that the
degree of cooperation is low or zero in any industry of course A high o togethegt
with a negative y which is small (in absolute terms) relative to a can give a
value of 61Y greater than one On the other hand if 61Y were less than one
o would have to be greater than zero
For y gt 0 two conditions may hold If y lt a it follows that 61 lt 0 since
(y - a ) lt 0 It also follows that 61Y lt 0 regardless of the value of o bull
In addition the relation is an if-and-only-if one ie if 61Y is negative
then y lt a Three industries have negative values for 61y dairy products
except milk (code 20 04) synthetic fibers (code 2806) and photographic equipment
(code 3808) In all three cases y is quite small so a does not have to be very
large to give a negative value to al
If y gt a the model shows that 61-- gt 0 whatever the value of 0 bull There are
12 cases where both y and 61 are positive with 61 ranging from near zero
(grain milling code 20 12) to 1 20 ( proprietary drugs code 2808) Several
coeffiei-en_rs
--
15
o f to zero
Tab le 3 - Consumer Goods
Indus tr ies
2Distributions of R and Producer Goods
Page 20a
Consumer Producer
Goods Goods
0 - 9 9 0 3
10 - 19 9 0 1 1
20 - 29 9 3 20
30 - 39 9 3 17
40 - 49 9 5 16
50 - 59 9 8
69 9 6 Ii 1060
--f 670 79 9 2
80 - 89 9 4 4
90 - 100 1 1
Totals 32 10 3
Notes
2is used R i s cal-
culated as F F + ((N - K- 1 ) K ] 1 b_ F is the F
statistic Lor the hypothesis that all o f the non - intercept terms are simul taneously equal
1
2
--
-- Since generalized least squ-area regress ion
Page 12
o f these cases have values o f 6 1 Y rea ter than 0 8 (pet food (20 10) bread amp cakes
(20 14 ) miscellaneous processed foods (20 29 ) men s and boys suit s and coats
(23 01 ) p rop riet ary drugs (28 08 ) elect ric housewares (36 1 2) radio and
IV sets 17 and games and toys (39 For be than(36 ) 04)) 6 1 Y to greater 0 8
both o and cry mus t b e less than 0 2 That is in these industries the results
shown in Tab le 2 would hold only with very strong dominance of marke t demand
effects over brand swit ching e ffects and a very low degree o f cooperation
Using the concentration rat ios given in Table 4 the average concentration
ratio for the eight indust ries is 38 1 For the remaining 24 industries it is
43 2 This result would lend some support t o the proposition that concentrat ion
and cooperation are related if the group of eight industries are identi fied
as having low degrees of cooperation
One f inal note on model predict ions S ince both cr and o are assumed to
be non-negative 60 should also be non-negat ive In only one o f the 32 cas es
is 60 negative and it is not signif icant bull
cor pJ
7 1
7 0 4
1 4 7 5 1
)
7 3 z a o a t s 1 1 3 3 5 ( i V Z 8 tJ1 2 1 - 1 1 0 7 3 5 c middot v 2 8 1 0 3 2 -0 0 1 4 3 0 C 1 V 9 8 3 2 8 1 5 1 6 - 0 0 6 t1 6 9 C _ V 1 11 0 2 9 0 1 29 0 00 7 6 1 C l k V 7 3 1 3 6 08 1 3 - o n 4 9 9 4 3 3 3 6 1 2 2 4 0 1 6 4 3
3 1 7 1 3 0 1 1 7 d 7 8 3 3 8 0 8 0 0 2 3 i 2 3 0 6
P 4
3 9 0 3 1 4
1 2
0 09 3 8 4 39 1)1 0 4 l 8 6 5
T A LE 4 e I NCU S T R Y C A T E G O R Y
F TC NOM e tOOE ass
C 1 I c 2 )
2 0 0 1 1 6 2 0 0r 1 7 zo o l Z 2 0 0 7 1 8
z o o 2 6
z o t a 1 6
2 1) 1 2 1 5 2 0 1 4 1 1 2 0 1 a 1 9 2 0 26 1 2 2 0 2 7 1 0 2 0 29 - 3 5
2 3 0 1 10
2 3 02 z o
2 3 03 1 9
2 5 5 1 2 4 Z 1 0 Z 1 1 2 0 3 1 9 2 7 4 1 0 z a o 1 3
2 A 0 7 z s
E LA S T I -C I T Y
c 3 )
- 0 09 7 2 7
0 00 449
- 0 0 2 8 9 3
- 0 3 2 7 70 - 0 OZ il I q
o 4 u t s a 0 0 1 3 7 3
o 1 4 o o-0 0 5 2 4o -0 1 94 0 1
-C l 6 l l 4 0 1 JJ 3 7 0 3 t 0 4 2 0 0 4 0 9
- 0 4 1 8 44 - O l 7 t ) 4 -ll 7 1 4 0 b -0 3 5 7 8 Z
- 0 1 9 7 7 0 1)0 3 3 4
-o n 9 2
IJ A T A
CONV
ON-C ONV
( 4 )
t 1NV
C 1t-V C NV c ) v C - JNV C rlNV c n v c a v C ON V ( i) l V
C CtV
C t1tV C IJNV
C 1 V
c 1 v
Pa ge 2La
HEQ F t Noe x
c 5 )
6 2q
6 89 1 48
6 1 4 ) q 3
1 9 7 1 9
q z t s 5 a e z
l l 9 1
4 o 7 1 2 0
c oqt 1 Z
4 4 )-q 1 z
5 5 9 1 0 64 1 3 3
n 3
Inter industry analysi
Page 23
Equation ( 3 6 ) may be used to fo rmulate a rough test o f the cooperat ion
vs e f ficiency controversy I f we rest ricted ourselves to some subset o f
indus tries where the relat ion be tween y and a i s the same we would be lookingbull
at a sample o f indus tries for which the dependence o f V on H and o is a
quadratic wi th no squared terms
I f we now sub s titute (26) into ( 36 ) we get
(38) v bull a + (y - a) R + (y - a) o (H) - (y - a) H o (H) + u
If it is only d if ferences in marke t shares that affect advertising inten-
s i ty levels i f concentration does not af fect cooperation and i f o bull 0
( 38 ) reduces t o
( 39 ) V bull a + (y - a) H + U
Let this b e the null hyp o thes is and note that V is a linear func-
tion o f H If y gt a and the null hyp othesis holds the expected relat ion
between V and H is as shown by the pos itively sloped straight line in Figure
6 b etween points A and C
In this context the alterna ive hypothesis the degree o f
cooperation ( o) is a positive function o f concentrat ion
is that
(H) I f the func-
tion is as described in (26) and Figure 4 the relation between V and H
would be as shown by the curved line connec t ing po ints A and C in Figure 6 One way to test the null hypothe s is against this al ternative is to test fo r
the hypothesized curvature in the function relating V and H
Fi gure
R
6
v
ID c
H 1 0
I
i I I
sfigned
clasJ s
in def ini g
effec t is the
I
Page middot2lf
One charac teristic of produc ts which may be useful
classes wi th respec t to significanc e of the brand-switching
conveniencenon-convenienc e distinc tion developed by Porter ( 1 97 )
Using his identif ication of IRS indus tries as a s tar t ing point
each of the 32 industries used in this s tudy to one of the two
The ass ignments are not ed in Tab le 4
Using the conveniencenon-convenienc e dis t inc tion the
divided into two sub-samples
sample was
Four regressions were then run bull ene for e ch
sub-sample one for the to tal sample and one for the total sample with
a dummy variable for convenience goods Each regression was run wi th LB s
and wi th indus tries as ob servations The results are given in Tab le 5
For the LB level equations ( 1A - 1D) equation(37) was used ome
rearranging of terms gives v bull y - ( 1-o ) (y - a) ( 1-z ) + ui i i
bull y - 6 1 ( 1 - z ) + ui When observations are pooled across indus trif s yi
is an industry variable In these regressions the values o f y es t imated
in the first stage were used Since industries have been grouped so as
to reduce differences in (y - a) that term is t reated as a cons tant
within each group Finally H the Herfindahl index has been subst ituted
for o the degree of cooperation The resultant equation is linear
y ( 1 - z ) and H x ( l - z ) This is the equation for whi ch results arei i
reported in Table 5 In all regress ions all variables were weighted by the
square roo t of sales
in
raquo
- 34
3 1
i l T 4 8LE S R E G E S S I ON R E SUL T S i X T f N S I ON S T O T Y PE Of
AND L E V E L OF C ONC E kT R A T I O
E O O tt S
tP bull T Y P E NR C ONV O B S tO N-CONV
E L A S T I shy
T I C J T Y
O E - t-41( T SliA R E
H( fl F
1 1 - K O x middot K O X E R F
- H S a 1 1 - H S t 1 8 1 1 9 1
R S O
1 1 0 1
q z
t - bull 9 f 2 1
Z t P D l 6 fC ONV - 1 9 1 s s c 2 1
e - l O I e bull u q a
2 3 1 0 1 1 Nl N- t O V 5 7 1 ( 1 2 19 1 6 1 - J l S 2 5 4
l l l l t z q o a 1 - J o v z I
l ( I NO l l A l l ol u l t o o c bull J u l l
d S
c 3 4 t t r a l
bull Z D 1 0 l l 4 l l 2 3 -gt 2 l l 2 0 - 4 lS 5 4 l C 4 1 o e 6 t l b b 5 1 1 0 1 1 1 c - 1 9 5 1 t d S 2 1 tV
OQ10 T-R A T I O S A E G I VE N l P A RE N T H E S E S l E V E L S f S I G 1 F l C ANC E A - 99 5 B - q 9 t - s NS J UC E NO I Nl E R C ( P T I S U (j f middot) I N f Hf t OUA TIOI S 6 E f 3R E A O J U S T H E N T F OR 1-t E T E R O SIC E O A S T f C I T V pIraquo
1
middot
I
R SO I S C A L C UL A T E D A S f I f bull 1 1 bull 1lt 1 K t t- E E N I S THf U A1 B E R O f fJe S E fi V A T I O S F t S HtE F - S T A T I S
I N DE P E ND E N T y II
T I C F O T H E U Y P O HIE S I S f A l l C JE F F I C H N T A R E QU A L HI l f RO A f40 K I S T H f UUM 8 E R OF Q ( A fL f S
I l l
1 A
1 6
t C
f 1 r R I l l
1 2 1
L Jt
L B
l q
L 1
on s T Y P E bull 2 )
P OUC T
X
H S I l l
l O l A I bull l 0 7 1
O O b O J
1 2 00 1
1 b 2 1 C I l l Z U
l l S b 1 -1 1 0 1
rlt F X l l - fi C f l
bull 1 I
l l
- ()4 2 3 l f H
1 - 2 7 9 1
K O X
1 1 - ti E R F I
l d l
S4 l i A I 2 l b
1 3 00
K O X -t f R F t S J
I 1 - HER F t
)
60 9
z o bull
0
t l l
3 8 8
1 7 6
5 64
64
N f bull
tl S bull
I l l
I 4 1
t ONV
NON-t ONV
A l l
A ll
c o v NO- ( ONV
1 4 1
1 5 1
- l S b
c - 1 1 2 1
5 8 2 C A I 1 6 7 0 1
middot I l 9 H
l 1 il
1 4 1 6 1
E L A C) T J shy
l t C I T Y
I S I
l b l
- 0 2 0
1 - 1 4 9 1
O Z 3 2 1 A I C b 6 b l
0 0 9 2 -J
1 1 b 4 1
o z t t t l 1 2 2 1 1
oru - Ht tt F
1 gt X bull b I
Page 25 bull
Mean val -as ior relevan t variables for the two sub-samples are
Conv Non-Conv
Adv Sales 4 7 2 3
iit Share 2 7 3 2
Ela cicity - 05 3 02 2
Herf indahl 7 3 1 860
That the dist nct ion between coft9enienee goods and non - convenience
goods mat ters is supported not only by these means but also by the regression
results Equat ions lA and lB are different in almost every respec t Only the
concentrat ionmarket share int eract ion term seems to mat ter for convenience
good s Jus t he opposite holds for non - convenience goods both elasticity1
and market sl e are highly significant but the concentrat ionmarket share L
int erac t ion t=rm is not
Given that one minus market share is used these equ3 tions the coeffici t
of market share is the negative of what is sh in col (6) of Table 5 For the
regress ion for non - convenience LB s reported in equation lB then LB s with larger
market shares have lower advertising to sales ratios
On the question of the impact of the degree of cooperation the story is mixed
For convenienc good s the coefficient is pos itive as equation (31) predicts when
there is a po sitive degree of cooperat ion Fo r the non-convenience goods LB s middot
concentration has no effec t
Aggregat ion t o the indus try level and then rer unning the equations $ives the
results shown in (2a-2d) of Tab le 5 As expected R2 goes up and levels of significanc e
fmiddotJr individual coefficients gJ down In add ition the coeffic ient of H ( l -H) inmiddot
quat ion 2A is no t significant though it is po s i tive Since 2A is the indus try level
I
--
all but one of them and that one is not significant For 18 of the 31
wo uld have on its adve rtising to sal es ratio
cognate of LA the variable H ( l-H) pl ays the same role in 2A as H ( l -z 1) p lays
in LA
Vconomies of scale in advert ising
The literature on advert ing contains a number of definitionbull o f economies
of scale and of findings eone rning its presence and magnitude Before I loo k
a t the data with the hope o f mmenting about scale effect s then I wan t to
define the term as I use it
By economies of scale I mean a decrease in the number o f units of adver-
tising which are needed to generate a unit of sales as advert ising is increased
Given the definition of the firm s own elast icity of demand with respect to
advertising (see the text foll wing equation (1) ) the technical definition
is that rii gt 1 Diseconomies of scale are def ined symmetrically i e
Yu lt 1
ith a cro s s - section oi observations on advertising intensity
(v bull a s1) and onmiddot advertising (a ) we may regress the former on the latter 1 i i
The results of such regress ions for the 32 industries are given in Table 2
lines lB ZB bull bull bull 32B For the 32 eases the regression coefficient is positive it
positive coefficients the coefficient i s signif icant
The que tion now is whether these results ar e evidence of pervasive dis-
economies of scale in advertislng I think no t Given my as sumption that the
data I observe are equilibrium results I may ri ghtly conclud e that almost every-
where high levels of advert ising are as sociated in equilibrium with high ratios
of advertising to sales I may no t conclud e anything however about the
impact that a move by some firm away from its equilibrium level o f advertising
middot- Page 27
Given the assumed relation in equation ( 37 ) there is a corre spondshy
ing equilibrium equation relating v and ai The equation is non - linear i
and I have not found a simp le way to characterize it It is po ssib le to
det ermi ne its deriviative with respect to a b y takin$ tne de riviative of ( 3 7 ) i
When that is done the result is
v - 1(40) i bull sl a a z ( 1-z )i i ira-l
Sinc e all the other terms in (40) are non-negative the sign dep ends only
on the sign of 6 1 which is in turn de pend ent only on the sign of (y - a)
That the signs of the coefficients of a in equations lB 2B bull bull bull 32B arei
not always equal to the signs of the coefficients of z in LA 2A bull bull bull 32A i
is still a mystery to me and something to be explored
bull
Summary
My purpose in writing this pa per wa s t o attempt to integrate explicit micro-
economic mode lling reasoned econome tric specification of error terms and high
quality data to exp lore some questions about adver tising by large firms
al quite important improvements could be made in the first two of those areas
and much data ma ssag ing is yet to be done in the third
Concerning the subst an ce of what I have done I think six things are im-
Page 2 8
Severshy
V and Conclusions
portant
1 Explicit modeling is worth the ef for t if for no other reas on than
that it provides a basis for choosing fr om a varie ty of funct ional
forms
2 The same is true for error specif ica t ion I did in fact look at scatshy
ter diagrams for all 32 industry categor ies and they vir tually all show
the het eroscedast icity which I as sumed and for which I correc ted
3 The predicted relation be tween advertis ing in tensity and market share
shows up clearly in on ly 25 of the case s examined I have ye t to
explore why that may b e the cas e
4 Some evidence concerning the presence and impact of coopera tion was
produced but it is not clear ly pervasive
5 The dis tinction between convenience goods and non-c onvenie nce goods is
unambiguously a good one
6 Virtually no evidence concerning econ omies of scale in adver tis ing can
be gleaned fr om this study given its assumpt ions
More work is called for on many of these issues my study seems to have
raised more questions than it has answer ed This is pr obab lly due to the richshy
ne ss of the data source since I had the oppor tunity to address several issue s
at once
- -- - Q --
Foo tnotes
Manager Line o f Business Program Federal Trade Commission The author depended heavily on several staf f members of the LB Pro gram for statistical and clerical support in particular Joe Cholka George Pascoe and Haro lene Jenkins Helpful comments were given b y Richard Caves Dennis Mueller Michael Porter and F M Scherer When the paper was in its early stages o f development both S teve Garber and Jon Rasmussen former economists in the LB Program provided eery useful cri tiques The views expressed here are my own o f course
1 The basic exp licit model of the role of advertising is Dorfman and S teiner ( 1954 ) and the models in this paper may be seen as ano ther extens ion of their work
2 For a review o f models of oligopoly which treat prices and quantities as relevant variables see Shubik ( 1 959) Extens ions to take advertis ing int o account are fairly s traightforward
3 Dorfman and S teiner dealt with a price-setting firm For a monopolis t inverting the demand relation to show Pbull P (Q A) _instead o f Q- (P A) has no effect on the results o f the analysis For oligopolists who face a sys tem of demand relations inversion does have some impac t on the results The dif ferences are not trivial nonetheless they will no t be explored in this study
4 Shubik ( 1 959) p
5 From a formal mathematical point o f view the function is similar to a1 i Legrangian funct ion of the form - r rr
L bull 11 + A ( j - (i i i+j
where the indicates some fixed level of the profits industry Following the same logic which resul ts in and income constraint as the marginal utility of the of the indus tr in the obj ective of firm i
6 Hall amp Weiss ( 1 967) p 323
7 See Schmallensee ( 1 97 2 ) Ch 2 pp 1 6-47
8 On this poin t see Imel amp Helmberger ( 1 9 7 1 )
9 Long ( 1 970) p 130 - 1 39
1 0 See Schmallensee ( 1 972) Ch 3 pp 48-8 7
rrj i+j ) of the rest of the
the identification profits of the rest
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980
15
943
middot
l A6l f 2 R(GRESSION RESULTS
EQ I NO
NR CODE cu 121
lA 2001 A (
2A 2004 r (
3A 2005 a
-R
ClS amp31
16
17
1 z
I TER-CEPT bull t
ooo9t21At 00022311 oooeos uu
OOl49tCAI 0010 2 middot 000960CCI
o049tl1At o04t55C61
MARKET SH ARE bull 5 I
-010670Ctl
-017600181
-001050
-0211AO
-0077 0
AOVER-TIS INti
lbl
000()39 000072
000018 0 0024liAI
-000079
SHARf
4ampgtVf Q bull
111
000140
-000170
RSQ
COl
66 485 862
607 6 3 t ()2
blti6 633
flASTI-
CI T Y PH
-007271(1
000441
-u0288l
clASTSHU (t)Ef
bull 10
tto
-234
210
( OOlltOCCI -126520 000922 181 001790
4A 200 7 18 00590liAI -038670 4e16 -032110 lell u c
OOOS74e 000700 025940
00067018) 0021001AI -02l100CAI
5 26
920
5A 6 c
200 8 26 0 03Sl9CA I 0012 26181 o 024511(
-005540
-0 81420 0001961AI O()Ol90CAI ooll240
441 604 891
-0020 19 274
6A 2010 16 Oe04224CAI o359lOIAI 9oq ll4015dCAI 061 6 -middot 005435CAI 00011 8 1AI 917 c 00535ltAI -009850((1 OOOA51J CAI -001610161 971
7A 2012 001862 000010 351 00187) oo1 a oooq45 ono1o6 439 c 001109 -027050 Oe0067SIAI -OOll60C81 897
- middotmiddot- -
8A 201t ll 000757 Ol4140CCI 814 ott9oocet 095 8 oooqo6 C C J 000069(81 811 c ootl2lampBI -080910CCI Oe009lUBI -OOtlOOCCI 960I
I DIIJIT pound11 1V OTIUTVTfAll-1pound1t- A _ 00 R - qq r - Q
1jl)OQCD
0
-middot middot--
l
_ __ _
I
i
0 00141 c c
91)7
I
T ABLE 2 REG ESSIOH RESULTS CCONTI
EO NR
IND CODE
iR bull
cos
I TEq-
CEPT
t1ARKE T
SHA E
AOVER-
Tl SING
SHARpound-cent
AuVER
RSO ILAST1-
C I T Y ELAST
SH CPtf (lt Cl) bull 11 I 41 I 5I (61 171 I 8 I f9) c 10)
9A 2 0 1 8 19 002397181 -007640 367 -005 246 le46 6 000605 000550CAI 597
-000029 G14180 OOl654CAJ -024li01At
lOA 2026 12 0028901AI -OlZl90 -0114 0 1 1 1 5 759
Ll (
llA 2021
f3 c
12A lOZ9
B
(
13A 2301
B
c
14A 2302
B
c
154 201
n (
16A 2551
8
10
35
10
20
19
24
001897(()
00141 6CCI
Oll9lHCI
010604 middot middot
Oll083
OOJ11JCAt
002059141
001411CIH
-000412
000293(81
000187
001159 (3)
ooo6a4cct
o00792CCI
OOl40dCAt
000006
-000076
OOlltl6fA)
o o0 5 64CCt
-034570
-o1middot o3o
-1659 90
Ol51lO
-046190CAt
Ol5450CA)
-005l60
002940
-030060(()
-0 43l50CCt
0 33650
-019080
00004
bullJ00741CAI
-000069
000466
0001 51CAt
000947CAt
O OOI)70CAt OOl750CA J
000175CCI
000671CAI
OOll5) f A t 002207CBI
0004b7CA I
- 007 1 70
001760
-001410CAt
- 0 16200CAt
-004210
- 1 207 70
705
96 0
542
466 805
460
614
921
832
9 61
996
424
550
9 15
56 1 651
884
-006114
019 037
03504lCA)
oo409Q
-141644((
491 -0 17664
607
871
lell
084
101
072
103
108 11IllOQfl)-- l
-037880fCJ OOl6l4CAJ -040l70fCI 0()
c
I
I
bull 9
9t
middot middot middotbullmiddot bull lraquo
TABLE 2 REGRE SION RESUlTS ICONTt
bull
EO I NO tHl I filER- f4A_K T AOVER- sbulluRE ttSQ flA Jl- iLAS T NR CODE cos CEPT SHARE JISING AOVfR bull (JTY SHR CiJCF
110Cl I f amp Itt 15J tbt bull 1c u caa
llA 2702 ll o079 6 81 Bt -019110 S6s -OlltO 111
a 004679 ooooo1 t69
( oOl26o -015160 001560 -022690 ltJr()I
ldA 270 3 19 0065lt -OIt2l60 220 -Ol57a2 11 8 e o 0 71t6IA t ol08 75tAt ( OOl983CiU -060llOC() OOl7191A t -O lltdOC A) middot 89
lqA 2704 10 005082 -Oil060 22b -l)i1977 lelO 6 -000080 oo2772C4 t 910 c 000907 -011 200 Oe04105CA t -046960 995
20A 2d06 1 II
O OlltStiAt -001120 837 00014 -))th O01380amp At -oooo 11 821 ( OOl526CAt -026700181 0002 5UAt 000530 969
2lA 2807 28 OOt9lt9CAt -00891 0 516 -0 03962 225I8 002095CCt OOOl lliAt 690 ( o02965CA t -056650((1 OOOt50CAJ -003l70ttl 911I
I 22A 2808 15 ot3l06 119730 655 1130) 090
0 - 007526CC t Oe003961At 835- middot-
c I oo7523 -067210 001l 2tA t -0091t80CCt 951
2lA 2809 21 Oe2067lCAt -1218101At 71 2 -10095((1 121 ttlIaOQ8 00797418) OOOI8lltt 686 Ill( 0092lU A t -Olltl70Ct t 000605CAt -OOit8701At 951 _
---- -
32 009506CAt
0 bull 06 981 I A J 006901 cA bull
00 24A 2810 -010960 5t 5 -OOltSO 7 56
8 -000011 s11 c -5811701At oooaoatAJ oot520IAI 888
-- 37 --- middot - -- -1 ) l lbull t5JJ r-l bullhmiddot1 --- - --
l5A
26A
27A
c
2RA
c
2 9A
31A
llA
24
13
597
Ci)55
14
l 6h
RSQ ELASTI- (lASTCOE f-
TABLE 2
EQ I NO NR CODE IU
REGRESSION RESULTS CCONTa
NR INTER- MARKET
(10S CEPT SHARE AOVER- SHARfcent
TIS lNG AOVER
) CITY StiR
c It) bull 5 co a I 1) bull 0 bull 9 c l 0 bull( 2 1
2815 16 003019 -009990 275 -o069bmiddotJ
6 001446 000224 282 c 002178 -054630 o02221CCI -0075 00 646
2901 29 00021t2 CAI 0)0520 592 0007 6 1 068
0 00105C8 t 000012141 60)g( OOOlJ51AJ -00126018) ooou271At -000u80CA) 974
36oa 1 3 003296(4) -oca29o -oo-q94
001755 000203 lt91t 002499((1 -04l820CCI OOl374CAI 0068)0 946
3612 0033281CI 011520 195 01684) oso
00 30 42C OI 00012 2 424 oo579ltAt -2 7l7801At 002 100IAI -003220 819
8 3617 btel 011789 081002259 009530
001448 oOOl OitCCI 732 OOl7 22CAI -0 46160CAI 0004961AI -OOll40181 989c
30A 3808 14 0021t 3 6fAI -000110 628 o023291C) -005 6 middot- _ middot- 0bull 02 04 1 IC I ooooo9 637 ( 001584((1 -0161)1)0 00021 91AI -0 00060
390) 14 002790CCt 006590 h8 t 0 0938 4 070
B OOl89bCCJ 000144 74
0021961(1 -OItlOlO OOl323CAI -004800 915
B
middot--- - middot 4 - middot- - dfll3904 12 004563181 0)8300 879 0428651(1 089 OQIll004 277CA I ooozstcea 919
c 0041)3181 -0-11250 0011881(1 -006270 972 0Ill
- J - bull bull bull bullbullbull middot - - middot - middot gt middotbull L)ofbullflbull middot middotfJltl -tn- middotmiddotmiddotmiddoto +f middot t middot bull
c
is
positive
2
Page 0 determineThe consumer goodproducer good split is sometimes difficult to
Sollle support for the split used in this paper is a comparison of goodness - of - fit
distributions of a
industries
set
advertising are
that industry
a sales increase of
the
highest aD10U3 those
measures for the two groups of industries Table 3 contains
measures for the 32- consumer goods industries and the 103 producer goods
2 Median R is 5 7 for the consumer goods set and 40 for the producer goods
Estimates - of y the elasticity of demand with respect to
given in column of Table 2 The hypothesis that y bull 0 with
alternative was applied for each ittdustry As shown in the
could not be rtjected for 24 of the 32 industries Of the
and 10 are positive For the remalnifg eight three are
and five are significantly positive
Of the threertegative elasetcities the largest (in absolute value)
cosmetics (code 2809) With all elasticity of -1 01 a one
industry advertising would gwerate a decrease in industry sales
more than one percetu
At the other extrema is toys and games (code 39 04) For
a one percent ittcrease in industry advertising would lead to
four-tenths of a pampreentmiddot Though not significantly different from zero
elasticity estimatbull for pt-bpuetary drugs (code 2808)
which are
Both drug industries have
to sales rafios (211 pe1 cent for 28 08 and 13 4 per cent
other hand flaIJo ing ext1ac s (mainly soft-drink SYUlS) also
is for
percent increase in
of slightly
a middottwo - tailed
table that hypothesis
24 14 are negative
significantly negative
9
the codetics and p-ropriacuy very high advertising
for 28 09) bull On the
has a high ind UY
advenisiltg t() sales ratio ( 8 0 per cent) bull but its estimated elasticity is very
small and not signifieantly different uom zero
middot It doe$ not seem unusual to find many elasticities near zero with a few
negative and a few positive If aggregate consumption is insensitive to aggregate
advertising as seems likely ld then what one industry gains must be lost by
others
Page 21
elasticity of demand does
(10) we can
if t 0 then
Finally a determination concerning the industry
not fix the individual firms elasticity Going back to equation
write y bull (1 - ) a + z y Since a gt 0 it follows thatii i i
Y gt 0 If Ylt 0 however y s sign is indeterminateij ij
Turning next to the coefficient of market share 20 of them are negative and
12 are positive with three of each sign being significantly different from zero
The largest positive value which is significant is for pet food (code 2010) and
a close second is mens amp boys suits and coats (code 23 01) The largest signifi
cant negative value is for cosmetics (code 28 09)
In the model some relations among y a 0 and 61 are implied If middota is
negative (1- y) is greater than one Since 61Y bull (1- a) (1 - ) 61Y is
pCsitive but it will be less than one if 0 is large enough Seventeen of the A
ys are negative in each of the seventeen industries 61 is also negative and les
1 (algebraicly) than y so 61Y is greater than one his is not evidence that the
degree of cooperation is low or zero in any industry of course A high o togethegt
with a negative y which is small (in absolute terms) relative to a can give a
value of 61Y greater than one On the other hand if 61Y were less than one
o would have to be greater than zero
For y gt 0 two conditions may hold If y lt a it follows that 61 lt 0 since
(y - a ) lt 0 It also follows that 61Y lt 0 regardless of the value of o bull
In addition the relation is an if-and-only-if one ie if 61Y is negative
then y lt a Three industries have negative values for 61y dairy products
except milk (code 20 04) synthetic fibers (code 2806) and photographic equipment
(code 3808) In all three cases y is quite small so a does not have to be very
large to give a negative value to al
If y gt a the model shows that 61-- gt 0 whatever the value of 0 bull There are
12 cases where both y and 61 are positive with 61 ranging from near zero
(grain milling code 20 12) to 1 20 ( proprietary drugs code 2808) Several
coeffiei-en_rs
--
15
o f to zero
Tab le 3 - Consumer Goods
Indus tr ies
2Distributions of R and Producer Goods
Page 20a
Consumer Producer
Goods Goods
0 - 9 9 0 3
10 - 19 9 0 1 1
20 - 29 9 3 20
30 - 39 9 3 17
40 - 49 9 5 16
50 - 59 9 8
69 9 6 Ii 1060
--f 670 79 9 2
80 - 89 9 4 4
90 - 100 1 1
Totals 32 10 3
Notes
2is used R i s cal-
culated as F F + ((N - K- 1 ) K ] 1 b_ F is the F
statistic Lor the hypothesis that all o f the non - intercept terms are simul taneously equal
1
2
--
-- Since generalized least squ-area regress ion
Page 12
o f these cases have values o f 6 1 Y rea ter than 0 8 (pet food (20 10) bread amp cakes
(20 14 ) miscellaneous processed foods (20 29 ) men s and boys suit s and coats
(23 01 ) p rop riet ary drugs (28 08 ) elect ric housewares (36 1 2) radio and
IV sets 17 and games and toys (39 For be than(36 ) 04)) 6 1 Y to greater 0 8
both o and cry mus t b e less than 0 2 That is in these industries the results
shown in Tab le 2 would hold only with very strong dominance of marke t demand
effects over brand swit ching e ffects and a very low degree o f cooperation
Using the concentration rat ios given in Table 4 the average concentration
ratio for the eight indust ries is 38 1 For the remaining 24 industries it is
43 2 This result would lend some support t o the proposition that concentrat ion
and cooperation are related if the group of eight industries are identi fied
as having low degrees of cooperation
One f inal note on model predict ions S ince both cr and o are assumed to
be non-negative 60 should also be non-negat ive In only one o f the 32 cas es
is 60 negative and it is not signif icant bull
cor pJ
7 1
7 0 4
1 4 7 5 1
)
7 3 z a o a t s 1 1 3 3 5 ( i V Z 8 tJ1 2 1 - 1 1 0 7 3 5 c middot v 2 8 1 0 3 2 -0 0 1 4 3 0 C 1 V 9 8 3 2 8 1 5 1 6 - 0 0 6 t1 6 9 C _ V 1 11 0 2 9 0 1 29 0 00 7 6 1 C l k V 7 3 1 3 6 08 1 3 - o n 4 9 9 4 3 3 3 6 1 2 2 4 0 1 6 4 3
3 1 7 1 3 0 1 1 7 d 7 8 3 3 8 0 8 0 0 2 3 i 2 3 0 6
P 4
3 9 0 3 1 4
1 2
0 09 3 8 4 39 1)1 0 4 l 8 6 5
T A LE 4 e I NCU S T R Y C A T E G O R Y
F TC NOM e tOOE ass
C 1 I c 2 )
2 0 0 1 1 6 2 0 0r 1 7 zo o l Z 2 0 0 7 1 8
z o o 2 6
z o t a 1 6
2 1) 1 2 1 5 2 0 1 4 1 1 2 0 1 a 1 9 2 0 26 1 2 2 0 2 7 1 0 2 0 29 - 3 5
2 3 0 1 10
2 3 02 z o
2 3 03 1 9
2 5 5 1 2 4 Z 1 0 Z 1 1 2 0 3 1 9 2 7 4 1 0 z a o 1 3
2 A 0 7 z s
E LA S T I -C I T Y
c 3 )
- 0 09 7 2 7
0 00 449
- 0 0 2 8 9 3
- 0 3 2 7 70 - 0 OZ il I q
o 4 u t s a 0 0 1 3 7 3
o 1 4 o o-0 0 5 2 4o -0 1 94 0 1
-C l 6 l l 4 0 1 JJ 3 7 0 3 t 0 4 2 0 0 4 0 9
- 0 4 1 8 44 - O l 7 t ) 4 -ll 7 1 4 0 b -0 3 5 7 8 Z
- 0 1 9 7 7 0 1)0 3 3 4
-o n 9 2
IJ A T A
CONV
ON-C ONV
( 4 )
t 1NV
C 1t-V C NV c ) v C - JNV C rlNV c n v c a v C ON V ( i) l V
C CtV
C t1tV C IJNV
C 1 V
c 1 v
Pa ge 2La
HEQ F t Noe x
c 5 )
6 2q
6 89 1 48
6 1 4 ) q 3
1 9 7 1 9
q z t s 5 a e z
l l 9 1
4 o 7 1 2 0
c oqt 1 Z
4 4 )-q 1 z
5 5 9 1 0 64 1 3 3
n 3
Inter industry analysi
Page 23
Equation ( 3 6 ) may be used to fo rmulate a rough test o f the cooperat ion
vs e f ficiency controversy I f we rest ricted ourselves to some subset o f
indus tries where the relat ion be tween y and a i s the same we would be lookingbull
at a sample o f indus tries for which the dependence o f V on H and o is a
quadratic wi th no squared terms
I f we now sub s titute (26) into ( 36 ) we get
(38) v bull a + (y - a) R + (y - a) o (H) - (y - a) H o (H) + u
If it is only d if ferences in marke t shares that affect advertising inten-
s i ty levels i f concentration does not af fect cooperation and i f o bull 0
( 38 ) reduces t o
( 39 ) V bull a + (y - a) H + U
Let this b e the null hyp o thes is and note that V is a linear func-
tion o f H If y gt a and the null hyp othesis holds the expected relat ion
between V and H is as shown by the pos itively sloped straight line in Figure
6 b etween points A and C
In this context the alterna ive hypothesis the degree o f
cooperation ( o) is a positive function o f concentrat ion
is that
(H) I f the func-
tion is as described in (26) and Figure 4 the relation between V and H
would be as shown by the curved line connec t ing po ints A and C in Figure 6 One way to test the null hypothe s is against this al ternative is to test fo r
the hypothesized curvature in the function relating V and H
Fi gure
R
6
v
ID c
H 1 0
I
i I I
sfigned
clasJ s
in def ini g
effec t is the
I
Page middot2lf
One charac teristic of produc ts which may be useful
classes wi th respec t to significanc e of the brand-switching
conveniencenon-convenienc e distinc tion developed by Porter ( 1 97 )
Using his identif ication of IRS indus tries as a s tar t ing point
each of the 32 industries used in this s tudy to one of the two
The ass ignments are not ed in Tab le 4
Using the conveniencenon-convenienc e dis t inc tion the
divided into two sub-samples
sample was
Four regressions were then run bull ene for e ch
sub-sample one for the to tal sample and one for the total sample with
a dummy variable for convenience goods Each regression was run wi th LB s
and wi th indus tries as ob servations The results are given in Tab le 5
For the LB level equations ( 1A - 1D) equation(37) was used ome
rearranging of terms gives v bull y - ( 1-o ) (y - a) ( 1-z ) + ui i i
bull y - 6 1 ( 1 - z ) + ui When observations are pooled across indus trif s yi
is an industry variable In these regressions the values o f y es t imated
in the first stage were used Since industries have been grouped so as
to reduce differences in (y - a) that term is t reated as a cons tant
within each group Finally H the Herfindahl index has been subst ituted
for o the degree of cooperation The resultant equation is linear
y ( 1 - z ) and H x ( l - z ) This is the equation for whi ch results arei i
reported in Table 5 In all regress ions all variables were weighted by the
square roo t of sales
in
raquo
- 34
3 1
i l T 4 8LE S R E G E S S I ON R E SUL T S i X T f N S I ON S T O T Y PE Of
AND L E V E L OF C ONC E kT R A T I O
E O O tt S
tP bull T Y P E NR C ONV O B S tO N-CONV
E L A S T I shy
T I C J T Y
O E - t-41( T SliA R E
H( fl F
1 1 - K O x middot K O X E R F
- H S a 1 1 - H S t 1 8 1 1 9 1
R S O
1 1 0 1
q z
t - bull 9 f 2 1
Z t P D l 6 fC ONV - 1 9 1 s s c 2 1
e - l O I e bull u q a
2 3 1 0 1 1 Nl N- t O V 5 7 1 ( 1 2 19 1 6 1 - J l S 2 5 4
l l l l t z q o a 1 - J o v z I
l ( I NO l l A l l ol u l t o o c bull J u l l
d S
c 3 4 t t r a l
bull Z D 1 0 l l 4 l l 2 3 -gt 2 l l 2 0 - 4 lS 5 4 l C 4 1 o e 6 t l b b 5 1 1 0 1 1 1 c - 1 9 5 1 t d S 2 1 tV
OQ10 T-R A T I O S A E G I VE N l P A RE N T H E S E S l E V E L S f S I G 1 F l C ANC E A - 99 5 B - q 9 t - s NS J UC E NO I Nl E R C ( P T I S U (j f middot) I N f Hf t OUA TIOI S 6 E f 3R E A O J U S T H E N T F OR 1-t E T E R O SIC E O A S T f C I T V pIraquo
1
middot
I
R SO I S C A L C UL A T E D A S f I f bull 1 1 bull 1lt 1 K t t- E E N I S THf U A1 B E R O f fJe S E fi V A T I O S F t S HtE F - S T A T I S
I N DE P E ND E N T y II
T I C F O T H E U Y P O HIE S I S f A l l C JE F F I C H N T A R E QU A L HI l f RO A f40 K I S T H f UUM 8 E R OF Q ( A fL f S
I l l
1 A
1 6
t C
f 1 r R I l l
1 2 1
L Jt
L B
l q
L 1
on s T Y P E bull 2 )
P OUC T
X
H S I l l
l O l A I bull l 0 7 1
O O b O J
1 2 00 1
1 b 2 1 C I l l Z U
l l S b 1 -1 1 0 1
rlt F X l l - fi C f l
bull 1 I
l l
- ()4 2 3 l f H
1 - 2 7 9 1
K O X
1 1 - ti E R F I
l d l
S4 l i A I 2 l b
1 3 00
K O X -t f R F t S J
I 1 - HER F t
)
60 9
z o bull
0
t l l
3 8 8
1 7 6
5 64
64
N f bull
tl S bull
I l l
I 4 1
t ONV
NON-t ONV
A l l
A ll
c o v NO- ( ONV
1 4 1
1 5 1
- l S b
c - 1 1 2 1
5 8 2 C A I 1 6 7 0 1
middot I l 9 H
l 1 il
1 4 1 6 1
E L A C) T J shy
l t C I T Y
I S I
l b l
- 0 2 0
1 - 1 4 9 1
O Z 3 2 1 A I C b 6 b l
0 0 9 2 -J
1 1 b 4 1
o z t t t l 1 2 2 1 1
oru - Ht tt F
1 gt X bull b I
Page 25 bull
Mean val -as ior relevan t variables for the two sub-samples are
Conv Non-Conv
Adv Sales 4 7 2 3
iit Share 2 7 3 2
Ela cicity - 05 3 02 2
Herf indahl 7 3 1 860
That the dist nct ion between coft9enienee goods and non - convenience
goods mat ters is supported not only by these means but also by the regression
results Equat ions lA and lB are different in almost every respec t Only the
concentrat ionmarket share int eract ion term seems to mat ter for convenience
good s Jus t he opposite holds for non - convenience goods both elasticity1
and market sl e are highly significant but the concentrat ionmarket share L
int erac t ion t=rm is not
Given that one minus market share is used these equ3 tions the coeffici t
of market share is the negative of what is sh in col (6) of Table 5 For the
regress ion for non - convenience LB s reported in equation lB then LB s with larger
market shares have lower advertising to sales ratios
On the question of the impact of the degree of cooperation the story is mixed
For convenienc good s the coefficient is pos itive as equation (31) predicts when
there is a po sitive degree of cooperat ion Fo r the non-convenience goods LB s middot
concentration has no effec t
Aggregat ion t o the indus try level and then rer unning the equations $ives the
results shown in (2a-2d) of Tab le 5 As expected R2 goes up and levels of significanc e
fmiddotJr individual coefficients gJ down In add ition the coeffic ient of H ( l -H) inmiddot
quat ion 2A is no t significant though it is po s i tive Since 2A is the indus try level
I
--
all but one of them and that one is not significant For 18 of the 31
wo uld have on its adve rtising to sal es ratio
cognate of LA the variable H ( l-H) pl ays the same role in 2A as H ( l -z 1) p lays
in LA
Vconomies of scale in advert ising
The literature on advert ing contains a number of definitionbull o f economies
of scale and of findings eone rning its presence and magnitude Before I loo k
a t the data with the hope o f mmenting about scale effect s then I wan t to
define the term as I use it
By economies of scale I mean a decrease in the number o f units of adver-
tising which are needed to generate a unit of sales as advert ising is increased
Given the definition of the firm s own elast icity of demand with respect to
advertising (see the text foll wing equation (1) ) the technical definition
is that rii gt 1 Diseconomies of scale are def ined symmetrically i e
Yu lt 1
ith a cro s s - section oi observations on advertising intensity
(v bull a s1) and onmiddot advertising (a ) we may regress the former on the latter 1 i i
The results of such regress ions for the 32 industries are given in Table 2
lines lB ZB bull bull bull 32B For the 32 eases the regression coefficient is positive it
positive coefficients the coefficient i s signif icant
The que tion now is whether these results ar e evidence of pervasive dis-
economies of scale in advertislng I think no t Given my as sumption that the
data I observe are equilibrium results I may ri ghtly conclud e that almost every-
where high levels of advert ising are as sociated in equilibrium with high ratios
of advertising to sales I may no t conclud e anything however about the
impact that a move by some firm away from its equilibrium level o f advertising
middot- Page 27
Given the assumed relation in equation ( 37 ) there is a corre spondshy
ing equilibrium equation relating v and ai The equation is non - linear i
and I have not found a simp le way to characterize it It is po ssib le to
det ermi ne its deriviative with respect to a b y takin$ tne de riviative of ( 3 7 ) i
When that is done the result is
v - 1(40) i bull sl a a z ( 1-z )i i ira-l
Sinc e all the other terms in (40) are non-negative the sign dep ends only
on the sign of 6 1 which is in turn de pend ent only on the sign of (y - a)
That the signs of the coefficients of a in equations lB 2B bull bull bull 32B arei
not always equal to the signs of the coefficients of z in LA 2A bull bull bull 32A i
is still a mystery to me and something to be explored
bull
Summary
My purpose in writing this pa per wa s t o attempt to integrate explicit micro-
economic mode lling reasoned econome tric specification of error terms and high
quality data to exp lore some questions about adver tising by large firms
al quite important improvements could be made in the first two of those areas
and much data ma ssag ing is yet to be done in the third
Concerning the subst an ce of what I have done I think six things are im-
Page 2 8
Severshy
V and Conclusions
portant
1 Explicit modeling is worth the ef for t if for no other reas on than
that it provides a basis for choosing fr om a varie ty of funct ional
forms
2 The same is true for error specif ica t ion I did in fact look at scatshy
ter diagrams for all 32 industry categor ies and they vir tually all show
the het eroscedast icity which I as sumed and for which I correc ted
3 The predicted relation be tween advertis ing in tensity and market share
shows up clearly in on ly 25 of the case s examined I have ye t to
explore why that may b e the cas e
4 Some evidence concerning the presence and impact of coopera tion was
produced but it is not clear ly pervasive
5 The dis tinction between convenience goods and non-c onvenie nce goods is
unambiguously a good one
6 Virtually no evidence concerning econ omies of scale in adver tis ing can
be gleaned fr om this study given its assumpt ions
More work is called for on many of these issues my study seems to have
raised more questions than it has answer ed This is pr obab lly due to the richshy
ne ss of the data source since I had the oppor tunity to address several issue s
at once
- -- - Q --
Foo tnotes
Manager Line o f Business Program Federal Trade Commission The author depended heavily on several staf f members of the LB Pro gram for statistical and clerical support in particular Joe Cholka George Pascoe and Haro lene Jenkins Helpful comments were given b y Richard Caves Dennis Mueller Michael Porter and F M Scherer When the paper was in its early stages o f development both S teve Garber and Jon Rasmussen former economists in the LB Program provided eery useful cri tiques The views expressed here are my own o f course
1 The basic exp licit model of the role of advertising is Dorfman and S teiner ( 1954 ) and the models in this paper may be seen as ano ther extens ion of their work
2 For a review o f models of oligopoly which treat prices and quantities as relevant variables see Shubik ( 1 959) Extens ions to take advertis ing int o account are fairly s traightforward
3 Dorfman and S teiner dealt with a price-setting firm For a monopolis t inverting the demand relation to show Pbull P (Q A) _instead o f Q- (P A) has no effect on the results o f the analysis For oligopolists who face a sys tem of demand relations inversion does have some impac t on the results The dif ferences are not trivial nonetheless they will no t be explored in this study
4 Shubik ( 1 959) p
5 From a formal mathematical point o f view the function is similar to a1 i Legrangian funct ion of the form - r rr
L bull 11 + A ( j - (i i i+j
where the indicates some fixed level of the profits industry Following the same logic which resul ts in and income constraint as the marginal utility of the of the indus tr in the obj ective of firm i
6 Hall amp Weiss ( 1 967) p 323
7 See Schmallensee ( 1 97 2 ) Ch 2 pp 1 6-47
8 On this poin t see Imel amp Helmberger ( 1 9 7 1 )
9 Long ( 1 970) p 130 - 1 39
1 0 See Schmallensee ( 1 972) Ch 3 pp 48-8 7
rrj i+j ) of the rest of the
the identification profits of the rest
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980
l
_ __ _
I
i
0 00141 c c
91)7
I
T ABLE 2 REG ESSIOH RESULTS CCONTI
EO NR
IND CODE
iR bull
cos
I TEq-
CEPT
t1ARKE T
SHA E
AOVER-
Tl SING
SHARpound-cent
AuVER
RSO ILAST1-
C I T Y ELAST
SH CPtf (lt Cl) bull 11 I 41 I 5I (61 171 I 8 I f9) c 10)
9A 2 0 1 8 19 002397181 -007640 367 -005 246 le46 6 000605 000550CAI 597
-000029 G14180 OOl654CAJ -024li01At
lOA 2026 12 0028901AI -OlZl90 -0114 0 1 1 1 5 759
Ll (
llA 2021
f3 c
12A lOZ9
B
(
13A 2301
B
c
14A 2302
B
c
154 201
n (
16A 2551
8
10
35
10
20
19
24
001897(()
00141 6CCI
Oll9lHCI
010604 middot middot
Oll083
OOJ11JCAt
002059141
001411CIH
-000412
000293(81
000187
001159 (3)
ooo6a4cct
o00792CCI
OOl40dCAt
000006
-000076
OOlltl6fA)
o o0 5 64CCt
-034570
-o1middot o3o
-1659 90
Ol51lO
-046190CAt
Ol5450CA)
-005l60
002940
-030060(()
-0 43l50CCt
0 33650
-019080
00004
bullJ00741CAI
-000069
000466
0001 51CAt
000947CAt
O OOI)70CAt OOl750CA J
000175CCI
000671CAI
OOll5) f A t 002207CBI
0004b7CA I
- 007 1 70
001760
-001410CAt
- 0 16200CAt
-004210
- 1 207 70
705
96 0
542
466 805
460
614
921
832
9 61
996
424
550
9 15
56 1 651
884
-006114
019 037
03504lCA)
oo409Q
-141644((
491 -0 17664
607
871
lell
084
101
072
103
108 11IllOQfl)-- l
-037880fCJ OOl6l4CAJ -040l70fCI 0()
c
I
I
bull 9
9t
middot middot middotbullmiddot bull lraquo
TABLE 2 REGRE SION RESUlTS ICONTt
bull
EO I NO tHl I filER- f4A_K T AOVER- sbulluRE ttSQ flA Jl- iLAS T NR CODE cos CEPT SHARE JISING AOVfR bull (JTY SHR CiJCF
110Cl I f amp Itt 15J tbt bull 1c u caa
llA 2702 ll o079 6 81 Bt -019110 S6s -OlltO 111
a 004679 ooooo1 t69
( oOl26o -015160 001560 -022690 ltJr()I
ldA 270 3 19 0065lt -OIt2l60 220 -Ol57a2 11 8 e o 0 71t6IA t ol08 75tAt ( OOl983CiU -060llOC() OOl7191A t -O lltdOC A) middot 89
lqA 2704 10 005082 -Oil060 22b -l)i1977 lelO 6 -000080 oo2772C4 t 910 c 000907 -011 200 Oe04105CA t -046960 995
20A 2d06 1 II
O OlltStiAt -001120 837 00014 -))th O01380amp At -oooo 11 821 ( OOl526CAt -026700181 0002 5UAt 000530 969
2lA 2807 28 OOt9lt9CAt -00891 0 516 -0 03962 225I8 002095CCt OOOl lliAt 690 ( o02965CA t -056650((1 OOOt50CAJ -003l70ttl 911I
I 22A 2808 15 ot3l06 119730 655 1130) 090
0 - 007526CC t Oe003961At 835- middot-
c I oo7523 -067210 001l 2tA t -0091t80CCt 951
2lA 2809 21 Oe2067lCAt -1218101At 71 2 -10095((1 121 ttlIaOQ8 00797418) OOOI8lltt 686 Ill( 0092lU A t -Olltl70Ct t 000605CAt -OOit8701At 951 _
---- -
32 009506CAt
0 bull 06 981 I A J 006901 cA bull
00 24A 2810 -010960 5t 5 -OOltSO 7 56
8 -000011 s11 c -5811701At oooaoatAJ oot520IAI 888
-- 37 --- middot - -- -1 ) l lbull t5JJ r-l bullhmiddot1 --- - --
l5A
26A
27A
c
2RA
c
2 9A
31A
llA
24
13
597
Ci)55
14
l 6h
RSQ ELASTI- (lASTCOE f-
TABLE 2
EQ I NO NR CODE IU
REGRESSION RESULTS CCONTa
NR INTER- MARKET
(10S CEPT SHARE AOVER- SHARfcent
TIS lNG AOVER
) CITY StiR
c It) bull 5 co a I 1) bull 0 bull 9 c l 0 bull( 2 1
2815 16 003019 -009990 275 -o069bmiddotJ
6 001446 000224 282 c 002178 -054630 o02221CCI -0075 00 646
2901 29 00021t2 CAI 0)0520 592 0007 6 1 068
0 00105C8 t 000012141 60)g( OOOlJ51AJ -00126018) ooou271At -000u80CA) 974
36oa 1 3 003296(4) -oca29o -oo-q94
001755 000203 lt91t 002499((1 -04l820CCI OOl374CAI 0068)0 946
3612 0033281CI 011520 195 01684) oso
00 30 42C OI 00012 2 424 oo579ltAt -2 7l7801At 002 100IAI -003220 819
8 3617 btel 011789 081002259 009530
001448 oOOl OitCCI 732 OOl7 22CAI -0 46160CAI 0004961AI -OOll40181 989c
30A 3808 14 0021t 3 6fAI -000110 628 o023291C) -005 6 middot- _ middot- 0bull 02 04 1 IC I ooooo9 637 ( 001584((1 -0161)1)0 00021 91AI -0 00060
390) 14 002790CCt 006590 h8 t 0 0938 4 070
B OOl89bCCJ 000144 74
0021961(1 -OItlOlO OOl323CAI -004800 915
B
middot--- - middot 4 - middot- - dfll3904 12 004563181 0)8300 879 0428651(1 089 OQIll004 277CA I ooozstcea 919
c 0041)3181 -0-11250 0011881(1 -006270 972 0Ill
- J - bull bull bull bullbullbull middot - - middot - middot gt middotbull L)ofbullflbull middot middotfJltl -tn- middotmiddotmiddotmiddoto +f middot t middot bull
c
is
positive
2
Page 0 determineThe consumer goodproducer good split is sometimes difficult to
Sollle support for the split used in this paper is a comparison of goodness - of - fit
distributions of a
industries
set
advertising are
that industry
a sales increase of
the
highest aD10U3 those
measures for the two groups of industries Table 3 contains
measures for the 32- consumer goods industries and the 103 producer goods
2 Median R is 5 7 for the consumer goods set and 40 for the producer goods
Estimates - of y the elasticity of demand with respect to
given in column of Table 2 The hypothesis that y bull 0 with
alternative was applied for each ittdustry As shown in the
could not be rtjected for 24 of the 32 industries Of the
and 10 are positive For the remalnifg eight three are
and five are significantly positive
Of the threertegative elasetcities the largest (in absolute value)
cosmetics (code 2809) With all elasticity of -1 01 a one
industry advertising would gwerate a decrease in industry sales
more than one percetu
At the other extrema is toys and games (code 39 04) For
a one percent ittcrease in industry advertising would lead to
four-tenths of a pampreentmiddot Though not significantly different from zero
elasticity estimatbull for pt-bpuetary drugs (code 2808)
which are
Both drug industries have
to sales rafios (211 pe1 cent for 28 08 and 13 4 per cent
other hand flaIJo ing ext1ac s (mainly soft-drink SYUlS) also
is for
percent increase in
of slightly
a middottwo - tailed
table that hypothesis
24 14 are negative
significantly negative
9
the codetics and p-ropriacuy very high advertising
for 28 09) bull On the
has a high ind UY
advenisiltg t() sales ratio ( 8 0 per cent) bull but its estimated elasticity is very
small and not signifieantly different uom zero
middot It doe$ not seem unusual to find many elasticities near zero with a few
negative and a few positive If aggregate consumption is insensitive to aggregate
advertising as seems likely ld then what one industry gains must be lost by
others
Page 21
elasticity of demand does
(10) we can
if t 0 then
Finally a determination concerning the industry
not fix the individual firms elasticity Going back to equation
write y bull (1 - ) a + z y Since a gt 0 it follows thatii i i
Y gt 0 If Ylt 0 however y s sign is indeterminateij ij
Turning next to the coefficient of market share 20 of them are negative and
12 are positive with three of each sign being significantly different from zero
The largest positive value which is significant is for pet food (code 2010) and
a close second is mens amp boys suits and coats (code 23 01) The largest signifi
cant negative value is for cosmetics (code 28 09)
In the model some relations among y a 0 and 61 are implied If middota is
negative (1- y) is greater than one Since 61Y bull (1- a) (1 - ) 61Y is
pCsitive but it will be less than one if 0 is large enough Seventeen of the A
ys are negative in each of the seventeen industries 61 is also negative and les
1 (algebraicly) than y so 61Y is greater than one his is not evidence that the
degree of cooperation is low or zero in any industry of course A high o togethegt
with a negative y which is small (in absolute terms) relative to a can give a
value of 61Y greater than one On the other hand if 61Y were less than one
o would have to be greater than zero
For y gt 0 two conditions may hold If y lt a it follows that 61 lt 0 since
(y - a ) lt 0 It also follows that 61Y lt 0 regardless of the value of o bull
In addition the relation is an if-and-only-if one ie if 61Y is negative
then y lt a Three industries have negative values for 61y dairy products
except milk (code 20 04) synthetic fibers (code 2806) and photographic equipment
(code 3808) In all three cases y is quite small so a does not have to be very
large to give a negative value to al
If y gt a the model shows that 61-- gt 0 whatever the value of 0 bull There are
12 cases where both y and 61 are positive with 61 ranging from near zero
(grain milling code 20 12) to 1 20 ( proprietary drugs code 2808) Several
coeffiei-en_rs
--
15
o f to zero
Tab le 3 - Consumer Goods
Indus tr ies
2Distributions of R and Producer Goods
Page 20a
Consumer Producer
Goods Goods
0 - 9 9 0 3
10 - 19 9 0 1 1
20 - 29 9 3 20
30 - 39 9 3 17
40 - 49 9 5 16
50 - 59 9 8
69 9 6 Ii 1060
--f 670 79 9 2
80 - 89 9 4 4
90 - 100 1 1
Totals 32 10 3
Notes
2is used R i s cal-
culated as F F + ((N - K- 1 ) K ] 1 b_ F is the F
statistic Lor the hypothesis that all o f the non - intercept terms are simul taneously equal
1
2
--
-- Since generalized least squ-area regress ion
Page 12
o f these cases have values o f 6 1 Y rea ter than 0 8 (pet food (20 10) bread amp cakes
(20 14 ) miscellaneous processed foods (20 29 ) men s and boys suit s and coats
(23 01 ) p rop riet ary drugs (28 08 ) elect ric housewares (36 1 2) radio and
IV sets 17 and games and toys (39 For be than(36 ) 04)) 6 1 Y to greater 0 8
both o and cry mus t b e less than 0 2 That is in these industries the results
shown in Tab le 2 would hold only with very strong dominance of marke t demand
effects over brand swit ching e ffects and a very low degree o f cooperation
Using the concentration rat ios given in Table 4 the average concentration
ratio for the eight indust ries is 38 1 For the remaining 24 industries it is
43 2 This result would lend some support t o the proposition that concentrat ion
and cooperation are related if the group of eight industries are identi fied
as having low degrees of cooperation
One f inal note on model predict ions S ince both cr and o are assumed to
be non-negative 60 should also be non-negat ive In only one o f the 32 cas es
is 60 negative and it is not signif icant bull
cor pJ
7 1
7 0 4
1 4 7 5 1
)
7 3 z a o a t s 1 1 3 3 5 ( i V Z 8 tJ1 2 1 - 1 1 0 7 3 5 c middot v 2 8 1 0 3 2 -0 0 1 4 3 0 C 1 V 9 8 3 2 8 1 5 1 6 - 0 0 6 t1 6 9 C _ V 1 11 0 2 9 0 1 29 0 00 7 6 1 C l k V 7 3 1 3 6 08 1 3 - o n 4 9 9 4 3 3 3 6 1 2 2 4 0 1 6 4 3
3 1 7 1 3 0 1 1 7 d 7 8 3 3 8 0 8 0 0 2 3 i 2 3 0 6
P 4
3 9 0 3 1 4
1 2
0 09 3 8 4 39 1)1 0 4 l 8 6 5
T A LE 4 e I NCU S T R Y C A T E G O R Y
F TC NOM e tOOE ass
C 1 I c 2 )
2 0 0 1 1 6 2 0 0r 1 7 zo o l Z 2 0 0 7 1 8
z o o 2 6
z o t a 1 6
2 1) 1 2 1 5 2 0 1 4 1 1 2 0 1 a 1 9 2 0 26 1 2 2 0 2 7 1 0 2 0 29 - 3 5
2 3 0 1 10
2 3 02 z o
2 3 03 1 9
2 5 5 1 2 4 Z 1 0 Z 1 1 2 0 3 1 9 2 7 4 1 0 z a o 1 3
2 A 0 7 z s
E LA S T I -C I T Y
c 3 )
- 0 09 7 2 7
0 00 449
- 0 0 2 8 9 3
- 0 3 2 7 70 - 0 OZ il I q
o 4 u t s a 0 0 1 3 7 3
o 1 4 o o-0 0 5 2 4o -0 1 94 0 1
-C l 6 l l 4 0 1 JJ 3 7 0 3 t 0 4 2 0 0 4 0 9
- 0 4 1 8 44 - O l 7 t ) 4 -ll 7 1 4 0 b -0 3 5 7 8 Z
- 0 1 9 7 7 0 1)0 3 3 4
-o n 9 2
IJ A T A
CONV
ON-C ONV
( 4 )
t 1NV
C 1t-V C NV c ) v C - JNV C rlNV c n v c a v C ON V ( i) l V
C CtV
C t1tV C IJNV
C 1 V
c 1 v
Pa ge 2La
HEQ F t Noe x
c 5 )
6 2q
6 89 1 48
6 1 4 ) q 3
1 9 7 1 9
q z t s 5 a e z
l l 9 1
4 o 7 1 2 0
c oqt 1 Z
4 4 )-q 1 z
5 5 9 1 0 64 1 3 3
n 3
Inter industry analysi
Page 23
Equation ( 3 6 ) may be used to fo rmulate a rough test o f the cooperat ion
vs e f ficiency controversy I f we rest ricted ourselves to some subset o f
indus tries where the relat ion be tween y and a i s the same we would be lookingbull
at a sample o f indus tries for which the dependence o f V on H and o is a
quadratic wi th no squared terms
I f we now sub s titute (26) into ( 36 ) we get
(38) v bull a + (y - a) R + (y - a) o (H) - (y - a) H o (H) + u
If it is only d if ferences in marke t shares that affect advertising inten-
s i ty levels i f concentration does not af fect cooperation and i f o bull 0
( 38 ) reduces t o
( 39 ) V bull a + (y - a) H + U
Let this b e the null hyp o thes is and note that V is a linear func-
tion o f H If y gt a and the null hyp othesis holds the expected relat ion
between V and H is as shown by the pos itively sloped straight line in Figure
6 b etween points A and C
In this context the alterna ive hypothesis the degree o f
cooperation ( o) is a positive function o f concentrat ion
is that
(H) I f the func-
tion is as described in (26) and Figure 4 the relation between V and H
would be as shown by the curved line connec t ing po ints A and C in Figure 6 One way to test the null hypothe s is against this al ternative is to test fo r
the hypothesized curvature in the function relating V and H
Fi gure
R
6
v
ID c
H 1 0
I
i I I
sfigned
clasJ s
in def ini g
effec t is the
I
Page middot2lf
One charac teristic of produc ts which may be useful
classes wi th respec t to significanc e of the brand-switching
conveniencenon-convenienc e distinc tion developed by Porter ( 1 97 )
Using his identif ication of IRS indus tries as a s tar t ing point
each of the 32 industries used in this s tudy to one of the two
The ass ignments are not ed in Tab le 4
Using the conveniencenon-convenienc e dis t inc tion the
divided into two sub-samples
sample was
Four regressions were then run bull ene for e ch
sub-sample one for the to tal sample and one for the total sample with
a dummy variable for convenience goods Each regression was run wi th LB s
and wi th indus tries as ob servations The results are given in Tab le 5
For the LB level equations ( 1A - 1D) equation(37) was used ome
rearranging of terms gives v bull y - ( 1-o ) (y - a) ( 1-z ) + ui i i
bull y - 6 1 ( 1 - z ) + ui When observations are pooled across indus trif s yi
is an industry variable In these regressions the values o f y es t imated
in the first stage were used Since industries have been grouped so as
to reduce differences in (y - a) that term is t reated as a cons tant
within each group Finally H the Herfindahl index has been subst ituted
for o the degree of cooperation The resultant equation is linear
y ( 1 - z ) and H x ( l - z ) This is the equation for whi ch results arei i
reported in Table 5 In all regress ions all variables were weighted by the
square roo t of sales
in
raquo
- 34
3 1
i l T 4 8LE S R E G E S S I ON R E SUL T S i X T f N S I ON S T O T Y PE Of
AND L E V E L OF C ONC E kT R A T I O
E O O tt S
tP bull T Y P E NR C ONV O B S tO N-CONV
E L A S T I shy
T I C J T Y
O E - t-41( T SliA R E
H( fl F
1 1 - K O x middot K O X E R F
- H S a 1 1 - H S t 1 8 1 1 9 1
R S O
1 1 0 1
q z
t - bull 9 f 2 1
Z t P D l 6 fC ONV - 1 9 1 s s c 2 1
e - l O I e bull u q a
2 3 1 0 1 1 Nl N- t O V 5 7 1 ( 1 2 19 1 6 1 - J l S 2 5 4
l l l l t z q o a 1 - J o v z I
l ( I NO l l A l l ol u l t o o c bull J u l l
d S
c 3 4 t t r a l
bull Z D 1 0 l l 4 l l 2 3 -gt 2 l l 2 0 - 4 lS 5 4 l C 4 1 o e 6 t l b b 5 1 1 0 1 1 1 c - 1 9 5 1 t d S 2 1 tV
OQ10 T-R A T I O S A E G I VE N l P A RE N T H E S E S l E V E L S f S I G 1 F l C ANC E A - 99 5 B - q 9 t - s NS J UC E NO I Nl E R C ( P T I S U (j f middot) I N f Hf t OUA TIOI S 6 E f 3R E A O J U S T H E N T F OR 1-t E T E R O SIC E O A S T f C I T V pIraquo
1
middot
I
R SO I S C A L C UL A T E D A S f I f bull 1 1 bull 1lt 1 K t t- E E N I S THf U A1 B E R O f fJe S E fi V A T I O S F t S HtE F - S T A T I S
I N DE P E ND E N T y II
T I C F O T H E U Y P O HIE S I S f A l l C JE F F I C H N T A R E QU A L HI l f RO A f40 K I S T H f UUM 8 E R OF Q ( A fL f S
I l l
1 A
1 6
t C
f 1 r R I l l
1 2 1
L Jt
L B
l q
L 1
on s T Y P E bull 2 )
P OUC T
X
H S I l l
l O l A I bull l 0 7 1
O O b O J
1 2 00 1
1 b 2 1 C I l l Z U
l l S b 1 -1 1 0 1
rlt F X l l - fi C f l
bull 1 I
l l
- ()4 2 3 l f H
1 - 2 7 9 1
K O X
1 1 - ti E R F I
l d l
S4 l i A I 2 l b
1 3 00
K O X -t f R F t S J
I 1 - HER F t
)
60 9
z o bull
0
t l l
3 8 8
1 7 6
5 64
64
N f bull
tl S bull
I l l
I 4 1
t ONV
NON-t ONV
A l l
A ll
c o v NO- ( ONV
1 4 1
1 5 1
- l S b
c - 1 1 2 1
5 8 2 C A I 1 6 7 0 1
middot I l 9 H
l 1 il
1 4 1 6 1
E L A C) T J shy
l t C I T Y
I S I
l b l
- 0 2 0
1 - 1 4 9 1
O Z 3 2 1 A I C b 6 b l
0 0 9 2 -J
1 1 b 4 1
o z t t t l 1 2 2 1 1
oru - Ht tt F
1 gt X bull b I
Page 25 bull
Mean val -as ior relevan t variables for the two sub-samples are
Conv Non-Conv
Adv Sales 4 7 2 3
iit Share 2 7 3 2
Ela cicity - 05 3 02 2
Herf indahl 7 3 1 860
That the dist nct ion between coft9enienee goods and non - convenience
goods mat ters is supported not only by these means but also by the regression
results Equat ions lA and lB are different in almost every respec t Only the
concentrat ionmarket share int eract ion term seems to mat ter for convenience
good s Jus t he opposite holds for non - convenience goods both elasticity1
and market sl e are highly significant but the concentrat ionmarket share L
int erac t ion t=rm is not
Given that one minus market share is used these equ3 tions the coeffici t
of market share is the negative of what is sh in col (6) of Table 5 For the
regress ion for non - convenience LB s reported in equation lB then LB s with larger
market shares have lower advertising to sales ratios
On the question of the impact of the degree of cooperation the story is mixed
For convenienc good s the coefficient is pos itive as equation (31) predicts when
there is a po sitive degree of cooperat ion Fo r the non-convenience goods LB s middot
concentration has no effec t
Aggregat ion t o the indus try level and then rer unning the equations $ives the
results shown in (2a-2d) of Tab le 5 As expected R2 goes up and levels of significanc e
fmiddotJr individual coefficients gJ down In add ition the coeffic ient of H ( l -H) inmiddot
quat ion 2A is no t significant though it is po s i tive Since 2A is the indus try level
I
--
all but one of them and that one is not significant For 18 of the 31
wo uld have on its adve rtising to sal es ratio
cognate of LA the variable H ( l-H) pl ays the same role in 2A as H ( l -z 1) p lays
in LA
Vconomies of scale in advert ising
The literature on advert ing contains a number of definitionbull o f economies
of scale and of findings eone rning its presence and magnitude Before I loo k
a t the data with the hope o f mmenting about scale effect s then I wan t to
define the term as I use it
By economies of scale I mean a decrease in the number o f units of adver-
tising which are needed to generate a unit of sales as advert ising is increased
Given the definition of the firm s own elast icity of demand with respect to
advertising (see the text foll wing equation (1) ) the technical definition
is that rii gt 1 Diseconomies of scale are def ined symmetrically i e
Yu lt 1
ith a cro s s - section oi observations on advertising intensity
(v bull a s1) and onmiddot advertising (a ) we may regress the former on the latter 1 i i
The results of such regress ions for the 32 industries are given in Table 2
lines lB ZB bull bull bull 32B For the 32 eases the regression coefficient is positive it
positive coefficients the coefficient i s signif icant
The que tion now is whether these results ar e evidence of pervasive dis-
economies of scale in advertislng I think no t Given my as sumption that the
data I observe are equilibrium results I may ri ghtly conclud e that almost every-
where high levels of advert ising are as sociated in equilibrium with high ratios
of advertising to sales I may no t conclud e anything however about the
impact that a move by some firm away from its equilibrium level o f advertising
middot- Page 27
Given the assumed relation in equation ( 37 ) there is a corre spondshy
ing equilibrium equation relating v and ai The equation is non - linear i
and I have not found a simp le way to characterize it It is po ssib le to
det ermi ne its deriviative with respect to a b y takin$ tne de riviative of ( 3 7 ) i
When that is done the result is
v - 1(40) i bull sl a a z ( 1-z )i i ira-l
Sinc e all the other terms in (40) are non-negative the sign dep ends only
on the sign of 6 1 which is in turn de pend ent only on the sign of (y - a)
That the signs of the coefficients of a in equations lB 2B bull bull bull 32B arei
not always equal to the signs of the coefficients of z in LA 2A bull bull bull 32A i
is still a mystery to me and something to be explored
bull
Summary
My purpose in writing this pa per wa s t o attempt to integrate explicit micro-
economic mode lling reasoned econome tric specification of error terms and high
quality data to exp lore some questions about adver tising by large firms
al quite important improvements could be made in the first two of those areas
and much data ma ssag ing is yet to be done in the third
Concerning the subst an ce of what I have done I think six things are im-
Page 2 8
Severshy
V and Conclusions
portant
1 Explicit modeling is worth the ef for t if for no other reas on than
that it provides a basis for choosing fr om a varie ty of funct ional
forms
2 The same is true for error specif ica t ion I did in fact look at scatshy
ter diagrams for all 32 industry categor ies and they vir tually all show
the het eroscedast icity which I as sumed and for which I correc ted
3 The predicted relation be tween advertis ing in tensity and market share
shows up clearly in on ly 25 of the case s examined I have ye t to
explore why that may b e the cas e
4 Some evidence concerning the presence and impact of coopera tion was
produced but it is not clear ly pervasive
5 The dis tinction between convenience goods and non-c onvenie nce goods is
unambiguously a good one
6 Virtually no evidence concerning econ omies of scale in adver tis ing can
be gleaned fr om this study given its assumpt ions
More work is called for on many of these issues my study seems to have
raised more questions than it has answer ed This is pr obab lly due to the richshy
ne ss of the data source since I had the oppor tunity to address several issue s
at once
- -- - Q --
Foo tnotes
Manager Line o f Business Program Federal Trade Commission The author depended heavily on several staf f members of the LB Pro gram for statistical and clerical support in particular Joe Cholka George Pascoe and Haro lene Jenkins Helpful comments were given b y Richard Caves Dennis Mueller Michael Porter and F M Scherer When the paper was in its early stages o f development both S teve Garber and Jon Rasmussen former economists in the LB Program provided eery useful cri tiques The views expressed here are my own o f course
1 The basic exp licit model of the role of advertising is Dorfman and S teiner ( 1954 ) and the models in this paper may be seen as ano ther extens ion of their work
2 For a review o f models of oligopoly which treat prices and quantities as relevant variables see Shubik ( 1 959) Extens ions to take advertis ing int o account are fairly s traightforward
3 Dorfman and S teiner dealt with a price-setting firm For a monopolis t inverting the demand relation to show Pbull P (Q A) _instead o f Q- (P A) has no effect on the results o f the analysis For oligopolists who face a sys tem of demand relations inversion does have some impac t on the results The dif ferences are not trivial nonetheless they will no t be explored in this study
4 Shubik ( 1 959) p
5 From a formal mathematical point o f view the function is similar to a1 i Legrangian funct ion of the form - r rr
L bull 11 + A ( j - (i i i+j
where the indicates some fixed level of the profits industry Following the same logic which resul ts in and income constraint as the marginal utility of the of the indus tr in the obj ective of firm i
6 Hall amp Weiss ( 1 967) p 323
7 See Schmallensee ( 1 97 2 ) Ch 2 pp 1 6-47
8 On this poin t see Imel amp Helmberger ( 1 9 7 1 )
9 Long ( 1 970) p 130 - 1 39
1 0 See Schmallensee ( 1 972) Ch 3 pp 48-8 7
rrj i+j ) of the rest of the
the identification profits of the rest
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980
I
I
bull 9
9t
middot middot middotbullmiddot bull lraquo
TABLE 2 REGRE SION RESUlTS ICONTt
bull
EO I NO tHl I filER- f4A_K T AOVER- sbulluRE ttSQ flA Jl- iLAS T NR CODE cos CEPT SHARE JISING AOVfR bull (JTY SHR CiJCF
110Cl I f amp Itt 15J tbt bull 1c u caa
llA 2702 ll o079 6 81 Bt -019110 S6s -OlltO 111
a 004679 ooooo1 t69
( oOl26o -015160 001560 -022690 ltJr()I
ldA 270 3 19 0065lt -OIt2l60 220 -Ol57a2 11 8 e o 0 71t6IA t ol08 75tAt ( OOl983CiU -060llOC() OOl7191A t -O lltdOC A) middot 89
lqA 2704 10 005082 -Oil060 22b -l)i1977 lelO 6 -000080 oo2772C4 t 910 c 000907 -011 200 Oe04105CA t -046960 995
20A 2d06 1 II
O OlltStiAt -001120 837 00014 -))th O01380amp At -oooo 11 821 ( OOl526CAt -026700181 0002 5UAt 000530 969
2lA 2807 28 OOt9lt9CAt -00891 0 516 -0 03962 225I8 002095CCt OOOl lliAt 690 ( o02965CA t -056650((1 OOOt50CAJ -003l70ttl 911I
I 22A 2808 15 ot3l06 119730 655 1130) 090
0 - 007526CC t Oe003961At 835- middot-
c I oo7523 -067210 001l 2tA t -0091t80CCt 951
2lA 2809 21 Oe2067lCAt -1218101At 71 2 -10095((1 121 ttlIaOQ8 00797418) OOOI8lltt 686 Ill( 0092lU A t -Olltl70Ct t 000605CAt -OOit8701At 951 _
---- -
32 009506CAt
0 bull 06 981 I A J 006901 cA bull
00 24A 2810 -010960 5t 5 -OOltSO 7 56
8 -000011 s11 c -5811701At oooaoatAJ oot520IAI 888
-- 37 --- middot - -- -1 ) l lbull t5JJ r-l bullhmiddot1 --- - --
l5A
26A
27A
c
2RA
c
2 9A
31A
llA
24
13
597
Ci)55
14
l 6h
RSQ ELASTI- (lASTCOE f-
TABLE 2
EQ I NO NR CODE IU
REGRESSION RESULTS CCONTa
NR INTER- MARKET
(10S CEPT SHARE AOVER- SHARfcent
TIS lNG AOVER
) CITY StiR
c It) bull 5 co a I 1) bull 0 bull 9 c l 0 bull( 2 1
2815 16 003019 -009990 275 -o069bmiddotJ
6 001446 000224 282 c 002178 -054630 o02221CCI -0075 00 646
2901 29 00021t2 CAI 0)0520 592 0007 6 1 068
0 00105C8 t 000012141 60)g( OOOlJ51AJ -00126018) ooou271At -000u80CA) 974
36oa 1 3 003296(4) -oca29o -oo-q94
001755 000203 lt91t 002499((1 -04l820CCI OOl374CAI 0068)0 946
3612 0033281CI 011520 195 01684) oso
00 30 42C OI 00012 2 424 oo579ltAt -2 7l7801At 002 100IAI -003220 819
8 3617 btel 011789 081002259 009530
001448 oOOl OitCCI 732 OOl7 22CAI -0 46160CAI 0004961AI -OOll40181 989c
30A 3808 14 0021t 3 6fAI -000110 628 o023291C) -005 6 middot- _ middot- 0bull 02 04 1 IC I ooooo9 637 ( 001584((1 -0161)1)0 00021 91AI -0 00060
390) 14 002790CCt 006590 h8 t 0 0938 4 070
B OOl89bCCJ 000144 74
0021961(1 -OItlOlO OOl323CAI -004800 915
B
middot--- - middot 4 - middot- - dfll3904 12 004563181 0)8300 879 0428651(1 089 OQIll004 277CA I ooozstcea 919
c 0041)3181 -0-11250 0011881(1 -006270 972 0Ill
- J - bull bull bull bullbullbull middot - - middot - middot gt middotbull L)ofbullflbull middot middotfJltl -tn- middotmiddotmiddotmiddoto +f middot t middot bull
c
is
positive
2
Page 0 determineThe consumer goodproducer good split is sometimes difficult to
Sollle support for the split used in this paper is a comparison of goodness - of - fit
distributions of a
industries
set
advertising are
that industry
a sales increase of
the
highest aD10U3 those
measures for the two groups of industries Table 3 contains
measures for the 32- consumer goods industries and the 103 producer goods
2 Median R is 5 7 for the consumer goods set and 40 for the producer goods
Estimates - of y the elasticity of demand with respect to
given in column of Table 2 The hypothesis that y bull 0 with
alternative was applied for each ittdustry As shown in the
could not be rtjected for 24 of the 32 industries Of the
and 10 are positive For the remalnifg eight three are
and five are significantly positive
Of the threertegative elasetcities the largest (in absolute value)
cosmetics (code 2809) With all elasticity of -1 01 a one
industry advertising would gwerate a decrease in industry sales
more than one percetu
At the other extrema is toys and games (code 39 04) For
a one percent ittcrease in industry advertising would lead to
four-tenths of a pampreentmiddot Though not significantly different from zero
elasticity estimatbull for pt-bpuetary drugs (code 2808)
which are
Both drug industries have
to sales rafios (211 pe1 cent for 28 08 and 13 4 per cent
other hand flaIJo ing ext1ac s (mainly soft-drink SYUlS) also
is for
percent increase in
of slightly
a middottwo - tailed
table that hypothesis
24 14 are negative
significantly negative
9
the codetics and p-ropriacuy very high advertising
for 28 09) bull On the
has a high ind UY
advenisiltg t() sales ratio ( 8 0 per cent) bull but its estimated elasticity is very
small and not signifieantly different uom zero
middot It doe$ not seem unusual to find many elasticities near zero with a few
negative and a few positive If aggregate consumption is insensitive to aggregate
advertising as seems likely ld then what one industry gains must be lost by
others
Page 21
elasticity of demand does
(10) we can
if t 0 then
Finally a determination concerning the industry
not fix the individual firms elasticity Going back to equation
write y bull (1 - ) a + z y Since a gt 0 it follows thatii i i
Y gt 0 If Ylt 0 however y s sign is indeterminateij ij
Turning next to the coefficient of market share 20 of them are negative and
12 are positive with three of each sign being significantly different from zero
The largest positive value which is significant is for pet food (code 2010) and
a close second is mens amp boys suits and coats (code 23 01) The largest signifi
cant negative value is for cosmetics (code 28 09)
In the model some relations among y a 0 and 61 are implied If middota is
negative (1- y) is greater than one Since 61Y bull (1- a) (1 - ) 61Y is
pCsitive but it will be less than one if 0 is large enough Seventeen of the A
ys are negative in each of the seventeen industries 61 is also negative and les
1 (algebraicly) than y so 61Y is greater than one his is not evidence that the
degree of cooperation is low or zero in any industry of course A high o togethegt
with a negative y which is small (in absolute terms) relative to a can give a
value of 61Y greater than one On the other hand if 61Y were less than one
o would have to be greater than zero
For y gt 0 two conditions may hold If y lt a it follows that 61 lt 0 since
(y - a ) lt 0 It also follows that 61Y lt 0 regardless of the value of o bull
In addition the relation is an if-and-only-if one ie if 61Y is negative
then y lt a Three industries have negative values for 61y dairy products
except milk (code 20 04) synthetic fibers (code 2806) and photographic equipment
(code 3808) In all three cases y is quite small so a does not have to be very
large to give a negative value to al
If y gt a the model shows that 61-- gt 0 whatever the value of 0 bull There are
12 cases where both y and 61 are positive with 61 ranging from near zero
(grain milling code 20 12) to 1 20 ( proprietary drugs code 2808) Several
coeffiei-en_rs
--
15
o f to zero
Tab le 3 - Consumer Goods
Indus tr ies
2Distributions of R and Producer Goods
Page 20a
Consumer Producer
Goods Goods
0 - 9 9 0 3
10 - 19 9 0 1 1
20 - 29 9 3 20
30 - 39 9 3 17
40 - 49 9 5 16
50 - 59 9 8
69 9 6 Ii 1060
--f 670 79 9 2
80 - 89 9 4 4
90 - 100 1 1
Totals 32 10 3
Notes
2is used R i s cal-
culated as F F + ((N - K- 1 ) K ] 1 b_ F is the F
statistic Lor the hypothesis that all o f the non - intercept terms are simul taneously equal
1
2
--
-- Since generalized least squ-area regress ion
Page 12
o f these cases have values o f 6 1 Y rea ter than 0 8 (pet food (20 10) bread amp cakes
(20 14 ) miscellaneous processed foods (20 29 ) men s and boys suit s and coats
(23 01 ) p rop riet ary drugs (28 08 ) elect ric housewares (36 1 2) radio and
IV sets 17 and games and toys (39 For be than(36 ) 04)) 6 1 Y to greater 0 8
both o and cry mus t b e less than 0 2 That is in these industries the results
shown in Tab le 2 would hold only with very strong dominance of marke t demand
effects over brand swit ching e ffects and a very low degree o f cooperation
Using the concentration rat ios given in Table 4 the average concentration
ratio for the eight indust ries is 38 1 For the remaining 24 industries it is
43 2 This result would lend some support t o the proposition that concentrat ion
and cooperation are related if the group of eight industries are identi fied
as having low degrees of cooperation
One f inal note on model predict ions S ince both cr and o are assumed to
be non-negative 60 should also be non-negat ive In only one o f the 32 cas es
is 60 negative and it is not signif icant bull
cor pJ
7 1
7 0 4
1 4 7 5 1
)
7 3 z a o a t s 1 1 3 3 5 ( i V Z 8 tJ1 2 1 - 1 1 0 7 3 5 c middot v 2 8 1 0 3 2 -0 0 1 4 3 0 C 1 V 9 8 3 2 8 1 5 1 6 - 0 0 6 t1 6 9 C _ V 1 11 0 2 9 0 1 29 0 00 7 6 1 C l k V 7 3 1 3 6 08 1 3 - o n 4 9 9 4 3 3 3 6 1 2 2 4 0 1 6 4 3
3 1 7 1 3 0 1 1 7 d 7 8 3 3 8 0 8 0 0 2 3 i 2 3 0 6
P 4
3 9 0 3 1 4
1 2
0 09 3 8 4 39 1)1 0 4 l 8 6 5
T A LE 4 e I NCU S T R Y C A T E G O R Y
F TC NOM e tOOE ass
C 1 I c 2 )
2 0 0 1 1 6 2 0 0r 1 7 zo o l Z 2 0 0 7 1 8
z o o 2 6
z o t a 1 6
2 1) 1 2 1 5 2 0 1 4 1 1 2 0 1 a 1 9 2 0 26 1 2 2 0 2 7 1 0 2 0 29 - 3 5
2 3 0 1 10
2 3 02 z o
2 3 03 1 9
2 5 5 1 2 4 Z 1 0 Z 1 1 2 0 3 1 9 2 7 4 1 0 z a o 1 3
2 A 0 7 z s
E LA S T I -C I T Y
c 3 )
- 0 09 7 2 7
0 00 449
- 0 0 2 8 9 3
- 0 3 2 7 70 - 0 OZ il I q
o 4 u t s a 0 0 1 3 7 3
o 1 4 o o-0 0 5 2 4o -0 1 94 0 1
-C l 6 l l 4 0 1 JJ 3 7 0 3 t 0 4 2 0 0 4 0 9
- 0 4 1 8 44 - O l 7 t ) 4 -ll 7 1 4 0 b -0 3 5 7 8 Z
- 0 1 9 7 7 0 1)0 3 3 4
-o n 9 2
IJ A T A
CONV
ON-C ONV
( 4 )
t 1NV
C 1t-V C NV c ) v C - JNV C rlNV c n v c a v C ON V ( i) l V
C CtV
C t1tV C IJNV
C 1 V
c 1 v
Pa ge 2La
HEQ F t Noe x
c 5 )
6 2q
6 89 1 48
6 1 4 ) q 3
1 9 7 1 9
q z t s 5 a e z
l l 9 1
4 o 7 1 2 0
c oqt 1 Z
4 4 )-q 1 z
5 5 9 1 0 64 1 3 3
n 3
Inter industry analysi
Page 23
Equation ( 3 6 ) may be used to fo rmulate a rough test o f the cooperat ion
vs e f ficiency controversy I f we rest ricted ourselves to some subset o f
indus tries where the relat ion be tween y and a i s the same we would be lookingbull
at a sample o f indus tries for which the dependence o f V on H and o is a
quadratic wi th no squared terms
I f we now sub s titute (26) into ( 36 ) we get
(38) v bull a + (y - a) R + (y - a) o (H) - (y - a) H o (H) + u
If it is only d if ferences in marke t shares that affect advertising inten-
s i ty levels i f concentration does not af fect cooperation and i f o bull 0
( 38 ) reduces t o
( 39 ) V bull a + (y - a) H + U
Let this b e the null hyp o thes is and note that V is a linear func-
tion o f H If y gt a and the null hyp othesis holds the expected relat ion
between V and H is as shown by the pos itively sloped straight line in Figure
6 b etween points A and C
In this context the alterna ive hypothesis the degree o f
cooperation ( o) is a positive function o f concentrat ion
is that
(H) I f the func-
tion is as described in (26) and Figure 4 the relation between V and H
would be as shown by the curved line connec t ing po ints A and C in Figure 6 One way to test the null hypothe s is against this al ternative is to test fo r
the hypothesized curvature in the function relating V and H
Fi gure
R
6
v
ID c
H 1 0
I
i I I
sfigned
clasJ s
in def ini g
effec t is the
I
Page middot2lf
One charac teristic of produc ts which may be useful
classes wi th respec t to significanc e of the brand-switching
conveniencenon-convenienc e distinc tion developed by Porter ( 1 97 )
Using his identif ication of IRS indus tries as a s tar t ing point
each of the 32 industries used in this s tudy to one of the two
The ass ignments are not ed in Tab le 4
Using the conveniencenon-convenienc e dis t inc tion the
divided into two sub-samples
sample was
Four regressions were then run bull ene for e ch
sub-sample one for the to tal sample and one for the total sample with
a dummy variable for convenience goods Each regression was run wi th LB s
and wi th indus tries as ob servations The results are given in Tab le 5
For the LB level equations ( 1A - 1D) equation(37) was used ome
rearranging of terms gives v bull y - ( 1-o ) (y - a) ( 1-z ) + ui i i
bull y - 6 1 ( 1 - z ) + ui When observations are pooled across indus trif s yi
is an industry variable In these regressions the values o f y es t imated
in the first stage were used Since industries have been grouped so as
to reduce differences in (y - a) that term is t reated as a cons tant
within each group Finally H the Herfindahl index has been subst ituted
for o the degree of cooperation The resultant equation is linear
y ( 1 - z ) and H x ( l - z ) This is the equation for whi ch results arei i
reported in Table 5 In all regress ions all variables were weighted by the
square roo t of sales
in
raquo
- 34
3 1
i l T 4 8LE S R E G E S S I ON R E SUL T S i X T f N S I ON S T O T Y PE Of
AND L E V E L OF C ONC E kT R A T I O
E O O tt S
tP bull T Y P E NR C ONV O B S tO N-CONV
E L A S T I shy
T I C J T Y
O E - t-41( T SliA R E
H( fl F
1 1 - K O x middot K O X E R F
- H S a 1 1 - H S t 1 8 1 1 9 1
R S O
1 1 0 1
q z
t - bull 9 f 2 1
Z t P D l 6 fC ONV - 1 9 1 s s c 2 1
e - l O I e bull u q a
2 3 1 0 1 1 Nl N- t O V 5 7 1 ( 1 2 19 1 6 1 - J l S 2 5 4
l l l l t z q o a 1 - J o v z I
l ( I NO l l A l l ol u l t o o c bull J u l l
d S
c 3 4 t t r a l
bull Z D 1 0 l l 4 l l 2 3 -gt 2 l l 2 0 - 4 lS 5 4 l C 4 1 o e 6 t l b b 5 1 1 0 1 1 1 c - 1 9 5 1 t d S 2 1 tV
OQ10 T-R A T I O S A E G I VE N l P A RE N T H E S E S l E V E L S f S I G 1 F l C ANC E A - 99 5 B - q 9 t - s NS J UC E NO I Nl E R C ( P T I S U (j f middot) I N f Hf t OUA TIOI S 6 E f 3R E A O J U S T H E N T F OR 1-t E T E R O SIC E O A S T f C I T V pIraquo
1
middot
I
R SO I S C A L C UL A T E D A S f I f bull 1 1 bull 1lt 1 K t t- E E N I S THf U A1 B E R O f fJe S E fi V A T I O S F t S HtE F - S T A T I S
I N DE P E ND E N T y II
T I C F O T H E U Y P O HIE S I S f A l l C JE F F I C H N T A R E QU A L HI l f RO A f40 K I S T H f UUM 8 E R OF Q ( A fL f S
I l l
1 A
1 6
t C
f 1 r R I l l
1 2 1
L Jt
L B
l q
L 1
on s T Y P E bull 2 )
P OUC T
X
H S I l l
l O l A I bull l 0 7 1
O O b O J
1 2 00 1
1 b 2 1 C I l l Z U
l l S b 1 -1 1 0 1
rlt F X l l - fi C f l
bull 1 I
l l
- ()4 2 3 l f H
1 - 2 7 9 1
K O X
1 1 - ti E R F I
l d l
S4 l i A I 2 l b
1 3 00
K O X -t f R F t S J
I 1 - HER F t
)
60 9
z o bull
0
t l l
3 8 8
1 7 6
5 64
64
N f bull
tl S bull
I l l
I 4 1
t ONV
NON-t ONV
A l l
A ll
c o v NO- ( ONV
1 4 1
1 5 1
- l S b
c - 1 1 2 1
5 8 2 C A I 1 6 7 0 1
middot I l 9 H
l 1 il
1 4 1 6 1
E L A C) T J shy
l t C I T Y
I S I
l b l
- 0 2 0
1 - 1 4 9 1
O Z 3 2 1 A I C b 6 b l
0 0 9 2 -J
1 1 b 4 1
o z t t t l 1 2 2 1 1
oru - Ht tt F
1 gt X bull b I
Page 25 bull
Mean val -as ior relevan t variables for the two sub-samples are
Conv Non-Conv
Adv Sales 4 7 2 3
iit Share 2 7 3 2
Ela cicity - 05 3 02 2
Herf indahl 7 3 1 860
That the dist nct ion between coft9enienee goods and non - convenience
goods mat ters is supported not only by these means but also by the regression
results Equat ions lA and lB are different in almost every respec t Only the
concentrat ionmarket share int eract ion term seems to mat ter for convenience
good s Jus t he opposite holds for non - convenience goods both elasticity1
and market sl e are highly significant but the concentrat ionmarket share L
int erac t ion t=rm is not
Given that one minus market share is used these equ3 tions the coeffici t
of market share is the negative of what is sh in col (6) of Table 5 For the
regress ion for non - convenience LB s reported in equation lB then LB s with larger
market shares have lower advertising to sales ratios
On the question of the impact of the degree of cooperation the story is mixed
For convenienc good s the coefficient is pos itive as equation (31) predicts when
there is a po sitive degree of cooperat ion Fo r the non-convenience goods LB s middot
concentration has no effec t
Aggregat ion t o the indus try level and then rer unning the equations $ives the
results shown in (2a-2d) of Tab le 5 As expected R2 goes up and levels of significanc e
fmiddotJr individual coefficients gJ down In add ition the coeffic ient of H ( l -H) inmiddot
quat ion 2A is no t significant though it is po s i tive Since 2A is the indus try level
I
--
all but one of them and that one is not significant For 18 of the 31
wo uld have on its adve rtising to sal es ratio
cognate of LA the variable H ( l-H) pl ays the same role in 2A as H ( l -z 1) p lays
in LA
Vconomies of scale in advert ising
The literature on advert ing contains a number of definitionbull o f economies
of scale and of findings eone rning its presence and magnitude Before I loo k
a t the data with the hope o f mmenting about scale effect s then I wan t to
define the term as I use it
By economies of scale I mean a decrease in the number o f units of adver-
tising which are needed to generate a unit of sales as advert ising is increased
Given the definition of the firm s own elast icity of demand with respect to
advertising (see the text foll wing equation (1) ) the technical definition
is that rii gt 1 Diseconomies of scale are def ined symmetrically i e
Yu lt 1
ith a cro s s - section oi observations on advertising intensity
(v bull a s1) and onmiddot advertising (a ) we may regress the former on the latter 1 i i
The results of such regress ions for the 32 industries are given in Table 2
lines lB ZB bull bull bull 32B For the 32 eases the regression coefficient is positive it
positive coefficients the coefficient i s signif icant
The que tion now is whether these results ar e evidence of pervasive dis-
economies of scale in advertislng I think no t Given my as sumption that the
data I observe are equilibrium results I may ri ghtly conclud e that almost every-
where high levels of advert ising are as sociated in equilibrium with high ratios
of advertising to sales I may no t conclud e anything however about the
impact that a move by some firm away from its equilibrium level o f advertising
middot- Page 27
Given the assumed relation in equation ( 37 ) there is a corre spondshy
ing equilibrium equation relating v and ai The equation is non - linear i
and I have not found a simp le way to characterize it It is po ssib le to
det ermi ne its deriviative with respect to a b y takin$ tne de riviative of ( 3 7 ) i
When that is done the result is
v - 1(40) i bull sl a a z ( 1-z )i i ira-l
Sinc e all the other terms in (40) are non-negative the sign dep ends only
on the sign of 6 1 which is in turn de pend ent only on the sign of (y - a)
That the signs of the coefficients of a in equations lB 2B bull bull bull 32B arei
not always equal to the signs of the coefficients of z in LA 2A bull bull bull 32A i
is still a mystery to me and something to be explored
bull
Summary
My purpose in writing this pa per wa s t o attempt to integrate explicit micro-
economic mode lling reasoned econome tric specification of error terms and high
quality data to exp lore some questions about adver tising by large firms
al quite important improvements could be made in the first two of those areas
and much data ma ssag ing is yet to be done in the third
Concerning the subst an ce of what I have done I think six things are im-
Page 2 8
Severshy
V and Conclusions
portant
1 Explicit modeling is worth the ef for t if for no other reas on than
that it provides a basis for choosing fr om a varie ty of funct ional
forms
2 The same is true for error specif ica t ion I did in fact look at scatshy
ter diagrams for all 32 industry categor ies and they vir tually all show
the het eroscedast icity which I as sumed and for which I correc ted
3 The predicted relation be tween advertis ing in tensity and market share
shows up clearly in on ly 25 of the case s examined I have ye t to
explore why that may b e the cas e
4 Some evidence concerning the presence and impact of coopera tion was
produced but it is not clear ly pervasive
5 The dis tinction between convenience goods and non-c onvenie nce goods is
unambiguously a good one
6 Virtually no evidence concerning econ omies of scale in adver tis ing can
be gleaned fr om this study given its assumpt ions
More work is called for on many of these issues my study seems to have
raised more questions than it has answer ed This is pr obab lly due to the richshy
ne ss of the data source since I had the oppor tunity to address several issue s
at once
- -- - Q --
Foo tnotes
Manager Line o f Business Program Federal Trade Commission The author depended heavily on several staf f members of the LB Pro gram for statistical and clerical support in particular Joe Cholka George Pascoe and Haro lene Jenkins Helpful comments were given b y Richard Caves Dennis Mueller Michael Porter and F M Scherer When the paper was in its early stages o f development both S teve Garber and Jon Rasmussen former economists in the LB Program provided eery useful cri tiques The views expressed here are my own o f course
1 The basic exp licit model of the role of advertising is Dorfman and S teiner ( 1954 ) and the models in this paper may be seen as ano ther extens ion of their work
2 For a review o f models of oligopoly which treat prices and quantities as relevant variables see Shubik ( 1 959) Extens ions to take advertis ing int o account are fairly s traightforward
3 Dorfman and S teiner dealt with a price-setting firm For a monopolis t inverting the demand relation to show Pbull P (Q A) _instead o f Q- (P A) has no effect on the results o f the analysis For oligopolists who face a sys tem of demand relations inversion does have some impac t on the results The dif ferences are not trivial nonetheless they will no t be explored in this study
4 Shubik ( 1 959) p
5 From a formal mathematical point o f view the function is similar to a1 i Legrangian funct ion of the form - r rr
L bull 11 + A ( j - (i i i+j
where the indicates some fixed level of the profits industry Following the same logic which resul ts in and income constraint as the marginal utility of the of the indus tr in the obj ective of firm i
6 Hall amp Weiss ( 1 967) p 323
7 See Schmallensee ( 1 97 2 ) Ch 2 pp 1 6-47
8 On this poin t see Imel amp Helmberger ( 1 9 7 1 )
9 Long ( 1 970) p 130 - 1 39
1 0 See Schmallensee ( 1 972) Ch 3 pp 48-8 7
rrj i+j ) of the rest of the
the identification profits of the rest
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980
-- 37 --- middot - -- -1 ) l lbull t5JJ r-l bullhmiddot1 --- - --
l5A
26A
27A
c
2RA
c
2 9A
31A
llA
24
13
597
Ci)55
14
l 6h
RSQ ELASTI- (lASTCOE f-
TABLE 2
EQ I NO NR CODE IU
REGRESSION RESULTS CCONTa
NR INTER- MARKET
(10S CEPT SHARE AOVER- SHARfcent
TIS lNG AOVER
) CITY StiR
c It) bull 5 co a I 1) bull 0 bull 9 c l 0 bull( 2 1
2815 16 003019 -009990 275 -o069bmiddotJ
6 001446 000224 282 c 002178 -054630 o02221CCI -0075 00 646
2901 29 00021t2 CAI 0)0520 592 0007 6 1 068
0 00105C8 t 000012141 60)g( OOOlJ51AJ -00126018) ooou271At -000u80CA) 974
36oa 1 3 003296(4) -oca29o -oo-q94
001755 000203 lt91t 002499((1 -04l820CCI OOl374CAI 0068)0 946
3612 0033281CI 011520 195 01684) oso
00 30 42C OI 00012 2 424 oo579ltAt -2 7l7801At 002 100IAI -003220 819
8 3617 btel 011789 081002259 009530
001448 oOOl OitCCI 732 OOl7 22CAI -0 46160CAI 0004961AI -OOll40181 989c
30A 3808 14 0021t 3 6fAI -000110 628 o023291C) -005 6 middot- _ middot- 0bull 02 04 1 IC I ooooo9 637 ( 001584((1 -0161)1)0 00021 91AI -0 00060
390) 14 002790CCt 006590 h8 t 0 0938 4 070
B OOl89bCCJ 000144 74
0021961(1 -OItlOlO OOl323CAI -004800 915
B
middot--- - middot 4 - middot- - dfll3904 12 004563181 0)8300 879 0428651(1 089 OQIll004 277CA I ooozstcea 919
c 0041)3181 -0-11250 0011881(1 -006270 972 0Ill
- J - bull bull bull bullbullbull middot - - middot - middot gt middotbull L)ofbullflbull middot middotfJltl -tn- middotmiddotmiddotmiddoto +f middot t middot bull
c
is
positive
2
Page 0 determineThe consumer goodproducer good split is sometimes difficult to
Sollle support for the split used in this paper is a comparison of goodness - of - fit
distributions of a
industries
set
advertising are
that industry
a sales increase of
the
highest aD10U3 those
measures for the two groups of industries Table 3 contains
measures for the 32- consumer goods industries and the 103 producer goods
2 Median R is 5 7 for the consumer goods set and 40 for the producer goods
Estimates - of y the elasticity of demand with respect to
given in column of Table 2 The hypothesis that y bull 0 with
alternative was applied for each ittdustry As shown in the
could not be rtjected for 24 of the 32 industries Of the
and 10 are positive For the remalnifg eight three are
and five are significantly positive
Of the threertegative elasetcities the largest (in absolute value)
cosmetics (code 2809) With all elasticity of -1 01 a one
industry advertising would gwerate a decrease in industry sales
more than one percetu
At the other extrema is toys and games (code 39 04) For
a one percent ittcrease in industry advertising would lead to
four-tenths of a pampreentmiddot Though not significantly different from zero
elasticity estimatbull for pt-bpuetary drugs (code 2808)
which are
Both drug industries have
to sales rafios (211 pe1 cent for 28 08 and 13 4 per cent
other hand flaIJo ing ext1ac s (mainly soft-drink SYUlS) also
is for
percent increase in
of slightly
a middottwo - tailed
table that hypothesis
24 14 are negative
significantly negative
9
the codetics and p-ropriacuy very high advertising
for 28 09) bull On the
has a high ind UY
advenisiltg t() sales ratio ( 8 0 per cent) bull but its estimated elasticity is very
small and not signifieantly different uom zero
middot It doe$ not seem unusual to find many elasticities near zero with a few
negative and a few positive If aggregate consumption is insensitive to aggregate
advertising as seems likely ld then what one industry gains must be lost by
others
Page 21
elasticity of demand does
(10) we can
if t 0 then
Finally a determination concerning the industry
not fix the individual firms elasticity Going back to equation
write y bull (1 - ) a + z y Since a gt 0 it follows thatii i i
Y gt 0 If Ylt 0 however y s sign is indeterminateij ij
Turning next to the coefficient of market share 20 of them are negative and
12 are positive with three of each sign being significantly different from zero
The largest positive value which is significant is for pet food (code 2010) and
a close second is mens amp boys suits and coats (code 23 01) The largest signifi
cant negative value is for cosmetics (code 28 09)
In the model some relations among y a 0 and 61 are implied If middota is
negative (1- y) is greater than one Since 61Y bull (1- a) (1 - ) 61Y is
pCsitive but it will be less than one if 0 is large enough Seventeen of the A
ys are negative in each of the seventeen industries 61 is also negative and les
1 (algebraicly) than y so 61Y is greater than one his is not evidence that the
degree of cooperation is low or zero in any industry of course A high o togethegt
with a negative y which is small (in absolute terms) relative to a can give a
value of 61Y greater than one On the other hand if 61Y were less than one
o would have to be greater than zero
For y gt 0 two conditions may hold If y lt a it follows that 61 lt 0 since
(y - a ) lt 0 It also follows that 61Y lt 0 regardless of the value of o bull
In addition the relation is an if-and-only-if one ie if 61Y is negative
then y lt a Three industries have negative values for 61y dairy products
except milk (code 20 04) synthetic fibers (code 2806) and photographic equipment
(code 3808) In all three cases y is quite small so a does not have to be very
large to give a negative value to al
If y gt a the model shows that 61-- gt 0 whatever the value of 0 bull There are
12 cases where both y and 61 are positive with 61 ranging from near zero
(grain milling code 20 12) to 1 20 ( proprietary drugs code 2808) Several
coeffiei-en_rs
--
15
o f to zero
Tab le 3 - Consumer Goods
Indus tr ies
2Distributions of R and Producer Goods
Page 20a
Consumer Producer
Goods Goods
0 - 9 9 0 3
10 - 19 9 0 1 1
20 - 29 9 3 20
30 - 39 9 3 17
40 - 49 9 5 16
50 - 59 9 8
69 9 6 Ii 1060
--f 670 79 9 2
80 - 89 9 4 4
90 - 100 1 1
Totals 32 10 3
Notes
2is used R i s cal-
culated as F F + ((N - K- 1 ) K ] 1 b_ F is the F
statistic Lor the hypothesis that all o f the non - intercept terms are simul taneously equal
1
2
--
-- Since generalized least squ-area regress ion
Page 12
o f these cases have values o f 6 1 Y rea ter than 0 8 (pet food (20 10) bread amp cakes
(20 14 ) miscellaneous processed foods (20 29 ) men s and boys suit s and coats
(23 01 ) p rop riet ary drugs (28 08 ) elect ric housewares (36 1 2) radio and
IV sets 17 and games and toys (39 For be than(36 ) 04)) 6 1 Y to greater 0 8
both o and cry mus t b e less than 0 2 That is in these industries the results
shown in Tab le 2 would hold only with very strong dominance of marke t demand
effects over brand swit ching e ffects and a very low degree o f cooperation
Using the concentration rat ios given in Table 4 the average concentration
ratio for the eight indust ries is 38 1 For the remaining 24 industries it is
43 2 This result would lend some support t o the proposition that concentrat ion
and cooperation are related if the group of eight industries are identi fied
as having low degrees of cooperation
One f inal note on model predict ions S ince both cr and o are assumed to
be non-negative 60 should also be non-negat ive In only one o f the 32 cas es
is 60 negative and it is not signif icant bull
cor pJ
7 1
7 0 4
1 4 7 5 1
)
7 3 z a o a t s 1 1 3 3 5 ( i V Z 8 tJ1 2 1 - 1 1 0 7 3 5 c middot v 2 8 1 0 3 2 -0 0 1 4 3 0 C 1 V 9 8 3 2 8 1 5 1 6 - 0 0 6 t1 6 9 C _ V 1 11 0 2 9 0 1 29 0 00 7 6 1 C l k V 7 3 1 3 6 08 1 3 - o n 4 9 9 4 3 3 3 6 1 2 2 4 0 1 6 4 3
3 1 7 1 3 0 1 1 7 d 7 8 3 3 8 0 8 0 0 2 3 i 2 3 0 6
P 4
3 9 0 3 1 4
1 2
0 09 3 8 4 39 1)1 0 4 l 8 6 5
T A LE 4 e I NCU S T R Y C A T E G O R Y
F TC NOM e tOOE ass
C 1 I c 2 )
2 0 0 1 1 6 2 0 0r 1 7 zo o l Z 2 0 0 7 1 8
z o o 2 6
z o t a 1 6
2 1) 1 2 1 5 2 0 1 4 1 1 2 0 1 a 1 9 2 0 26 1 2 2 0 2 7 1 0 2 0 29 - 3 5
2 3 0 1 10
2 3 02 z o
2 3 03 1 9
2 5 5 1 2 4 Z 1 0 Z 1 1 2 0 3 1 9 2 7 4 1 0 z a o 1 3
2 A 0 7 z s
E LA S T I -C I T Y
c 3 )
- 0 09 7 2 7
0 00 449
- 0 0 2 8 9 3
- 0 3 2 7 70 - 0 OZ il I q
o 4 u t s a 0 0 1 3 7 3
o 1 4 o o-0 0 5 2 4o -0 1 94 0 1
-C l 6 l l 4 0 1 JJ 3 7 0 3 t 0 4 2 0 0 4 0 9
- 0 4 1 8 44 - O l 7 t ) 4 -ll 7 1 4 0 b -0 3 5 7 8 Z
- 0 1 9 7 7 0 1)0 3 3 4
-o n 9 2
IJ A T A
CONV
ON-C ONV
( 4 )
t 1NV
C 1t-V C NV c ) v C - JNV C rlNV c n v c a v C ON V ( i) l V
C CtV
C t1tV C IJNV
C 1 V
c 1 v
Pa ge 2La
HEQ F t Noe x
c 5 )
6 2q
6 89 1 48
6 1 4 ) q 3
1 9 7 1 9
q z t s 5 a e z
l l 9 1
4 o 7 1 2 0
c oqt 1 Z
4 4 )-q 1 z
5 5 9 1 0 64 1 3 3
n 3
Inter industry analysi
Page 23
Equation ( 3 6 ) may be used to fo rmulate a rough test o f the cooperat ion
vs e f ficiency controversy I f we rest ricted ourselves to some subset o f
indus tries where the relat ion be tween y and a i s the same we would be lookingbull
at a sample o f indus tries for which the dependence o f V on H and o is a
quadratic wi th no squared terms
I f we now sub s titute (26) into ( 36 ) we get
(38) v bull a + (y - a) R + (y - a) o (H) - (y - a) H o (H) + u
If it is only d if ferences in marke t shares that affect advertising inten-
s i ty levels i f concentration does not af fect cooperation and i f o bull 0
( 38 ) reduces t o
( 39 ) V bull a + (y - a) H + U
Let this b e the null hyp o thes is and note that V is a linear func-
tion o f H If y gt a and the null hyp othesis holds the expected relat ion
between V and H is as shown by the pos itively sloped straight line in Figure
6 b etween points A and C
In this context the alterna ive hypothesis the degree o f
cooperation ( o) is a positive function o f concentrat ion
is that
(H) I f the func-
tion is as described in (26) and Figure 4 the relation between V and H
would be as shown by the curved line connec t ing po ints A and C in Figure 6 One way to test the null hypothe s is against this al ternative is to test fo r
the hypothesized curvature in the function relating V and H
Fi gure
R
6
v
ID c
H 1 0
I
i I I
sfigned
clasJ s
in def ini g
effec t is the
I
Page middot2lf
One charac teristic of produc ts which may be useful
classes wi th respec t to significanc e of the brand-switching
conveniencenon-convenienc e distinc tion developed by Porter ( 1 97 )
Using his identif ication of IRS indus tries as a s tar t ing point
each of the 32 industries used in this s tudy to one of the two
The ass ignments are not ed in Tab le 4
Using the conveniencenon-convenienc e dis t inc tion the
divided into two sub-samples
sample was
Four regressions were then run bull ene for e ch
sub-sample one for the to tal sample and one for the total sample with
a dummy variable for convenience goods Each regression was run wi th LB s
and wi th indus tries as ob servations The results are given in Tab le 5
For the LB level equations ( 1A - 1D) equation(37) was used ome
rearranging of terms gives v bull y - ( 1-o ) (y - a) ( 1-z ) + ui i i
bull y - 6 1 ( 1 - z ) + ui When observations are pooled across indus trif s yi
is an industry variable In these regressions the values o f y es t imated
in the first stage were used Since industries have been grouped so as
to reduce differences in (y - a) that term is t reated as a cons tant
within each group Finally H the Herfindahl index has been subst ituted
for o the degree of cooperation The resultant equation is linear
y ( 1 - z ) and H x ( l - z ) This is the equation for whi ch results arei i
reported in Table 5 In all regress ions all variables were weighted by the
square roo t of sales
in
raquo
- 34
3 1
i l T 4 8LE S R E G E S S I ON R E SUL T S i X T f N S I ON S T O T Y PE Of
AND L E V E L OF C ONC E kT R A T I O
E O O tt S
tP bull T Y P E NR C ONV O B S tO N-CONV
E L A S T I shy
T I C J T Y
O E - t-41( T SliA R E
H( fl F
1 1 - K O x middot K O X E R F
- H S a 1 1 - H S t 1 8 1 1 9 1
R S O
1 1 0 1
q z
t - bull 9 f 2 1
Z t P D l 6 fC ONV - 1 9 1 s s c 2 1
e - l O I e bull u q a
2 3 1 0 1 1 Nl N- t O V 5 7 1 ( 1 2 19 1 6 1 - J l S 2 5 4
l l l l t z q o a 1 - J o v z I
l ( I NO l l A l l ol u l t o o c bull J u l l
d S
c 3 4 t t r a l
bull Z D 1 0 l l 4 l l 2 3 -gt 2 l l 2 0 - 4 lS 5 4 l C 4 1 o e 6 t l b b 5 1 1 0 1 1 1 c - 1 9 5 1 t d S 2 1 tV
OQ10 T-R A T I O S A E G I VE N l P A RE N T H E S E S l E V E L S f S I G 1 F l C ANC E A - 99 5 B - q 9 t - s NS J UC E NO I Nl E R C ( P T I S U (j f middot) I N f Hf t OUA TIOI S 6 E f 3R E A O J U S T H E N T F OR 1-t E T E R O SIC E O A S T f C I T V pIraquo
1
middot
I
R SO I S C A L C UL A T E D A S f I f bull 1 1 bull 1lt 1 K t t- E E N I S THf U A1 B E R O f fJe S E fi V A T I O S F t S HtE F - S T A T I S
I N DE P E ND E N T y II
T I C F O T H E U Y P O HIE S I S f A l l C JE F F I C H N T A R E QU A L HI l f RO A f40 K I S T H f UUM 8 E R OF Q ( A fL f S
I l l
1 A
1 6
t C
f 1 r R I l l
1 2 1
L Jt
L B
l q
L 1
on s T Y P E bull 2 )
P OUC T
X
H S I l l
l O l A I bull l 0 7 1
O O b O J
1 2 00 1
1 b 2 1 C I l l Z U
l l S b 1 -1 1 0 1
rlt F X l l - fi C f l
bull 1 I
l l
- ()4 2 3 l f H
1 - 2 7 9 1
K O X
1 1 - ti E R F I
l d l
S4 l i A I 2 l b
1 3 00
K O X -t f R F t S J
I 1 - HER F t
)
60 9
z o bull
0
t l l
3 8 8
1 7 6
5 64
64
N f bull
tl S bull
I l l
I 4 1
t ONV
NON-t ONV
A l l
A ll
c o v NO- ( ONV
1 4 1
1 5 1
- l S b
c - 1 1 2 1
5 8 2 C A I 1 6 7 0 1
middot I l 9 H
l 1 il
1 4 1 6 1
E L A C) T J shy
l t C I T Y
I S I
l b l
- 0 2 0
1 - 1 4 9 1
O Z 3 2 1 A I C b 6 b l
0 0 9 2 -J
1 1 b 4 1
o z t t t l 1 2 2 1 1
oru - Ht tt F
1 gt X bull b I
Page 25 bull
Mean val -as ior relevan t variables for the two sub-samples are
Conv Non-Conv
Adv Sales 4 7 2 3
iit Share 2 7 3 2
Ela cicity - 05 3 02 2
Herf indahl 7 3 1 860
That the dist nct ion between coft9enienee goods and non - convenience
goods mat ters is supported not only by these means but also by the regression
results Equat ions lA and lB are different in almost every respec t Only the
concentrat ionmarket share int eract ion term seems to mat ter for convenience
good s Jus t he opposite holds for non - convenience goods both elasticity1
and market sl e are highly significant but the concentrat ionmarket share L
int erac t ion t=rm is not
Given that one minus market share is used these equ3 tions the coeffici t
of market share is the negative of what is sh in col (6) of Table 5 For the
regress ion for non - convenience LB s reported in equation lB then LB s with larger
market shares have lower advertising to sales ratios
On the question of the impact of the degree of cooperation the story is mixed
For convenienc good s the coefficient is pos itive as equation (31) predicts when
there is a po sitive degree of cooperat ion Fo r the non-convenience goods LB s middot
concentration has no effec t
Aggregat ion t o the indus try level and then rer unning the equations $ives the
results shown in (2a-2d) of Tab le 5 As expected R2 goes up and levels of significanc e
fmiddotJr individual coefficients gJ down In add ition the coeffic ient of H ( l -H) inmiddot
quat ion 2A is no t significant though it is po s i tive Since 2A is the indus try level
I
--
all but one of them and that one is not significant For 18 of the 31
wo uld have on its adve rtising to sal es ratio
cognate of LA the variable H ( l-H) pl ays the same role in 2A as H ( l -z 1) p lays
in LA
Vconomies of scale in advert ising
The literature on advert ing contains a number of definitionbull o f economies
of scale and of findings eone rning its presence and magnitude Before I loo k
a t the data with the hope o f mmenting about scale effect s then I wan t to
define the term as I use it
By economies of scale I mean a decrease in the number o f units of adver-
tising which are needed to generate a unit of sales as advert ising is increased
Given the definition of the firm s own elast icity of demand with respect to
advertising (see the text foll wing equation (1) ) the technical definition
is that rii gt 1 Diseconomies of scale are def ined symmetrically i e
Yu lt 1
ith a cro s s - section oi observations on advertising intensity
(v bull a s1) and onmiddot advertising (a ) we may regress the former on the latter 1 i i
The results of such regress ions for the 32 industries are given in Table 2
lines lB ZB bull bull bull 32B For the 32 eases the regression coefficient is positive it
positive coefficients the coefficient i s signif icant
The que tion now is whether these results ar e evidence of pervasive dis-
economies of scale in advertislng I think no t Given my as sumption that the
data I observe are equilibrium results I may ri ghtly conclud e that almost every-
where high levels of advert ising are as sociated in equilibrium with high ratios
of advertising to sales I may no t conclud e anything however about the
impact that a move by some firm away from its equilibrium level o f advertising
middot- Page 27
Given the assumed relation in equation ( 37 ) there is a corre spondshy
ing equilibrium equation relating v and ai The equation is non - linear i
and I have not found a simp le way to characterize it It is po ssib le to
det ermi ne its deriviative with respect to a b y takin$ tne de riviative of ( 3 7 ) i
When that is done the result is
v - 1(40) i bull sl a a z ( 1-z )i i ira-l
Sinc e all the other terms in (40) are non-negative the sign dep ends only
on the sign of 6 1 which is in turn de pend ent only on the sign of (y - a)
That the signs of the coefficients of a in equations lB 2B bull bull bull 32B arei
not always equal to the signs of the coefficients of z in LA 2A bull bull bull 32A i
is still a mystery to me and something to be explored
bull
Summary
My purpose in writing this pa per wa s t o attempt to integrate explicit micro-
economic mode lling reasoned econome tric specification of error terms and high
quality data to exp lore some questions about adver tising by large firms
al quite important improvements could be made in the first two of those areas
and much data ma ssag ing is yet to be done in the third
Concerning the subst an ce of what I have done I think six things are im-
Page 2 8
Severshy
V and Conclusions
portant
1 Explicit modeling is worth the ef for t if for no other reas on than
that it provides a basis for choosing fr om a varie ty of funct ional
forms
2 The same is true for error specif ica t ion I did in fact look at scatshy
ter diagrams for all 32 industry categor ies and they vir tually all show
the het eroscedast icity which I as sumed and for which I correc ted
3 The predicted relation be tween advertis ing in tensity and market share
shows up clearly in on ly 25 of the case s examined I have ye t to
explore why that may b e the cas e
4 Some evidence concerning the presence and impact of coopera tion was
produced but it is not clear ly pervasive
5 The dis tinction between convenience goods and non-c onvenie nce goods is
unambiguously a good one
6 Virtually no evidence concerning econ omies of scale in adver tis ing can
be gleaned fr om this study given its assumpt ions
More work is called for on many of these issues my study seems to have
raised more questions than it has answer ed This is pr obab lly due to the richshy
ne ss of the data source since I had the oppor tunity to address several issue s
at once
- -- - Q --
Foo tnotes
Manager Line o f Business Program Federal Trade Commission The author depended heavily on several staf f members of the LB Pro gram for statistical and clerical support in particular Joe Cholka George Pascoe and Haro lene Jenkins Helpful comments were given b y Richard Caves Dennis Mueller Michael Porter and F M Scherer When the paper was in its early stages o f development both S teve Garber and Jon Rasmussen former economists in the LB Program provided eery useful cri tiques The views expressed here are my own o f course
1 The basic exp licit model of the role of advertising is Dorfman and S teiner ( 1954 ) and the models in this paper may be seen as ano ther extens ion of their work
2 For a review o f models of oligopoly which treat prices and quantities as relevant variables see Shubik ( 1 959) Extens ions to take advertis ing int o account are fairly s traightforward
3 Dorfman and S teiner dealt with a price-setting firm For a monopolis t inverting the demand relation to show Pbull P (Q A) _instead o f Q- (P A) has no effect on the results o f the analysis For oligopolists who face a sys tem of demand relations inversion does have some impac t on the results The dif ferences are not trivial nonetheless they will no t be explored in this study
4 Shubik ( 1 959) p
5 From a formal mathematical point o f view the function is similar to a1 i Legrangian funct ion of the form - r rr
L bull 11 + A ( j - (i i i+j
where the indicates some fixed level of the profits industry Following the same logic which resul ts in and income constraint as the marginal utility of the of the indus tr in the obj ective of firm i
6 Hall amp Weiss ( 1 967) p 323
7 See Schmallensee ( 1 97 2 ) Ch 2 pp 1 6-47
8 On this poin t see Imel amp Helmberger ( 1 9 7 1 )
9 Long ( 1 970) p 130 - 1 39
1 0 See Schmallensee ( 1 972) Ch 3 pp 48-8 7
rrj i+j ) of the rest of the
the identification profits of the rest
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980
is
positive
2
Page 0 determineThe consumer goodproducer good split is sometimes difficult to
Sollle support for the split used in this paper is a comparison of goodness - of - fit
distributions of a
industries
set
advertising are
that industry
a sales increase of
the
highest aD10U3 those
measures for the two groups of industries Table 3 contains
measures for the 32- consumer goods industries and the 103 producer goods
2 Median R is 5 7 for the consumer goods set and 40 for the producer goods
Estimates - of y the elasticity of demand with respect to
given in column of Table 2 The hypothesis that y bull 0 with
alternative was applied for each ittdustry As shown in the
could not be rtjected for 24 of the 32 industries Of the
and 10 are positive For the remalnifg eight three are
and five are significantly positive
Of the threertegative elasetcities the largest (in absolute value)
cosmetics (code 2809) With all elasticity of -1 01 a one
industry advertising would gwerate a decrease in industry sales
more than one percetu
At the other extrema is toys and games (code 39 04) For
a one percent ittcrease in industry advertising would lead to
four-tenths of a pampreentmiddot Though not significantly different from zero
elasticity estimatbull for pt-bpuetary drugs (code 2808)
which are
Both drug industries have
to sales rafios (211 pe1 cent for 28 08 and 13 4 per cent
other hand flaIJo ing ext1ac s (mainly soft-drink SYUlS) also
is for
percent increase in
of slightly
a middottwo - tailed
table that hypothesis
24 14 are negative
significantly negative
9
the codetics and p-ropriacuy very high advertising
for 28 09) bull On the
has a high ind UY
advenisiltg t() sales ratio ( 8 0 per cent) bull but its estimated elasticity is very
small and not signifieantly different uom zero
middot It doe$ not seem unusual to find many elasticities near zero with a few
negative and a few positive If aggregate consumption is insensitive to aggregate
advertising as seems likely ld then what one industry gains must be lost by
others
Page 21
elasticity of demand does
(10) we can
if t 0 then
Finally a determination concerning the industry
not fix the individual firms elasticity Going back to equation
write y bull (1 - ) a + z y Since a gt 0 it follows thatii i i
Y gt 0 If Ylt 0 however y s sign is indeterminateij ij
Turning next to the coefficient of market share 20 of them are negative and
12 are positive with three of each sign being significantly different from zero
The largest positive value which is significant is for pet food (code 2010) and
a close second is mens amp boys suits and coats (code 23 01) The largest signifi
cant negative value is for cosmetics (code 28 09)
In the model some relations among y a 0 and 61 are implied If middota is
negative (1- y) is greater than one Since 61Y bull (1- a) (1 - ) 61Y is
pCsitive but it will be less than one if 0 is large enough Seventeen of the A
ys are negative in each of the seventeen industries 61 is also negative and les
1 (algebraicly) than y so 61Y is greater than one his is not evidence that the
degree of cooperation is low or zero in any industry of course A high o togethegt
with a negative y which is small (in absolute terms) relative to a can give a
value of 61Y greater than one On the other hand if 61Y were less than one
o would have to be greater than zero
For y gt 0 two conditions may hold If y lt a it follows that 61 lt 0 since
(y - a ) lt 0 It also follows that 61Y lt 0 regardless of the value of o bull
In addition the relation is an if-and-only-if one ie if 61Y is negative
then y lt a Three industries have negative values for 61y dairy products
except milk (code 20 04) synthetic fibers (code 2806) and photographic equipment
(code 3808) In all three cases y is quite small so a does not have to be very
large to give a negative value to al
If y gt a the model shows that 61-- gt 0 whatever the value of 0 bull There are
12 cases where both y and 61 are positive with 61 ranging from near zero
(grain milling code 20 12) to 1 20 ( proprietary drugs code 2808) Several
coeffiei-en_rs
--
15
o f to zero
Tab le 3 - Consumer Goods
Indus tr ies
2Distributions of R and Producer Goods
Page 20a
Consumer Producer
Goods Goods
0 - 9 9 0 3
10 - 19 9 0 1 1
20 - 29 9 3 20
30 - 39 9 3 17
40 - 49 9 5 16
50 - 59 9 8
69 9 6 Ii 1060
--f 670 79 9 2
80 - 89 9 4 4
90 - 100 1 1
Totals 32 10 3
Notes
2is used R i s cal-
culated as F F + ((N - K- 1 ) K ] 1 b_ F is the F
statistic Lor the hypothesis that all o f the non - intercept terms are simul taneously equal
1
2
--
-- Since generalized least squ-area regress ion
Page 12
o f these cases have values o f 6 1 Y rea ter than 0 8 (pet food (20 10) bread amp cakes
(20 14 ) miscellaneous processed foods (20 29 ) men s and boys suit s and coats
(23 01 ) p rop riet ary drugs (28 08 ) elect ric housewares (36 1 2) radio and
IV sets 17 and games and toys (39 For be than(36 ) 04)) 6 1 Y to greater 0 8
both o and cry mus t b e less than 0 2 That is in these industries the results
shown in Tab le 2 would hold only with very strong dominance of marke t demand
effects over brand swit ching e ffects and a very low degree o f cooperation
Using the concentration rat ios given in Table 4 the average concentration
ratio for the eight indust ries is 38 1 For the remaining 24 industries it is
43 2 This result would lend some support t o the proposition that concentrat ion
and cooperation are related if the group of eight industries are identi fied
as having low degrees of cooperation
One f inal note on model predict ions S ince both cr and o are assumed to
be non-negative 60 should also be non-negat ive In only one o f the 32 cas es
is 60 negative and it is not signif icant bull
cor pJ
7 1
7 0 4
1 4 7 5 1
)
7 3 z a o a t s 1 1 3 3 5 ( i V Z 8 tJ1 2 1 - 1 1 0 7 3 5 c middot v 2 8 1 0 3 2 -0 0 1 4 3 0 C 1 V 9 8 3 2 8 1 5 1 6 - 0 0 6 t1 6 9 C _ V 1 11 0 2 9 0 1 29 0 00 7 6 1 C l k V 7 3 1 3 6 08 1 3 - o n 4 9 9 4 3 3 3 6 1 2 2 4 0 1 6 4 3
3 1 7 1 3 0 1 1 7 d 7 8 3 3 8 0 8 0 0 2 3 i 2 3 0 6
P 4
3 9 0 3 1 4
1 2
0 09 3 8 4 39 1)1 0 4 l 8 6 5
T A LE 4 e I NCU S T R Y C A T E G O R Y
F TC NOM e tOOE ass
C 1 I c 2 )
2 0 0 1 1 6 2 0 0r 1 7 zo o l Z 2 0 0 7 1 8
z o o 2 6
z o t a 1 6
2 1) 1 2 1 5 2 0 1 4 1 1 2 0 1 a 1 9 2 0 26 1 2 2 0 2 7 1 0 2 0 29 - 3 5
2 3 0 1 10
2 3 02 z o
2 3 03 1 9
2 5 5 1 2 4 Z 1 0 Z 1 1 2 0 3 1 9 2 7 4 1 0 z a o 1 3
2 A 0 7 z s
E LA S T I -C I T Y
c 3 )
- 0 09 7 2 7
0 00 449
- 0 0 2 8 9 3
- 0 3 2 7 70 - 0 OZ il I q
o 4 u t s a 0 0 1 3 7 3
o 1 4 o o-0 0 5 2 4o -0 1 94 0 1
-C l 6 l l 4 0 1 JJ 3 7 0 3 t 0 4 2 0 0 4 0 9
- 0 4 1 8 44 - O l 7 t ) 4 -ll 7 1 4 0 b -0 3 5 7 8 Z
- 0 1 9 7 7 0 1)0 3 3 4
-o n 9 2
IJ A T A
CONV
ON-C ONV
( 4 )
t 1NV
C 1t-V C NV c ) v C - JNV C rlNV c n v c a v C ON V ( i) l V
C CtV
C t1tV C IJNV
C 1 V
c 1 v
Pa ge 2La
HEQ F t Noe x
c 5 )
6 2q
6 89 1 48
6 1 4 ) q 3
1 9 7 1 9
q z t s 5 a e z
l l 9 1
4 o 7 1 2 0
c oqt 1 Z
4 4 )-q 1 z
5 5 9 1 0 64 1 3 3
n 3
Inter industry analysi
Page 23
Equation ( 3 6 ) may be used to fo rmulate a rough test o f the cooperat ion
vs e f ficiency controversy I f we rest ricted ourselves to some subset o f
indus tries where the relat ion be tween y and a i s the same we would be lookingbull
at a sample o f indus tries for which the dependence o f V on H and o is a
quadratic wi th no squared terms
I f we now sub s titute (26) into ( 36 ) we get
(38) v bull a + (y - a) R + (y - a) o (H) - (y - a) H o (H) + u
If it is only d if ferences in marke t shares that affect advertising inten-
s i ty levels i f concentration does not af fect cooperation and i f o bull 0
( 38 ) reduces t o
( 39 ) V bull a + (y - a) H + U
Let this b e the null hyp o thes is and note that V is a linear func-
tion o f H If y gt a and the null hyp othesis holds the expected relat ion
between V and H is as shown by the pos itively sloped straight line in Figure
6 b etween points A and C
In this context the alterna ive hypothesis the degree o f
cooperation ( o) is a positive function o f concentrat ion
is that
(H) I f the func-
tion is as described in (26) and Figure 4 the relation between V and H
would be as shown by the curved line connec t ing po ints A and C in Figure 6 One way to test the null hypothe s is against this al ternative is to test fo r
the hypothesized curvature in the function relating V and H
Fi gure
R
6
v
ID c
H 1 0
I
i I I
sfigned
clasJ s
in def ini g
effec t is the
I
Page middot2lf
One charac teristic of produc ts which may be useful
classes wi th respec t to significanc e of the brand-switching
conveniencenon-convenienc e distinc tion developed by Porter ( 1 97 )
Using his identif ication of IRS indus tries as a s tar t ing point
each of the 32 industries used in this s tudy to one of the two
The ass ignments are not ed in Tab le 4
Using the conveniencenon-convenienc e dis t inc tion the
divided into two sub-samples
sample was
Four regressions were then run bull ene for e ch
sub-sample one for the to tal sample and one for the total sample with
a dummy variable for convenience goods Each regression was run wi th LB s
and wi th indus tries as ob servations The results are given in Tab le 5
For the LB level equations ( 1A - 1D) equation(37) was used ome
rearranging of terms gives v bull y - ( 1-o ) (y - a) ( 1-z ) + ui i i
bull y - 6 1 ( 1 - z ) + ui When observations are pooled across indus trif s yi
is an industry variable In these regressions the values o f y es t imated
in the first stage were used Since industries have been grouped so as
to reduce differences in (y - a) that term is t reated as a cons tant
within each group Finally H the Herfindahl index has been subst ituted
for o the degree of cooperation The resultant equation is linear
y ( 1 - z ) and H x ( l - z ) This is the equation for whi ch results arei i
reported in Table 5 In all regress ions all variables were weighted by the
square roo t of sales
in
raquo
- 34
3 1
i l T 4 8LE S R E G E S S I ON R E SUL T S i X T f N S I ON S T O T Y PE Of
AND L E V E L OF C ONC E kT R A T I O
E O O tt S
tP bull T Y P E NR C ONV O B S tO N-CONV
E L A S T I shy
T I C J T Y
O E - t-41( T SliA R E
H( fl F
1 1 - K O x middot K O X E R F
- H S a 1 1 - H S t 1 8 1 1 9 1
R S O
1 1 0 1
q z
t - bull 9 f 2 1
Z t P D l 6 fC ONV - 1 9 1 s s c 2 1
e - l O I e bull u q a
2 3 1 0 1 1 Nl N- t O V 5 7 1 ( 1 2 19 1 6 1 - J l S 2 5 4
l l l l t z q o a 1 - J o v z I
l ( I NO l l A l l ol u l t o o c bull J u l l
d S
c 3 4 t t r a l
bull Z D 1 0 l l 4 l l 2 3 -gt 2 l l 2 0 - 4 lS 5 4 l C 4 1 o e 6 t l b b 5 1 1 0 1 1 1 c - 1 9 5 1 t d S 2 1 tV
OQ10 T-R A T I O S A E G I VE N l P A RE N T H E S E S l E V E L S f S I G 1 F l C ANC E A - 99 5 B - q 9 t - s NS J UC E NO I Nl E R C ( P T I S U (j f middot) I N f Hf t OUA TIOI S 6 E f 3R E A O J U S T H E N T F OR 1-t E T E R O SIC E O A S T f C I T V pIraquo
1
middot
I
R SO I S C A L C UL A T E D A S f I f bull 1 1 bull 1lt 1 K t t- E E N I S THf U A1 B E R O f fJe S E fi V A T I O S F t S HtE F - S T A T I S
I N DE P E ND E N T y II
T I C F O T H E U Y P O HIE S I S f A l l C JE F F I C H N T A R E QU A L HI l f RO A f40 K I S T H f UUM 8 E R OF Q ( A fL f S
I l l
1 A
1 6
t C
f 1 r R I l l
1 2 1
L Jt
L B
l q
L 1
on s T Y P E bull 2 )
P OUC T
X
H S I l l
l O l A I bull l 0 7 1
O O b O J
1 2 00 1
1 b 2 1 C I l l Z U
l l S b 1 -1 1 0 1
rlt F X l l - fi C f l
bull 1 I
l l
- ()4 2 3 l f H
1 - 2 7 9 1
K O X
1 1 - ti E R F I
l d l
S4 l i A I 2 l b
1 3 00
K O X -t f R F t S J
I 1 - HER F t
)
60 9
z o bull
0
t l l
3 8 8
1 7 6
5 64
64
N f bull
tl S bull
I l l
I 4 1
t ONV
NON-t ONV
A l l
A ll
c o v NO- ( ONV
1 4 1
1 5 1
- l S b
c - 1 1 2 1
5 8 2 C A I 1 6 7 0 1
middot I l 9 H
l 1 il
1 4 1 6 1
E L A C) T J shy
l t C I T Y
I S I
l b l
- 0 2 0
1 - 1 4 9 1
O Z 3 2 1 A I C b 6 b l
0 0 9 2 -J
1 1 b 4 1
o z t t t l 1 2 2 1 1
oru - Ht tt F
1 gt X bull b I
Page 25 bull
Mean val -as ior relevan t variables for the two sub-samples are
Conv Non-Conv
Adv Sales 4 7 2 3
iit Share 2 7 3 2
Ela cicity - 05 3 02 2
Herf indahl 7 3 1 860
That the dist nct ion between coft9enienee goods and non - convenience
goods mat ters is supported not only by these means but also by the regression
results Equat ions lA and lB are different in almost every respec t Only the
concentrat ionmarket share int eract ion term seems to mat ter for convenience
good s Jus t he opposite holds for non - convenience goods both elasticity1
and market sl e are highly significant but the concentrat ionmarket share L
int erac t ion t=rm is not
Given that one minus market share is used these equ3 tions the coeffici t
of market share is the negative of what is sh in col (6) of Table 5 For the
regress ion for non - convenience LB s reported in equation lB then LB s with larger
market shares have lower advertising to sales ratios
On the question of the impact of the degree of cooperation the story is mixed
For convenienc good s the coefficient is pos itive as equation (31) predicts when
there is a po sitive degree of cooperat ion Fo r the non-convenience goods LB s middot
concentration has no effec t
Aggregat ion t o the indus try level and then rer unning the equations $ives the
results shown in (2a-2d) of Tab le 5 As expected R2 goes up and levels of significanc e
fmiddotJr individual coefficients gJ down In add ition the coeffic ient of H ( l -H) inmiddot
quat ion 2A is no t significant though it is po s i tive Since 2A is the indus try level
I
--
all but one of them and that one is not significant For 18 of the 31
wo uld have on its adve rtising to sal es ratio
cognate of LA the variable H ( l-H) pl ays the same role in 2A as H ( l -z 1) p lays
in LA
Vconomies of scale in advert ising
The literature on advert ing contains a number of definitionbull o f economies
of scale and of findings eone rning its presence and magnitude Before I loo k
a t the data with the hope o f mmenting about scale effect s then I wan t to
define the term as I use it
By economies of scale I mean a decrease in the number o f units of adver-
tising which are needed to generate a unit of sales as advert ising is increased
Given the definition of the firm s own elast icity of demand with respect to
advertising (see the text foll wing equation (1) ) the technical definition
is that rii gt 1 Diseconomies of scale are def ined symmetrically i e
Yu lt 1
ith a cro s s - section oi observations on advertising intensity
(v bull a s1) and onmiddot advertising (a ) we may regress the former on the latter 1 i i
The results of such regress ions for the 32 industries are given in Table 2
lines lB ZB bull bull bull 32B For the 32 eases the regression coefficient is positive it
positive coefficients the coefficient i s signif icant
The que tion now is whether these results ar e evidence of pervasive dis-
economies of scale in advertislng I think no t Given my as sumption that the
data I observe are equilibrium results I may ri ghtly conclud e that almost every-
where high levels of advert ising are as sociated in equilibrium with high ratios
of advertising to sales I may no t conclud e anything however about the
impact that a move by some firm away from its equilibrium level o f advertising
middot- Page 27
Given the assumed relation in equation ( 37 ) there is a corre spondshy
ing equilibrium equation relating v and ai The equation is non - linear i
and I have not found a simp le way to characterize it It is po ssib le to
det ermi ne its deriviative with respect to a b y takin$ tne de riviative of ( 3 7 ) i
When that is done the result is
v - 1(40) i bull sl a a z ( 1-z )i i ira-l
Sinc e all the other terms in (40) are non-negative the sign dep ends only
on the sign of 6 1 which is in turn de pend ent only on the sign of (y - a)
That the signs of the coefficients of a in equations lB 2B bull bull bull 32B arei
not always equal to the signs of the coefficients of z in LA 2A bull bull bull 32A i
is still a mystery to me and something to be explored
bull
Summary
My purpose in writing this pa per wa s t o attempt to integrate explicit micro-
economic mode lling reasoned econome tric specification of error terms and high
quality data to exp lore some questions about adver tising by large firms
al quite important improvements could be made in the first two of those areas
and much data ma ssag ing is yet to be done in the third
Concerning the subst an ce of what I have done I think six things are im-
Page 2 8
Severshy
V and Conclusions
portant
1 Explicit modeling is worth the ef for t if for no other reas on than
that it provides a basis for choosing fr om a varie ty of funct ional
forms
2 The same is true for error specif ica t ion I did in fact look at scatshy
ter diagrams for all 32 industry categor ies and they vir tually all show
the het eroscedast icity which I as sumed and for which I correc ted
3 The predicted relation be tween advertis ing in tensity and market share
shows up clearly in on ly 25 of the case s examined I have ye t to
explore why that may b e the cas e
4 Some evidence concerning the presence and impact of coopera tion was
produced but it is not clear ly pervasive
5 The dis tinction between convenience goods and non-c onvenie nce goods is
unambiguously a good one
6 Virtually no evidence concerning econ omies of scale in adver tis ing can
be gleaned fr om this study given its assumpt ions
More work is called for on many of these issues my study seems to have
raised more questions than it has answer ed This is pr obab lly due to the richshy
ne ss of the data source since I had the oppor tunity to address several issue s
at once
- -- - Q --
Foo tnotes
Manager Line o f Business Program Federal Trade Commission The author depended heavily on several staf f members of the LB Pro gram for statistical and clerical support in particular Joe Cholka George Pascoe and Haro lene Jenkins Helpful comments were given b y Richard Caves Dennis Mueller Michael Porter and F M Scherer When the paper was in its early stages o f development both S teve Garber and Jon Rasmussen former economists in the LB Program provided eery useful cri tiques The views expressed here are my own o f course
1 The basic exp licit model of the role of advertising is Dorfman and S teiner ( 1954 ) and the models in this paper may be seen as ano ther extens ion of their work
2 For a review o f models of oligopoly which treat prices and quantities as relevant variables see Shubik ( 1 959) Extens ions to take advertis ing int o account are fairly s traightforward
3 Dorfman and S teiner dealt with a price-setting firm For a monopolis t inverting the demand relation to show Pbull P (Q A) _instead o f Q- (P A) has no effect on the results o f the analysis For oligopolists who face a sys tem of demand relations inversion does have some impac t on the results The dif ferences are not trivial nonetheless they will no t be explored in this study
4 Shubik ( 1 959) p
5 From a formal mathematical point o f view the function is similar to a1 i Legrangian funct ion of the form - r rr
L bull 11 + A ( j - (i i i+j
where the indicates some fixed level of the profits industry Following the same logic which resul ts in and income constraint as the marginal utility of the of the indus tr in the obj ective of firm i
6 Hall amp Weiss ( 1 967) p 323
7 See Schmallensee ( 1 97 2 ) Ch 2 pp 1 6-47
8 On this poin t see Imel amp Helmberger ( 1 9 7 1 )
9 Long ( 1 970) p 130 - 1 39
1 0 See Schmallensee ( 1 972) Ch 3 pp 48-8 7
rrj i+j ) of the rest of the
the identification profits of the rest
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980
Page 21
elasticity of demand does
(10) we can
if t 0 then
Finally a determination concerning the industry
not fix the individual firms elasticity Going back to equation
write y bull (1 - ) a + z y Since a gt 0 it follows thatii i i
Y gt 0 If Ylt 0 however y s sign is indeterminateij ij
Turning next to the coefficient of market share 20 of them are negative and
12 are positive with three of each sign being significantly different from zero
The largest positive value which is significant is for pet food (code 2010) and
a close second is mens amp boys suits and coats (code 23 01) The largest signifi
cant negative value is for cosmetics (code 28 09)
In the model some relations among y a 0 and 61 are implied If middota is
negative (1- y) is greater than one Since 61Y bull (1- a) (1 - ) 61Y is
pCsitive but it will be less than one if 0 is large enough Seventeen of the A
ys are negative in each of the seventeen industries 61 is also negative and les
1 (algebraicly) than y so 61Y is greater than one his is not evidence that the
degree of cooperation is low or zero in any industry of course A high o togethegt
with a negative y which is small (in absolute terms) relative to a can give a
value of 61Y greater than one On the other hand if 61Y were less than one
o would have to be greater than zero
For y gt 0 two conditions may hold If y lt a it follows that 61 lt 0 since
(y - a ) lt 0 It also follows that 61Y lt 0 regardless of the value of o bull
In addition the relation is an if-and-only-if one ie if 61Y is negative
then y lt a Three industries have negative values for 61y dairy products
except milk (code 20 04) synthetic fibers (code 2806) and photographic equipment
(code 3808) In all three cases y is quite small so a does not have to be very
large to give a negative value to al
If y gt a the model shows that 61-- gt 0 whatever the value of 0 bull There are
12 cases where both y and 61 are positive with 61 ranging from near zero
(grain milling code 20 12) to 1 20 ( proprietary drugs code 2808) Several
coeffiei-en_rs
--
15
o f to zero
Tab le 3 - Consumer Goods
Indus tr ies
2Distributions of R and Producer Goods
Page 20a
Consumer Producer
Goods Goods
0 - 9 9 0 3
10 - 19 9 0 1 1
20 - 29 9 3 20
30 - 39 9 3 17
40 - 49 9 5 16
50 - 59 9 8
69 9 6 Ii 1060
--f 670 79 9 2
80 - 89 9 4 4
90 - 100 1 1
Totals 32 10 3
Notes
2is used R i s cal-
culated as F F + ((N - K- 1 ) K ] 1 b_ F is the F
statistic Lor the hypothesis that all o f the non - intercept terms are simul taneously equal
1
2
--
-- Since generalized least squ-area regress ion
Page 12
o f these cases have values o f 6 1 Y rea ter than 0 8 (pet food (20 10) bread amp cakes
(20 14 ) miscellaneous processed foods (20 29 ) men s and boys suit s and coats
(23 01 ) p rop riet ary drugs (28 08 ) elect ric housewares (36 1 2) radio and
IV sets 17 and games and toys (39 For be than(36 ) 04)) 6 1 Y to greater 0 8
both o and cry mus t b e less than 0 2 That is in these industries the results
shown in Tab le 2 would hold only with very strong dominance of marke t demand
effects over brand swit ching e ffects and a very low degree o f cooperation
Using the concentration rat ios given in Table 4 the average concentration
ratio for the eight indust ries is 38 1 For the remaining 24 industries it is
43 2 This result would lend some support t o the proposition that concentrat ion
and cooperation are related if the group of eight industries are identi fied
as having low degrees of cooperation
One f inal note on model predict ions S ince both cr and o are assumed to
be non-negative 60 should also be non-negat ive In only one o f the 32 cas es
is 60 negative and it is not signif icant bull
cor pJ
7 1
7 0 4
1 4 7 5 1
)
7 3 z a o a t s 1 1 3 3 5 ( i V Z 8 tJ1 2 1 - 1 1 0 7 3 5 c middot v 2 8 1 0 3 2 -0 0 1 4 3 0 C 1 V 9 8 3 2 8 1 5 1 6 - 0 0 6 t1 6 9 C _ V 1 11 0 2 9 0 1 29 0 00 7 6 1 C l k V 7 3 1 3 6 08 1 3 - o n 4 9 9 4 3 3 3 6 1 2 2 4 0 1 6 4 3
3 1 7 1 3 0 1 1 7 d 7 8 3 3 8 0 8 0 0 2 3 i 2 3 0 6
P 4
3 9 0 3 1 4
1 2
0 09 3 8 4 39 1)1 0 4 l 8 6 5
T A LE 4 e I NCU S T R Y C A T E G O R Y
F TC NOM e tOOE ass
C 1 I c 2 )
2 0 0 1 1 6 2 0 0r 1 7 zo o l Z 2 0 0 7 1 8
z o o 2 6
z o t a 1 6
2 1) 1 2 1 5 2 0 1 4 1 1 2 0 1 a 1 9 2 0 26 1 2 2 0 2 7 1 0 2 0 29 - 3 5
2 3 0 1 10
2 3 02 z o
2 3 03 1 9
2 5 5 1 2 4 Z 1 0 Z 1 1 2 0 3 1 9 2 7 4 1 0 z a o 1 3
2 A 0 7 z s
E LA S T I -C I T Y
c 3 )
- 0 09 7 2 7
0 00 449
- 0 0 2 8 9 3
- 0 3 2 7 70 - 0 OZ il I q
o 4 u t s a 0 0 1 3 7 3
o 1 4 o o-0 0 5 2 4o -0 1 94 0 1
-C l 6 l l 4 0 1 JJ 3 7 0 3 t 0 4 2 0 0 4 0 9
- 0 4 1 8 44 - O l 7 t ) 4 -ll 7 1 4 0 b -0 3 5 7 8 Z
- 0 1 9 7 7 0 1)0 3 3 4
-o n 9 2
IJ A T A
CONV
ON-C ONV
( 4 )
t 1NV
C 1t-V C NV c ) v C - JNV C rlNV c n v c a v C ON V ( i) l V
C CtV
C t1tV C IJNV
C 1 V
c 1 v
Pa ge 2La
HEQ F t Noe x
c 5 )
6 2q
6 89 1 48
6 1 4 ) q 3
1 9 7 1 9
q z t s 5 a e z
l l 9 1
4 o 7 1 2 0
c oqt 1 Z
4 4 )-q 1 z
5 5 9 1 0 64 1 3 3
n 3
Inter industry analysi
Page 23
Equation ( 3 6 ) may be used to fo rmulate a rough test o f the cooperat ion
vs e f ficiency controversy I f we rest ricted ourselves to some subset o f
indus tries where the relat ion be tween y and a i s the same we would be lookingbull
at a sample o f indus tries for which the dependence o f V on H and o is a
quadratic wi th no squared terms
I f we now sub s titute (26) into ( 36 ) we get
(38) v bull a + (y - a) R + (y - a) o (H) - (y - a) H o (H) + u
If it is only d if ferences in marke t shares that affect advertising inten-
s i ty levels i f concentration does not af fect cooperation and i f o bull 0
( 38 ) reduces t o
( 39 ) V bull a + (y - a) H + U
Let this b e the null hyp o thes is and note that V is a linear func-
tion o f H If y gt a and the null hyp othesis holds the expected relat ion
between V and H is as shown by the pos itively sloped straight line in Figure
6 b etween points A and C
In this context the alterna ive hypothesis the degree o f
cooperation ( o) is a positive function o f concentrat ion
is that
(H) I f the func-
tion is as described in (26) and Figure 4 the relation between V and H
would be as shown by the curved line connec t ing po ints A and C in Figure 6 One way to test the null hypothe s is against this al ternative is to test fo r
the hypothesized curvature in the function relating V and H
Fi gure
R
6
v
ID c
H 1 0
I
i I I
sfigned
clasJ s
in def ini g
effec t is the
I
Page middot2lf
One charac teristic of produc ts which may be useful
classes wi th respec t to significanc e of the brand-switching
conveniencenon-convenienc e distinc tion developed by Porter ( 1 97 )
Using his identif ication of IRS indus tries as a s tar t ing point
each of the 32 industries used in this s tudy to one of the two
The ass ignments are not ed in Tab le 4
Using the conveniencenon-convenienc e dis t inc tion the
divided into two sub-samples
sample was
Four regressions were then run bull ene for e ch
sub-sample one for the to tal sample and one for the total sample with
a dummy variable for convenience goods Each regression was run wi th LB s
and wi th indus tries as ob servations The results are given in Tab le 5
For the LB level equations ( 1A - 1D) equation(37) was used ome
rearranging of terms gives v bull y - ( 1-o ) (y - a) ( 1-z ) + ui i i
bull y - 6 1 ( 1 - z ) + ui When observations are pooled across indus trif s yi
is an industry variable In these regressions the values o f y es t imated
in the first stage were used Since industries have been grouped so as
to reduce differences in (y - a) that term is t reated as a cons tant
within each group Finally H the Herfindahl index has been subst ituted
for o the degree of cooperation The resultant equation is linear
y ( 1 - z ) and H x ( l - z ) This is the equation for whi ch results arei i
reported in Table 5 In all regress ions all variables were weighted by the
square roo t of sales
in
raquo
- 34
3 1
i l T 4 8LE S R E G E S S I ON R E SUL T S i X T f N S I ON S T O T Y PE Of
AND L E V E L OF C ONC E kT R A T I O
E O O tt S
tP bull T Y P E NR C ONV O B S tO N-CONV
E L A S T I shy
T I C J T Y
O E - t-41( T SliA R E
H( fl F
1 1 - K O x middot K O X E R F
- H S a 1 1 - H S t 1 8 1 1 9 1
R S O
1 1 0 1
q z
t - bull 9 f 2 1
Z t P D l 6 fC ONV - 1 9 1 s s c 2 1
e - l O I e bull u q a
2 3 1 0 1 1 Nl N- t O V 5 7 1 ( 1 2 19 1 6 1 - J l S 2 5 4
l l l l t z q o a 1 - J o v z I
l ( I NO l l A l l ol u l t o o c bull J u l l
d S
c 3 4 t t r a l
bull Z D 1 0 l l 4 l l 2 3 -gt 2 l l 2 0 - 4 lS 5 4 l C 4 1 o e 6 t l b b 5 1 1 0 1 1 1 c - 1 9 5 1 t d S 2 1 tV
OQ10 T-R A T I O S A E G I VE N l P A RE N T H E S E S l E V E L S f S I G 1 F l C ANC E A - 99 5 B - q 9 t - s NS J UC E NO I Nl E R C ( P T I S U (j f middot) I N f Hf t OUA TIOI S 6 E f 3R E A O J U S T H E N T F OR 1-t E T E R O SIC E O A S T f C I T V pIraquo
1
middot
I
R SO I S C A L C UL A T E D A S f I f bull 1 1 bull 1lt 1 K t t- E E N I S THf U A1 B E R O f fJe S E fi V A T I O S F t S HtE F - S T A T I S
I N DE P E ND E N T y II
T I C F O T H E U Y P O HIE S I S f A l l C JE F F I C H N T A R E QU A L HI l f RO A f40 K I S T H f UUM 8 E R OF Q ( A fL f S
I l l
1 A
1 6
t C
f 1 r R I l l
1 2 1
L Jt
L B
l q
L 1
on s T Y P E bull 2 )
P OUC T
X
H S I l l
l O l A I bull l 0 7 1
O O b O J
1 2 00 1
1 b 2 1 C I l l Z U
l l S b 1 -1 1 0 1
rlt F X l l - fi C f l
bull 1 I
l l
- ()4 2 3 l f H
1 - 2 7 9 1
K O X
1 1 - ti E R F I
l d l
S4 l i A I 2 l b
1 3 00
K O X -t f R F t S J
I 1 - HER F t
)
60 9
z o bull
0
t l l
3 8 8
1 7 6
5 64
64
N f bull
tl S bull
I l l
I 4 1
t ONV
NON-t ONV
A l l
A ll
c o v NO- ( ONV
1 4 1
1 5 1
- l S b
c - 1 1 2 1
5 8 2 C A I 1 6 7 0 1
middot I l 9 H
l 1 il
1 4 1 6 1
E L A C) T J shy
l t C I T Y
I S I
l b l
- 0 2 0
1 - 1 4 9 1
O Z 3 2 1 A I C b 6 b l
0 0 9 2 -J
1 1 b 4 1
o z t t t l 1 2 2 1 1
oru - Ht tt F
1 gt X bull b I
Page 25 bull
Mean val -as ior relevan t variables for the two sub-samples are
Conv Non-Conv
Adv Sales 4 7 2 3
iit Share 2 7 3 2
Ela cicity - 05 3 02 2
Herf indahl 7 3 1 860
That the dist nct ion between coft9enienee goods and non - convenience
goods mat ters is supported not only by these means but also by the regression
results Equat ions lA and lB are different in almost every respec t Only the
concentrat ionmarket share int eract ion term seems to mat ter for convenience
good s Jus t he opposite holds for non - convenience goods both elasticity1
and market sl e are highly significant but the concentrat ionmarket share L
int erac t ion t=rm is not
Given that one minus market share is used these equ3 tions the coeffici t
of market share is the negative of what is sh in col (6) of Table 5 For the
regress ion for non - convenience LB s reported in equation lB then LB s with larger
market shares have lower advertising to sales ratios
On the question of the impact of the degree of cooperation the story is mixed
For convenienc good s the coefficient is pos itive as equation (31) predicts when
there is a po sitive degree of cooperat ion Fo r the non-convenience goods LB s middot
concentration has no effec t
Aggregat ion t o the indus try level and then rer unning the equations $ives the
results shown in (2a-2d) of Tab le 5 As expected R2 goes up and levels of significanc e
fmiddotJr individual coefficients gJ down In add ition the coeffic ient of H ( l -H) inmiddot
quat ion 2A is no t significant though it is po s i tive Since 2A is the indus try level
I
--
all but one of them and that one is not significant For 18 of the 31
wo uld have on its adve rtising to sal es ratio
cognate of LA the variable H ( l-H) pl ays the same role in 2A as H ( l -z 1) p lays
in LA
Vconomies of scale in advert ising
The literature on advert ing contains a number of definitionbull o f economies
of scale and of findings eone rning its presence and magnitude Before I loo k
a t the data with the hope o f mmenting about scale effect s then I wan t to
define the term as I use it
By economies of scale I mean a decrease in the number o f units of adver-
tising which are needed to generate a unit of sales as advert ising is increased
Given the definition of the firm s own elast icity of demand with respect to
advertising (see the text foll wing equation (1) ) the technical definition
is that rii gt 1 Diseconomies of scale are def ined symmetrically i e
Yu lt 1
ith a cro s s - section oi observations on advertising intensity
(v bull a s1) and onmiddot advertising (a ) we may regress the former on the latter 1 i i
The results of such regress ions for the 32 industries are given in Table 2
lines lB ZB bull bull bull 32B For the 32 eases the regression coefficient is positive it
positive coefficients the coefficient i s signif icant
The que tion now is whether these results ar e evidence of pervasive dis-
economies of scale in advertislng I think no t Given my as sumption that the
data I observe are equilibrium results I may ri ghtly conclud e that almost every-
where high levels of advert ising are as sociated in equilibrium with high ratios
of advertising to sales I may no t conclud e anything however about the
impact that a move by some firm away from its equilibrium level o f advertising
middot- Page 27
Given the assumed relation in equation ( 37 ) there is a corre spondshy
ing equilibrium equation relating v and ai The equation is non - linear i
and I have not found a simp le way to characterize it It is po ssib le to
det ermi ne its deriviative with respect to a b y takin$ tne de riviative of ( 3 7 ) i
When that is done the result is
v - 1(40) i bull sl a a z ( 1-z )i i ira-l
Sinc e all the other terms in (40) are non-negative the sign dep ends only
on the sign of 6 1 which is in turn de pend ent only on the sign of (y - a)
That the signs of the coefficients of a in equations lB 2B bull bull bull 32B arei
not always equal to the signs of the coefficients of z in LA 2A bull bull bull 32A i
is still a mystery to me and something to be explored
bull
Summary
My purpose in writing this pa per wa s t o attempt to integrate explicit micro-
economic mode lling reasoned econome tric specification of error terms and high
quality data to exp lore some questions about adver tising by large firms
al quite important improvements could be made in the first two of those areas
and much data ma ssag ing is yet to be done in the third
Concerning the subst an ce of what I have done I think six things are im-
Page 2 8
Severshy
V and Conclusions
portant
1 Explicit modeling is worth the ef for t if for no other reas on than
that it provides a basis for choosing fr om a varie ty of funct ional
forms
2 The same is true for error specif ica t ion I did in fact look at scatshy
ter diagrams for all 32 industry categor ies and they vir tually all show
the het eroscedast icity which I as sumed and for which I correc ted
3 The predicted relation be tween advertis ing in tensity and market share
shows up clearly in on ly 25 of the case s examined I have ye t to
explore why that may b e the cas e
4 Some evidence concerning the presence and impact of coopera tion was
produced but it is not clear ly pervasive
5 The dis tinction between convenience goods and non-c onvenie nce goods is
unambiguously a good one
6 Virtually no evidence concerning econ omies of scale in adver tis ing can
be gleaned fr om this study given its assumpt ions
More work is called for on many of these issues my study seems to have
raised more questions than it has answer ed This is pr obab lly due to the richshy
ne ss of the data source since I had the oppor tunity to address several issue s
at once
- -- - Q --
Foo tnotes
Manager Line o f Business Program Federal Trade Commission The author depended heavily on several staf f members of the LB Pro gram for statistical and clerical support in particular Joe Cholka George Pascoe and Haro lene Jenkins Helpful comments were given b y Richard Caves Dennis Mueller Michael Porter and F M Scherer When the paper was in its early stages o f development both S teve Garber and Jon Rasmussen former economists in the LB Program provided eery useful cri tiques The views expressed here are my own o f course
1 The basic exp licit model of the role of advertising is Dorfman and S teiner ( 1954 ) and the models in this paper may be seen as ano ther extens ion of their work
2 For a review o f models of oligopoly which treat prices and quantities as relevant variables see Shubik ( 1 959) Extens ions to take advertis ing int o account are fairly s traightforward
3 Dorfman and S teiner dealt with a price-setting firm For a monopolis t inverting the demand relation to show Pbull P (Q A) _instead o f Q- (P A) has no effect on the results o f the analysis For oligopolists who face a sys tem of demand relations inversion does have some impac t on the results The dif ferences are not trivial nonetheless they will no t be explored in this study
4 Shubik ( 1 959) p
5 From a formal mathematical point o f view the function is similar to a1 i Legrangian funct ion of the form - r rr
L bull 11 + A ( j - (i i i+j
where the indicates some fixed level of the profits industry Following the same logic which resul ts in and income constraint as the marginal utility of the of the indus tr in the obj ective of firm i
6 Hall amp Weiss ( 1 967) p 323
7 See Schmallensee ( 1 97 2 ) Ch 2 pp 1 6-47
8 On this poin t see Imel amp Helmberger ( 1 9 7 1 )
9 Long ( 1 970) p 130 - 1 39
1 0 See Schmallensee ( 1 972) Ch 3 pp 48-8 7
rrj i+j ) of the rest of the
the identification profits of the rest
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980
coeffiei-en_rs
--
15
o f to zero
Tab le 3 - Consumer Goods
Indus tr ies
2Distributions of R and Producer Goods
Page 20a
Consumer Producer
Goods Goods
0 - 9 9 0 3
10 - 19 9 0 1 1
20 - 29 9 3 20
30 - 39 9 3 17
40 - 49 9 5 16
50 - 59 9 8
69 9 6 Ii 1060
--f 670 79 9 2
80 - 89 9 4 4
90 - 100 1 1
Totals 32 10 3
Notes
2is used R i s cal-
culated as F F + ((N - K- 1 ) K ] 1 b_ F is the F
statistic Lor the hypothesis that all o f the non - intercept terms are simul taneously equal
1
2
--
-- Since generalized least squ-area regress ion
Page 12
o f these cases have values o f 6 1 Y rea ter than 0 8 (pet food (20 10) bread amp cakes
(20 14 ) miscellaneous processed foods (20 29 ) men s and boys suit s and coats
(23 01 ) p rop riet ary drugs (28 08 ) elect ric housewares (36 1 2) radio and
IV sets 17 and games and toys (39 For be than(36 ) 04)) 6 1 Y to greater 0 8
both o and cry mus t b e less than 0 2 That is in these industries the results
shown in Tab le 2 would hold only with very strong dominance of marke t demand
effects over brand swit ching e ffects and a very low degree o f cooperation
Using the concentration rat ios given in Table 4 the average concentration
ratio for the eight indust ries is 38 1 For the remaining 24 industries it is
43 2 This result would lend some support t o the proposition that concentrat ion
and cooperation are related if the group of eight industries are identi fied
as having low degrees of cooperation
One f inal note on model predict ions S ince both cr and o are assumed to
be non-negative 60 should also be non-negat ive In only one o f the 32 cas es
is 60 negative and it is not signif icant bull
cor pJ
7 1
7 0 4
1 4 7 5 1
)
7 3 z a o a t s 1 1 3 3 5 ( i V Z 8 tJ1 2 1 - 1 1 0 7 3 5 c middot v 2 8 1 0 3 2 -0 0 1 4 3 0 C 1 V 9 8 3 2 8 1 5 1 6 - 0 0 6 t1 6 9 C _ V 1 11 0 2 9 0 1 29 0 00 7 6 1 C l k V 7 3 1 3 6 08 1 3 - o n 4 9 9 4 3 3 3 6 1 2 2 4 0 1 6 4 3
3 1 7 1 3 0 1 1 7 d 7 8 3 3 8 0 8 0 0 2 3 i 2 3 0 6
P 4
3 9 0 3 1 4
1 2
0 09 3 8 4 39 1)1 0 4 l 8 6 5
T A LE 4 e I NCU S T R Y C A T E G O R Y
F TC NOM e tOOE ass
C 1 I c 2 )
2 0 0 1 1 6 2 0 0r 1 7 zo o l Z 2 0 0 7 1 8
z o o 2 6
z o t a 1 6
2 1) 1 2 1 5 2 0 1 4 1 1 2 0 1 a 1 9 2 0 26 1 2 2 0 2 7 1 0 2 0 29 - 3 5
2 3 0 1 10
2 3 02 z o
2 3 03 1 9
2 5 5 1 2 4 Z 1 0 Z 1 1 2 0 3 1 9 2 7 4 1 0 z a o 1 3
2 A 0 7 z s
E LA S T I -C I T Y
c 3 )
- 0 09 7 2 7
0 00 449
- 0 0 2 8 9 3
- 0 3 2 7 70 - 0 OZ il I q
o 4 u t s a 0 0 1 3 7 3
o 1 4 o o-0 0 5 2 4o -0 1 94 0 1
-C l 6 l l 4 0 1 JJ 3 7 0 3 t 0 4 2 0 0 4 0 9
- 0 4 1 8 44 - O l 7 t ) 4 -ll 7 1 4 0 b -0 3 5 7 8 Z
- 0 1 9 7 7 0 1)0 3 3 4
-o n 9 2
IJ A T A
CONV
ON-C ONV
( 4 )
t 1NV
C 1t-V C NV c ) v C - JNV C rlNV c n v c a v C ON V ( i) l V
C CtV
C t1tV C IJNV
C 1 V
c 1 v
Pa ge 2La
HEQ F t Noe x
c 5 )
6 2q
6 89 1 48
6 1 4 ) q 3
1 9 7 1 9
q z t s 5 a e z
l l 9 1
4 o 7 1 2 0
c oqt 1 Z
4 4 )-q 1 z
5 5 9 1 0 64 1 3 3
n 3
Inter industry analysi
Page 23
Equation ( 3 6 ) may be used to fo rmulate a rough test o f the cooperat ion
vs e f ficiency controversy I f we rest ricted ourselves to some subset o f
indus tries where the relat ion be tween y and a i s the same we would be lookingbull
at a sample o f indus tries for which the dependence o f V on H and o is a
quadratic wi th no squared terms
I f we now sub s titute (26) into ( 36 ) we get
(38) v bull a + (y - a) R + (y - a) o (H) - (y - a) H o (H) + u
If it is only d if ferences in marke t shares that affect advertising inten-
s i ty levels i f concentration does not af fect cooperation and i f o bull 0
( 38 ) reduces t o
( 39 ) V bull a + (y - a) H + U
Let this b e the null hyp o thes is and note that V is a linear func-
tion o f H If y gt a and the null hyp othesis holds the expected relat ion
between V and H is as shown by the pos itively sloped straight line in Figure
6 b etween points A and C
In this context the alterna ive hypothesis the degree o f
cooperation ( o) is a positive function o f concentrat ion
is that
(H) I f the func-
tion is as described in (26) and Figure 4 the relation between V and H
would be as shown by the curved line connec t ing po ints A and C in Figure 6 One way to test the null hypothe s is against this al ternative is to test fo r
the hypothesized curvature in the function relating V and H
Fi gure
R
6
v
ID c
H 1 0
I
i I I
sfigned
clasJ s
in def ini g
effec t is the
I
Page middot2lf
One charac teristic of produc ts which may be useful
classes wi th respec t to significanc e of the brand-switching
conveniencenon-convenienc e distinc tion developed by Porter ( 1 97 )
Using his identif ication of IRS indus tries as a s tar t ing point
each of the 32 industries used in this s tudy to one of the two
The ass ignments are not ed in Tab le 4
Using the conveniencenon-convenienc e dis t inc tion the
divided into two sub-samples
sample was
Four regressions were then run bull ene for e ch
sub-sample one for the to tal sample and one for the total sample with
a dummy variable for convenience goods Each regression was run wi th LB s
and wi th indus tries as ob servations The results are given in Tab le 5
For the LB level equations ( 1A - 1D) equation(37) was used ome
rearranging of terms gives v bull y - ( 1-o ) (y - a) ( 1-z ) + ui i i
bull y - 6 1 ( 1 - z ) + ui When observations are pooled across indus trif s yi
is an industry variable In these regressions the values o f y es t imated
in the first stage were used Since industries have been grouped so as
to reduce differences in (y - a) that term is t reated as a cons tant
within each group Finally H the Herfindahl index has been subst ituted
for o the degree of cooperation The resultant equation is linear
y ( 1 - z ) and H x ( l - z ) This is the equation for whi ch results arei i
reported in Table 5 In all regress ions all variables were weighted by the
square roo t of sales
in
raquo
- 34
3 1
i l T 4 8LE S R E G E S S I ON R E SUL T S i X T f N S I ON S T O T Y PE Of
AND L E V E L OF C ONC E kT R A T I O
E O O tt S
tP bull T Y P E NR C ONV O B S tO N-CONV
E L A S T I shy
T I C J T Y
O E - t-41( T SliA R E
H( fl F
1 1 - K O x middot K O X E R F
- H S a 1 1 - H S t 1 8 1 1 9 1
R S O
1 1 0 1
q z
t - bull 9 f 2 1
Z t P D l 6 fC ONV - 1 9 1 s s c 2 1
e - l O I e bull u q a
2 3 1 0 1 1 Nl N- t O V 5 7 1 ( 1 2 19 1 6 1 - J l S 2 5 4
l l l l t z q o a 1 - J o v z I
l ( I NO l l A l l ol u l t o o c bull J u l l
d S
c 3 4 t t r a l
bull Z D 1 0 l l 4 l l 2 3 -gt 2 l l 2 0 - 4 lS 5 4 l C 4 1 o e 6 t l b b 5 1 1 0 1 1 1 c - 1 9 5 1 t d S 2 1 tV
OQ10 T-R A T I O S A E G I VE N l P A RE N T H E S E S l E V E L S f S I G 1 F l C ANC E A - 99 5 B - q 9 t - s NS J UC E NO I Nl E R C ( P T I S U (j f middot) I N f Hf t OUA TIOI S 6 E f 3R E A O J U S T H E N T F OR 1-t E T E R O SIC E O A S T f C I T V pIraquo
1
middot
I
R SO I S C A L C UL A T E D A S f I f bull 1 1 bull 1lt 1 K t t- E E N I S THf U A1 B E R O f fJe S E fi V A T I O S F t S HtE F - S T A T I S
I N DE P E ND E N T y II
T I C F O T H E U Y P O HIE S I S f A l l C JE F F I C H N T A R E QU A L HI l f RO A f40 K I S T H f UUM 8 E R OF Q ( A fL f S
I l l
1 A
1 6
t C
f 1 r R I l l
1 2 1
L Jt
L B
l q
L 1
on s T Y P E bull 2 )
P OUC T
X
H S I l l
l O l A I bull l 0 7 1
O O b O J
1 2 00 1
1 b 2 1 C I l l Z U
l l S b 1 -1 1 0 1
rlt F X l l - fi C f l
bull 1 I
l l
- ()4 2 3 l f H
1 - 2 7 9 1
K O X
1 1 - ti E R F I
l d l
S4 l i A I 2 l b
1 3 00
K O X -t f R F t S J
I 1 - HER F t
)
60 9
z o bull
0
t l l
3 8 8
1 7 6
5 64
64
N f bull
tl S bull
I l l
I 4 1
t ONV
NON-t ONV
A l l
A ll
c o v NO- ( ONV
1 4 1
1 5 1
- l S b
c - 1 1 2 1
5 8 2 C A I 1 6 7 0 1
middot I l 9 H
l 1 il
1 4 1 6 1
E L A C) T J shy
l t C I T Y
I S I
l b l
- 0 2 0
1 - 1 4 9 1
O Z 3 2 1 A I C b 6 b l
0 0 9 2 -J
1 1 b 4 1
o z t t t l 1 2 2 1 1
oru - Ht tt F
1 gt X bull b I
Page 25 bull
Mean val -as ior relevan t variables for the two sub-samples are
Conv Non-Conv
Adv Sales 4 7 2 3
iit Share 2 7 3 2
Ela cicity - 05 3 02 2
Herf indahl 7 3 1 860
That the dist nct ion between coft9enienee goods and non - convenience
goods mat ters is supported not only by these means but also by the regression
results Equat ions lA and lB are different in almost every respec t Only the
concentrat ionmarket share int eract ion term seems to mat ter for convenience
good s Jus t he opposite holds for non - convenience goods both elasticity1
and market sl e are highly significant but the concentrat ionmarket share L
int erac t ion t=rm is not
Given that one minus market share is used these equ3 tions the coeffici t
of market share is the negative of what is sh in col (6) of Table 5 For the
regress ion for non - convenience LB s reported in equation lB then LB s with larger
market shares have lower advertising to sales ratios
On the question of the impact of the degree of cooperation the story is mixed
For convenienc good s the coefficient is pos itive as equation (31) predicts when
there is a po sitive degree of cooperat ion Fo r the non-convenience goods LB s middot
concentration has no effec t
Aggregat ion t o the indus try level and then rer unning the equations $ives the
results shown in (2a-2d) of Tab le 5 As expected R2 goes up and levels of significanc e
fmiddotJr individual coefficients gJ down In add ition the coeffic ient of H ( l -H) inmiddot
quat ion 2A is no t significant though it is po s i tive Since 2A is the indus try level
I
--
all but one of them and that one is not significant For 18 of the 31
wo uld have on its adve rtising to sal es ratio
cognate of LA the variable H ( l-H) pl ays the same role in 2A as H ( l -z 1) p lays
in LA
Vconomies of scale in advert ising
The literature on advert ing contains a number of definitionbull o f economies
of scale and of findings eone rning its presence and magnitude Before I loo k
a t the data with the hope o f mmenting about scale effect s then I wan t to
define the term as I use it
By economies of scale I mean a decrease in the number o f units of adver-
tising which are needed to generate a unit of sales as advert ising is increased
Given the definition of the firm s own elast icity of demand with respect to
advertising (see the text foll wing equation (1) ) the technical definition
is that rii gt 1 Diseconomies of scale are def ined symmetrically i e
Yu lt 1
ith a cro s s - section oi observations on advertising intensity
(v bull a s1) and onmiddot advertising (a ) we may regress the former on the latter 1 i i
The results of such regress ions for the 32 industries are given in Table 2
lines lB ZB bull bull bull 32B For the 32 eases the regression coefficient is positive it
positive coefficients the coefficient i s signif icant
The que tion now is whether these results ar e evidence of pervasive dis-
economies of scale in advertislng I think no t Given my as sumption that the
data I observe are equilibrium results I may ri ghtly conclud e that almost every-
where high levels of advert ising are as sociated in equilibrium with high ratios
of advertising to sales I may no t conclud e anything however about the
impact that a move by some firm away from its equilibrium level o f advertising
middot- Page 27
Given the assumed relation in equation ( 37 ) there is a corre spondshy
ing equilibrium equation relating v and ai The equation is non - linear i
and I have not found a simp le way to characterize it It is po ssib le to
det ermi ne its deriviative with respect to a b y takin$ tne de riviative of ( 3 7 ) i
When that is done the result is
v - 1(40) i bull sl a a z ( 1-z )i i ira-l
Sinc e all the other terms in (40) are non-negative the sign dep ends only
on the sign of 6 1 which is in turn de pend ent only on the sign of (y - a)
That the signs of the coefficients of a in equations lB 2B bull bull bull 32B arei
not always equal to the signs of the coefficients of z in LA 2A bull bull bull 32A i
is still a mystery to me and something to be explored
bull
Summary
My purpose in writing this pa per wa s t o attempt to integrate explicit micro-
economic mode lling reasoned econome tric specification of error terms and high
quality data to exp lore some questions about adver tising by large firms
al quite important improvements could be made in the first two of those areas
and much data ma ssag ing is yet to be done in the third
Concerning the subst an ce of what I have done I think six things are im-
Page 2 8
Severshy
V and Conclusions
portant
1 Explicit modeling is worth the ef for t if for no other reas on than
that it provides a basis for choosing fr om a varie ty of funct ional
forms
2 The same is true for error specif ica t ion I did in fact look at scatshy
ter diagrams for all 32 industry categor ies and they vir tually all show
the het eroscedast icity which I as sumed and for which I correc ted
3 The predicted relation be tween advertis ing in tensity and market share
shows up clearly in on ly 25 of the case s examined I have ye t to
explore why that may b e the cas e
4 Some evidence concerning the presence and impact of coopera tion was
produced but it is not clear ly pervasive
5 The dis tinction between convenience goods and non-c onvenie nce goods is
unambiguously a good one
6 Virtually no evidence concerning econ omies of scale in adver tis ing can
be gleaned fr om this study given its assumpt ions
More work is called for on many of these issues my study seems to have
raised more questions than it has answer ed This is pr obab lly due to the richshy
ne ss of the data source since I had the oppor tunity to address several issue s
at once
- -- - Q --
Foo tnotes
Manager Line o f Business Program Federal Trade Commission The author depended heavily on several staf f members of the LB Pro gram for statistical and clerical support in particular Joe Cholka George Pascoe and Haro lene Jenkins Helpful comments were given b y Richard Caves Dennis Mueller Michael Porter and F M Scherer When the paper was in its early stages o f development both S teve Garber and Jon Rasmussen former economists in the LB Program provided eery useful cri tiques The views expressed here are my own o f course
1 The basic exp licit model of the role of advertising is Dorfman and S teiner ( 1954 ) and the models in this paper may be seen as ano ther extens ion of their work
2 For a review o f models of oligopoly which treat prices and quantities as relevant variables see Shubik ( 1 959) Extens ions to take advertis ing int o account are fairly s traightforward
3 Dorfman and S teiner dealt with a price-setting firm For a monopolis t inverting the demand relation to show Pbull P (Q A) _instead o f Q- (P A) has no effect on the results o f the analysis For oligopolists who face a sys tem of demand relations inversion does have some impac t on the results The dif ferences are not trivial nonetheless they will no t be explored in this study
4 Shubik ( 1 959) p
5 From a formal mathematical point o f view the function is similar to a1 i Legrangian funct ion of the form - r rr
L bull 11 + A ( j - (i i i+j
where the indicates some fixed level of the profits industry Following the same logic which resul ts in and income constraint as the marginal utility of the of the indus tr in the obj ective of firm i
6 Hall amp Weiss ( 1 967) p 323
7 See Schmallensee ( 1 97 2 ) Ch 2 pp 1 6-47
8 On this poin t see Imel amp Helmberger ( 1 9 7 1 )
9 Long ( 1 970) p 130 - 1 39
1 0 See Schmallensee ( 1 972) Ch 3 pp 48-8 7
rrj i+j ) of the rest of the
the identification profits of the rest
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980
Page 12
o f these cases have values o f 6 1 Y rea ter than 0 8 (pet food (20 10) bread amp cakes
(20 14 ) miscellaneous processed foods (20 29 ) men s and boys suit s and coats
(23 01 ) p rop riet ary drugs (28 08 ) elect ric housewares (36 1 2) radio and
IV sets 17 and games and toys (39 For be than(36 ) 04)) 6 1 Y to greater 0 8
both o and cry mus t b e less than 0 2 That is in these industries the results
shown in Tab le 2 would hold only with very strong dominance of marke t demand
effects over brand swit ching e ffects and a very low degree o f cooperation
Using the concentration rat ios given in Table 4 the average concentration
ratio for the eight indust ries is 38 1 For the remaining 24 industries it is
43 2 This result would lend some support t o the proposition that concentrat ion
and cooperation are related if the group of eight industries are identi fied
as having low degrees of cooperation
One f inal note on model predict ions S ince both cr and o are assumed to
be non-negative 60 should also be non-negat ive In only one o f the 32 cas es
is 60 negative and it is not signif icant bull
cor pJ
7 1
7 0 4
1 4 7 5 1
)
7 3 z a o a t s 1 1 3 3 5 ( i V Z 8 tJ1 2 1 - 1 1 0 7 3 5 c middot v 2 8 1 0 3 2 -0 0 1 4 3 0 C 1 V 9 8 3 2 8 1 5 1 6 - 0 0 6 t1 6 9 C _ V 1 11 0 2 9 0 1 29 0 00 7 6 1 C l k V 7 3 1 3 6 08 1 3 - o n 4 9 9 4 3 3 3 6 1 2 2 4 0 1 6 4 3
3 1 7 1 3 0 1 1 7 d 7 8 3 3 8 0 8 0 0 2 3 i 2 3 0 6
P 4
3 9 0 3 1 4
1 2
0 09 3 8 4 39 1)1 0 4 l 8 6 5
T A LE 4 e I NCU S T R Y C A T E G O R Y
F TC NOM e tOOE ass
C 1 I c 2 )
2 0 0 1 1 6 2 0 0r 1 7 zo o l Z 2 0 0 7 1 8
z o o 2 6
z o t a 1 6
2 1) 1 2 1 5 2 0 1 4 1 1 2 0 1 a 1 9 2 0 26 1 2 2 0 2 7 1 0 2 0 29 - 3 5
2 3 0 1 10
2 3 02 z o
2 3 03 1 9
2 5 5 1 2 4 Z 1 0 Z 1 1 2 0 3 1 9 2 7 4 1 0 z a o 1 3
2 A 0 7 z s
E LA S T I -C I T Y
c 3 )
- 0 09 7 2 7
0 00 449
- 0 0 2 8 9 3
- 0 3 2 7 70 - 0 OZ il I q
o 4 u t s a 0 0 1 3 7 3
o 1 4 o o-0 0 5 2 4o -0 1 94 0 1
-C l 6 l l 4 0 1 JJ 3 7 0 3 t 0 4 2 0 0 4 0 9
- 0 4 1 8 44 - O l 7 t ) 4 -ll 7 1 4 0 b -0 3 5 7 8 Z
- 0 1 9 7 7 0 1)0 3 3 4
-o n 9 2
IJ A T A
CONV
ON-C ONV
( 4 )
t 1NV
C 1t-V C NV c ) v C - JNV C rlNV c n v c a v C ON V ( i) l V
C CtV
C t1tV C IJNV
C 1 V
c 1 v
Pa ge 2La
HEQ F t Noe x
c 5 )
6 2q
6 89 1 48
6 1 4 ) q 3
1 9 7 1 9
q z t s 5 a e z
l l 9 1
4 o 7 1 2 0
c oqt 1 Z
4 4 )-q 1 z
5 5 9 1 0 64 1 3 3
n 3
Inter industry analysi
Page 23
Equation ( 3 6 ) may be used to fo rmulate a rough test o f the cooperat ion
vs e f ficiency controversy I f we rest ricted ourselves to some subset o f
indus tries where the relat ion be tween y and a i s the same we would be lookingbull
at a sample o f indus tries for which the dependence o f V on H and o is a
quadratic wi th no squared terms
I f we now sub s titute (26) into ( 36 ) we get
(38) v bull a + (y - a) R + (y - a) o (H) - (y - a) H o (H) + u
If it is only d if ferences in marke t shares that affect advertising inten-
s i ty levels i f concentration does not af fect cooperation and i f o bull 0
( 38 ) reduces t o
( 39 ) V bull a + (y - a) H + U
Let this b e the null hyp o thes is and note that V is a linear func-
tion o f H If y gt a and the null hyp othesis holds the expected relat ion
between V and H is as shown by the pos itively sloped straight line in Figure
6 b etween points A and C
In this context the alterna ive hypothesis the degree o f
cooperation ( o) is a positive function o f concentrat ion
is that
(H) I f the func-
tion is as described in (26) and Figure 4 the relation between V and H
would be as shown by the curved line connec t ing po ints A and C in Figure 6 One way to test the null hypothe s is against this al ternative is to test fo r
the hypothesized curvature in the function relating V and H
Fi gure
R
6
v
ID c
H 1 0
I
i I I
sfigned
clasJ s
in def ini g
effec t is the
I
Page middot2lf
One charac teristic of produc ts which may be useful
classes wi th respec t to significanc e of the brand-switching
conveniencenon-convenienc e distinc tion developed by Porter ( 1 97 )
Using his identif ication of IRS indus tries as a s tar t ing point
each of the 32 industries used in this s tudy to one of the two
The ass ignments are not ed in Tab le 4
Using the conveniencenon-convenienc e dis t inc tion the
divided into two sub-samples
sample was
Four regressions were then run bull ene for e ch
sub-sample one for the to tal sample and one for the total sample with
a dummy variable for convenience goods Each regression was run wi th LB s
and wi th indus tries as ob servations The results are given in Tab le 5
For the LB level equations ( 1A - 1D) equation(37) was used ome
rearranging of terms gives v bull y - ( 1-o ) (y - a) ( 1-z ) + ui i i
bull y - 6 1 ( 1 - z ) + ui When observations are pooled across indus trif s yi
is an industry variable In these regressions the values o f y es t imated
in the first stage were used Since industries have been grouped so as
to reduce differences in (y - a) that term is t reated as a cons tant
within each group Finally H the Herfindahl index has been subst ituted
for o the degree of cooperation The resultant equation is linear
y ( 1 - z ) and H x ( l - z ) This is the equation for whi ch results arei i
reported in Table 5 In all regress ions all variables were weighted by the
square roo t of sales
in
raquo
- 34
3 1
i l T 4 8LE S R E G E S S I ON R E SUL T S i X T f N S I ON S T O T Y PE Of
AND L E V E L OF C ONC E kT R A T I O
E O O tt S
tP bull T Y P E NR C ONV O B S tO N-CONV
E L A S T I shy
T I C J T Y
O E - t-41( T SliA R E
H( fl F
1 1 - K O x middot K O X E R F
- H S a 1 1 - H S t 1 8 1 1 9 1
R S O
1 1 0 1
q z
t - bull 9 f 2 1
Z t P D l 6 fC ONV - 1 9 1 s s c 2 1
e - l O I e bull u q a
2 3 1 0 1 1 Nl N- t O V 5 7 1 ( 1 2 19 1 6 1 - J l S 2 5 4
l l l l t z q o a 1 - J o v z I
l ( I NO l l A l l ol u l t o o c bull J u l l
d S
c 3 4 t t r a l
bull Z D 1 0 l l 4 l l 2 3 -gt 2 l l 2 0 - 4 lS 5 4 l C 4 1 o e 6 t l b b 5 1 1 0 1 1 1 c - 1 9 5 1 t d S 2 1 tV
OQ10 T-R A T I O S A E G I VE N l P A RE N T H E S E S l E V E L S f S I G 1 F l C ANC E A - 99 5 B - q 9 t - s NS J UC E NO I Nl E R C ( P T I S U (j f middot) I N f Hf t OUA TIOI S 6 E f 3R E A O J U S T H E N T F OR 1-t E T E R O SIC E O A S T f C I T V pIraquo
1
middot
I
R SO I S C A L C UL A T E D A S f I f bull 1 1 bull 1lt 1 K t t- E E N I S THf U A1 B E R O f fJe S E fi V A T I O S F t S HtE F - S T A T I S
I N DE P E ND E N T y II
T I C F O T H E U Y P O HIE S I S f A l l C JE F F I C H N T A R E QU A L HI l f RO A f40 K I S T H f UUM 8 E R OF Q ( A fL f S
I l l
1 A
1 6
t C
f 1 r R I l l
1 2 1
L Jt
L B
l q
L 1
on s T Y P E bull 2 )
P OUC T
X
H S I l l
l O l A I bull l 0 7 1
O O b O J
1 2 00 1
1 b 2 1 C I l l Z U
l l S b 1 -1 1 0 1
rlt F X l l - fi C f l
bull 1 I
l l
- ()4 2 3 l f H
1 - 2 7 9 1
K O X
1 1 - ti E R F I
l d l
S4 l i A I 2 l b
1 3 00
K O X -t f R F t S J
I 1 - HER F t
)
60 9
z o bull
0
t l l
3 8 8
1 7 6
5 64
64
N f bull
tl S bull
I l l
I 4 1
t ONV
NON-t ONV
A l l
A ll
c o v NO- ( ONV
1 4 1
1 5 1
- l S b
c - 1 1 2 1
5 8 2 C A I 1 6 7 0 1
middot I l 9 H
l 1 il
1 4 1 6 1
E L A C) T J shy
l t C I T Y
I S I
l b l
- 0 2 0
1 - 1 4 9 1
O Z 3 2 1 A I C b 6 b l
0 0 9 2 -J
1 1 b 4 1
o z t t t l 1 2 2 1 1
oru - Ht tt F
1 gt X bull b I
Page 25 bull
Mean val -as ior relevan t variables for the two sub-samples are
Conv Non-Conv
Adv Sales 4 7 2 3
iit Share 2 7 3 2
Ela cicity - 05 3 02 2
Herf indahl 7 3 1 860
That the dist nct ion between coft9enienee goods and non - convenience
goods mat ters is supported not only by these means but also by the regression
results Equat ions lA and lB are different in almost every respec t Only the
concentrat ionmarket share int eract ion term seems to mat ter for convenience
good s Jus t he opposite holds for non - convenience goods both elasticity1
and market sl e are highly significant but the concentrat ionmarket share L
int erac t ion t=rm is not
Given that one minus market share is used these equ3 tions the coeffici t
of market share is the negative of what is sh in col (6) of Table 5 For the
regress ion for non - convenience LB s reported in equation lB then LB s with larger
market shares have lower advertising to sales ratios
On the question of the impact of the degree of cooperation the story is mixed
For convenienc good s the coefficient is pos itive as equation (31) predicts when
there is a po sitive degree of cooperat ion Fo r the non-convenience goods LB s middot
concentration has no effec t
Aggregat ion t o the indus try level and then rer unning the equations $ives the
results shown in (2a-2d) of Tab le 5 As expected R2 goes up and levels of significanc e
fmiddotJr individual coefficients gJ down In add ition the coeffic ient of H ( l -H) inmiddot
quat ion 2A is no t significant though it is po s i tive Since 2A is the indus try level
I
--
all but one of them and that one is not significant For 18 of the 31
wo uld have on its adve rtising to sal es ratio
cognate of LA the variable H ( l-H) pl ays the same role in 2A as H ( l -z 1) p lays
in LA
Vconomies of scale in advert ising
The literature on advert ing contains a number of definitionbull o f economies
of scale and of findings eone rning its presence and magnitude Before I loo k
a t the data with the hope o f mmenting about scale effect s then I wan t to
define the term as I use it
By economies of scale I mean a decrease in the number o f units of adver-
tising which are needed to generate a unit of sales as advert ising is increased
Given the definition of the firm s own elast icity of demand with respect to
advertising (see the text foll wing equation (1) ) the technical definition
is that rii gt 1 Diseconomies of scale are def ined symmetrically i e
Yu lt 1
ith a cro s s - section oi observations on advertising intensity
(v bull a s1) and onmiddot advertising (a ) we may regress the former on the latter 1 i i
The results of such regress ions for the 32 industries are given in Table 2
lines lB ZB bull bull bull 32B For the 32 eases the regression coefficient is positive it
positive coefficients the coefficient i s signif icant
The que tion now is whether these results ar e evidence of pervasive dis-
economies of scale in advertislng I think no t Given my as sumption that the
data I observe are equilibrium results I may ri ghtly conclud e that almost every-
where high levels of advert ising are as sociated in equilibrium with high ratios
of advertising to sales I may no t conclud e anything however about the
impact that a move by some firm away from its equilibrium level o f advertising
middot- Page 27
Given the assumed relation in equation ( 37 ) there is a corre spondshy
ing equilibrium equation relating v and ai The equation is non - linear i
and I have not found a simp le way to characterize it It is po ssib le to
det ermi ne its deriviative with respect to a b y takin$ tne de riviative of ( 3 7 ) i
When that is done the result is
v - 1(40) i bull sl a a z ( 1-z )i i ira-l
Sinc e all the other terms in (40) are non-negative the sign dep ends only
on the sign of 6 1 which is in turn de pend ent only on the sign of (y - a)
That the signs of the coefficients of a in equations lB 2B bull bull bull 32B arei
not always equal to the signs of the coefficients of z in LA 2A bull bull bull 32A i
is still a mystery to me and something to be explored
bull
Summary
My purpose in writing this pa per wa s t o attempt to integrate explicit micro-
economic mode lling reasoned econome tric specification of error terms and high
quality data to exp lore some questions about adver tising by large firms
al quite important improvements could be made in the first two of those areas
and much data ma ssag ing is yet to be done in the third
Concerning the subst an ce of what I have done I think six things are im-
Page 2 8
Severshy
V and Conclusions
portant
1 Explicit modeling is worth the ef for t if for no other reas on than
that it provides a basis for choosing fr om a varie ty of funct ional
forms
2 The same is true for error specif ica t ion I did in fact look at scatshy
ter diagrams for all 32 industry categor ies and they vir tually all show
the het eroscedast icity which I as sumed and for which I correc ted
3 The predicted relation be tween advertis ing in tensity and market share
shows up clearly in on ly 25 of the case s examined I have ye t to
explore why that may b e the cas e
4 Some evidence concerning the presence and impact of coopera tion was
produced but it is not clear ly pervasive
5 The dis tinction between convenience goods and non-c onvenie nce goods is
unambiguously a good one
6 Virtually no evidence concerning econ omies of scale in adver tis ing can
be gleaned fr om this study given its assumpt ions
More work is called for on many of these issues my study seems to have
raised more questions than it has answer ed This is pr obab lly due to the richshy
ne ss of the data source since I had the oppor tunity to address several issue s
at once
- -- - Q --
Foo tnotes
Manager Line o f Business Program Federal Trade Commission The author depended heavily on several staf f members of the LB Pro gram for statistical and clerical support in particular Joe Cholka George Pascoe and Haro lene Jenkins Helpful comments were given b y Richard Caves Dennis Mueller Michael Porter and F M Scherer When the paper was in its early stages o f development both S teve Garber and Jon Rasmussen former economists in the LB Program provided eery useful cri tiques The views expressed here are my own o f course
1 The basic exp licit model of the role of advertising is Dorfman and S teiner ( 1954 ) and the models in this paper may be seen as ano ther extens ion of their work
2 For a review o f models of oligopoly which treat prices and quantities as relevant variables see Shubik ( 1 959) Extens ions to take advertis ing int o account are fairly s traightforward
3 Dorfman and S teiner dealt with a price-setting firm For a monopolis t inverting the demand relation to show Pbull P (Q A) _instead o f Q- (P A) has no effect on the results o f the analysis For oligopolists who face a sys tem of demand relations inversion does have some impac t on the results The dif ferences are not trivial nonetheless they will no t be explored in this study
4 Shubik ( 1 959) p
5 From a formal mathematical point o f view the function is similar to a1 i Legrangian funct ion of the form - r rr
L bull 11 + A ( j - (i i i+j
where the indicates some fixed level of the profits industry Following the same logic which resul ts in and income constraint as the marginal utility of the of the indus tr in the obj ective of firm i
6 Hall amp Weiss ( 1 967) p 323
7 See Schmallensee ( 1 97 2 ) Ch 2 pp 1 6-47
8 On this poin t see Imel amp Helmberger ( 1 9 7 1 )
9 Long ( 1 970) p 130 - 1 39
1 0 See Schmallensee ( 1 972) Ch 3 pp 48-8 7
rrj i+j ) of the rest of the
the identification profits of the rest
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980
cor pJ
7 1
7 0 4
1 4 7 5 1
)
7 3 z a o a t s 1 1 3 3 5 ( i V Z 8 tJ1 2 1 - 1 1 0 7 3 5 c middot v 2 8 1 0 3 2 -0 0 1 4 3 0 C 1 V 9 8 3 2 8 1 5 1 6 - 0 0 6 t1 6 9 C _ V 1 11 0 2 9 0 1 29 0 00 7 6 1 C l k V 7 3 1 3 6 08 1 3 - o n 4 9 9 4 3 3 3 6 1 2 2 4 0 1 6 4 3
3 1 7 1 3 0 1 1 7 d 7 8 3 3 8 0 8 0 0 2 3 i 2 3 0 6
P 4
3 9 0 3 1 4
1 2
0 09 3 8 4 39 1)1 0 4 l 8 6 5
T A LE 4 e I NCU S T R Y C A T E G O R Y
F TC NOM e tOOE ass
C 1 I c 2 )
2 0 0 1 1 6 2 0 0r 1 7 zo o l Z 2 0 0 7 1 8
z o o 2 6
z o t a 1 6
2 1) 1 2 1 5 2 0 1 4 1 1 2 0 1 a 1 9 2 0 26 1 2 2 0 2 7 1 0 2 0 29 - 3 5
2 3 0 1 10
2 3 02 z o
2 3 03 1 9
2 5 5 1 2 4 Z 1 0 Z 1 1 2 0 3 1 9 2 7 4 1 0 z a o 1 3
2 A 0 7 z s
E LA S T I -C I T Y
c 3 )
- 0 09 7 2 7
0 00 449
- 0 0 2 8 9 3
- 0 3 2 7 70 - 0 OZ il I q
o 4 u t s a 0 0 1 3 7 3
o 1 4 o o-0 0 5 2 4o -0 1 94 0 1
-C l 6 l l 4 0 1 JJ 3 7 0 3 t 0 4 2 0 0 4 0 9
- 0 4 1 8 44 - O l 7 t ) 4 -ll 7 1 4 0 b -0 3 5 7 8 Z
- 0 1 9 7 7 0 1)0 3 3 4
-o n 9 2
IJ A T A
CONV
ON-C ONV
( 4 )
t 1NV
C 1t-V C NV c ) v C - JNV C rlNV c n v c a v C ON V ( i) l V
C CtV
C t1tV C IJNV
C 1 V
c 1 v
Pa ge 2La
HEQ F t Noe x
c 5 )
6 2q
6 89 1 48
6 1 4 ) q 3
1 9 7 1 9
q z t s 5 a e z
l l 9 1
4 o 7 1 2 0
c oqt 1 Z
4 4 )-q 1 z
5 5 9 1 0 64 1 3 3
n 3
Inter industry analysi
Page 23
Equation ( 3 6 ) may be used to fo rmulate a rough test o f the cooperat ion
vs e f ficiency controversy I f we rest ricted ourselves to some subset o f
indus tries where the relat ion be tween y and a i s the same we would be lookingbull
at a sample o f indus tries for which the dependence o f V on H and o is a
quadratic wi th no squared terms
I f we now sub s titute (26) into ( 36 ) we get
(38) v bull a + (y - a) R + (y - a) o (H) - (y - a) H o (H) + u
If it is only d if ferences in marke t shares that affect advertising inten-
s i ty levels i f concentration does not af fect cooperation and i f o bull 0
( 38 ) reduces t o
( 39 ) V bull a + (y - a) H + U
Let this b e the null hyp o thes is and note that V is a linear func-
tion o f H If y gt a and the null hyp othesis holds the expected relat ion
between V and H is as shown by the pos itively sloped straight line in Figure
6 b etween points A and C
In this context the alterna ive hypothesis the degree o f
cooperation ( o) is a positive function o f concentrat ion
is that
(H) I f the func-
tion is as described in (26) and Figure 4 the relation between V and H
would be as shown by the curved line connec t ing po ints A and C in Figure 6 One way to test the null hypothe s is against this al ternative is to test fo r
the hypothesized curvature in the function relating V and H
Fi gure
R
6
v
ID c
H 1 0
I
i I I
sfigned
clasJ s
in def ini g
effec t is the
I
Page middot2lf
One charac teristic of produc ts which may be useful
classes wi th respec t to significanc e of the brand-switching
conveniencenon-convenienc e distinc tion developed by Porter ( 1 97 )
Using his identif ication of IRS indus tries as a s tar t ing point
each of the 32 industries used in this s tudy to one of the two
The ass ignments are not ed in Tab le 4
Using the conveniencenon-convenienc e dis t inc tion the
divided into two sub-samples
sample was
Four regressions were then run bull ene for e ch
sub-sample one for the to tal sample and one for the total sample with
a dummy variable for convenience goods Each regression was run wi th LB s
and wi th indus tries as ob servations The results are given in Tab le 5
For the LB level equations ( 1A - 1D) equation(37) was used ome
rearranging of terms gives v bull y - ( 1-o ) (y - a) ( 1-z ) + ui i i
bull y - 6 1 ( 1 - z ) + ui When observations are pooled across indus trif s yi
is an industry variable In these regressions the values o f y es t imated
in the first stage were used Since industries have been grouped so as
to reduce differences in (y - a) that term is t reated as a cons tant
within each group Finally H the Herfindahl index has been subst ituted
for o the degree of cooperation The resultant equation is linear
y ( 1 - z ) and H x ( l - z ) This is the equation for whi ch results arei i
reported in Table 5 In all regress ions all variables were weighted by the
square roo t of sales
in
raquo
- 34
3 1
i l T 4 8LE S R E G E S S I ON R E SUL T S i X T f N S I ON S T O T Y PE Of
AND L E V E L OF C ONC E kT R A T I O
E O O tt S
tP bull T Y P E NR C ONV O B S tO N-CONV
E L A S T I shy
T I C J T Y
O E - t-41( T SliA R E
H( fl F
1 1 - K O x middot K O X E R F
- H S a 1 1 - H S t 1 8 1 1 9 1
R S O
1 1 0 1
q z
t - bull 9 f 2 1
Z t P D l 6 fC ONV - 1 9 1 s s c 2 1
e - l O I e bull u q a
2 3 1 0 1 1 Nl N- t O V 5 7 1 ( 1 2 19 1 6 1 - J l S 2 5 4
l l l l t z q o a 1 - J o v z I
l ( I NO l l A l l ol u l t o o c bull J u l l
d S
c 3 4 t t r a l
bull Z D 1 0 l l 4 l l 2 3 -gt 2 l l 2 0 - 4 lS 5 4 l C 4 1 o e 6 t l b b 5 1 1 0 1 1 1 c - 1 9 5 1 t d S 2 1 tV
OQ10 T-R A T I O S A E G I VE N l P A RE N T H E S E S l E V E L S f S I G 1 F l C ANC E A - 99 5 B - q 9 t - s NS J UC E NO I Nl E R C ( P T I S U (j f middot) I N f Hf t OUA TIOI S 6 E f 3R E A O J U S T H E N T F OR 1-t E T E R O SIC E O A S T f C I T V pIraquo
1
middot
I
R SO I S C A L C UL A T E D A S f I f bull 1 1 bull 1lt 1 K t t- E E N I S THf U A1 B E R O f fJe S E fi V A T I O S F t S HtE F - S T A T I S
I N DE P E ND E N T y II
T I C F O T H E U Y P O HIE S I S f A l l C JE F F I C H N T A R E QU A L HI l f RO A f40 K I S T H f UUM 8 E R OF Q ( A fL f S
I l l
1 A
1 6
t C
f 1 r R I l l
1 2 1
L Jt
L B
l q
L 1
on s T Y P E bull 2 )
P OUC T
X
H S I l l
l O l A I bull l 0 7 1
O O b O J
1 2 00 1
1 b 2 1 C I l l Z U
l l S b 1 -1 1 0 1
rlt F X l l - fi C f l
bull 1 I
l l
- ()4 2 3 l f H
1 - 2 7 9 1
K O X
1 1 - ti E R F I
l d l
S4 l i A I 2 l b
1 3 00
K O X -t f R F t S J
I 1 - HER F t
)
60 9
z o bull
0
t l l
3 8 8
1 7 6
5 64
64
N f bull
tl S bull
I l l
I 4 1
t ONV
NON-t ONV
A l l
A ll
c o v NO- ( ONV
1 4 1
1 5 1
- l S b
c - 1 1 2 1
5 8 2 C A I 1 6 7 0 1
middot I l 9 H
l 1 il
1 4 1 6 1
E L A C) T J shy
l t C I T Y
I S I
l b l
- 0 2 0
1 - 1 4 9 1
O Z 3 2 1 A I C b 6 b l
0 0 9 2 -J
1 1 b 4 1
o z t t t l 1 2 2 1 1
oru - Ht tt F
1 gt X bull b I
Page 25 bull
Mean val -as ior relevan t variables for the two sub-samples are
Conv Non-Conv
Adv Sales 4 7 2 3
iit Share 2 7 3 2
Ela cicity - 05 3 02 2
Herf indahl 7 3 1 860
That the dist nct ion between coft9enienee goods and non - convenience
goods mat ters is supported not only by these means but also by the regression
results Equat ions lA and lB are different in almost every respec t Only the
concentrat ionmarket share int eract ion term seems to mat ter for convenience
good s Jus t he opposite holds for non - convenience goods both elasticity1
and market sl e are highly significant but the concentrat ionmarket share L
int erac t ion t=rm is not
Given that one minus market share is used these equ3 tions the coeffici t
of market share is the negative of what is sh in col (6) of Table 5 For the
regress ion for non - convenience LB s reported in equation lB then LB s with larger
market shares have lower advertising to sales ratios
On the question of the impact of the degree of cooperation the story is mixed
For convenienc good s the coefficient is pos itive as equation (31) predicts when
there is a po sitive degree of cooperat ion Fo r the non-convenience goods LB s middot
concentration has no effec t
Aggregat ion t o the indus try level and then rer unning the equations $ives the
results shown in (2a-2d) of Tab le 5 As expected R2 goes up and levels of significanc e
fmiddotJr individual coefficients gJ down In add ition the coeffic ient of H ( l -H) inmiddot
quat ion 2A is no t significant though it is po s i tive Since 2A is the indus try level
I
--
all but one of them and that one is not significant For 18 of the 31
wo uld have on its adve rtising to sal es ratio
cognate of LA the variable H ( l-H) pl ays the same role in 2A as H ( l -z 1) p lays
in LA
Vconomies of scale in advert ising
The literature on advert ing contains a number of definitionbull o f economies
of scale and of findings eone rning its presence and magnitude Before I loo k
a t the data with the hope o f mmenting about scale effect s then I wan t to
define the term as I use it
By economies of scale I mean a decrease in the number o f units of adver-
tising which are needed to generate a unit of sales as advert ising is increased
Given the definition of the firm s own elast icity of demand with respect to
advertising (see the text foll wing equation (1) ) the technical definition
is that rii gt 1 Diseconomies of scale are def ined symmetrically i e
Yu lt 1
ith a cro s s - section oi observations on advertising intensity
(v bull a s1) and onmiddot advertising (a ) we may regress the former on the latter 1 i i
The results of such regress ions for the 32 industries are given in Table 2
lines lB ZB bull bull bull 32B For the 32 eases the regression coefficient is positive it
positive coefficients the coefficient i s signif icant
The que tion now is whether these results ar e evidence of pervasive dis-
economies of scale in advertislng I think no t Given my as sumption that the
data I observe are equilibrium results I may ri ghtly conclud e that almost every-
where high levels of advert ising are as sociated in equilibrium with high ratios
of advertising to sales I may no t conclud e anything however about the
impact that a move by some firm away from its equilibrium level o f advertising
middot- Page 27
Given the assumed relation in equation ( 37 ) there is a corre spondshy
ing equilibrium equation relating v and ai The equation is non - linear i
and I have not found a simp le way to characterize it It is po ssib le to
det ermi ne its deriviative with respect to a b y takin$ tne de riviative of ( 3 7 ) i
When that is done the result is
v - 1(40) i bull sl a a z ( 1-z )i i ira-l
Sinc e all the other terms in (40) are non-negative the sign dep ends only
on the sign of 6 1 which is in turn de pend ent only on the sign of (y - a)
That the signs of the coefficients of a in equations lB 2B bull bull bull 32B arei
not always equal to the signs of the coefficients of z in LA 2A bull bull bull 32A i
is still a mystery to me and something to be explored
bull
Summary
My purpose in writing this pa per wa s t o attempt to integrate explicit micro-
economic mode lling reasoned econome tric specification of error terms and high
quality data to exp lore some questions about adver tising by large firms
al quite important improvements could be made in the first two of those areas
and much data ma ssag ing is yet to be done in the third
Concerning the subst an ce of what I have done I think six things are im-
Page 2 8
Severshy
V and Conclusions
portant
1 Explicit modeling is worth the ef for t if for no other reas on than
that it provides a basis for choosing fr om a varie ty of funct ional
forms
2 The same is true for error specif ica t ion I did in fact look at scatshy
ter diagrams for all 32 industry categor ies and they vir tually all show
the het eroscedast icity which I as sumed and for which I correc ted
3 The predicted relation be tween advertis ing in tensity and market share
shows up clearly in on ly 25 of the case s examined I have ye t to
explore why that may b e the cas e
4 Some evidence concerning the presence and impact of coopera tion was
produced but it is not clear ly pervasive
5 The dis tinction between convenience goods and non-c onvenie nce goods is
unambiguously a good one
6 Virtually no evidence concerning econ omies of scale in adver tis ing can
be gleaned fr om this study given its assumpt ions
More work is called for on many of these issues my study seems to have
raised more questions than it has answer ed This is pr obab lly due to the richshy
ne ss of the data source since I had the oppor tunity to address several issue s
at once
- -- - Q --
Foo tnotes
Manager Line o f Business Program Federal Trade Commission The author depended heavily on several staf f members of the LB Pro gram for statistical and clerical support in particular Joe Cholka George Pascoe and Haro lene Jenkins Helpful comments were given b y Richard Caves Dennis Mueller Michael Porter and F M Scherer When the paper was in its early stages o f development both S teve Garber and Jon Rasmussen former economists in the LB Program provided eery useful cri tiques The views expressed here are my own o f course
1 The basic exp licit model of the role of advertising is Dorfman and S teiner ( 1954 ) and the models in this paper may be seen as ano ther extens ion of their work
2 For a review o f models of oligopoly which treat prices and quantities as relevant variables see Shubik ( 1 959) Extens ions to take advertis ing int o account are fairly s traightforward
3 Dorfman and S teiner dealt with a price-setting firm For a monopolis t inverting the demand relation to show Pbull P (Q A) _instead o f Q- (P A) has no effect on the results o f the analysis For oligopolists who face a sys tem of demand relations inversion does have some impac t on the results The dif ferences are not trivial nonetheless they will no t be explored in this study
4 Shubik ( 1 959) p
5 From a formal mathematical point o f view the function is similar to a1 i Legrangian funct ion of the form - r rr
L bull 11 + A ( j - (i i i+j
where the indicates some fixed level of the profits industry Following the same logic which resul ts in and income constraint as the marginal utility of the of the indus tr in the obj ective of firm i
6 Hall amp Weiss ( 1 967) p 323
7 See Schmallensee ( 1 97 2 ) Ch 2 pp 1 6-47
8 On this poin t see Imel amp Helmberger ( 1 9 7 1 )
9 Long ( 1 970) p 130 - 1 39
1 0 See Schmallensee ( 1 972) Ch 3 pp 48-8 7
rrj i+j ) of the rest of the
the identification profits of the rest
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980
Inter industry analysi
Page 23
Equation ( 3 6 ) may be used to fo rmulate a rough test o f the cooperat ion
vs e f ficiency controversy I f we rest ricted ourselves to some subset o f
indus tries where the relat ion be tween y and a i s the same we would be lookingbull
at a sample o f indus tries for which the dependence o f V on H and o is a
quadratic wi th no squared terms
I f we now sub s titute (26) into ( 36 ) we get
(38) v bull a + (y - a) R + (y - a) o (H) - (y - a) H o (H) + u
If it is only d if ferences in marke t shares that affect advertising inten-
s i ty levels i f concentration does not af fect cooperation and i f o bull 0
( 38 ) reduces t o
( 39 ) V bull a + (y - a) H + U
Let this b e the null hyp o thes is and note that V is a linear func-
tion o f H If y gt a and the null hyp othesis holds the expected relat ion
between V and H is as shown by the pos itively sloped straight line in Figure
6 b etween points A and C
In this context the alterna ive hypothesis the degree o f
cooperation ( o) is a positive function o f concentrat ion
is that
(H) I f the func-
tion is as described in (26) and Figure 4 the relation between V and H
would be as shown by the curved line connec t ing po ints A and C in Figure 6 One way to test the null hypothe s is against this al ternative is to test fo r
the hypothesized curvature in the function relating V and H
Fi gure
R
6
v
ID c
H 1 0
I
i I I
sfigned
clasJ s
in def ini g
effec t is the
I
Page middot2lf
One charac teristic of produc ts which may be useful
classes wi th respec t to significanc e of the brand-switching
conveniencenon-convenienc e distinc tion developed by Porter ( 1 97 )
Using his identif ication of IRS indus tries as a s tar t ing point
each of the 32 industries used in this s tudy to one of the two
The ass ignments are not ed in Tab le 4
Using the conveniencenon-convenienc e dis t inc tion the
divided into two sub-samples
sample was
Four regressions were then run bull ene for e ch
sub-sample one for the to tal sample and one for the total sample with
a dummy variable for convenience goods Each regression was run wi th LB s
and wi th indus tries as ob servations The results are given in Tab le 5
For the LB level equations ( 1A - 1D) equation(37) was used ome
rearranging of terms gives v bull y - ( 1-o ) (y - a) ( 1-z ) + ui i i
bull y - 6 1 ( 1 - z ) + ui When observations are pooled across indus trif s yi
is an industry variable In these regressions the values o f y es t imated
in the first stage were used Since industries have been grouped so as
to reduce differences in (y - a) that term is t reated as a cons tant
within each group Finally H the Herfindahl index has been subst ituted
for o the degree of cooperation The resultant equation is linear
y ( 1 - z ) and H x ( l - z ) This is the equation for whi ch results arei i
reported in Table 5 In all regress ions all variables were weighted by the
square roo t of sales
in
raquo
- 34
3 1
i l T 4 8LE S R E G E S S I ON R E SUL T S i X T f N S I ON S T O T Y PE Of
AND L E V E L OF C ONC E kT R A T I O
E O O tt S
tP bull T Y P E NR C ONV O B S tO N-CONV
E L A S T I shy
T I C J T Y
O E - t-41( T SliA R E
H( fl F
1 1 - K O x middot K O X E R F
- H S a 1 1 - H S t 1 8 1 1 9 1
R S O
1 1 0 1
q z
t - bull 9 f 2 1
Z t P D l 6 fC ONV - 1 9 1 s s c 2 1
e - l O I e bull u q a
2 3 1 0 1 1 Nl N- t O V 5 7 1 ( 1 2 19 1 6 1 - J l S 2 5 4
l l l l t z q o a 1 - J o v z I
l ( I NO l l A l l ol u l t o o c bull J u l l
d S
c 3 4 t t r a l
bull Z D 1 0 l l 4 l l 2 3 -gt 2 l l 2 0 - 4 lS 5 4 l C 4 1 o e 6 t l b b 5 1 1 0 1 1 1 c - 1 9 5 1 t d S 2 1 tV
OQ10 T-R A T I O S A E G I VE N l P A RE N T H E S E S l E V E L S f S I G 1 F l C ANC E A - 99 5 B - q 9 t - s NS J UC E NO I Nl E R C ( P T I S U (j f middot) I N f Hf t OUA TIOI S 6 E f 3R E A O J U S T H E N T F OR 1-t E T E R O SIC E O A S T f C I T V pIraquo
1
middot
I
R SO I S C A L C UL A T E D A S f I f bull 1 1 bull 1lt 1 K t t- E E N I S THf U A1 B E R O f fJe S E fi V A T I O S F t S HtE F - S T A T I S
I N DE P E ND E N T y II
T I C F O T H E U Y P O HIE S I S f A l l C JE F F I C H N T A R E QU A L HI l f RO A f40 K I S T H f UUM 8 E R OF Q ( A fL f S
I l l
1 A
1 6
t C
f 1 r R I l l
1 2 1
L Jt
L B
l q
L 1
on s T Y P E bull 2 )
P OUC T
X
H S I l l
l O l A I bull l 0 7 1
O O b O J
1 2 00 1
1 b 2 1 C I l l Z U
l l S b 1 -1 1 0 1
rlt F X l l - fi C f l
bull 1 I
l l
- ()4 2 3 l f H
1 - 2 7 9 1
K O X
1 1 - ti E R F I
l d l
S4 l i A I 2 l b
1 3 00
K O X -t f R F t S J
I 1 - HER F t
)
60 9
z o bull
0
t l l
3 8 8
1 7 6
5 64
64
N f bull
tl S bull
I l l
I 4 1
t ONV
NON-t ONV
A l l
A ll
c o v NO- ( ONV
1 4 1
1 5 1
- l S b
c - 1 1 2 1
5 8 2 C A I 1 6 7 0 1
middot I l 9 H
l 1 il
1 4 1 6 1
E L A C) T J shy
l t C I T Y
I S I
l b l
- 0 2 0
1 - 1 4 9 1
O Z 3 2 1 A I C b 6 b l
0 0 9 2 -J
1 1 b 4 1
o z t t t l 1 2 2 1 1
oru - Ht tt F
1 gt X bull b I
Page 25 bull
Mean val -as ior relevan t variables for the two sub-samples are
Conv Non-Conv
Adv Sales 4 7 2 3
iit Share 2 7 3 2
Ela cicity - 05 3 02 2
Herf indahl 7 3 1 860
That the dist nct ion between coft9enienee goods and non - convenience
goods mat ters is supported not only by these means but also by the regression
results Equat ions lA and lB are different in almost every respec t Only the
concentrat ionmarket share int eract ion term seems to mat ter for convenience
good s Jus t he opposite holds for non - convenience goods both elasticity1
and market sl e are highly significant but the concentrat ionmarket share L
int erac t ion t=rm is not
Given that one minus market share is used these equ3 tions the coeffici t
of market share is the negative of what is sh in col (6) of Table 5 For the
regress ion for non - convenience LB s reported in equation lB then LB s with larger
market shares have lower advertising to sales ratios
On the question of the impact of the degree of cooperation the story is mixed
For convenienc good s the coefficient is pos itive as equation (31) predicts when
there is a po sitive degree of cooperat ion Fo r the non-convenience goods LB s middot
concentration has no effec t
Aggregat ion t o the indus try level and then rer unning the equations $ives the
results shown in (2a-2d) of Tab le 5 As expected R2 goes up and levels of significanc e
fmiddotJr individual coefficients gJ down In add ition the coeffic ient of H ( l -H) inmiddot
quat ion 2A is no t significant though it is po s i tive Since 2A is the indus try level
I
--
all but one of them and that one is not significant For 18 of the 31
wo uld have on its adve rtising to sal es ratio
cognate of LA the variable H ( l-H) pl ays the same role in 2A as H ( l -z 1) p lays
in LA
Vconomies of scale in advert ising
The literature on advert ing contains a number of definitionbull o f economies
of scale and of findings eone rning its presence and magnitude Before I loo k
a t the data with the hope o f mmenting about scale effect s then I wan t to
define the term as I use it
By economies of scale I mean a decrease in the number o f units of adver-
tising which are needed to generate a unit of sales as advert ising is increased
Given the definition of the firm s own elast icity of demand with respect to
advertising (see the text foll wing equation (1) ) the technical definition
is that rii gt 1 Diseconomies of scale are def ined symmetrically i e
Yu lt 1
ith a cro s s - section oi observations on advertising intensity
(v bull a s1) and onmiddot advertising (a ) we may regress the former on the latter 1 i i
The results of such regress ions for the 32 industries are given in Table 2
lines lB ZB bull bull bull 32B For the 32 eases the regression coefficient is positive it
positive coefficients the coefficient i s signif icant
The que tion now is whether these results ar e evidence of pervasive dis-
economies of scale in advertislng I think no t Given my as sumption that the
data I observe are equilibrium results I may ri ghtly conclud e that almost every-
where high levels of advert ising are as sociated in equilibrium with high ratios
of advertising to sales I may no t conclud e anything however about the
impact that a move by some firm away from its equilibrium level o f advertising
middot- Page 27
Given the assumed relation in equation ( 37 ) there is a corre spondshy
ing equilibrium equation relating v and ai The equation is non - linear i
and I have not found a simp le way to characterize it It is po ssib le to
det ermi ne its deriviative with respect to a b y takin$ tne de riviative of ( 3 7 ) i
When that is done the result is
v - 1(40) i bull sl a a z ( 1-z )i i ira-l
Sinc e all the other terms in (40) are non-negative the sign dep ends only
on the sign of 6 1 which is in turn de pend ent only on the sign of (y - a)
That the signs of the coefficients of a in equations lB 2B bull bull bull 32B arei
not always equal to the signs of the coefficients of z in LA 2A bull bull bull 32A i
is still a mystery to me and something to be explored
bull
Summary
My purpose in writing this pa per wa s t o attempt to integrate explicit micro-
economic mode lling reasoned econome tric specification of error terms and high
quality data to exp lore some questions about adver tising by large firms
al quite important improvements could be made in the first two of those areas
and much data ma ssag ing is yet to be done in the third
Concerning the subst an ce of what I have done I think six things are im-
Page 2 8
Severshy
V and Conclusions
portant
1 Explicit modeling is worth the ef for t if for no other reas on than
that it provides a basis for choosing fr om a varie ty of funct ional
forms
2 The same is true for error specif ica t ion I did in fact look at scatshy
ter diagrams for all 32 industry categor ies and they vir tually all show
the het eroscedast icity which I as sumed and for which I correc ted
3 The predicted relation be tween advertis ing in tensity and market share
shows up clearly in on ly 25 of the case s examined I have ye t to
explore why that may b e the cas e
4 Some evidence concerning the presence and impact of coopera tion was
produced but it is not clear ly pervasive
5 The dis tinction between convenience goods and non-c onvenie nce goods is
unambiguously a good one
6 Virtually no evidence concerning econ omies of scale in adver tis ing can
be gleaned fr om this study given its assumpt ions
More work is called for on many of these issues my study seems to have
raised more questions than it has answer ed This is pr obab lly due to the richshy
ne ss of the data source since I had the oppor tunity to address several issue s
at once
- -- - Q --
Foo tnotes
Manager Line o f Business Program Federal Trade Commission The author depended heavily on several staf f members of the LB Pro gram for statistical and clerical support in particular Joe Cholka George Pascoe and Haro lene Jenkins Helpful comments were given b y Richard Caves Dennis Mueller Michael Porter and F M Scherer When the paper was in its early stages o f development both S teve Garber and Jon Rasmussen former economists in the LB Program provided eery useful cri tiques The views expressed here are my own o f course
1 The basic exp licit model of the role of advertising is Dorfman and S teiner ( 1954 ) and the models in this paper may be seen as ano ther extens ion of their work
2 For a review o f models of oligopoly which treat prices and quantities as relevant variables see Shubik ( 1 959) Extens ions to take advertis ing int o account are fairly s traightforward
3 Dorfman and S teiner dealt with a price-setting firm For a monopolis t inverting the demand relation to show Pbull P (Q A) _instead o f Q- (P A) has no effect on the results o f the analysis For oligopolists who face a sys tem of demand relations inversion does have some impac t on the results The dif ferences are not trivial nonetheless they will no t be explored in this study
4 Shubik ( 1 959) p
5 From a formal mathematical point o f view the function is similar to a1 i Legrangian funct ion of the form - r rr
L bull 11 + A ( j - (i i i+j
where the indicates some fixed level of the profits industry Following the same logic which resul ts in and income constraint as the marginal utility of the of the indus tr in the obj ective of firm i
6 Hall amp Weiss ( 1 967) p 323
7 See Schmallensee ( 1 97 2 ) Ch 2 pp 1 6-47
8 On this poin t see Imel amp Helmberger ( 1 9 7 1 )
9 Long ( 1 970) p 130 - 1 39
1 0 See Schmallensee ( 1 972) Ch 3 pp 48-8 7
rrj i+j ) of the rest of the
the identification profits of the rest
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980
Fi gure
R
6
v
ID c
H 1 0
I
i I I
sfigned
clasJ s
in def ini g
effec t is the
I
Page middot2lf
One charac teristic of produc ts which may be useful
classes wi th respec t to significanc e of the brand-switching
conveniencenon-convenienc e distinc tion developed by Porter ( 1 97 )
Using his identif ication of IRS indus tries as a s tar t ing point
each of the 32 industries used in this s tudy to one of the two
The ass ignments are not ed in Tab le 4
Using the conveniencenon-convenienc e dis t inc tion the
divided into two sub-samples
sample was
Four regressions were then run bull ene for e ch
sub-sample one for the to tal sample and one for the total sample with
a dummy variable for convenience goods Each regression was run wi th LB s
and wi th indus tries as ob servations The results are given in Tab le 5
For the LB level equations ( 1A - 1D) equation(37) was used ome
rearranging of terms gives v bull y - ( 1-o ) (y - a) ( 1-z ) + ui i i
bull y - 6 1 ( 1 - z ) + ui When observations are pooled across indus trif s yi
is an industry variable In these regressions the values o f y es t imated
in the first stage were used Since industries have been grouped so as
to reduce differences in (y - a) that term is t reated as a cons tant
within each group Finally H the Herfindahl index has been subst ituted
for o the degree of cooperation The resultant equation is linear
y ( 1 - z ) and H x ( l - z ) This is the equation for whi ch results arei i
reported in Table 5 In all regress ions all variables were weighted by the
square roo t of sales
in
raquo
- 34
3 1
i l T 4 8LE S R E G E S S I ON R E SUL T S i X T f N S I ON S T O T Y PE Of
AND L E V E L OF C ONC E kT R A T I O
E O O tt S
tP bull T Y P E NR C ONV O B S tO N-CONV
E L A S T I shy
T I C J T Y
O E - t-41( T SliA R E
H( fl F
1 1 - K O x middot K O X E R F
- H S a 1 1 - H S t 1 8 1 1 9 1
R S O
1 1 0 1
q z
t - bull 9 f 2 1
Z t P D l 6 fC ONV - 1 9 1 s s c 2 1
e - l O I e bull u q a
2 3 1 0 1 1 Nl N- t O V 5 7 1 ( 1 2 19 1 6 1 - J l S 2 5 4
l l l l t z q o a 1 - J o v z I
l ( I NO l l A l l ol u l t o o c bull J u l l
d S
c 3 4 t t r a l
bull Z D 1 0 l l 4 l l 2 3 -gt 2 l l 2 0 - 4 lS 5 4 l C 4 1 o e 6 t l b b 5 1 1 0 1 1 1 c - 1 9 5 1 t d S 2 1 tV
OQ10 T-R A T I O S A E G I VE N l P A RE N T H E S E S l E V E L S f S I G 1 F l C ANC E A - 99 5 B - q 9 t - s NS J UC E NO I Nl E R C ( P T I S U (j f middot) I N f Hf t OUA TIOI S 6 E f 3R E A O J U S T H E N T F OR 1-t E T E R O SIC E O A S T f C I T V pIraquo
1
middot
I
R SO I S C A L C UL A T E D A S f I f bull 1 1 bull 1lt 1 K t t- E E N I S THf U A1 B E R O f fJe S E fi V A T I O S F t S HtE F - S T A T I S
I N DE P E ND E N T y II
T I C F O T H E U Y P O HIE S I S f A l l C JE F F I C H N T A R E QU A L HI l f RO A f40 K I S T H f UUM 8 E R OF Q ( A fL f S
I l l
1 A
1 6
t C
f 1 r R I l l
1 2 1
L Jt
L B
l q
L 1
on s T Y P E bull 2 )
P OUC T
X
H S I l l
l O l A I bull l 0 7 1
O O b O J
1 2 00 1
1 b 2 1 C I l l Z U
l l S b 1 -1 1 0 1
rlt F X l l - fi C f l
bull 1 I
l l
- ()4 2 3 l f H
1 - 2 7 9 1
K O X
1 1 - ti E R F I
l d l
S4 l i A I 2 l b
1 3 00
K O X -t f R F t S J
I 1 - HER F t
)
60 9
z o bull
0
t l l
3 8 8
1 7 6
5 64
64
N f bull
tl S bull
I l l
I 4 1
t ONV
NON-t ONV
A l l
A ll
c o v NO- ( ONV
1 4 1
1 5 1
- l S b
c - 1 1 2 1
5 8 2 C A I 1 6 7 0 1
middot I l 9 H
l 1 il
1 4 1 6 1
E L A C) T J shy
l t C I T Y
I S I
l b l
- 0 2 0
1 - 1 4 9 1
O Z 3 2 1 A I C b 6 b l
0 0 9 2 -J
1 1 b 4 1
o z t t t l 1 2 2 1 1
oru - Ht tt F
1 gt X bull b I
Page 25 bull
Mean val -as ior relevan t variables for the two sub-samples are
Conv Non-Conv
Adv Sales 4 7 2 3
iit Share 2 7 3 2
Ela cicity - 05 3 02 2
Herf indahl 7 3 1 860
That the dist nct ion between coft9enienee goods and non - convenience
goods mat ters is supported not only by these means but also by the regression
results Equat ions lA and lB are different in almost every respec t Only the
concentrat ionmarket share int eract ion term seems to mat ter for convenience
good s Jus t he opposite holds for non - convenience goods both elasticity1
and market sl e are highly significant but the concentrat ionmarket share L
int erac t ion t=rm is not
Given that one minus market share is used these equ3 tions the coeffici t
of market share is the negative of what is sh in col (6) of Table 5 For the
regress ion for non - convenience LB s reported in equation lB then LB s with larger
market shares have lower advertising to sales ratios
On the question of the impact of the degree of cooperation the story is mixed
For convenienc good s the coefficient is pos itive as equation (31) predicts when
there is a po sitive degree of cooperat ion Fo r the non-convenience goods LB s middot
concentration has no effec t
Aggregat ion t o the indus try level and then rer unning the equations $ives the
results shown in (2a-2d) of Tab le 5 As expected R2 goes up and levels of significanc e
fmiddotJr individual coefficients gJ down In add ition the coeffic ient of H ( l -H) inmiddot
quat ion 2A is no t significant though it is po s i tive Since 2A is the indus try level
I
--
all but one of them and that one is not significant For 18 of the 31
wo uld have on its adve rtising to sal es ratio
cognate of LA the variable H ( l-H) pl ays the same role in 2A as H ( l -z 1) p lays
in LA
Vconomies of scale in advert ising
The literature on advert ing contains a number of definitionbull o f economies
of scale and of findings eone rning its presence and magnitude Before I loo k
a t the data with the hope o f mmenting about scale effect s then I wan t to
define the term as I use it
By economies of scale I mean a decrease in the number o f units of adver-
tising which are needed to generate a unit of sales as advert ising is increased
Given the definition of the firm s own elast icity of demand with respect to
advertising (see the text foll wing equation (1) ) the technical definition
is that rii gt 1 Diseconomies of scale are def ined symmetrically i e
Yu lt 1
ith a cro s s - section oi observations on advertising intensity
(v bull a s1) and onmiddot advertising (a ) we may regress the former on the latter 1 i i
The results of such regress ions for the 32 industries are given in Table 2
lines lB ZB bull bull bull 32B For the 32 eases the regression coefficient is positive it
positive coefficients the coefficient i s signif icant
The que tion now is whether these results ar e evidence of pervasive dis-
economies of scale in advertislng I think no t Given my as sumption that the
data I observe are equilibrium results I may ri ghtly conclud e that almost every-
where high levels of advert ising are as sociated in equilibrium with high ratios
of advertising to sales I may no t conclud e anything however about the
impact that a move by some firm away from its equilibrium level o f advertising
middot- Page 27
Given the assumed relation in equation ( 37 ) there is a corre spondshy
ing equilibrium equation relating v and ai The equation is non - linear i
and I have not found a simp le way to characterize it It is po ssib le to
det ermi ne its deriviative with respect to a b y takin$ tne de riviative of ( 3 7 ) i
When that is done the result is
v - 1(40) i bull sl a a z ( 1-z )i i ira-l
Sinc e all the other terms in (40) are non-negative the sign dep ends only
on the sign of 6 1 which is in turn de pend ent only on the sign of (y - a)
That the signs of the coefficients of a in equations lB 2B bull bull bull 32B arei
not always equal to the signs of the coefficients of z in LA 2A bull bull bull 32A i
is still a mystery to me and something to be explored
bull
Summary
My purpose in writing this pa per wa s t o attempt to integrate explicit micro-
economic mode lling reasoned econome tric specification of error terms and high
quality data to exp lore some questions about adver tising by large firms
al quite important improvements could be made in the first two of those areas
and much data ma ssag ing is yet to be done in the third
Concerning the subst an ce of what I have done I think six things are im-
Page 2 8
Severshy
V and Conclusions
portant
1 Explicit modeling is worth the ef for t if for no other reas on than
that it provides a basis for choosing fr om a varie ty of funct ional
forms
2 The same is true for error specif ica t ion I did in fact look at scatshy
ter diagrams for all 32 industry categor ies and they vir tually all show
the het eroscedast icity which I as sumed and for which I correc ted
3 The predicted relation be tween advertis ing in tensity and market share
shows up clearly in on ly 25 of the case s examined I have ye t to
explore why that may b e the cas e
4 Some evidence concerning the presence and impact of coopera tion was
produced but it is not clear ly pervasive
5 The dis tinction between convenience goods and non-c onvenie nce goods is
unambiguously a good one
6 Virtually no evidence concerning econ omies of scale in adver tis ing can
be gleaned fr om this study given its assumpt ions
More work is called for on many of these issues my study seems to have
raised more questions than it has answer ed This is pr obab lly due to the richshy
ne ss of the data source since I had the oppor tunity to address several issue s
at once
- -- - Q --
Foo tnotes
Manager Line o f Business Program Federal Trade Commission The author depended heavily on several staf f members of the LB Pro gram for statistical and clerical support in particular Joe Cholka George Pascoe and Haro lene Jenkins Helpful comments were given b y Richard Caves Dennis Mueller Michael Porter and F M Scherer When the paper was in its early stages o f development both S teve Garber and Jon Rasmussen former economists in the LB Program provided eery useful cri tiques The views expressed here are my own o f course
1 The basic exp licit model of the role of advertising is Dorfman and S teiner ( 1954 ) and the models in this paper may be seen as ano ther extens ion of their work
2 For a review o f models of oligopoly which treat prices and quantities as relevant variables see Shubik ( 1 959) Extens ions to take advertis ing int o account are fairly s traightforward
3 Dorfman and S teiner dealt with a price-setting firm For a monopolis t inverting the demand relation to show Pbull P (Q A) _instead o f Q- (P A) has no effect on the results o f the analysis For oligopolists who face a sys tem of demand relations inversion does have some impac t on the results The dif ferences are not trivial nonetheless they will no t be explored in this study
4 Shubik ( 1 959) p
5 From a formal mathematical point o f view the function is similar to a1 i Legrangian funct ion of the form - r rr
L bull 11 + A ( j - (i i i+j
where the indicates some fixed level of the profits industry Following the same logic which resul ts in and income constraint as the marginal utility of the of the indus tr in the obj ective of firm i
6 Hall amp Weiss ( 1 967) p 323
7 See Schmallensee ( 1 97 2 ) Ch 2 pp 1 6-47
8 On this poin t see Imel amp Helmberger ( 1 9 7 1 )
9 Long ( 1 970) p 130 - 1 39
1 0 See Schmallensee ( 1 972) Ch 3 pp 48-8 7
rrj i+j ) of the rest of the
the identification profits of the rest
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980
i I I
sfigned
clasJ s
in def ini g
effec t is the
I
Page middot2lf
One charac teristic of produc ts which may be useful
classes wi th respec t to significanc e of the brand-switching
conveniencenon-convenienc e distinc tion developed by Porter ( 1 97 )
Using his identif ication of IRS indus tries as a s tar t ing point
each of the 32 industries used in this s tudy to one of the two
The ass ignments are not ed in Tab le 4
Using the conveniencenon-convenienc e dis t inc tion the
divided into two sub-samples
sample was
Four regressions were then run bull ene for e ch
sub-sample one for the to tal sample and one for the total sample with
a dummy variable for convenience goods Each regression was run wi th LB s
and wi th indus tries as ob servations The results are given in Tab le 5
For the LB level equations ( 1A - 1D) equation(37) was used ome
rearranging of terms gives v bull y - ( 1-o ) (y - a) ( 1-z ) + ui i i
bull y - 6 1 ( 1 - z ) + ui When observations are pooled across indus trif s yi
is an industry variable In these regressions the values o f y es t imated
in the first stage were used Since industries have been grouped so as
to reduce differences in (y - a) that term is t reated as a cons tant
within each group Finally H the Herfindahl index has been subst ituted
for o the degree of cooperation The resultant equation is linear
y ( 1 - z ) and H x ( l - z ) This is the equation for whi ch results arei i
reported in Table 5 In all regress ions all variables were weighted by the
square roo t of sales
in
raquo
- 34
3 1
i l T 4 8LE S R E G E S S I ON R E SUL T S i X T f N S I ON S T O T Y PE Of
AND L E V E L OF C ONC E kT R A T I O
E O O tt S
tP bull T Y P E NR C ONV O B S tO N-CONV
E L A S T I shy
T I C J T Y
O E - t-41( T SliA R E
H( fl F
1 1 - K O x middot K O X E R F
- H S a 1 1 - H S t 1 8 1 1 9 1
R S O
1 1 0 1
q z
t - bull 9 f 2 1
Z t P D l 6 fC ONV - 1 9 1 s s c 2 1
e - l O I e bull u q a
2 3 1 0 1 1 Nl N- t O V 5 7 1 ( 1 2 19 1 6 1 - J l S 2 5 4
l l l l t z q o a 1 - J o v z I
l ( I NO l l A l l ol u l t o o c bull J u l l
d S
c 3 4 t t r a l
bull Z D 1 0 l l 4 l l 2 3 -gt 2 l l 2 0 - 4 lS 5 4 l C 4 1 o e 6 t l b b 5 1 1 0 1 1 1 c - 1 9 5 1 t d S 2 1 tV
OQ10 T-R A T I O S A E G I VE N l P A RE N T H E S E S l E V E L S f S I G 1 F l C ANC E A - 99 5 B - q 9 t - s NS J UC E NO I Nl E R C ( P T I S U (j f middot) I N f Hf t OUA TIOI S 6 E f 3R E A O J U S T H E N T F OR 1-t E T E R O SIC E O A S T f C I T V pIraquo
1
middot
I
R SO I S C A L C UL A T E D A S f I f bull 1 1 bull 1lt 1 K t t- E E N I S THf U A1 B E R O f fJe S E fi V A T I O S F t S HtE F - S T A T I S
I N DE P E ND E N T y II
T I C F O T H E U Y P O HIE S I S f A l l C JE F F I C H N T A R E QU A L HI l f RO A f40 K I S T H f UUM 8 E R OF Q ( A fL f S
I l l
1 A
1 6
t C
f 1 r R I l l
1 2 1
L Jt
L B
l q
L 1
on s T Y P E bull 2 )
P OUC T
X
H S I l l
l O l A I bull l 0 7 1
O O b O J
1 2 00 1
1 b 2 1 C I l l Z U
l l S b 1 -1 1 0 1
rlt F X l l - fi C f l
bull 1 I
l l
- ()4 2 3 l f H
1 - 2 7 9 1
K O X
1 1 - ti E R F I
l d l
S4 l i A I 2 l b
1 3 00
K O X -t f R F t S J
I 1 - HER F t
)
60 9
z o bull
0
t l l
3 8 8
1 7 6
5 64
64
N f bull
tl S bull
I l l
I 4 1
t ONV
NON-t ONV
A l l
A ll
c o v NO- ( ONV
1 4 1
1 5 1
- l S b
c - 1 1 2 1
5 8 2 C A I 1 6 7 0 1
middot I l 9 H
l 1 il
1 4 1 6 1
E L A C) T J shy
l t C I T Y
I S I
l b l
- 0 2 0
1 - 1 4 9 1
O Z 3 2 1 A I C b 6 b l
0 0 9 2 -J
1 1 b 4 1
o z t t t l 1 2 2 1 1
oru - Ht tt F
1 gt X bull b I
Page 25 bull
Mean val -as ior relevan t variables for the two sub-samples are
Conv Non-Conv
Adv Sales 4 7 2 3
iit Share 2 7 3 2
Ela cicity - 05 3 02 2
Herf indahl 7 3 1 860
That the dist nct ion between coft9enienee goods and non - convenience
goods mat ters is supported not only by these means but also by the regression
results Equat ions lA and lB are different in almost every respec t Only the
concentrat ionmarket share int eract ion term seems to mat ter for convenience
good s Jus t he opposite holds for non - convenience goods both elasticity1
and market sl e are highly significant but the concentrat ionmarket share L
int erac t ion t=rm is not
Given that one minus market share is used these equ3 tions the coeffici t
of market share is the negative of what is sh in col (6) of Table 5 For the
regress ion for non - convenience LB s reported in equation lB then LB s with larger
market shares have lower advertising to sales ratios
On the question of the impact of the degree of cooperation the story is mixed
For convenienc good s the coefficient is pos itive as equation (31) predicts when
there is a po sitive degree of cooperat ion Fo r the non-convenience goods LB s middot
concentration has no effec t
Aggregat ion t o the indus try level and then rer unning the equations $ives the
results shown in (2a-2d) of Tab le 5 As expected R2 goes up and levels of significanc e
fmiddotJr individual coefficients gJ down In add ition the coeffic ient of H ( l -H) inmiddot
quat ion 2A is no t significant though it is po s i tive Since 2A is the indus try level
I
--
all but one of them and that one is not significant For 18 of the 31
wo uld have on its adve rtising to sal es ratio
cognate of LA the variable H ( l-H) pl ays the same role in 2A as H ( l -z 1) p lays
in LA
Vconomies of scale in advert ising
The literature on advert ing contains a number of definitionbull o f economies
of scale and of findings eone rning its presence and magnitude Before I loo k
a t the data with the hope o f mmenting about scale effect s then I wan t to
define the term as I use it
By economies of scale I mean a decrease in the number o f units of adver-
tising which are needed to generate a unit of sales as advert ising is increased
Given the definition of the firm s own elast icity of demand with respect to
advertising (see the text foll wing equation (1) ) the technical definition
is that rii gt 1 Diseconomies of scale are def ined symmetrically i e
Yu lt 1
ith a cro s s - section oi observations on advertising intensity
(v bull a s1) and onmiddot advertising (a ) we may regress the former on the latter 1 i i
The results of such regress ions for the 32 industries are given in Table 2
lines lB ZB bull bull bull 32B For the 32 eases the regression coefficient is positive it
positive coefficients the coefficient i s signif icant
The que tion now is whether these results ar e evidence of pervasive dis-
economies of scale in advertislng I think no t Given my as sumption that the
data I observe are equilibrium results I may ri ghtly conclud e that almost every-
where high levels of advert ising are as sociated in equilibrium with high ratios
of advertising to sales I may no t conclud e anything however about the
impact that a move by some firm away from its equilibrium level o f advertising
middot- Page 27
Given the assumed relation in equation ( 37 ) there is a corre spondshy
ing equilibrium equation relating v and ai The equation is non - linear i
and I have not found a simp le way to characterize it It is po ssib le to
det ermi ne its deriviative with respect to a b y takin$ tne de riviative of ( 3 7 ) i
When that is done the result is
v - 1(40) i bull sl a a z ( 1-z )i i ira-l
Sinc e all the other terms in (40) are non-negative the sign dep ends only
on the sign of 6 1 which is in turn de pend ent only on the sign of (y - a)
That the signs of the coefficients of a in equations lB 2B bull bull bull 32B arei
not always equal to the signs of the coefficients of z in LA 2A bull bull bull 32A i
is still a mystery to me and something to be explored
bull
Summary
My purpose in writing this pa per wa s t o attempt to integrate explicit micro-
economic mode lling reasoned econome tric specification of error terms and high
quality data to exp lore some questions about adver tising by large firms
al quite important improvements could be made in the first two of those areas
and much data ma ssag ing is yet to be done in the third
Concerning the subst an ce of what I have done I think six things are im-
Page 2 8
Severshy
V and Conclusions
portant
1 Explicit modeling is worth the ef for t if for no other reas on than
that it provides a basis for choosing fr om a varie ty of funct ional
forms
2 The same is true for error specif ica t ion I did in fact look at scatshy
ter diagrams for all 32 industry categor ies and they vir tually all show
the het eroscedast icity which I as sumed and for which I correc ted
3 The predicted relation be tween advertis ing in tensity and market share
shows up clearly in on ly 25 of the case s examined I have ye t to
explore why that may b e the cas e
4 Some evidence concerning the presence and impact of coopera tion was
produced but it is not clear ly pervasive
5 The dis tinction between convenience goods and non-c onvenie nce goods is
unambiguously a good one
6 Virtually no evidence concerning econ omies of scale in adver tis ing can
be gleaned fr om this study given its assumpt ions
More work is called for on many of these issues my study seems to have
raised more questions than it has answer ed This is pr obab lly due to the richshy
ne ss of the data source since I had the oppor tunity to address several issue s
at once
- -- - Q --
Foo tnotes
Manager Line o f Business Program Federal Trade Commission The author depended heavily on several staf f members of the LB Pro gram for statistical and clerical support in particular Joe Cholka George Pascoe and Haro lene Jenkins Helpful comments were given b y Richard Caves Dennis Mueller Michael Porter and F M Scherer When the paper was in its early stages o f development both S teve Garber and Jon Rasmussen former economists in the LB Program provided eery useful cri tiques The views expressed here are my own o f course
1 The basic exp licit model of the role of advertising is Dorfman and S teiner ( 1954 ) and the models in this paper may be seen as ano ther extens ion of their work
2 For a review o f models of oligopoly which treat prices and quantities as relevant variables see Shubik ( 1 959) Extens ions to take advertis ing int o account are fairly s traightforward
3 Dorfman and S teiner dealt with a price-setting firm For a monopolis t inverting the demand relation to show Pbull P (Q A) _instead o f Q- (P A) has no effect on the results o f the analysis For oligopolists who face a sys tem of demand relations inversion does have some impac t on the results The dif ferences are not trivial nonetheless they will no t be explored in this study
4 Shubik ( 1 959) p
5 From a formal mathematical point o f view the function is similar to a1 i Legrangian funct ion of the form - r rr
L bull 11 + A ( j - (i i i+j
where the indicates some fixed level of the profits industry Following the same logic which resul ts in and income constraint as the marginal utility of the of the indus tr in the obj ective of firm i
6 Hall amp Weiss ( 1 967) p 323
7 See Schmallensee ( 1 97 2 ) Ch 2 pp 1 6-47
8 On this poin t see Imel amp Helmberger ( 1 9 7 1 )
9 Long ( 1 970) p 130 - 1 39
1 0 See Schmallensee ( 1 972) Ch 3 pp 48-8 7
rrj i+j ) of the rest of the
the identification profits of the rest
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980
raquo
- 34
3 1
i l T 4 8LE S R E G E S S I ON R E SUL T S i X T f N S I ON S T O T Y PE Of
AND L E V E L OF C ONC E kT R A T I O
E O O tt S
tP bull T Y P E NR C ONV O B S tO N-CONV
E L A S T I shy
T I C J T Y
O E - t-41( T SliA R E
H( fl F
1 1 - K O x middot K O X E R F
- H S a 1 1 - H S t 1 8 1 1 9 1
R S O
1 1 0 1
q z
t - bull 9 f 2 1
Z t P D l 6 fC ONV - 1 9 1 s s c 2 1
e - l O I e bull u q a
2 3 1 0 1 1 Nl N- t O V 5 7 1 ( 1 2 19 1 6 1 - J l S 2 5 4
l l l l t z q o a 1 - J o v z I
l ( I NO l l A l l ol u l t o o c bull J u l l
d S
c 3 4 t t r a l
bull Z D 1 0 l l 4 l l 2 3 -gt 2 l l 2 0 - 4 lS 5 4 l C 4 1 o e 6 t l b b 5 1 1 0 1 1 1 c - 1 9 5 1 t d S 2 1 tV
OQ10 T-R A T I O S A E G I VE N l P A RE N T H E S E S l E V E L S f S I G 1 F l C ANC E A - 99 5 B - q 9 t - s NS J UC E NO I Nl E R C ( P T I S U (j f middot) I N f Hf t OUA TIOI S 6 E f 3R E A O J U S T H E N T F OR 1-t E T E R O SIC E O A S T f C I T V pIraquo
1
middot
I
R SO I S C A L C UL A T E D A S f I f bull 1 1 bull 1lt 1 K t t- E E N I S THf U A1 B E R O f fJe S E fi V A T I O S F t S HtE F - S T A T I S
I N DE P E ND E N T y II
T I C F O T H E U Y P O HIE S I S f A l l C JE F F I C H N T A R E QU A L HI l f RO A f40 K I S T H f UUM 8 E R OF Q ( A fL f S
I l l
1 A
1 6
t C
f 1 r R I l l
1 2 1
L Jt
L B
l q
L 1
on s T Y P E bull 2 )
P OUC T
X
H S I l l
l O l A I bull l 0 7 1
O O b O J
1 2 00 1
1 b 2 1 C I l l Z U
l l S b 1 -1 1 0 1
rlt F X l l - fi C f l
bull 1 I
l l
- ()4 2 3 l f H
1 - 2 7 9 1
K O X
1 1 - ti E R F I
l d l
S4 l i A I 2 l b
1 3 00
K O X -t f R F t S J
I 1 - HER F t
)
60 9
z o bull
0
t l l
3 8 8
1 7 6
5 64
64
N f bull
tl S bull
I l l
I 4 1
t ONV
NON-t ONV
A l l
A ll
c o v NO- ( ONV
1 4 1
1 5 1
- l S b
c - 1 1 2 1
5 8 2 C A I 1 6 7 0 1
middot I l 9 H
l 1 il
1 4 1 6 1
E L A C) T J shy
l t C I T Y
I S I
l b l
- 0 2 0
1 - 1 4 9 1
O Z 3 2 1 A I C b 6 b l
0 0 9 2 -J
1 1 b 4 1
o z t t t l 1 2 2 1 1
oru - Ht tt F
1 gt X bull b I
Page 25 bull
Mean val -as ior relevan t variables for the two sub-samples are
Conv Non-Conv
Adv Sales 4 7 2 3
iit Share 2 7 3 2
Ela cicity - 05 3 02 2
Herf indahl 7 3 1 860
That the dist nct ion between coft9enienee goods and non - convenience
goods mat ters is supported not only by these means but also by the regression
results Equat ions lA and lB are different in almost every respec t Only the
concentrat ionmarket share int eract ion term seems to mat ter for convenience
good s Jus t he opposite holds for non - convenience goods both elasticity1
and market sl e are highly significant but the concentrat ionmarket share L
int erac t ion t=rm is not
Given that one minus market share is used these equ3 tions the coeffici t
of market share is the negative of what is sh in col (6) of Table 5 For the
regress ion for non - convenience LB s reported in equation lB then LB s with larger
market shares have lower advertising to sales ratios
On the question of the impact of the degree of cooperation the story is mixed
For convenienc good s the coefficient is pos itive as equation (31) predicts when
there is a po sitive degree of cooperat ion Fo r the non-convenience goods LB s middot
concentration has no effec t
Aggregat ion t o the indus try level and then rer unning the equations $ives the
results shown in (2a-2d) of Tab le 5 As expected R2 goes up and levels of significanc e
fmiddotJr individual coefficients gJ down In add ition the coeffic ient of H ( l -H) inmiddot
quat ion 2A is no t significant though it is po s i tive Since 2A is the indus try level
I
--
all but one of them and that one is not significant For 18 of the 31
wo uld have on its adve rtising to sal es ratio
cognate of LA the variable H ( l-H) pl ays the same role in 2A as H ( l -z 1) p lays
in LA
Vconomies of scale in advert ising
The literature on advert ing contains a number of definitionbull o f economies
of scale and of findings eone rning its presence and magnitude Before I loo k
a t the data with the hope o f mmenting about scale effect s then I wan t to
define the term as I use it
By economies of scale I mean a decrease in the number o f units of adver-
tising which are needed to generate a unit of sales as advert ising is increased
Given the definition of the firm s own elast icity of demand with respect to
advertising (see the text foll wing equation (1) ) the technical definition
is that rii gt 1 Diseconomies of scale are def ined symmetrically i e
Yu lt 1
ith a cro s s - section oi observations on advertising intensity
(v bull a s1) and onmiddot advertising (a ) we may regress the former on the latter 1 i i
The results of such regress ions for the 32 industries are given in Table 2
lines lB ZB bull bull bull 32B For the 32 eases the regression coefficient is positive it
positive coefficients the coefficient i s signif icant
The que tion now is whether these results ar e evidence of pervasive dis-
economies of scale in advertislng I think no t Given my as sumption that the
data I observe are equilibrium results I may ri ghtly conclud e that almost every-
where high levels of advert ising are as sociated in equilibrium with high ratios
of advertising to sales I may no t conclud e anything however about the
impact that a move by some firm away from its equilibrium level o f advertising
middot- Page 27
Given the assumed relation in equation ( 37 ) there is a corre spondshy
ing equilibrium equation relating v and ai The equation is non - linear i
and I have not found a simp le way to characterize it It is po ssib le to
det ermi ne its deriviative with respect to a b y takin$ tne de riviative of ( 3 7 ) i
When that is done the result is
v - 1(40) i bull sl a a z ( 1-z )i i ira-l
Sinc e all the other terms in (40) are non-negative the sign dep ends only
on the sign of 6 1 which is in turn de pend ent only on the sign of (y - a)
That the signs of the coefficients of a in equations lB 2B bull bull bull 32B arei
not always equal to the signs of the coefficients of z in LA 2A bull bull bull 32A i
is still a mystery to me and something to be explored
bull
Summary
My purpose in writing this pa per wa s t o attempt to integrate explicit micro-
economic mode lling reasoned econome tric specification of error terms and high
quality data to exp lore some questions about adver tising by large firms
al quite important improvements could be made in the first two of those areas
and much data ma ssag ing is yet to be done in the third
Concerning the subst an ce of what I have done I think six things are im-
Page 2 8
Severshy
V and Conclusions
portant
1 Explicit modeling is worth the ef for t if for no other reas on than
that it provides a basis for choosing fr om a varie ty of funct ional
forms
2 The same is true for error specif ica t ion I did in fact look at scatshy
ter diagrams for all 32 industry categor ies and they vir tually all show
the het eroscedast icity which I as sumed and for which I correc ted
3 The predicted relation be tween advertis ing in tensity and market share
shows up clearly in on ly 25 of the case s examined I have ye t to
explore why that may b e the cas e
4 Some evidence concerning the presence and impact of coopera tion was
produced but it is not clear ly pervasive
5 The dis tinction between convenience goods and non-c onvenie nce goods is
unambiguously a good one
6 Virtually no evidence concerning econ omies of scale in adver tis ing can
be gleaned fr om this study given its assumpt ions
More work is called for on many of these issues my study seems to have
raised more questions than it has answer ed This is pr obab lly due to the richshy
ne ss of the data source since I had the oppor tunity to address several issue s
at once
- -- - Q --
Foo tnotes
Manager Line o f Business Program Federal Trade Commission The author depended heavily on several staf f members of the LB Pro gram for statistical and clerical support in particular Joe Cholka George Pascoe and Haro lene Jenkins Helpful comments were given b y Richard Caves Dennis Mueller Michael Porter and F M Scherer When the paper was in its early stages o f development both S teve Garber and Jon Rasmussen former economists in the LB Program provided eery useful cri tiques The views expressed here are my own o f course
1 The basic exp licit model of the role of advertising is Dorfman and S teiner ( 1954 ) and the models in this paper may be seen as ano ther extens ion of their work
2 For a review o f models of oligopoly which treat prices and quantities as relevant variables see Shubik ( 1 959) Extens ions to take advertis ing int o account are fairly s traightforward
3 Dorfman and S teiner dealt with a price-setting firm For a monopolis t inverting the demand relation to show Pbull P (Q A) _instead o f Q- (P A) has no effect on the results o f the analysis For oligopolists who face a sys tem of demand relations inversion does have some impac t on the results The dif ferences are not trivial nonetheless they will no t be explored in this study
4 Shubik ( 1 959) p
5 From a formal mathematical point o f view the function is similar to a1 i Legrangian funct ion of the form - r rr
L bull 11 + A ( j - (i i i+j
where the indicates some fixed level of the profits industry Following the same logic which resul ts in and income constraint as the marginal utility of the of the indus tr in the obj ective of firm i
6 Hall amp Weiss ( 1 967) p 323
7 See Schmallensee ( 1 97 2 ) Ch 2 pp 1 6-47
8 On this poin t see Imel amp Helmberger ( 1 9 7 1 )
9 Long ( 1 970) p 130 - 1 39
1 0 See Schmallensee ( 1 972) Ch 3 pp 48-8 7
rrj i+j ) of the rest of the
the identification profits of the rest
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980
Page 25 bull
Mean val -as ior relevan t variables for the two sub-samples are
Conv Non-Conv
Adv Sales 4 7 2 3
iit Share 2 7 3 2
Ela cicity - 05 3 02 2
Herf indahl 7 3 1 860
That the dist nct ion between coft9enienee goods and non - convenience
goods mat ters is supported not only by these means but also by the regression
results Equat ions lA and lB are different in almost every respec t Only the
concentrat ionmarket share int eract ion term seems to mat ter for convenience
good s Jus t he opposite holds for non - convenience goods both elasticity1
and market sl e are highly significant but the concentrat ionmarket share L
int erac t ion t=rm is not
Given that one minus market share is used these equ3 tions the coeffici t
of market share is the negative of what is sh in col (6) of Table 5 For the
regress ion for non - convenience LB s reported in equation lB then LB s with larger
market shares have lower advertising to sales ratios
On the question of the impact of the degree of cooperation the story is mixed
For convenienc good s the coefficient is pos itive as equation (31) predicts when
there is a po sitive degree of cooperat ion Fo r the non-convenience goods LB s middot
concentration has no effec t
Aggregat ion t o the indus try level and then rer unning the equations $ives the
results shown in (2a-2d) of Tab le 5 As expected R2 goes up and levels of significanc e
fmiddotJr individual coefficients gJ down In add ition the coeffic ient of H ( l -H) inmiddot
quat ion 2A is no t significant though it is po s i tive Since 2A is the indus try level
I
--
all but one of them and that one is not significant For 18 of the 31
wo uld have on its adve rtising to sal es ratio
cognate of LA the variable H ( l-H) pl ays the same role in 2A as H ( l -z 1) p lays
in LA
Vconomies of scale in advert ising
The literature on advert ing contains a number of definitionbull o f economies
of scale and of findings eone rning its presence and magnitude Before I loo k
a t the data with the hope o f mmenting about scale effect s then I wan t to
define the term as I use it
By economies of scale I mean a decrease in the number o f units of adver-
tising which are needed to generate a unit of sales as advert ising is increased
Given the definition of the firm s own elast icity of demand with respect to
advertising (see the text foll wing equation (1) ) the technical definition
is that rii gt 1 Diseconomies of scale are def ined symmetrically i e
Yu lt 1
ith a cro s s - section oi observations on advertising intensity
(v bull a s1) and onmiddot advertising (a ) we may regress the former on the latter 1 i i
The results of such regress ions for the 32 industries are given in Table 2
lines lB ZB bull bull bull 32B For the 32 eases the regression coefficient is positive it
positive coefficients the coefficient i s signif icant
The que tion now is whether these results ar e evidence of pervasive dis-
economies of scale in advertislng I think no t Given my as sumption that the
data I observe are equilibrium results I may ri ghtly conclud e that almost every-
where high levels of advert ising are as sociated in equilibrium with high ratios
of advertising to sales I may no t conclud e anything however about the
impact that a move by some firm away from its equilibrium level o f advertising
middot- Page 27
Given the assumed relation in equation ( 37 ) there is a corre spondshy
ing equilibrium equation relating v and ai The equation is non - linear i
and I have not found a simp le way to characterize it It is po ssib le to
det ermi ne its deriviative with respect to a b y takin$ tne de riviative of ( 3 7 ) i
When that is done the result is
v - 1(40) i bull sl a a z ( 1-z )i i ira-l
Sinc e all the other terms in (40) are non-negative the sign dep ends only
on the sign of 6 1 which is in turn de pend ent only on the sign of (y - a)
That the signs of the coefficients of a in equations lB 2B bull bull bull 32B arei
not always equal to the signs of the coefficients of z in LA 2A bull bull bull 32A i
is still a mystery to me and something to be explored
bull
Summary
My purpose in writing this pa per wa s t o attempt to integrate explicit micro-
economic mode lling reasoned econome tric specification of error terms and high
quality data to exp lore some questions about adver tising by large firms
al quite important improvements could be made in the first two of those areas
and much data ma ssag ing is yet to be done in the third
Concerning the subst an ce of what I have done I think six things are im-
Page 2 8
Severshy
V and Conclusions
portant
1 Explicit modeling is worth the ef for t if for no other reas on than
that it provides a basis for choosing fr om a varie ty of funct ional
forms
2 The same is true for error specif ica t ion I did in fact look at scatshy
ter diagrams for all 32 industry categor ies and they vir tually all show
the het eroscedast icity which I as sumed and for which I correc ted
3 The predicted relation be tween advertis ing in tensity and market share
shows up clearly in on ly 25 of the case s examined I have ye t to
explore why that may b e the cas e
4 Some evidence concerning the presence and impact of coopera tion was
produced but it is not clear ly pervasive
5 The dis tinction between convenience goods and non-c onvenie nce goods is
unambiguously a good one
6 Virtually no evidence concerning econ omies of scale in adver tis ing can
be gleaned fr om this study given its assumpt ions
More work is called for on many of these issues my study seems to have
raised more questions than it has answer ed This is pr obab lly due to the richshy
ne ss of the data source since I had the oppor tunity to address several issue s
at once
- -- - Q --
Foo tnotes
Manager Line o f Business Program Federal Trade Commission The author depended heavily on several staf f members of the LB Pro gram for statistical and clerical support in particular Joe Cholka George Pascoe and Haro lene Jenkins Helpful comments were given b y Richard Caves Dennis Mueller Michael Porter and F M Scherer When the paper was in its early stages o f development both S teve Garber and Jon Rasmussen former economists in the LB Program provided eery useful cri tiques The views expressed here are my own o f course
1 The basic exp licit model of the role of advertising is Dorfman and S teiner ( 1954 ) and the models in this paper may be seen as ano ther extens ion of their work
2 For a review o f models of oligopoly which treat prices and quantities as relevant variables see Shubik ( 1 959) Extens ions to take advertis ing int o account are fairly s traightforward
3 Dorfman and S teiner dealt with a price-setting firm For a monopolis t inverting the demand relation to show Pbull P (Q A) _instead o f Q- (P A) has no effect on the results o f the analysis For oligopolists who face a sys tem of demand relations inversion does have some impac t on the results The dif ferences are not trivial nonetheless they will no t be explored in this study
4 Shubik ( 1 959) p
5 From a formal mathematical point o f view the function is similar to a1 i Legrangian funct ion of the form - r rr
L bull 11 + A ( j - (i i i+j
where the indicates some fixed level of the profits industry Following the same logic which resul ts in and income constraint as the marginal utility of the of the indus tr in the obj ective of firm i
6 Hall amp Weiss ( 1 967) p 323
7 See Schmallensee ( 1 97 2 ) Ch 2 pp 1 6-47
8 On this poin t see Imel amp Helmberger ( 1 9 7 1 )
9 Long ( 1 970) p 130 - 1 39
1 0 See Schmallensee ( 1 972) Ch 3 pp 48-8 7
rrj i+j ) of the rest of the
the identification profits of the rest
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980
--
all but one of them and that one is not significant For 18 of the 31
wo uld have on its adve rtising to sal es ratio
cognate of LA the variable H ( l-H) pl ays the same role in 2A as H ( l -z 1) p lays
in LA
Vconomies of scale in advert ising
The literature on advert ing contains a number of definitionbull o f economies
of scale and of findings eone rning its presence and magnitude Before I loo k
a t the data with the hope o f mmenting about scale effect s then I wan t to
define the term as I use it
By economies of scale I mean a decrease in the number o f units of adver-
tising which are needed to generate a unit of sales as advert ising is increased
Given the definition of the firm s own elast icity of demand with respect to
advertising (see the text foll wing equation (1) ) the technical definition
is that rii gt 1 Diseconomies of scale are def ined symmetrically i e
Yu lt 1
ith a cro s s - section oi observations on advertising intensity
(v bull a s1) and onmiddot advertising (a ) we may regress the former on the latter 1 i i
The results of such regress ions for the 32 industries are given in Table 2
lines lB ZB bull bull bull 32B For the 32 eases the regression coefficient is positive it
positive coefficients the coefficient i s signif icant
The que tion now is whether these results ar e evidence of pervasive dis-
economies of scale in advertislng I think no t Given my as sumption that the
data I observe are equilibrium results I may ri ghtly conclud e that almost every-
where high levels of advert ising are as sociated in equilibrium with high ratios
of advertising to sales I may no t conclud e anything however about the
impact that a move by some firm away from its equilibrium level o f advertising
middot- Page 27
Given the assumed relation in equation ( 37 ) there is a corre spondshy
ing equilibrium equation relating v and ai The equation is non - linear i
and I have not found a simp le way to characterize it It is po ssib le to
det ermi ne its deriviative with respect to a b y takin$ tne de riviative of ( 3 7 ) i
When that is done the result is
v - 1(40) i bull sl a a z ( 1-z )i i ira-l
Sinc e all the other terms in (40) are non-negative the sign dep ends only
on the sign of 6 1 which is in turn de pend ent only on the sign of (y - a)
That the signs of the coefficients of a in equations lB 2B bull bull bull 32B arei
not always equal to the signs of the coefficients of z in LA 2A bull bull bull 32A i
is still a mystery to me and something to be explored
bull
Summary
My purpose in writing this pa per wa s t o attempt to integrate explicit micro-
economic mode lling reasoned econome tric specification of error terms and high
quality data to exp lore some questions about adver tising by large firms
al quite important improvements could be made in the first two of those areas
and much data ma ssag ing is yet to be done in the third
Concerning the subst an ce of what I have done I think six things are im-
Page 2 8
Severshy
V and Conclusions
portant
1 Explicit modeling is worth the ef for t if for no other reas on than
that it provides a basis for choosing fr om a varie ty of funct ional
forms
2 The same is true for error specif ica t ion I did in fact look at scatshy
ter diagrams for all 32 industry categor ies and they vir tually all show
the het eroscedast icity which I as sumed and for which I correc ted
3 The predicted relation be tween advertis ing in tensity and market share
shows up clearly in on ly 25 of the case s examined I have ye t to
explore why that may b e the cas e
4 Some evidence concerning the presence and impact of coopera tion was
produced but it is not clear ly pervasive
5 The dis tinction between convenience goods and non-c onvenie nce goods is
unambiguously a good one
6 Virtually no evidence concerning econ omies of scale in adver tis ing can
be gleaned fr om this study given its assumpt ions
More work is called for on many of these issues my study seems to have
raised more questions than it has answer ed This is pr obab lly due to the richshy
ne ss of the data source since I had the oppor tunity to address several issue s
at once
- -- - Q --
Foo tnotes
Manager Line o f Business Program Federal Trade Commission The author depended heavily on several staf f members of the LB Pro gram for statistical and clerical support in particular Joe Cholka George Pascoe and Haro lene Jenkins Helpful comments were given b y Richard Caves Dennis Mueller Michael Porter and F M Scherer When the paper was in its early stages o f development both S teve Garber and Jon Rasmussen former economists in the LB Program provided eery useful cri tiques The views expressed here are my own o f course
1 The basic exp licit model of the role of advertising is Dorfman and S teiner ( 1954 ) and the models in this paper may be seen as ano ther extens ion of their work
2 For a review o f models of oligopoly which treat prices and quantities as relevant variables see Shubik ( 1 959) Extens ions to take advertis ing int o account are fairly s traightforward
3 Dorfman and S teiner dealt with a price-setting firm For a monopolis t inverting the demand relation to show Pbull P (Q A) _instead o f Q- (P A) has no effect on the results o f the analysis For oligopolists who face a sys tem of demand relations inversion does have some impac t on the results The dif ferences are not trivial nonetheless they will no t be explored in this study
4 Shubik ( 1 959) p
5 From a formal mathematical point o f view the function is similar to a1 i Legrangian funct ion of the form - r rr
L bull 11 + A ( j - (i i i+j
where the indicates some fixed level of the profits industry Following the same logic which resul ts in and income constraint as the marginal utility of the of the indus tr in the obj ective of firm i
6 Hall amp Weiss ( 1 967) p 323
7 See Schmallensee ( 1 97 2 ) Ch 2 pp 1 6-47
8 On this poin t see Imel amp Helmberger ( 1 9 7 1 )
9 Long ( 1 970) p 130 - 1 39
1 0 See Schmallensee ( 1 972) Ch 3 pp 48-8 7
rrj i+j ) of the rest of the
the identification profits of the rest
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980
middot- Page 27
Given the assumed relation in equation ( 37 ) there is a corre spondshy
ing equilibrium equation relating v and ai The equation is non - linear i
and I have not found a simp le way to characterize it It is po ssib le to
det ermi ne its deriviative with respect to a b y takin$ tne de riviative of ( 3 7 ) i
When that is done the result is
v - 1(40) i bull sl a a z ( 1-z )i i ira-l
Sinc e all the other terms in (40) are non-negative the sign dep ends only
on the sign of 6 1 which is in turn de pend ent only on the sign of (y - a)
That the signs of the coefficients of a in equations lB 2B bull bull bull 32B arei
not always equal to the signs of the coefficients of z in LA 2A bull bull bull 32A i
is still a mystery to me and something to be explored
bull
Summary
My purpose in writing this pa per wa s t o attempt to integrate explicit micro-
economic mode lling reasoned econome tric specification of error terms and high
quality data to exp lore some questions about adver tising by large firms
al quite important improvements could be made in the first two of those areas
and much data ma ssag ing is yet to be done in the third
Concerning the subst an ce of what I have done I think six things are im-
Page 2 8
Severshy
V and Conclusions
portant
1 Explicit modeling is worth the ef for t if for no other reas on than
that it provides a basis for choosing fr om a varie ty of funct ional
forms
2 The same is true for error specif ica t ion I did in fact look at scatshy
ter diagrams for all 32 industry categor ies and they vir tually all show
the het eroscedast icity which I as sumed and for which I correc ted
3 The predicted relation be tween advertis ing in tensity and market share
shows up clearly in on ly 25 of the case s examined I have ye t to
explore why that may b e the cas e
4 Some evidence concerning the presence and impact of coopera tion was
produced but it is not clear ly pervasive
5 The dis tinction between convenience goods and non-c onvenie nce goods is
unambiguously a good one
6 Virtually no evidence concerning econ omies of scale in adver tis ing can
be gleaned fr om this study given its assumpt ions
More work is called for on many of these issues my study seems to have
raised more questions than it has answer ed This is pr obab lly due to the richshy
ne ss of the data source since I had the oppor tunity to address several issue s
at once
- -- - Q --
Foo tnotes
Manager Line o f Business Program Federal Trade Commission The author depended heavily on several staf f members of the LB Pro gram for statistical and clerical support in particular Joe Cholka George Pascoe and Haro lene Jenkins Helpful comments were given b y Richard Caves Dennis Mueller Michael Porter and F M Scherer When the paper was in its early stages o f development both S teve Garber and Jon Rasmussen former economists in the LB Program provided eery useful cri tiques The views expressed here are my own o f course
1 The basic exp licit model of the role of advertising is Dorfman and S teiner ( 1954 ) and the models in this paper may be seen as ano ther extens ion of their work
2 For a review o f models of oligopoly which treat prices and quantities as relevant variables see Shubik ( 1 959) Extens ions to take advertis ing int o account are fairly s traightforward
3 Dorfman and S teiner dealt with a price-setting firm For a monopolis t inverting the demand relation to show Pbull P (Q A) _instead o f Q- (P A) has no effect on the results o f the analysis For oligopolists who face a sys tem of demand relations inversion does have some impac t on the results The dif ferences are not trivial nonetheless they will no t be explored in this study
4 Shubik ( 1 959) p
5 From a formal mathematical point o f view the function is similar to a1 i Legrangian funct ion of the form - r rr
L bull 11 + A ( j - (i i i+j
where the indicates some fixed level of the profits industry Following the same logic which resul ts in and income constraint as the marginal utility of the of the indus tr in the obj ective of firm i
6 Hall amp Weiss ( 1 967) p 323
7 See Schmallensee ( 1 97 2 ) Ch 2 pp 1 6-47
8 On this poin t see Imel amp Helmberger ( 1 9 7 1 )
9 Long ( 1 970) p 130 - 1 39
1 0 See Schmallensee ( 1 972) Ch 3 pp 48-8 7
rrj i+j ) of the rest of the
the identification profits of the rest
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980
Summary
My purpose in writing this pa per wa s t o attempt to integrate explicit micro-
economic mode lling reasoned econome tric specification of error terms and high
quality data to exp lore some questions about adver tising by large firms
al quite important improvements could be made in the first two of those areas
and much data ma ssag ing is yet to be done in the third
Concerning the subst an ce of what I have done I think six things are im-
Page 2 8
Severshy
V and Conclusions
portant
1 Explicit modeling is worth the ef for t if for no other reas on than
that it provides a basis for choosing fr om a varie ty of funct ional
forms
2 The same is true for error specif ica t ion I did in fact look at scatshy
ter diagrams for all 32 industry categor ies and they vir tually all show
the het eroscedast icity which I as sumed and for which I correc ted
3 The predicted relation be tween advertis ing in tensity and market share
shows up clearly in on ly 25 of the case s examined I have ye t to
explore why that may b e the cas e
4 Some evidence concerning the presence and impact of coopera tion was
produced but it is not clear ly pervasive
5 The dis tinction between convenience goods and non-c onvenie nce goods is
unambiguously a good one
6 Virtually no evidence concerning econ omies of scale in adver tis ing can
be gleaned fr om this study given its assumpt ions
More work is called for on many of these issues my study seems to have
raised more questions than it has answer ed This is pr obab lly due to the richshy
ne ss of the data source since I had the oppor tunity to address several issue s
at once
- -- - Q --
Foo tnotes
Manager Line o f Business Program Federal Trade Commission The author depended heavily on several staf f members of the LB Pro gram for statistical and clerical support in particular Joe Cholka George Pascoe and Haro lene Jenkins Helpful comments were given b y Richard Caves Dennis Mueller Michael Porter and F M Scherer When the paper was in its early stages o f development both S teve Garber and Jon Rasmussen former economists in the LB Program provided eery useful cri tiques The views expressed here are my own o f course
1 The basic exp licit model of the role of advertising is Dorfman and S teiner ( 1954 ) and the models in this paper may be seen as ano ther extens ion of their work
2 For a review o f models of oligopoly which treat prices and quantities as relevant variables see Shubik ( 1 959) Extens ions to take advertis ing int o account are fairly s traightforward
3 Dorfman and S teiner dealt with a price-setting firm For a monopolis t inverting the demand relation to show Pbull P (Q A) _instead o f Q- (P A) has no effect on the results o f the analysis For oligopolists who face a sys tem of demand relations inversion does have some impac t on the results The dif ferences are not trivial nonetheless they will no t be explored in this study
4 Shubik ( 1 959) p
5 From a formal mathematical point o f view the function is similar to a1 i Legrangian funct ion of the form - r rr
L bull 11 + A ( j - (i i i+j
where the indicates some fixed level of the profits industry Following the same logic which resul ts in and income constraint as the marginal utility of the of the indus tr in the obj ective of firm i
6 Hall amp Weiss ( 1 967) p 323
7 See Schmallensee ( 1 97 2 ) Ch 2 pp 1 6-47
8 On this poin t see Imel amp Helmberger ( 1 9 7 1 )
9 Long ( 1 970) p 130 - 1 39
1 0 See Schmallensee ( 1 972) Ch 3 pp 48-8 7
rrj i+j ) of the rest of the
the identification profits of the rest
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980
- -- - Q --
Foo tnotes
Manager Line o f Business Program Federal Trade Commission The author depended heavily on several staf f members of the LB Pro gram for statistical and clerical support in particular Joe Cholka George Pascoe and Haro lene Jenkins Helpful comments were given b y Richard Caves Dennis Mueller Michael Porter and F M Scherer When the paper was in its early stages o f development both S teve Garber and Jon Rasmussen former economists in the LB Program provided eery useful cri tiques The views expressed here are my own o f course
1 The basic exp licit model of the role of advertising is Dorfman and S teiner ( 1954 ) and the models in this paper may be seen as ano ther extens ion of their work
2 For a review o f models of oligopoly which treat prices and quantities as relevant variables see Shubik ( 1 959) Extens ions to take advertis ing int o account are fairly s traightforward
3 Dorfman and S teiner dealt with a price-setting firm For a monopolis t inverting the demand relation to show Pbull P (Q A) _instead o f Q- (P A) has no effect on the results o f the analysis For oligopolists who face a sys tem of demand relations inversion does have some impac t on the results The dif ferences are not trivial nonetheless they will no t be explored in this study
4 Shubik ( 1 959) p
5 From a formal mathematical point o f view the function is similar to a1 i Legrangian funct ion of the form - r rr
L bull 11 + A ( j - (i i i+j
where the indicates some fixed level of the profits industry Following the same logic which resul ts in and income constraint as the marginal utility of the of the indus tr in the obj ective of firm i
6 Hall amp Weiss ( 1 967) p 323
7 See Schmallensee ( 1 97 2 ) Ch 2 pp 1 6-47
8 On this poin t see Imel amp Helmberger ( 1 9 7 1 )
9 Long ( 1 970) p 130 - 1 39
1 0 See Schmallensee ( 1 972) Ch 3 pp 48-8 7
rrj i+j ) of the rest of the
the identification profits of the rest
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980
Advertising
- -- -middot
J Econ Lit
Rev
Rev
Study Manufacturing Industry
Henning Econ
Journal
Advertising
Strategy
Econ
Report
BIBLIOGRAPHY
Brow Randall S Estimating Advantages to Large-Scale Advertising Rev Econ Stat ist August 1978 60 pp 428-3 7
Comancr William S and WUson Thomas A and Market P ower Cambridge Harvard University Press 1 974
The Effect of Advertising on Competition A Survey June 1979 17 pp 453-476
Dorfman Robert and Steiner Peter 0 Optimal Advertis ing and Optimal Quality Amer Econ Dec 1954 44 pp 826-36
Hall Marshall amp Weiss Leonard W Firm Size and Profitab ility Rev Econ Statist August 1967 4 9 pp 3 1 9-33 1
mel Blake and Helmberger Peter Estimation of Structure-Profit Relationship s with Application to the Food Proc es sion Sector Amer Econ Sep t 1971 6 1 (4) pp 6 14-27
Long William F An Econometric of Performance in American 1955-1958 Unpublished Ph D Dissertation
Berkeley University of California 1 970
Mann H Michael J A and Meehan J W Jr Advertising and Concentration An Empirical Investigation J Ind Nov 1967 1 6 pp 34-45
Martin Stephen Advertising Concentration and Prof itability the Simultaniety Problem Bell Autumn 1979 10middot pp 639-647
Schmalensee Richard The Economics o f Amsterdam North-Holland 1972
Shubik Martin and Market Structure New York Wil y 1959
Strickland Allyn D and Weiss Leonard W Advertising Concentration Price-ltost Margins Polit Oct 197 6 84 (5) pp 1 109-2 1
U S Federal Trade Commis sion Annual Line of Bus ine ss 1 97 4 -
Washington 1980