Econometrics I 21

Embed Size (px)

Citation preview

  • 7/25/2019 Econometrics I 21

    1/67

    Part 21: Generalized Method of Moments1-1/67

    Econometrics I

    Professor William Greene

    Stern School of Business

    Department of Economics

  • 7/25/2019 Econometrics I 21

    2/67

    Part 21: Generalized Method of Moments1-2/67

    Econometrics I

    Part 21 Generalized

    Method of Moments

  • 7/25/2019 Econometrics I 21

    3/67

    Part 21: Generalized Method of Moments1-3/67

    I also have a questions about nonlinear GMM - which is more or less nonlinear IV technique

    I suppose.

    I am running a panel non-linear regression (non-linear in the parameters) and I have L

    parameters and K eogenous variables with L!K.

    In particular m" model loo#s #ind o$ li#e this% & ' b*+b, e and so I am tr"ing to

    estimate the etra b,that don/t usuall" appear in a regression.

    0rom what I am reading to run nonlinear GMM I can use the K eogenous variables toconstruct the orthogonalit" conditions but what should I use $or the etra b,coe$$icients1

    2ust some more possible IVs (li#e lags) o$ the eogenous variables1

    I agree that b" adding more IVs "ou will get a more e$$icient estimation but isn/t it onl" the

    case when "ou believe the IVs are trul" uncorrelated with the error term1

    3o b" adding more 4instruments4 "ou are more or less imposing more and more restrictive

    assumptions about the model (which might not actuall" be true).

    I am as#ing because I have not $ound sources comparing nonlinear GMM5IV to nonlinear

    least squares. I$ there is no homoscadesticit"5serial correlation what is more e$$icient5give

    tighter estimates1

  • 7/25/2019 Econometrics I 21

    4/67

    Part 21: Generalized Method of Moments1-4/67

    Im trin! to minimize a nonlinear pro!ram "ith theleast s#uare under nonlinear constraints$ Its firstintroduced % &n' ( Geman )2***+$ It consisted onthe minimization of the sum of s#uared difference%et"een the moment !eneratin! function and thetheoretical moment !eneratin! function

  • 7/25/2019 Econometrics I 21

    5/67

    Part 21: Generalized Method of Moments1-5/67

    , ,,

    , ,,

    Method o$ Moment Generating 0unctions

    0or the normal distribution the MG0 is

    M(t6 )'78ep(t)9'ep8t 9

    Moment 7quations% ep( ) ep8t 9 : ,.

    ;hoose two values o$ t

    n

    j i j ji

    t

    t x tn

    =

    = =

    , , , ,

    , ,

    and solve the two moment equations $or and .

    Miture o$ se the method o$ moment generating $unctions with ? values o$ t.

    M(t6 )'7

    N N

    = +

    , , , , , ,, ,

    8ep(t)9' ep8t 9 ( )ep8t 9t t +

  • 7/25/2019 Econometrics I 21

    6/67

    Part 21: Generalized Method of Moments1-6/67

    ( )

    ,

    , ,

    ? , , , , , ,, ,

    0inding the solutions to the moment equations% Least squares@M(t ) ep( ) and li#ewise $or t ...

    MinimiAe( )

    @M(t ) ep8t 9 ( )ep8t 9

    Bltern

    n

    ii

    jj

    t xn

    t t

    =

    =

    =

    +

    { } , , , ,

    ative estimator% Maimum Li#elihood

    L( ) log 8 6 9 ( ) 8 6 9

    N

    i ii N N == +

  • 7/25/2019 Econometrics I 21

    7/67Part 21: Generalized Method of Moments1-7/67

    ,G-S

    ,easi%le G-S is %ased on findin! an estimator

    "hich has the same properties as the true G-S$

    E.ample /ar0i 20E.p)zi+2$

    3rue G-S "ould re!ress 40E.p)zi+on the same transformation of xi$

    With a consistent estimator of 055 sa 0s5c5 "e dothe same computation "ith our estimates$

    So lon! as plim 0s5c 055 ,G-S is as 6!ood7 astrue G-S$ 8onsistent Same &smptotic /ariance Same &smptotic 9ormal Distri%ution

  • 7/25/2019 Econometrics I 21

    8/67Part 21: Generalized Method of Moments1-8/67

    3he Method of Moments

    1 2 ;

    9 91 1 i 1 i i 1 i9 9

    Estimatin! Parameters of Distri%utions Moment E#uation> ) 15$$$5;+

    m . f ) 5 5$$$5 +

    Method of Moments

    ? ! )m 5$$$5m +5 15$$$5;

    =

    = =

    = =

    =

  • 7/25/2019 Econometrics I 21

    9/67Part 21: Generalized Method of Moments1-/67

    Estimatin! a Parameter Mean of Poisson

    p)+e.p)=@+ @4 A

    E0 @$plim )149+ii @$

    3his is the estimator

    Mean of E.ponential

    p)+ @ e.p)= @+ E0 14 @$

    plim )149+ii 14@

  • 7/25/2019 Econometrics I 21

    10/67Part 21: Generalized Method of Moments1-1!/67

    Mean and /ariance of a

    9ormal Distri%ution

    = =

    = =

    =

    = = +

    = = +

    = =

    2

    2

    2 2 2

    9 9 2 2 21 1i 1 i i 1 i9 9

    2 9 2 2 9 21 1i 1 i i 1 i9 n

    1 ) +p)+ e.p

    22

    Population MomentsE0 5 E0

    Moment E#uations

    5

    Method of Moments Estimators

    ?5 ? ) + ) +

  • 7/25/2019 Econometrics I 21

    11/67Part 21: Generalized Method of Moments1-11/67

    Gamma Distri%utionP P 1

    2

    2

    e.p) +p)+

    )P+

    PE0

    P)P 1+E0

    E014 P 1

    E0lo! )P+ lo! 5 )P+dln )P+4dP

    )Each pair !iCes a different ans"er$ Is there a >%est> pairD es5

    the ones that are >sufficient> statistics$

    =

    =

    +=

    =

    =

    E0 and E0lo!$ ,or a

    different course$$$$+

  • 7/25/2019 Econometrics I 21

    12/67Part 21: Generalized Method of Moments1-12/67

    3he -inear Fe!ression Modeli i i

    i i

    9

    i i1 1 i2 2 i; ; i1i 1

    9

    i i1 1 i2 2 i; ; i2i 1

    9

    i i1 1 i2 2 i; ; i;i 1

    Population

    .

    Population E.pectation

    E0 . *

    Moment E#uations1

    ) . . $$$ . +. *9

    1) . . $$$ . +. *

    9

    $$$

    1) . . $$$ . +. *

    9

    Solution : -inea

    =

    =

    =

    = +

    =

    =

    =

    =

    r sstem of ; e#uations in ; unno"ns$

    -east S#uares

  • 7/25/2019 Econometrics I 21

    13/67Part 21: Generalized Method of Moments1-13/67

    Instrumental /aria%les

    i i i

    i i 1 ;

    9

    i i1 1 i2 2 i; ; i1i 1

    9

    i i1 1 i2 2 i; ; i2i 1

    i i1 1i 1

    Population

    .

    Population E.pectation

    E0 z * for instrumental Caria%les z $$$ z $

    Moment E#uations

    1) . . $$$ . +z *

    9

    1) . . $$$ . +z *

    9

    $$$1

    ) .9

    =

    =

    =

    = +

    =

    =

    =

    9

    i2 2 i; ; i;

    =1

    I/

    . $$$ . +z *

    Solution : &lso a linear sstem of ; e#uations in ; unno"ns$

    % ) +

    =

    Z'X Z'y/n ( /n)

  • 7/25/2019 Econometrics I 21

    14/67Part 21: Generalized Method of Moments1-14/67

    Ma.imum -ielihood91i 1 i i 1 ;9

    9 i i 1 ;

    i 1

    -o! lielihood function5 lo!- lo!f) G . 5 5$$$5 +

    Population E.pectations

    lo!-E *5 15$$$5;

    Sample Moments

    lo!f) G . 5 5$$$5 +1*

    9

    Solution : ; nonlinear e#uations in ; un

    =

    =

    =

    =

    9 i i 15M-E ;5M-E

    i 15M-E

    no"ns$

    ? ?lo!f) G . 5 5$$$5 +1*

    ?9 =

    =

  • 7/25/2019 Econometrics I 21

    15/67Part 21: Generalized Method of Moments1-15/67

    BehaCioral &pplication

    +

    +

    +

    + = + =

    =

    =

    =

    + =

    t 1t t

    t

    t

    t

    tH1 t 1 t

    t t 1 t

    -ife 8cle 8onsumption )te.t5 pa!e JJ+

    c1E )1 r+ 1 *

    1 c

    discount rate

    c consumption

    information at time t

    -et 14)1H +5 F c 4 c 5 =

    E 0 )1 r+F 1 *

    What is in the information set Each piece of >information>

    proCides a moment e#uation for estimation of the t"o parameters$

  • 7/25/2019 Econometrics I 21

    16/67Part 21: Generalized Method of Moments1-16/67

    Identification

    8an the parameters %e estimated

    9ot a sample Kpropert

    &ssume an infinite sample Is there sufficient information in a sample to reCealconsistent estimators of the parameters

    8an the Kmoment e#uations %e solCed for the

    population parameters

  • 7/25/2019 Econometrics I 21

    17/67Part 21: Generalized Method of Moments1-17/67

    Identification

    E.actl Identified 8ase: ; population moment e#uations

    in ; unno"n parameters$ Lur familiar cases5 L-S5 I/5 M-5 the MLM estimators

    Is the countin! rule sufficient

    What else is needed

    LCeridentified 8ase Instrumental /aria%les

    & coCariance structures model

  • 7/25/2019 Econometrics I 21

    18/67Part 21: Generalized Method of Moments1-18/67

    LCeridentification

    i i i 1 ;

    i i 1 M

    9

    i i1 1 i2 2 i; ; i1i 1

    Population

    . 5 5$$$5

    Population E.pectation

    E0 z * for instrumental Caria%les z $$$ z M ;$

    3here are M ; Moment E#uations = more than necessar1

    ) . . $$$ . +z *9

    1

    =

    = +

    =

    =9

    i i1 1 i2 2 i; ; i2i 1

    9

    i i1 1 i2 2 i; ; iMi 1

    ) . . $$$ . +z *9

    $$$1

    ) . . $$$ . +z *9

    Solution: & linear sstem of M e#uations in ; unno"ns$ DDDDD

    =

    =

    =

    =

  • 7/25/2019 Econometrics I 21

    19/67Part 21: Generalized Method of Moments1-1/67

    LCeridentification

    = +

    = +

    =

    =

    1 1 1

    2 2 2

    1 1 1

    2 2 2

    3"o E#uation 8oCariance Structures Model

    8ountr 1:

    8ountr 2:

    3"o Population Moment 8onditions:E0)143+ >) +

    E0)143+ >) +

    )1+ No" do "e com%ine the t"o sets of e#

    " #

    " #

    # " # !

    # " # !

    =

    1 2

    1 2 2

    uations

    )2+ GiCen t"o L-S estimates5 and 5 ho" do "e

    reconcile them

    9ote: 3here are eCen more$ E0)143+ >) + $

    $ $

    # " # !

  • 7/25/2019 Econometrics I 21

    20/67Part 21: Generalized Method of Moments1-2!/67

  • 7/25/2019 Econometrics I 21

    21/67Part 21: Generalized Method of Moments1-21/67

    = +

    =

    z

    x

    i i i

    i i 0 i0

    i

    Consider the Mover - Stayer Model

    Binary choice for whether an individual 'moves' or 'stays'

    d ( u 0)

    !utcome e"uation for the individual# conditional on the state$

    y % (d 0) &

    y % (d ) +

    xi i' '

    i0 i 0 0

    &

    ( # ) *(0#0)#( # # )+

    ,n individual either moves or stays# ut not oth (or neither).he arameter cannot e estimated with the oserved data

    re1ardless of the samle si2e. 3t is unidenti4ed.

  • 7/25/2019 Econometrics I 21

    22/67

    Part 21: Generalized Method of Moments1-22/67

  • 7/25/2019 Econometrics I 21

    23/67

    Part 21: Generalized Method of Moments1-23/67

  • 7/25/2019 Econometrics I 21

    24/67

    Part 21: Generalized Method of Moments1-24/67

    9onlinear -east S#uares--> NAMELIST ; x = one,age,educ,female,hhkids,married --> !alc ; k=col"x# --> NLS$ ; Lhs = hhninc ; %cn = & ' ex&"()*x#

    ; la(els = k+(,& ; sar = k+,) ; maxi = .Moment matrix has become nonpositive definite.

    Switching to BFGS algorithm

    Normal exit: 16 iterations. Stats!". F! #$1.1"%$

    &&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&

    'ser (efined )ptimi*ation.........................

    Nonlinear least s+ares regression ............

    ,-S!--NN/ Mean ! .#0%"$

    Standard deviation ! .1621

    Nmber of observs. ! %#%6Model si*e 3arameters !

    (egrees of freedom ! %#12

    4esidals Sm of s+ares ! 6%.%"001

    Standard error of e ! .16"1

    &&&&&&&&5&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&

    ariable7 /oefficient Standard 8rror b9St.8r. 37;7% B#7 &."00%@@@ .""1"0 &0%.$"2 .""""

    B>7 &."1$>#@@@ .""0$" .1$" .""10

    B07 ."0>>0@@@ .""660 $.1$> .""""

    B67 &.%6>%>@@@ .""$%# %.1"2 .""""

    37 .6#%#2 2"00.>2# .""" .2222 ?!!!!!!!

    &&&&&&&&5&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&

    Nonlinear least s+ares did not worA. hat is the implication of the

    infinite standard errors for B1 Cthe constantD and 3.

  • 7/25/2019 Econometrics I 21

    25/67

    Part 21: Generalized Method of Moments1-25/67

    Ma.imum -ielihood&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&Gamma C,oglinearD 4egression Model

    (ependent variable --NN/

    ,og liAelihood fnction 1>%2#.""%1>4estricted log liAelihood 1120."620#/hi s+ared 6 d.f.= %6120.$60%%

    Significance level ."""""

    McFadden 3sedo 4&s+ared &1".20220# C> observations with income ! "8stimation based on N ! %#%%E ! were deleted so log, was comptable.D

    &&&&&&&&5&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&

    ariable7 /oefficient Standard 8rror b9St.8r. 37;7"$>1@@@ ."%10> 10$.%1# .""""

    HG87 .""%"0@@@ ."""%$ .>1# ."""" >#.0%%

    8('/7 &."00%@@@ .""1%" &>6.>26 ."""" 11.#%"% F8MH,87 &.""0>% .""0>0 &.220 .#12$ .>$$1

    --(S7 ."601%@@@ .""61$ 1".0>% ."""" .>"%%MH448(7 &.%6#>1@@@ .""62% $.">1 ."""" .0$62

    7Scale parameter for gamma model3Iscale7 0.1%>$6@@@ .">%0" 1%".02> .""""&&&&&&&&5&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&

    M,8 apparentlJ worAed fine. KhJ did one method CnlsD fail and anotherconsistent estimator worA withot difficltJL

  • 7/25/2019 Econometrics I 21

    26/67

    Part 21: Generalized Method of Moments1-26/67

    Moment E#uations: 9-S

    ( ) ,

    8 6 9 5 ep( )

    / 5 ep( )

    ,/E

    ep( )

    ,/

    ep( );onsider the term $or the constant in the model .

  • 7/25/2019 Econometrics I 21

    27/67

    Part 21: Generalized Method of Moments1-27/67

    Moment E#uations M-E

    ( )

    Fhe log li#elihood $unction and li#elihood equations are

    logL' log log ( ) ( ) log

    log log ( )log ( ) log E ( )

    log using .

    Hecall

    N

    i i i ii

    N

    i ii

    N ii i i i ii

    i i

    P P y P y

    L d PP y P

    P dP

    L Py

    =

    =

    =

    +

    = + = =

    = =

    x

    the epected values o$ the derivatives o$ the log li#elihood equal

    Aero. 3o a loo# at the $irst equation reveals that the moment equation in

    use $or estimating = is 78log" 6 9 ( ) log and anothei i iP= x r K moment

    equations 7 " E are also in use. 3o the ML7 uses K

    $unctionall" independent moment equations $or K parameters while

  • 7/25/2019 Econometrics I 21

    28/67

    Part 21: Generalized Method of Moments1-28/67

    &!enda

    3he Method of Moments$ SolCin! the moment e#uationsE.actl identified cases

    LCeridentified cases

    8onsistenc$ No" do "e no" the method of moments is

    consistent&smptotic coCariance matri.$

    8onsistent Cs$ Efficient estimation

    & "ei!htin! matri.

    3he minimum distance estimator

    What is the efficient "ei!htin! matri.

    Estimatin! the "ei!htin! matri.$3he Generalized method of moments estimator = ho" it is computed$

    8omputin! the appropriate asmptotic coCariance matri.

  • 7/25/2019 Econometrics I 21

    29/67

    Part 21: Generalized Method of Moments1-2/67

    3he Method of Moments

    Moment E#uation: Defines a sample statistic that

    mimics a population e.pectation:

    %he &o&'lation ex&ectation ortho(onalit"

    condition:

    E0 mi)

    + !$ Su%script i indicates it dependson data Cector inde.ed % >i> )or >t> for a time

    series settin!+

  • 7/25/2019 Econometrics I 21

    30/67

    Part 21: Generalized Method of Moments1-3!/67

    3he Method of Moments = E.ample

    P P 1

    i ii

    i i

    9

    1 i1 i

    9

    2 i1 i

    Gamma Distri%ution Parameters

    e.p) +p) +

    )P+Population Moment 8onditions

    PE0 5 E0lo! )P+ lo!

    Moment E#uations:

    E0m ) 5P+ E0Q)14n+ R P 4 *

    E0m ) 5P+ E0Q)14n+ lo! R )

    =

    = =

    =

    )P+ lo! + * =

  • 7/25/2019 Econometrics I 21

    31/67

    Part 21: Generalized Method of Moments1-31/67

    &pplication

    7lot of 7si(7)8unction

    P

    -10

    -8

    -6

    -4

    -2

    0

    2

    -12

    1 2 3 4 5 60

    PSI

    2 21 1 2 2

    2 2

    1 2

    1

    2

    SolCin! the moment e#uations

  • 7/25/2019 Econometrics I 21

    32/67

    Part 21: Generalized Method of Moments1-32/67

    Method of Moments Solution

    creae ; /)=/ ; /.=log"/#calc ; m)=x(r"/)# ; ms=x(r"/.#minimi0e; sar = .1, 12 ; la(els = &,l ; fcn = "l3m)-.

    5 "ms - &si"log"l## 4. 5---------------------------------------------5

    6 7ser 8efined 9&imi0aion 66 8e&enden :aria(le %uncion 66 Num(er of o(ser:aions ) 66 Ieraions com&leed 2 66 Log likelihood funcion 12.

  • 7/25/2019 Econometrics I 21

    33/67

    Part 21: Generalized Method of Moments1-33/67

    9onlinear Instrumental /aria%les

    3here are ; parameters5

    i f)xi5+ H i$3here e.ists a set of ; instrumental Caria%les5 zisuch that

    E0zii *$3he sample counterpart is the moment e)'ation

    )14n+izii )14n+i zi0i= f)xi5+ )14n+imi)+ )+ !$

    3he method of moments estimatoris the solution to themoment e#uation)s+$

    )No" the solution is o%tained is not al"as o%Cious5 andCaries from pro%lem to pro%lem$+

    m

  • 7/25/2019 Econometrics I 21

    34/67

    Part 21: Generalized Method of Moments1-34/67

    3he MLM Solution

    3here are ; e#uations in ; unno"ns in ) + If there is a solution5 there is an e.act solution

    &t the solution5 ) + and 0 ) +>0 ) + *

    Since 0 ) +>0 ) + *5 the solution can %e fou

    m !

    m !* m m

    m m

    nd

    % solCin! the pro!rammin! pro%lem Minimize "rt : 0 ) +>0 ) +

    ,or this pro%lem5

    0 ) +>0 ) + 0)14n+ 0)14n+

    3he solution is defined %

    0 ) +>0 ) + 0)14n+

    m m

    m m

    m m

    'Z Z'

    'Z 0)14n+

    Z'

  • 7/25/2019 Econometrics I 21

    35/67

    Part 21: Generalized Method of Moments1-35/67

    MLM Solution

    [ ] =

    =

    =

    i

    i

    i i

    0)14n+ 0)14n+ 2 )14 n+ 0)14n+

    f) 5 +

    n ; matri. "ith ro" i e#ual to

    ,or the classical linear re!ression model5

    f) 5 + > 5 5 and the ,L8 are =20)1

    x

    G (

    x x + # G #

    'Z Z' G'Z Z'

    4n+) + 0)14n+ >

    ?

    #,# #

    #,# #,"

    -

    + &0

    which has uni"ue solution &( )

  • 7/25/2019 Econometrics I 21

    36/67

    Part 21: Generalized Method of Moments1-36/67

    /ariance of the Method

    of Moments Estimator

    n

    ii 1

    i i

    =1 =1

    3he MLM estimator solCes m) +

    1 1) + ) + so the Cariance is for some

    n 9Generall5 E0 ) + ) +

    3he asmptotic coCariance matri. of the estimator is

    1&s$/ar0 ) + ) + "9

    =

    !

    MOM

    m m

    m m '

    G G') +

    here

    m

    G= '

  • 7/25/2019 Econometrics I 21

    37/67

    Part 21: Generalized Method of Moments1-37/67

    E.ample 1: Gamma Distri%ution

    =

    =

    =

    =

    = +

    =

    =

    2

    n1 P1 i 1 in

    n12 i 1 in

    i i i

    i i i

    1 P9

    i 1 1

    m ) +

    m )lo! )P+ lo! +

    /ar) + 8oC) 5lo! +1 1

    8oC) 5lo! + /ar)lo! +n n

    1

    n >)P+

    G

  • 7/25/2019 Econometrics I 21

    38/67

    Part 21: Generalized Method of Moments1-38/67

    E.ample 2: 9onlinear I/ -east S#uares

    =

    = +

    = =

    =

    =

    =

    i i i

    2i

    i i i

    2

    i

    n2 2 2 2

    i 1

    *

    i i

    f) 5 + 5 the set of ; instrumental Caria%les

    /ar0

    /ar0

    With independent o%serCations5 o%serCations are uncorrelated

    /ar0 ) +)14n + ) 4n + >

    )14n+ >

    i

    i i

    i i

    x z

    m z

    m zz'

    m zz' Z Z

    G zx=

    =

    =

    =

    n *ii 1* ii i

    *

    1 1 * 1 2 2 * 1

    2

    "here is the Cector of >pseudo=re!ressors5>

    f) 5 +$ In the linear model5 this is Vust $

    )14n+ > $

    ) + ) +> 0 )14n+ > 0) 4 n + > 0 )14n+ >

    0

    x

    xx x

    G Z X

    G V G Z X Z Z X Z

    Z * 1 * 1> 0 > 0 > X Z Z X Z

  • 7/25/2019 Econometrics I 21

    39/67

    Part 21: Generalized Method of Moments1-3/67

    /ariance of the Moments

    =

    i

    n

    i i i ii1

    n

    i ii1

    No" to estimate )14n+ /ar0 ) +

    /ar0 ) +)14n+/ar0 ) + )14n+

    Estimate /ar0 ) + "ith Est$/ar0 ) + )14n+ ) + ) +>

    3hen5

    ? ? ?)14 n+ )14 n+ ) + ) +>

    ,or the linear re!ression

    m

    m m

    m m m m

    m m

    =

    = =

    =

    =

    i i i

    n n 2

    i i i i i i ii1 i1

    n=1 2 =1

    MLM i i ii1

    n=1 2 =1

    i i ii1

    model5

    5

    ? )14 n+ )14 n+ e e )14 n+ )14 n+ e

    )14 n+

    Est$/ar0% 0)14 n+ 0)14 n+ )14 n+ e 0)14 n+

    0 0 e 0

    m x

    V x x' x x'

    G X'X

    X'X x x' X'X

    X'X x x' X'X )familiar+

  • 7/25/2019 Econometrics I 21

    40/67

    Part 21: Generalized Method of Moments1-4!/67

    Properties of the MLM Estimator

    8onsistent 3he --9 implies that the moments are consistent estimators of

    their population counterparts )zero+

  • 7/25/2019 Econometrics I 21

    41/67

    Part 21: Generalized Method of Moments1-41/67

    Generalizin! the Method

    of Moments Estimator

    More moments than parameters

    the oCeridentified case E.ample: Instrumental Caria%le

    case5 M ; instruments

  • 7/25/2019 Econometrics I 21

    42/67

    Part 21: Generalized Method of Moments1-42/67

    3"o Sta!e -east S#uares

    No" to use an 6e.cess7 of instrumental Caria%les

    )1+ #is ; Caria%les$ Some )at least one+ of the ;

    Caria%les in #are correlated "ith .$

    )2+ +is M ; Caria%les$ Some of the Caria%les in

    +are also in #5 some are not$ 9one of the

    Caria%les in +are correlated "ith .

    )+ Which ; Caria%les to use to compute +0#and

    +0"

  • 7/25/2019 Econometrics I 21

    43/67

    Part 21: Generalized Method of Moments1-43/67

    8hoosin! the Instruments

    8hoose ; randoml

    8hoose the included s and the remainder randoml

  • 7/25/2019 Econometrics I 21

    44/67

    Part 21: Generalized Method of Moments1-44/67

    2S-S &l!e%ra

    9

    9 9 9( )

    But# & ( ) and ( ) is idemotent.9 9 ( )( ) ( ) so

    9 9( ) & a real 3: estimator y the de4nition.

    9ote# lim( /n) &

    =

    =

    =

    -1

    2SLS

    -1

    Z Z

    Z Z Z

    2SLS

    X Z(Z'Z Z'X

    ! X'X X'y

    Z(Z'Z Z'X I -M X I -MX'X= X' I - M I - M X= X' I - M X

    ! X'X X'y

    X' 0

    -

    9since columns of are linear cominations

    of the columns of # all of which are uncorrelated with

    ( ) + ( )

    2SLS Z Z

    X

    Z

    ! X' I -M X X' I -M y

    = [

  • 7/25/2019 Econometrics I 21

    45/67

    Part 21: Generalized Method of Moments1-45/67

    Method of Moments Estimation

    m 0

    m 0"

    Same Moment ;"uation

    ( )&

    ow# M moment e"uations# < arameters. here is no

    uni"ue solution. here is also no e6act solution to

    ( )&

    We 1et as close as we can.

    =ow to cho

    m 'm 'Z Z' 'ZZ'

    ose the estimator? >east s"uares is an ovious choice.

    Minimi2e wrt $ ( ) ( );.1.# Minimi2e wrt $ *(/n) ( ) +*(/n) ( )+&(/n ) ( ) ( )

  • 7/25/2019 Econometrics I 21

    46/67

    Part 21: Generalized Method of Moments1-46/67

    ,L8 for MLM

    m 'm = G 'm

    'ZZ' = - X Z Z' y - X

    0

    8irst order conditions

    () eneral

    ( ) ( )/ ( ) ( ) & 0

    () he 3nstrumental :ariales 7rolem(/n ) ( ) ( )/ (/n )( ' )* ( )+

    &

    !r#

    X Z Z' y - X 0

    0

    X Z Z' y - X 0

    Z' y - X 0

    ( ' )* ( )+ &

    (< M) (M )( ) &

    ,t the solution# ( ' )* ( )+ &

    But# * ( )+ as it was efore.

  • 7/25/2019 Econometrics I 21

    47/67

    Part 21: Generalized Method of Moments1-47/67

    8omputin! the Estimator

    Pro!rammin! Pro!ram

    9o all purpose solution

    9onlinear optimization pro%lem

    solution Caries from settin! to settin!$

  • 7/25/2019 Econometrics I 21

    48/67

    Part 21: Generalized Method of Moments1-48/67

    &smptotic 8oCariance Matri.

    -1

    m 0

    G 'm 0 G '

    G ' V G V= m-

    eneral @esult for Method of Moments when M

  • 7/25/2019 Econometrics I 21

    49/67

    Part 21: Generalized Method of Moments1-4/67

    More Efficient Estimation

    m 'm

    Mi#im$m %i&(#)* +&

    We have used least s"uares#

    Minimi2e wrt $ ( ) ( )

    to 4nd the estimator of . 3s this the most eAcient

    way to roceed?enerally not$ We consider a more 1eneral aroach

    M%

    im(i,#

    . / = m ' m

    >et e any ositive de4nite matri6$

    9>et & the solution to Minimi2e wrt( ) ( )

    /his is a minimum distance (in the metric of ) estimator.

  • 7/25/2019 Econometrics I 21

    50/67

    Part 21: Generalized Method of Moments1-5!/67

    Minimum Distance Estimation

    M%

    .

    / = m ' m

    m

    >et e any ositive de4nite matri6$

    9>et & the solution to Minimi2e wrt

    ( ) ( )

    where ;* ( )+ & 0 (the usual moment conditions).

    his is a minimum distance (in th

    M%

    M%

    e metric of ) estimator.

    9 is consistent

    9 is asymtotically normally distriuted.

    Same ar1uments as for the MM estimator. ;Aciency of

    the estimator deends on the choice of .

  • 7/25/2019 Econometrics I 21

    51/67

    Part 21: Generalized Method of Moments1-51/67

    MDE Estimation: &pplication

    = + =i iy X X

    !

    i i i

    i

    units# oservations er unit# S country y country# roduces estimators of

    () =ow to comine the estimators?

    We hav

    =

    2

    !

    !0

    """!

    e 'moment' e"uation$ ;

    =ow can 3 comine the estimators of

  • 7/25/2019 Econometrics I 21

    52/67

    Part 21: Generalized Method of Moments1-52/67

    -east S#uares

    =

    =

    =

    = = =

    2 2

    2

    ! !

    ! ! 0 m

    """ """

    ! !

    m m ! ' !

    !

    !m m I I I ! 0

    """!

    !

    i ii

    ii

    i

    ; . ( )&

    o minimi2e ( )' ( ) & ( ) ( )

    ( )' ( )* # #...# + ( ) .

    he solution is ( )= =

    = = 0 = ! ! ii i

    or

  • 7/25/2019 Econometrics I 21

    53/67

    Part 21: Generalized Method of Moments1-53/67

    Generalized -east S#uares

    X X

    X X

    he recedin1 used !>S - simle unwei1hted least s"uares.

    3 0 ... 00 3 ... 0

    ,lso# it uses & . Suose we use wei1hted# >S.... ... ... ...

    0 0 ... 3

    * ( ) + 0 ... 0

    0 * (hen# &

    ( )

    X X

    m m

    XX !

    = XX

    i i i ii&

    i i i ii&

    ) + ... 0

    ... ... ... ...

    0 0 ... * ( ) +

    he 4rst order condition for minimi2in1 ( )' ( ) is

    * ( ) + D( ) & 0

    or * ( ) + D * (

    XX !

    !

    i i ii&

    i ii&

    ) + D

    & & a matri6 wei1hted avera1e.

  • 7/25/2019 Econometrics I 21

    54/67

    Part 21: Generalized Method of Moments1-54/67

    Minimum Distance Estimation

    he minimum distance estimator minimi2es

    ( ) ( )

    he estimator is() Consistent

    () ,symtotically normally distriuted

    (E) =as asymtotic covariance matri6

    9,sy.:ar* + * ( )=M%

    / = m ' m

    G

    ( )+ * ( ) ( )+* ( ) ( )+

    'G G 'VG G 'G

  • 7/25/2019 Econometrics I 21

    55/67

    Part 21: Generalized Method of Moments1-55/67

    Lptimal Wei!htin! Matri.

    is the Wei1htin1 Matri6 of the minimum distance estimator.

    ,re some 's etter than others? (Fes)

    3s there a est choice for ? Fes

    he variance of the MG; is minimi2ed when

    & ,sy.

    3*#*4(5iz*6 m*7,6 ,8 m,m*#& *&im(,4

    m

    V

    -

    -

    :ar* ( )+D

    his de4nes the .

    &

  • 7/25/2019 Econometrics I 21

    56/67

    Part 21: Generalized Method of Moments1-56/67

    GMM Estimation

    =

    =

    = =

    =

    i

    i i

    i i

    GM

    m m 9x 9

    m 2 m 9x 9 m 9x 9

    m 9x 9 '2 m 9x 9

    i i i

    i i i i i

    i i i i i i

    ( )& (y )

    ,sy.:ar* ( )+ estimated with & (y ) (y )

    he MM estimator of then minimi2es

    " (y ) (y ) .

    9;st.,sy.:ar*

    =

    -1

    M

    mG'2 G G=

    ( )+ * + #

  • 7/25/2019 Econometrics I 21

    57/67

    Part 21: Generalized Method of Moments1-57/67

    GMM Estimation

    =

    = =

    =

    =

    i

    i

    m m 9x 9 0

    m 9x 9 '2 m 9x

    i i i

    i i i i i i

    ;6actly identi4ed MM rolems

    When ( ) & (y ) is < e"uations in

    < un5nown arameters (the e6actly identi4ed case)#the wei1htin1 matri6 in

    " (y ) (y

    =

    =

    i

    -1 -1 -1

    9

    m 0

    G'2 G G 2G'

    )

    is irrelevant to the solution# since we can set e6actly( ) so " & 0. ,nd# the asymtotic covariance matri6

    (estimator) is the roduct of E s"uare matrices and ecomes

    * +

  • 7/25/2019 Econometrics I 21

    58/67

    Part 21: Generalized Method of Moments1-58/67

    & Practical Pro%lem

    i i i i i

    i i i i i i

    ,sy.:ar* ( )+ estimated with

    & (y ) (y )

    he MM estimator of then minimi2es

    " (y ) (y ) .

    3n order to comute # you need to 5now

    =

    = =

    =

    i i

    i i

    m

    2 m 9x 9 m 9x 9

    m 9x 9 '2 m 9x 9

    2 # and you are

    tryin1 to estimate . =ow to roceed?

    yically two stes$

    () Hse & Simle least s"uares# to 1et a reliminary

    estimator of . his is consistent# thou1h not eAcient.

    (

    I"

    ) Comute the wei1htin1 matri6# then use MM.

  • 7/25/2019 Econometrics I 21

    59/67

    Part 21: Generalized Method of Moments1-5/67

    Inference

    3estin! hpotheses a%out the parameters:

    Wald test

    & counterpart to the lielihood ratio test

    3estin! the oCeridentifin! restrictions

  • 7/25/2019 Econometrics I 21

    60/67

    Part 21: Generalized Method of Moments1-6!/67

    3estin! Npotheses

    m '2 m

    restricted unrestrict

    () Wald ests in the usual fashion.

    () , counterart to li5elihood ratio tests

    MM criterion is " & ( ) ( ) when restrictions are imosed on

    " increases.

    " " d

    ed chi s"uared*I+(he wei1htin1 matri6 must e the same for oth.)

  • 7/25/2019 Econometrics I 21

    61/67

    Part 21: Generalized Method of Moments1-61/67

    &pplication: Dnamic Panel Data Model

    = + + +xi#t i#t i#t i#t i

    (,rellano/Bond/Bover# I ournal of ;conometrics# JJK)

    y y u

    Gynamic random eLects model for anel data.

    Can't use least s"uares to estimate consistently. Can't use 8>S without

    esti

    + =

    + =

    x

    x

    i# i i#

    i# i i#

    mates of arameters.Many moment conditions$ What is ortho1onal to the eriod disturance?

    ;*( u) + 0 & < ortho1onality conditions#

  • 7/25/2019 Econometrics I 21

    62/67

    Part 21: Generalized Method of Moments1-62/67

    &&lication nnoation

  • 7/25/2019 Econometrics I 21

    63/67

    Part 21: Generalized Method of Moments1-63/67

    &&lication nnoation

  • 7/25/2019 Econometrics I 21

    64/67

    Part 21: Generalized Method of Moments1-64/67

    &pplication: MultiCariate Pro%it Model

    ? J ,

    ? , J ?

    ? - variate =robit Model

    " " 8" E9

    log 8K(, ) ...? 9

    Hequires ? dimensional integration o$ the :oint no

    i i i i i

    it it it it it

    i it it i i i i iL y s t ds ds ds ds ds

    = + = >

    = = x x x x x

    x

    , , ,

    J J J

    rmal densit". Ver" hard

    Nut 78" 6 9 ( ).

    Orthogonalit" ;onditions% 78K" - ( )

    K" - ( )

    K" - ( )

    Moment 7quations% K" - ( )K" - ( )

    it it it

    it it it

    i i i

    i i in

    i i ii

    i i

    n =

    =

    =

    x x

    x x 0

    x x

    x x

    x x

    x

    ? ? ?

    I? equations in P parameters.

    K" - ( )

    i

    i i i

    =

    0

    x

    x x

  • 7/25/2019 Econometrics I 21

    65/67

    Part 21: Generalized Method of Moments1-65/67

    Pooled Pro$it (norin( orrelation

  • 7/25/2019 Econometrics I 21

    66/67

    Part 21: Generalized Method of Moments1-66/67

    andom ffects 9:;1-

  • 7/25/2019 Econometrics I 21

    67/67

    ?nrestricted orrelation Matrix