39
Bank Ownership and Expansion of the Financial System in Thailand Juliano Assunção Sergey Mityakov Robert Townsend Abstract This paper studies bank ownership and different aspects of the financial expansion in Thailand through a dynamic spatial competition model. Data suggest that ownership matters. BAAC, a government development bank in Thailand, expands into poorer and less populated villages and villages more distant from the markets, while commercial banks' entry has positive correlation with wealth/population and proximity to markets. Assuming that commercial banks maximize profits and BAAC maximizes total access to finance, the model not only matches the profile of bank expansion in data, but also explains key characteristics of the interaction between BAAC and commercial banks – the existence of commercial banks affect the behavior of the BAAC, and commercial banks locate in clusters while BAAC doesn’t. (VERY PRELIMINARY – PLEASE DO NOT QUOTE) PUCRio, Department of Economics, [email protected] Clemson University, The John E. Walker Department of Economics, [email protected] University of Chicago and MIT, Department of Economics, [email protected]

Bank!Ownership!and!Expansion!of!the!Financial!Systemin

  • Upload
    others

  • View
    2

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Bank!Ownership!and!Expansion!of!the!Financial!Systemin

 

Bank  Ownership  and  Expansion  of  the  Financial  System  in  

Thailand    

Juliano  Assunção∗  

Sergey  Mityakov⊥  

Robert  Townsend⊕  

 

Abstract  

 

This   paper   studies   bank   ownership   and   different   aspects   of   the   financial  

expansion   in   Thailand   through   a   dynamic   spatial   competition  model.   Data  

suggest  that  ownership  matters.  BAAC,  a  government  development  bank  in  

Thailand,  expands  into  poorer  and  less  populated  villages  and  villages  more  

distant   from   the   markets,   while   commercial   banks'   entry   has   positive  

correlation   with   wealth/population   and   proximity   to   markets.   Assuming  

that  commercial  banks  maximize  profits  and  BAAC  maximizes  total  access  to  

finance,   the  model  not  only  matches   the  profile  of  bank  expansion   in  data,  

but   also   explains   key   characteristics   of   the   interaction   between  BAAC   and  

commercial  banks  –  the  existence  of  commercial  banks  affect  the  behavior  of  

the  BAAC,  and  commercial  banks  locate  in  clusters  while  BAAC  doesn’t.    

 

(VERY  PRELIMINARY  –  PLEASE  DO  NOT  QUOTE)  

                                                                                                               ∗  PUC-­‐Rio,  Department  of  Economics,  [email protected]­‐rio.br ⊥  Clemson  University,  The  John  E.  Walker  Department  of  Economics,  [email protected] ⊕  University  of  Chicago  and  MIT,  Department  of  Economics,  [email protected]

Page 2: Bank!Ownership!and!Expansion!of!the!Financial!Systemin

1   Introduction  

 

Does  bank  ownership  affect  the  expansion  of  the  financial  system?  How  do  private  banks  

change  the  behavior  of  government  development  banks?    

 

Government  ownership  of  banks  is  a  common  phenomenon  around  the  world  -­‐  on  average,  

42   percent   of   the   equity   of   the   10   largest   banks   in   92   countries   was   owned   by   the  

government  in  1995  (La  Porta,  López-­‐de-­‐Silanes  and  Shleifer  2002).  The  literature  on  the  

consequences  of  government  ownership  of  banks  is  still  thin  and  primarily  focused  on  the  

influence   of   political   objectives   in   the   determination   of   lending   policies   of   state-­‐owned  

banks  (Sapienza  2004;  Khwaja  and  Mian  2005;  Dinç  2005;  Micco,  Panizza  and  Yañes  2007).  

 

This   paper   studies   geographical   location   as   another   potential   channel   through   which  

government   ownership   matters.   The   analysis   is   based   on   the   comparison   between   the  

behavior   of   a   government   development   bank   (BAAC)   and   that   of   commercial   banks   in   a  

period  of  financial  expansion  in  Thailand.  We  assume  that  the  locations  of  both  BAAC  and  

commercial   banks   are   endogenous.   We   build   a   structural   model   to   investigate   the  

geographical  expansion  of  BAAC  and  commercial  banks.  The  model  is  a  dynamic  sequential  

entry  game  between  BAAC  and  commercial  bank(s),  where  the  players  compete  by  opening  

offices  in  different  locations.  We  explicitly  incorporate  geography  in  the  model  by  assuming  

that   villages   constitute   a   weighted   graph   with   weights   reflecting   village   characteristics  

influencing  profits  of  the  banks  (e.g.  wealth,  population,  etc).    

 

The  model  interacts  with  the  data  in  different  aspects  of  the  main  assumptions  adopted  and  

the  empirical  implications  derived.    

 

First,  the  model  incorporates  the  evidence  that  BAAC  and  commercial  banks  are  aiming  at  

different   tasks.   Commercial   banks’   access   is   positively   correlated   with   wealth   and  

proximity   to   markets,   evidence   that   lead   us   to   assume   that   commercial   banks   are  

maximizing  profits.  BAAC,  on  the  other  hand,  expands  into  poorer  villages  and  villages  that  

are   more   distant   from   the   markets.   Even   controlling   for   (per   capita)   wealth   and   other  

Page 3: Bank!Ownership!and!Expansion!of!the!Financial!Systemin

village  characteristics,  BAAC  tends  to  expand  into  less  populated  areas.  Thus,  data  make  it  

difficult  to  consider  that  BAAC  is  maximizing  profits  or  targeting  poverty.  We  then  assume  

that   BAAC  maximizes   the   total   access   to   finance   in   the  model   -­‐   an   assumption  which   is  

compatible  with  the  observed  patterns.      

 

Second,  the  model  matches  the  empirical  pattern  of  change  in  the  behavior  of  BAAC  when  

facing   commercial   banks.     Data   show   that   the   presence   of   BAAC   is   positively   correlated  

with  population   in  provinces  with   low  penetration  of  commercial  banks.  This  correlation  

decreases   according   to   the   presence   of   commercial   banks.   The   model   generates   this  

pattern.   In   an   economy   without   commercial   banks,   BAAC   expands   first   into   the   more  

populated  areas.    This  behavior  changes  substantially  when  commercial  banks  are  present.  

If  the  BAAC  puts  enough  weight  on  the  future,  BAAC  has  an  incentive  not  to  serve  the  best  

markets  first,  since  it  anticipates  these  markets  are  quickly  attended  by  commercial  banks.  

The  behavior  of   commercial  banks,  on   the  other  hand,   is   less  affected  by   the  presence  of  

BAAC.   With   or   without   BAAC,   commercial   banks   serve   first   the   best   locations.   BAAC   is  

treated  as  a  competitor  that  prefers  to  attend  the  less  profitable  locations  first  and  do  not  

affect  the  choices  of  commercial  banks  in  the  first  stages  of  interaction.    

 

Third,   the   model   also   explains   the   differences   we   observe   in   the   pattern   of   spatial  

distribution  of  BAAC  and  commercial  banks.  We  approximate  the  reaction  functions  of  the  

model  based  on  a  spatial   regression  approach  suggested  by  Chen  and  Conley  (2001).  We  

show  that  BAAC  expansion  in  each  village  is  correlated  with  changes  in  villages  at  a  wide  

range  of   distances,  while   commercial   banks   changes   are   correlated  only  with   changes   in  

nearby   villages.   Commercial   banks   seem   to   be   clustered   and   BAAC   is   distributed   more  

uniformly   across  Thailand.  This   evidence   is   compatible  with   some  key   examples  derived  

from  the  model,  where  commercial  banks  are  clustered  around  large  markets  while  BAAC  

goes  to  more  isolated  areas.    

 

Our   paper   is   also   related   to   the   literature   of   dynamic   spatial   competition   that   studies  

similar  questions  though  in  different  contexts.    

 

Page 4: Bank!Ownership!and!Expansion!of!the!Financial!Systemin

Chan,  Padmanabhan,  and  Seetharaman  (2006)  use  geographical  location  model  to  estimate  

the  spatial  interaction  between  different  gasoline  stations  in  Singapore.  However,  they  do  

not   model   location   choice   per   se:   gas   stations   locations   are   chosen   by   the   benevolent  

government.   Contrary   to   that   in   our   paper   we   take   demand   for   financial   services   and  

structure  of  bank  competition  as  given  and  try  to  explain  the  location  choice  of  BAAC  and  

commercial  bank  offices.    In  this  sense  this  paper  is  complementary  to  ours.  Aguirregabiria,  

and   Vicentini   (2006)   employs  multiperiod   simultaneous  move   game   to   analyze   location  

choice  of  multistore  firms.    

 

Holmes  (2005)   looks  at   the  expansion  of  Wal-­‐Mart   through   the   lens  of  a  dynamic  model.  

The  author  estimates  economies  of  density  which  he  argues  is  an  important  determinant  of  

Wal-­‐Mart's  choice  of  new  stores  locations.  However,  in  that  paper  there  is  only  one  player  

(Wal-­‐Mart)  and  its  competition  with  other  chains  is  not  modeled  explicitly.  Whereas  in  our  

paper   competition   between   commercial   banks   and   BAAC   is   crucial   in   explaining   BAAC  

behavior.  

 

Paper   by   Schmidt-­‐Dengler   (2006)   is   the   closest   reference   to   our  model.   The   author   also  

studies  a  dynamic  sequential  move  game  to  analyze  new  technology  (MRI)  adoption  by  the  

US   hospitals   taking   hospital   payoff   structure   as   exogenous.   However,   there   is   rather  

important  difference:  players  in  his  paper  have  the  same  objective  (profit  maximization).  In  

our   paper   on   the   contrary   we   are   trying   to   reconcile   observed   differences   in   behavior  

between  BAAC  and  commercial  banks  by  assuming  that  objective  functions  of  these  banks  

are  different.   In  particular,  we  assume  that  commercial  banks  maximize  profits,  and  then  

choose  objective  function  for  BAAC  trying  to  match  observed  empirical  patterns.  

 

2   Background    

 

We  study  the  behavior  of  commercial  banks  and  the  Bank  for  Agriculture  and  Agricultural  

Cooperative  (BAAC).    

 

Page 5: Bank!Ownership!and!Expansion!of!the!Financial!Systemin

BAAC  is  a  State  Enterprise  created  in  1966  to  finance  agricultural  activities  of  farmers  and  

farmer   institutions.   BAAC   is   organized   in   a   three   tier   structure   comprising   head   office,  

branch   offices   and   field   offices.   The   head   office   is   located   in   Bangkok.   Branch   offices   are  

located  in  the  provincial  centers  and  district  towns.  And  field  offices,  which  are  the  bank’s  

front  line  of  contact  with  the  farming  communities,  are  located  in  towns.  

 

The   operation   of   BAAC   is   decentralized.   Each   field   office   has   an   average   of   three   to   five  

officers,  each  one  responding   for  around  700  client   farmers.  Field  officers  are  supervised  

by  a  field  office  chief,  who  reports  directly  to  the  respective  branch  manager.  

 

BAAC  offer  a  package  of  services  that  includes  savings,  credit,  and  payment  and  transfers  of  

funds.   Credit   conditions   and   requirements   are   flexible,   accepting   different   kinds   of  

collateral  -­‐   land  titles,  government  securities,  saving  deposits  or  joint  liability  agreements  

(Tambunlertchai  2004).  

   

We   analyzed   the   period   from   1986   to   1996,   which   was   characterized   by   a   substantial  

financial  expansion.  Figure  1  shows  a  sharp  increase  in  the  number  of  branches,  especially  

for   the  case  of  BAAC.  Number  of  branches   for  BAAC  has   increased   from  79   to  502   in   the  

whole  country,  while  the  ratio  population  per  BAAC  branch  decreased  from  462  thousand  

individuals  per  branch  to  67  thousand   individuals  per  branch.  Commercial  banks  opened  

more  than  2,000  branches  in  the  same  period.    

 

[Figure  1  -­‐  Number  of  branches  and  Outreach]  

 

Data   depicted   in   figure  1   shows   an   increase   in   the  number   of   bank  branches.   There   is   a  

monotonic  increase  in  the  number  of  branches  of  both  BAAC  and  commercial  banks,  which  

have  determined  a   substantial   reduction   in   the  population   to  branch   ratio.  Although   this  

clearly  shows  the  expansion  of   the   financial  system   in  Thailand,   the   information  on  bank  

branches  represents  only  one  dimension  of   the  supply  of   financial  services   in   the  period.  

Branches  are  concentrated  in  some  areas  –  for  example,  27%  of  the  branches  are  located  in  

Page 6: Bank!Ownership!and!Expansion!of!the!Financial!Systemin

5   provinces   (out   of   53   provinces   in   the   country).   Population   in   Thailand   is   relatively  

dispersed   in   approximately   50,000   villages   and,   consequently,   there   is   another   aspect   of  

the  provision  of  financial  services  through  credit  officers.    

 

In  order   to  better  characterize   the  change   in   the  access   to   financial   services,  we  also  use  

data  from  the  Thai  Community  Development  Department  (CDD)  survey  that  is  presented  in  

the  following  section.  This  data  indicate  that  the  fraction  of  villages  without  access  to  BAAC  

decreased  from  20%  in  1986  to  5%  in  1996.  For  the  case  of  commercial  banks,   the  same  

indicator  decreased  from  73%  to  56%.  Both  the  branch  and  credit  use  perspectives  show  a  

significant  financial  expansion  in  the  period  from  1986  to  1996.      

 

The   comparison   between   the   branch   data   and   the   CDD   survey   also   illustrates   how  

important  field  officers  are  for  the  BAAC  operation.  Even  with  a  lower  number  of  branches,  

BAAC  attends  a  substantial  higher  number  of  villages.  

 

3   Data  description  

 

As  mentioned  above,  we  have  data  from  two  main  sources.  The  first  one  is  a  list  of  all  bank  

branches   in   Thailand   with   address,   date   of   entry   and   exits,   obtained   from   the   Bank   of  

Thailand.   This   data   comprise   all   branches   opened   from   1943.   All   branches   were   geo-­‐

located.  

 

[Table  1  -­‐  Descriptive  statistics  -­‐  CDD  data]  

 

The   second   data   source   is   the   Thai   Community  Development  Department   (CDD)   survey,  

conducted  every   two  year   from  1986  to  1996.  Table  1  depicts  summary  statistics   for   the  

1986  and  1996  years.  There  are  binary  variables   indicating  the  use  of   financial  providers  

(BAAC   and   commercial   banks).   We   also   have   information   on   population,   education,   a  

wealth  index  based  on  the  number  motorcycles,  pick-­‐up  trucks  and  flush  toilets  per  1000  

villagers,  and  distance  to  the  market.  

 

Page 7: Bank!Ownership!and!Expansion!of!the!Financial!Systemin

[Figure  2  -­‐  Thai  regions  and  provinces]  

 

[Figure  3  -­‐  CDD  selected  villages  and  amphoe  district  center  locations]  

 

CDD  data  are  organized  in  samples  with  two  levels  of  aggregation  -­‐  amphoe-­‐level  data  and  

village-­‐level   data.   The   amphoe-­‐level   sample   comprises   the  whole   country,  with  704   geo-­‐

located  observations.  The  village-­‐level  data  consists  in  two  samples.  The  full  sample  covers  

villages   from  the  whole  country   for  which  we  have  data   -­‐  more   than  24,000  villages.  But  

this  full  sample  is  not  geo-­‐located.  In  order  to  characterize  the  geographical  patterns  at  the  

village-­‐level,   we   use   a   restricted   sample   of   villages   from   four   chosen   provinces,   out   of  

Thailand's   73   provinces   as   shown   in   figure   2.   The   selected   provinces   are   presented   in  

figure   3   and  were   chosen   to   represent   the   strong   regional   economic   and   environmental  

variation   exhibited   in   Thailand.   We   have   a   total   of   3,391   geo-­‐located   villages   -­‐   1,300  

villages  in  Sisaket,  1,230  in  Buriram,  419  in  Lop  Buri  and  442  in  Chachoengsao.    

 4   The  profile  of  financial  expansion  for  BAAC  and  commercial  banks  

 

This  section  characterizes  the  profile  of  financial  expansion  in  Thailand.  We  start  showing  a  

comparison  between  our  two  main  data  sources.  Table  2  presents  regressions  of  the  credit  

use,  as  measure  in  CDD  data,  on  the  respective  number  of  branches  on  that  category,  at  the  

amphoe-­‐level  of  aggregation,  for  levels  and  changes  of  1986  and  1996.  The  main  conclusion  

of   table   2   is   that   CDD   data   and   branches   are   capturing   different   aspects   of   the   financial  

services   in   Thailand.   The   R-­‐squared   are   negligible   and   the   only   two   out   of   the   six  

coefficients   are   significant,   with   one   of   them   negative.   The   point   estimates   for   the  

regressions  of  the  1996  levels  and  the  change  between  1986  and  1996  are  all  negative.  One  

possible   explanation   for   this   discrepancy   is   that   lending   activities   are   provided   by   other  

distribution  channels,   as   credit  officers   for  example.  We   thus  consider   the  CDD  data  as  a  

better  characterization  of  the  location  of  financial  provision  in  Thailand.        

 

[Table  2  –  Branch  data  vs  CDD  data]  

 

Page 8: Bank!Ownership!and!Expansion!of!the!Financial!Systemin

We  now  show  evidence   that  ownership  may  matter  –   there  are  visible  differences   in   the  

profile  of  the  expansion  of  BAAC  and  commercial  banks.  Table  3  presents  the  profile  of  the  

changes  for  the  period  from  1986  to  1996.  Columns  (1)  to  (4)  are  based  on  the  CDD  sample,  

while  column  (5)  depict  similar  results  for  the  branch  data.    

 

[Table  3  –  Change  in  access  to  financial  providers  from  1986  to  1996]  

 

In   the   following   regressions,   we   interpret   data   as   if   market   conditions   (population,   per  

capita  wealth,  education  and  distance  to  markets)  are  exogenous  to  the  location  of  financial  

providers.  Although  this  assumption  may  be  too  strong  for  the  empirical  analysis,  it  allows  

us   to   characterize   overall   patterns   that   are   incorporated   in   the   model   of   the   following  

section.  We   consider   specification   in  which   the   dependent   variables   are   changes   for   the  

1986-­‐1996  period  and  independent  variables  are  changes  in  the  other  provider  plus  a  set  

of  variables  characterizing   initial  market  conditions.  The  objective   is   to  characterize  how  

market  conditions  are  correlated  with  the  subsequent  expansion  of  BAAC  and  commercial  

banks.  The  analysis  is  focused  on  the  cross-­‐section  variation  of  markets  at  both  village  and  

amphoe   levels   of   aggregation.   This   is   why   we   do   not   consider   fixed-­‐effects   panel   data  

models.      

 

CDD   data   (columns   1   to   4)   suggest   that   commercial   banks   expansion   is   positively  

correlated  to  population  and  wealth,  while  these  correlations  are  negative  for  BAAC,  when  

significant.  For  the  case  of  distance  to  markets,  table  3  shows  that  the  changes  in  access  to  

commercial  banks  are  higher  in  villages  closer  to  markets.  BAAC,  again,  shows  the  opposite,  

going  to  more  distant  villages.  

 

The  comparison  between  columns  (3)  and  (4)  also  reveals  differences  in  the  profile  of  the  

expansion  within  and  across  amphoes.  This  comparison  may  shed  light  on  the  way  BAAC  is  

reaching   the  poor,   less  populated  and  more   isolated  areas.   In  column  (4),   the   increase   in  

BAAC  access  is  higher  for  poor  amphoes  and  not  related  to  population,  or  average  distance  

to  markets.  Thus,   it   seems   that  BAAC  concentrates  on  poorer  amphoes  and,  within   these  

amphoes,  focuses  on  the  less  populated  and  more  isolated  villages.      

Page 9: Bank!Ownership!and!Expansion!of!the!Financial!Systemin

 

Column   (5)   presents   the   same   exercise   considering   the   branch   data.   First,   there   is   a  

difference  with  the  patterns  depicted  by  the  CDD  data,  although  they  are  not  incompatible.  

Second,  we  also  see  a  difference  between  the  profile  of  changes  in  branches  of  commercial  

banks   and  BAAC.   Changes   in  BAAC   are  negatively   related   to   education,  while   changes   in  

commercial  banks  are  positively  related  to  population.  

 

We  thus  end  up  in  a  situation  where  commercial  banks  and  BAAC  seem  to  aim  at  different  

targets.  While  commercial  banks  seem  to  maximize  profits  in  the  choice  of  locations,  data  

make  it  difficult  to  reconcile  BAAC  with  profit  maximization  or  targeting  poverty  –  BAAC  is  

negatively   correlated  with   population   even   conditioning   on  wealth.   Next   section   tries   to  

rationalize   this   difference   assuming   that   commercial   banks  maximizes   profits   and   BAAC  

maximizes  the  population  access  to  financial  services.    

 

5   Model  

 

We   now   present   a   dynamic   spatial   model   that   incorporates   the   patterns   we   observe   in  

data.   In   particular,   we   assume   that   BAAC   behaves   in   order   to   maximize   the   population  

access   to   financial   services   over   time,   while   commercial   banks   maximize   intertemporal  

profits.  

 

5.1  Demand  for  financial  services  

 

Consider   a   finite   number  

M   of  markets   for   financial   services   (villages).   Each   village   can  

have  a  bank  office  open  in  it.  Each  village  

i = 1,...,M  is  characterized  by  the  population  size  

ni   and   potential   profits   per   person  

pi   (denote  

π i = ni pi   the   total   potential   profit   from  

village  

i).  We  consider  

ni  and  

pi  as  exogenous  variables.  

 

Page 10: Bank!Ownership!and!Expansion!of!the!Financial!Systemin

Villages  are  connected  by  roads,  which  are  represented  by  

R = rij( ) ,  where  

rij = 1  if  and  only  

if    villages  

i  and  

j  are  directly  connected  by  a  road  and  

rij = 0  otherwise.  Notice  that  

R  is  a  

symmetric  matrix.  

 

If  there  is  a  bank  branch  in  village  

i  then  people  from  any  directly  connected  village  

k  can  

travel  to  get  access  to  financial  services,  bringing  profits  

pk  to  village  

i .  We  assume  that  

travel   is   costly   and   that   only   a   fraction  

τ −1   (with  

τ > 1)   of   people   from   such   villages   can  

travel.   If   village  

k   has   no   bank   office   but   is   directly   connected   to   several   villages   with  

banks’  offices  then  the  travelers  

τ −1  split  equally  between  those  banks.  

 

5.2  Supply  of  financial  services  

 

For   the   sake   of   simplicity,   we   assume   there   are   two   banks,   which   we   name   BAAC   and  

commercial   bank.   These   banks   compete   by   opening   branches   in   different   villages.   Each  

village  can  have  at  most  one  office  of  any  bank.  Banks  move  sequentially  by  opening  offices  

in  villages  without  banks.    

 

We  summarize  their  interaction  in  the  form  of  the  following  multi-­‐period  game:  

 

1. New  period  begins.  

2. Nature   selects  which   bank  will  move   in   the   current   period.  We   call   this   bank  

“active”  in  this  period.  

3. “Active”  bank  selects  a  village  to  open  a  new  branch  in.  

4. All   banks   receive   profits   from   the   people   who   come   to   their   branches   to   get  

access  to  financial  services,  according  to  the  profit  and  road  structure  

π i{ },R( ) .  5. Current  period  ends.  If  there  are  empty  villages  the  game  continues  from  stage  

1.  If  all  villages  are  occupied  the  game  ends.  

 

We  do  not  allow  banks  to  close  their  branches.  Since  there  are  a  finite  number  of  villages,  it  

is  clear  that  banks  will  stop  opening  branches.  After  that,  all  banks  receive  the  same  profits  

Page 11: Bank!Ownership!and!Expansion!of!the!Financial!Systemin

over   time.  This   focus  on   entries  not   only   simplifies   the   analysis  but   it   is   also   compatible  

with  the  investigated  period  in  Thailand,  as  described  in  section  2.  

 

The  state  of  access   to   finance   in   the  economy  is  characterized  by  the  set  of  existing  bank  

offices  in  each  period  –  access  to  financial  services  is  represented  by  the  vectors  

F1,F2 ∈RM  

(

M   is   the   number   of   villages).   The   state   variable   of   the   economy   is   represented   by  

F = Fk{ }k=1,2 .   Each   vector  

Fk   has  

M   components   which   are   zeros   or   ones,   with   ones  

corresponding   to   the   villages  where   bank  

k   has   offices   and   zeros   for   the   other   villages.  

More  formally,  

Fk = fk1,..., fk

M( ) ,  where  

fki = 1  if  bank  

k  has  an  office  in  village  

i  and  

fki = 0  

otherwise.  

 

We  denote  

G F( ) = F1 + F2   the  vector  of  aggregate  access   to  banking  services.  Also,  

Gi F( )  shows  whether  village  

i  has  access  to  banking  services  at  all  in  state  

F .  

 

Since   people   are   allowed   to   travel   between   adjacent   villages   (i.e.   villages   connected   by  

roads)  to  get  access  to  financial  services,  it  is  useful  for  each  village  

i  to  count  the  number  

of  banking  options  residents  of  that  village  have  if  they  decide  to  travel.  For  each  village  

i ,  

we  denote  

N i F( )   the  number  of  adjacent   locations  with  banks’  offices   in   them  when   the  

state  of   the  economy  is  

F ,   i.e.  

N i F( ) =# j : j = 1,...,M, j ≠ i,Ri, j = 1, f1j + f2

j = 1{ } .  We  assume  that  people  who  travel  from  village  

i  to  get  access  to  finance  in  adjacent  locations  spread  

equally  among  those  

N i F( )  available  options.    

Banks  are  interested  in  how  much  profits  traveling  people  would  bring  to  their  respective  

offices.  For  each  village  

i ,  we  define  

N i k,F( )  as  the  number  of  adjacent  locations  with  bank  

k   offices,   i.e.  

N i F( ) =# j : j = 1,...,M, j ≠ i,Ri, j = 1, fkj = 1{ }.   Evidently,   we   have   that  

N i F( ) = N i 1,F( ) + N i 2,F( ) .   Thus,   from   any   village  

i ,  which   residents   travel   outside   to   get  

access  to  banking  services,  bank’s  

k  payoff  will  be:  

 

Page 12: Bank!Ownership!and!Expansion!of!the!Financial!Systemin

   

1τπ iN i k,F( )N i F( ) .  

 

Thus,  we  can  write  the  profit  of  bank  

k  in  a  given  period  as:  

   

Π k,F( ) = π ii: fk

i F( ) =1∑ +

π iN i k,F( )N i F( )i:Gi F( ) =0

∑ .   (1)  

 

The  first  term  captures  profits  obtained  by  bank  

k  in  villages  it  has  offices  in.  The  second  

term  represents  what  is  obtained  from  the  directly  connected  villages.  

 

We  are  now  ready  to  write  the  objective  functions  of  BAAC  and  commercial  banks.  In  each  

period,   each   bank   can   be   called   to   open   a   new   office.  We   denote   this   bank   as   active   in  

period  

t .   If   bank  

1   is   active,   bank  

0   is   non-­active,   and   vice-­‐versa.   We   assume   a   simple  

Markov  structure  for  transition  of  the  banks  between  active  and  non-­active  states.  Namely,  

we  denote   by  

PA |Ak   and  

PNA |Ak   the   probabilities   that   an   active   bank  

k   in   a   given  period   is  

active  and  non-­‐active,  respectively.  Similarly,  

PA |NAk  and  

PNA |NAk  represent  the  probability  of  a  

non-­‐active  bank  

k  be  active  or  not  in  the  following  period.  

 

We  start  with   the  problem  of   the  active  commercial  bank,   for  which  

k = 1.  Denote  

VA1 F( )  

the  value  function  of  commercial  bank  conditional  on  the  fact  that  it  is  active  in  this  period  

and  the  aggregate  financial  access  state  is  

F .  In  this  case,  commercial  bank  is  called  to  open  

an   office   in   any   of   the   empty   locations.   The   state   of   financial   access,   after   the   choice,  

changes   from  

F = F1,F2( )   to   a   feasible  

F '= ′ F 1,F2( ) ,   where  

F1   and  

′ F 1   differ   only   in   one  

component  

i0  indicating  the  empty  village  in  state  

F  where  commercial  bank  opens  a  new  

office.   We   formally   define   the   set   of   feasible   moves   for   bank  

k   in   state  

F   as  

Ωk F( ) = ′ F k :∀i0 = 1,...,M, ′ F ki0 − Fk

i0 = 1⇒ Gi0 F( ) = 0{ }.   Then,   the   problem   of   an   active  commercial  bank  can  be  written  as:  

 

   

VA1 F( ) = max

′ F 1∈Ω1 F( )Π 1, ′ F ( ) + β VS

1 ′ F 1,F2( )PS |A1

S∈ A ,NA{ }∑

⎣ ⎢

⎦ ⎥ ,   (2)  

Page 13: Bank!Ownership!and!Expansion!of!the!Financial!Systemin

 

where  

β ∈ 0,1[ ]   is   the   discount   factor.   Commercial   bank   choose   where   to   open   the   new  office   maximizing   its   current   period   payoff  

Π 1, ′ F ( )   plus   the   discounted   expected  continuation  value.  

 

Let   us   now   define   the   value   function   of   commercial   bank   when   it   is   non-­‐active.   If  

commercial  bank  is  not  active,  BAAC  choose  the  next  state  

′ F F( ) = F1, ′ F 2 F( )( ),  where  

′ F 2 F( )  

is  the  optimal  choice  of  BAAC  when  facing  state  

F .  The  value  function  for  the  non-­active  

commercial  bank  is  given  by:  

 

   

VNA1 F( ) = Π 1, ′ F F( )( ) + β VS

1 F1, ′ F 2 F( )( )PS |NA1

S∈ A ,NA{ }∑

⎣ ⎢

⎦ ⎥ .   (3)  

 

We   assume   that,   different   from   the  profit-­‐maximizing   commercial   bank,  BAAC   (the   bank  

labeled  as  2)  cares  about  the  total  financial  access  of  people  to  finance  services  no  matter  

which  bank   is  provider.  The  simplest  way  to   incorporate   this  behavior   in   the  model   is   to  

assume  that  BAAC  current  period  payoff  includes  payoffs  from  all  customers  getting  access  

to  finance  in  offices  of  both  BAAC  and  commercial  bank.  In  fact,  if  the  profit  per  person  is  

uniform  across  villages  (which  implies  in  

π i = pni),  maximizing  total  profits  is  equivalent  to  

maximize  the  number  of  customers  with  access  to  financial  services.  Thus,  we  can  write  the  

value  function  for  the  active  BAAC  as:  

 

   

VA2 F( ) = max

′ F 2∈Ω2 F( )Π k, ′ F ( )

k =1,2∑ + β VS

2 F1, ′ F 2( )PS |A2

S∈ A ,NA{ }∑

⎣ ⎢

⎦ ⎥ ,   (4)  

 

where  

′ F = F1, ′ F 2( ) .  We  have  an  analogous  value   function   for   the  non-­active  BAAC,  given  by:  

 

   

VNA2 F( ) = Π k, ′ F F( )( )

k =1,2∑ + β VS

2 ′ F 1 F( ),F2( )PS |NA2

S∈ A ,NA{ }∑

⎣ ⎢

⎦ ⎥ .   (5)  

Page 14: Bank!Ownership!and!Expansion!of!the!Financial!Systemin

 

The  only  difference  between  the  objective   functions  of  commercial  bank  and  BAAC   is   the  

sum  of  payoffs  in  the  first  term  of  (4)  and  (5).  Will  such  change  be  enough  to  induce  BAAC  

to  go  to  worse  and  more  distant  locations  as  observed  in  the  data?    

 

It  appears  that  there  are  two  effects  at  work.  On  the  one  hand,  since  all  locations  occupied  

(and  customers  served)  by  the  commercial  bank  in  the  current  period  are  treated  by  BAAC  

as   its   own   locations   (customers),   BAAC   acts   as   a   big   commercial   bank   owning   all   these  

locations.  In  particular  it  tries  not  to  interfere  with  its  current  market  and  tries  to  open  new  

branches  in  locations  with  are  far  from  both  its  own  and  commercial  bank  existing  offices.  

 

On  the  other  hand,  there  is  also  an  effect  related  to  the  anticipation  of  future  expansion  of  

commercial  banks.  If  BAAC  for  some  reason  knows  that  some  (more  profitable)  market  will  

be  served  by  commercial  banks   in  the  near   future,   it  does  not  open  a  branch  there  today  

going  to  less  lucrative  markets  instead.  If  BAAC  were  to  go  to  the  (next)  best  market  in  the  

current   period   (even   if   this   does   not   interfere   with   the   current   markets   of   commercial  

banks)  then  in  the  future  commercial  banks  would  come  to  this  market  as  well.  Since  each  

additional  $  received  by  commercial  banks  expanding  into  this  market  would  mean  only  a  

fraction  of  a  $  to  BAAC,  this  would  be  not  optimal   from  viewpoint  of  maximization  of  the  

total  access  to  financial  services.  That's  why  this  effect  might  push  BAAC  to  go  into  distant  

and  less  profitable/populated  locations  even  if  there  are  no  commercial  banks  currently  in  

better  locations  in  the  current  period.  

 

5.3  Equilibrium  

 

The  model  described  above  has  an  equilibrium  that  is  almost  always  unique.  This  result  can  

be  demonstrated  by  backward  induction.  Consider  the  last  stage  of  the  game  played  by  the  

banks.  After  this  stage  banks  will  be  receiving  constant  profits  over  time.  Then  the  game  is  

trivial  the  bank  which  is  active  fills  the  last  village,  which  has  no  bank  office  in  it.  

 

Page 15: Bank!Ownership!and!Expansion!of!the!Financial!Systemin

Consider   the  point  before  uncertainty  about  who  moves   last   is   realized  (the  stage  before  

the  last  one).  This  uncertainty  would  give  expected  continuation  payoffs  to  the  bank  that  is  

active  in  the  last  but  one  stage.  Now  each  active  bank  has  two  choices.  Assume  that  profits  

for  each  location  are  such  that  for  any  final  node  of  the  game  resulting  continuation  profits  

are  different.  Consider  the  set  of  parameters  of  the  game  

π i( )  such  that  the  active  bank  is  indifferent  between  locations  in  at  least  one  of  the  final  nodes.  Define  this  set  as  

Q M −1( ) ,  where  

M   is  total  number  of  villages/periods  in  the  game,  and  denote  the  number  of  final  

nodes  of  the  game  

x .  Notice  that  

x < ∞ .    

 

We  can  represent  

Q M −1( )  as  the  union  of  

x  sets  

Qx M −1( ) ,  where  

Qx M −1( )   is  the  set  of  parameters   for  which   the  player  active   in  node  

x   is   indifferent  between   its   two  possible  

actions.  Then,  

Qx  places  only  one  linear  restriction  on  

π i( )  and,  therefore,  it  is  a  set  of  zero  measure.  Consequently,  

Q  is  also  a  set  of  zero  measure.  

 

Consider  now  a  stage  where  the  active  bank  has  

z  possible  options.  Then,  analogously  to  

the  case  above,  we  have  

z  linear  restrictions  on  a  corresponding  set  

Q M − z +1( ) .  This  set  is  defined   by   the   indifference   of   the   active   bank   at   least   between   two   options.   The   set  

Q M − z +1( )  can  also  be  written  as  a  finite  union  of  the  sets  

Q M − z +1( )i, j  defined  by  linear  

restrictions  representing  the   indifference  of   the  active  bank  going  to  village  

i  or  village

j .  

Therefore,  

Q M − z +1( )i, j  has  zero  measure  and,  consequently,  

Q M − z +1( )   is  also  a  set  of  

zero  measure.  

 

Thus,  the  (finite)  union  of  all  sets  

Q M − z +1( )  with  respect  to  all  possible  values  of  

z  is  also  

a   set   of   zero   measure   and   the   game   has   a   unique   equilibrium   for   almost   all   values   of  

parameters  

π i( ) .    

5.4  Discussion  

 

We  make  two  restrictive  assumptions  to  make  our  game  solvable.  

Page 16: Bank!Ownership!and!Expansion!of!the!Financial!Systemin

 

First,  we  assume   that  banks  are   called   to   choose   locations   sequentially   and  exogenously.  

We  do  not  adopt  simultaneous  move  game  to  avoid  multiple  equilibria.  Sequential  game  we  

analyze  here  has  an  advantage  that  equilibrium  is  almost  always  unique,  as  showed  in  the  

previous  sections.    

 

Second,  we  also  assume  that  banks  cannot  open  offices  in  the  same  location.  We  make  this  

assumption  for  computational  reasons,  because  it  reduces  the  number  of  nodes  in  the  game  

tree.  The  (unique)  equilibrium  of  the  game  described  above  can  easily  be  found  using  the  

algorithm  of  backward  induction.  Unfortunately,  computing  this  equilibrium  could  become  

rather  difficult  computation  task  because  of  dimensionality  curse.  For  example,  considering  

a  deterministic  version  of   the  model  where   the  BAAC  and  the  commercial  bank  alternate  

choosing  locations,  the  game  tree  of  a  game  with  

M  villages  has  

M!  possible  paths.      

The  computational  burden  is  of  order  of  

M!.  In  practice,  computing  game  even  for  9  villages  on  a  workstation  could  take  several  minutes.  Computing  it  for  12-­‐15  nodes  requires  several  

days.  That's  why  in  our  analysis  below  we  have  to  severely  restrict  the  number  of  villages  

(6-­‐7  nodes)  to  make  the  problem  solvable  in  reasonable  time.  This  is  also  the  reason  why  

we  do  not  allow  several  banks  to  open  branches  in  the  same  village  because  this  essentially  

means  adding  more  nodes  to  the  game  making  it  intractable  computationally.  

 

6   Some  key  examples  

 

This  section  solves  the  model  for  a  set  of  artificial  examples  in  order  to  shed  some  light  on  

the   differences   in   the   behavior   of   BAAC   and   commercial   banks   and   the   role   of   key  

parameters.  

 

Consider  a  particular  configuration  of  villages  on  a  map.  Namely,  assume  that  there  are  six  

villages  having  coordinates  (0,0),  (1,1),  (2,2),  (3,3),  (3,4),  (4,4).  Further  assume  that  there  is  

a  road  going  through  villages  (1,1),  (2,2),  (3,3),  (3,4),  whereas  villages  (0,0)  and  (4,4)  are  

not  connected  to  any  village  at  all.  Figure  4  depicts  this  economy.  

Page 17: Bank!Ownership!and!Expansion!of!the!Financial!Systemin

 

[Figure  4  –  Example  economy]  

 

We  populate  villages  by  people  with  the  assumption  that  more  central  locations  have  more  

people,   which   is   consistent   with   empirical   patterns   we   observe   in   the   data.   We   also  

consider  

τ = 2,  a  parameter  that  determines  that  50%  of   the  population  of  each  village   is  

able  to  travel.  

 

Finally,   we   consider   a   deterministic   version   of   the  model   where   each   bank   alternate   in  

choosing  locations.  This  allows  us  to  build  more  informative  examples  in  a  computationally  

feasible  number  of  villages.  

 

6.1  Basic  intuition  of  the  model  

 

Let's   see   how   the   BAAC   and   the   commercial   bank   enter   into   this   economy.   We   start  

considering  

β = 1,   a   parameter   representing   that   both   providers   are   forward   looking.  

Figure  5  shows  expansion  paths  of   two  banks:  red  and  green  dots  shows  the   locations  of  

offices  of  commercial  bank  and  BAAC  at  any  period,  respectively.  

 

 [Figure  5  –  Commercial  bank  (red)  x  BAAC  (green)  

β = 1,τ = 2( ) ]    

In   this  case,   commercial  bank   immediately  occupies   the  best   location  (2,2).  BAAC,  on   the  

other  hand,   chooses   first   the   least   profitable   and  more   isolated   village.  BAAC  anticipates  

future  entry  of  commercial  bank  in  location  (3,3)  in  the  next  period,  which  also  serves  part  

of  location  (3,4).  Instead  of  going  to  the  most  populated  available  villages,  it  prefers  to  go  to  

places   that   will   not   be   occupied   by   commercial   bank   in   the   immediate   future   (isolated  

locations  (0,0)  and  (4,4)).  

 

In  

T = 3,  when  BAAC  is  called  to  play,  the  location  that  maximizes  the  current  payoff  is  the  

location   (3,3),   which   brings   $4.15   in   profits   ($2.8   from   its   market   plus   $1.35   from   the  

Page 18: Bank!Ownership!and!Expansion!of!the!Financial!Systemin

travelers  of  location  (3,4)).  On  the  other  hand,  half  of  the  population  of  location  (3,3)  was  

already  with  access   to   financial   services   travelling   to   location   (2,2).  Thus,   the  net  gain  of  

locating   in   (3,3)   is   reduced   from   $4.15   to   $2.65.   Contrary   to   that   expanding   into   either  

location   (0,0)   or   (4,4)   would   bring   $2   every   period   on   top   of   the   expansion   by   the  

commercial   bank   into   location   (3,3).   Thus,  when   discount   rate   is   sufficiently   high,   BAAC  

prefers   going   into   those   isolated   and  poorer   locations   instead.   In   the   last   period,  we   see  

commercial   bank   clustered   in   the  most   populated   villages,   while   BAAC   scattered   in   less  

populated  and  more  isolated  villages.  

 

In   order   to   better   illustrate   this   argument,   figure   6   shows   the   same   exercise   but  

considering  

β = 0 .   Both   providers   now   only   concern   about   the   current   profits   in   each  

period.  

 

[Figure  6  –  Commercial  bank  (red)  x  BAAC  (green)  

β = 0,τ = 2( )]    

Commercial  bank  immediately  occupies  the  best  location  (location  (2,2))  as  in  the  previous  

example.   But   now,   BAAC   opens   a   branch   right   next   to   the   commercial   bank   in   location  

(3,3).  Remember  that  in  figure  5  BAAC  goes  to  (0,0)  because  it  incorporates  the  benefits  of  

waiting  for  commercial  banks  to  attend  the  most  populated  villages.  Here,  this  effect  does  

not  exist  because  BAAC  is  totally  myopic.  The  only  thing  that  matters  is  the  current  payoff.  

Opening   a   branch   in   (3,3)   provides   access   to   its   own   market   (half   of   which   was   not  

attended  by  commercial  bank  in  (2,2))  plus  part  of  the  market  of  village  (3,4).  In  this  case,  

we   see   a   quite   different   pattern   in   the   last   period.   Cross   section   data   from   this   example  

would   show   commercial   bank   and   BAAC   operating   in   villages   with   the   same   average  

population,  with  quite  similar  geographical  distribution.  

 

The  comparison  between  these   two  simple  examples   indicates   that   the  differences   in   the  

behavior   of   the   two   providers   increase   with   the   discount   factor.   In   particular,   for  

β  

sufficiently  high,  we  might  observe  BAAC  going  to  less  populated  and  more  isolated  areas,  

anticipating  the  profit  maximizing  behavior  of  the  commercial  bank.  

Page 19: Bank!Ownership!and!Expansion!of!the!Financial!Systemin

 

These  patterns  help   to   interpret   the  differences   in   the  behavior  of  BAAC  and  commercial  

banks  in  Thailand  from  1986  to  1996,  as  described  in  section  4.  

 

6.2  Interaction  between  BAAC  and  commercial  bank    

 

We   now   study   the   interaction   between   BAAC   and   commercial   bank.   We   would   like   to  

develop   the   intuition   about   how   the   behavior   of   BAAC   changes   in   the   presence   of  

commercial   banks.   Applying   the   argument   developed   above   to   a   higher   level   of  

aggregation,   we  might   expect   BAAC   going   first   to   less   populated   provinces   and   start   to  

operate   alone   in   villages   from   these   provinces.   As   time   goes   by,   with   the   increasing  

penetration  of  financial  services,  commercial  bank  arrives  in  later  periods.  The  question  we  

would  like  to  investigate  is  whether  the  behavior  of  BAAC  changes  or  not  at  this  moment,  

when  commercial  banks  start  to  play  in  these  provinces.  

 

[Figure  7  –  BAAC  (red)  x  BAAC  (green)  

β = 1,τ = 2( ) ]    

Figure   7   shows   a   situation   in  which  BAAC   is   playing   alone   (against   itself),   considering   a  

discount   factor  

β = 1.   In   this   case,   different   from   what   we   see   in   figure   5.   BAAC   starts  

choosing   the  most   populated   and   central   village   (location   (2,2)).   At   this   point,   the   cross  

section   variation   of   access   to   financial   services   depicts   a   positive   correlation   between  

population  and  access  to  BAAC.  Then,  after  assuring  that  the  most  populated  villages  are  at  

least  partially  attended,  BAAC  starts  to  operate  in  isolated  villages.    

 

[Table  4  –  Interaction  between  BAAC  and  commercial  banks]  

 

The   change   in   the   behavior   of   BAAC   when   facing   or   not   commercial   banks   is   also  

compatible  with   our   data,   as   shown   in   table   4.  We   focus   on   the   year   of   1986,  when   the  

penetration  of  commercial  banks  was  particularly  low.  The  main  idea  is  to  test  whether  the  

profile  of  BAAC  changes  with  the  penetration  of  commercial  banks  at  the  province  level.  

Page 20: Bank!Ownership!and!Expansion!of!the!Financial!Systemin

 

For  each  village,  we  compute  the  percentage  of  villages  with  commercial  bank  access  in  the  

respective  province.  This   is  a  way  of   characterizing   the  presence  of   commercial  banks  at  

the  province  level,  when  considering  the  village-­‐level  regressions.  

 

Table  4  show  a  positive  and  statistically  significant  coefficient  for  population.  This  suggests  

that,   in  provinces  where  the  percentage  of  villages  with  commercial  banks  access   is  zero,  

the  presence  of  BAAC  is  positively  correlated  to  the  population.  The  coefficient  related  to  

the   interaction   between   population   and   the   penetration   of   commercial   banks   at   the  

province  level  is  negative  and  statistically  significant.  As  we  move  from  provinces  with  low  

to   those  with  high  penetration  of   commercial  banks,   the  presence  of  BAAC  becomes   less  

related   to   population.   The   coefficients   of   both   level   and   the   interaction   of   the  per   capita  

wealth   are  not   significant,   suggesting   that  BAAC  behavior   is  more   sensible   to  population  

than  to  wealth.  

 

Thus,  not  only  the  profile  of  BAAC  is  different  from  that  of  commercial  bank,  as  showed  in  

section  4,  but  also  it  changes  with  the  presence  of  commercial  banks,  as  suggested  by  the  

model.    

 

7   Simulation  (to  be  done)  

 

This  section  is  devoted  to  simulation  exercises  from  the  model.  The  idea  is  to  draw  random  

samples  of  6  to  8  villages  from  our  data,  compute  the  equilibrium,  and  analyze  the  profile  of  

the  outcomes,  comparing  BAAC  with  the  commercial  bank.  

 

8   Estimation  (to  be  done)  

 

This   section   aims   at   proposing   an   algorithm   to   estimate   the   parameters   of   the   model  

β,τ ,λ( ) ,  based  on  the  approach  of  Bajari,  Benkard  and  Levin  (Econometrica,  2007).    

Page 21: Bank!Ownership!and!Expansion!of!the!Financial!Systemin

9   Empirical   approximations   of   the   reaction   function   of   BAAC   and   commercial  

banks  

 

The   model   outlined   in   section   5   determines   optimal   entry   policies   for   BAAC   and  

commercial  banks.  As  already  mentioned,  solving  the  model  is  a  computationally  too  costly  

for   larger  economies.   In  particular,   it   is  not   feasible  to  solve  the  model   for  the  number  of  

villages  we  observe  in  data,  even  restricting  the  analysis  to  a  specific  province.    

 

This   section   considers   reduced-­‐form   approximations   for   the   optimal   policy   functions   of  

BAAC   and   commercial   banks,   based   on   a   semi-­‐parametric   spatial   econometric   approach  

developed   by   Chen   and   Conley   (2001).   This   approach   allows   us   to   identify   geographical  

patterns  of  the  observed  expansion  of  BAAC  and  commercial  banks.    

 

The  optimal  policy  function  of  bank  

k  can  be  represented  by:  

 

    Fk* t +1( ) = σ k ,t Fk

* t( ),F−k* t( ) R, π i( ),β,τ( ) ,   (6)  

 

where  

F * 0( ) = 0 .  We  take  a  quasi-­‐linear  and  stationary  approximation  for  the  village-­‐level  

version  of  equation  (6),  considering  the  following  specification:  

 

    fki t +1( ) = wk

i, j fkj t( )

j=1

M

∑ + w−ki, j f−k

j t( )j=1

M

∑ + γ R, π i( ),β,τ( ) + εki t +1( ) ,   (7)  

 

where   εki t +1( )  is  an  approximation  error  and   γ ⋅( )  is  any  (potentially  non-­‐linear)  function.  

Notice  that  the  weights  wkj  and  w−k

j  do  not  vary  according  to  the  time  periods.  Stationarity  

is  a  strong  assumption  in  our  case,  since  we  focus  exclusively  on  entries  in  a  finite  number  

of  markets.  However,   this  assumption   is   less   strong   for  our  data,  where  we  are  quite   far  

apart   from   the   end   of   the   game.   Still,   this   exercise   shows   us   another   dimensions   of   the  

differences  we  observe  in  the  behavior  of  BAAC  and  commercial  banks.    

 

Page 22: Bank!Ownership!and!Expansion!of!the!Financial!Systemin

Considering  equation  (7)  for  the  periods   t +1  and  T +1,  where  T > t ,  we  can  write:  

 

    Δfki T( ) = wk

i, jΔfkj T( )

j=1

M

∑ + w−ki, jΔf−k

j T( )j=1

M

∑ + Δεki T( ) ,   (8)  

 

where   we   also   take   that   Δfki T +1( ) ≈ Δfk

i T( ) ,   since   they   differ   only   in   one   component.  Actually,  the  error  from  this  approximation  tends  to  be  smaller  the  greater  is  the  difference  

between   t  and   T .  Equation  (8)  represents  how  the  changes  in  the  presence  of  bank   k  in  

village   i  is  related  to  the  changes  in  other  villages  and  to  the  changes  in  the  presence  of  the  

other  provider.  

 

In  principle,  we  would  like  to  estimate  (8)  and  compare  the  estimated  weights  of  BAAC  and  

commercial   bank.   Unfortunately,   coefficients   of   the   regression   related   to   (8)   cannot   be  

identified   –   there   are   more   coefficients   than   observations.   Thus,   we   need   to   impose  

restrictions  on   the  weights   wki, j   and   w−k

i, j .  We   thus   take   the  approach  of  Chen  and  Conley  

(2001)   and   consider   weights   structured   according   to   a   spatial   distance   matrix,   with  

wki, j = gk Di, j( )  and  w−k

i, j = g−k Di, j( ) ,  where  Di, j  is  the  distance  between  villages   i  and   j .  

 Error!   Reference   source   not   found.Error!   Reference   source   not   found.Error!  

Reference  source  not  found.Error!  Reference  source  not  found.  

Therefore,  we  consider  two  different  specifications  to  for  equation  (8),  for  the  period  from  

1986   to   1996.   In   order   to   focus   on   the   variation   observed   in   the   period,   we   focus   on  

specifications  related  to  the  changes  in  access  as  following:  

 

    Δfki T( ) = gk Di, j( )Δfkj T( )

j=1

M

∑ +αΔf−kj T( ) + Δεk

i T( ) ,   (9)  

and  

    Δfki T( ) = αΔfkj T( ) + g−k Di, j( )Δf−kj T( )

j=1

M

∑ + Δεki T( ) .   (10)  

 

Page 23: Bank!Ownership!and!Expansion!of!the!Financial!Systemin

The  models   are   straightforward   adaptations   of   the  methodology   presented   in   Chen   and  

Conley  (2001).  Notice  that  the  only  difference  between  (9)  and  (10)  is  related  to  the  term  

in  the  summation.  In  the  first  case,  the  function   gk D( )  represents  how  changes  in  access  at  village  

i  are  correlated  with  changes  at  village  

j  for  a  given  provider   k  (geographical  own-­‐

effects).   In   the   second   case,   the   function   g−k D( )   shows   how   changes   in   one   provider   at  village  

i   are   related   to   changes   of   the   other   provider   at   village  

j   (geographical   cross-­‐

effects).  The  differences  in  the  behavior  of  BAAC  and  commercial  banks  can  be  captured  by  

the  comparison  of  functions   gk  and   g−k .    

 

Equations   (9)   and   (10)   are   estimated   for   two   samples   for   which   we   have   geo-­‐located  

observations.  We   consider   the   amphoe-­‐level   data   for   the  whole   country   and   the   village-­‐

level  data  for  the  four  selected  provinces.  Results  are  depicted  in  figures  8  and  9.    

 

In   all   cases,   we   get   flat  

g   functions   for   BAAC   and   decaying  

g   functions   for   commercial  

banks.  The  presence  of  BAAC  in  a  given  village  (amphoe)  is  connected  with  the  presence  of  

BAAC  at  a  wider  range  of  distances.  The  presence  of  commercial  banks   in  a  given  village  

(amphoe),   on   the   other   hand,   is   related   only   with   the   presence   of   commercial   banks   in  

nearby  villages  (amphoes).  

 

In  the  case  of  the  specification  with  geographical  cross-­‐effects,  we  study  how  the  changes  in  

BAAC   for  village  

i   are  affected  by   the  changes   in  commercial  bank   in  village  

j   and  vice-­‐

versa.  We   focus   now   on   the   geographical   interaction   across   different   banks   rather   than  

self-­‐related  geographical  interaction  of  each  bank.  

 

At  the  amphoe-­‐level,  each  provider  interacts  with  the  other  in  the  same  way  it  interacts  to  

itself.   BAAC   presents   a   flat   pattern  with   respect   to   commercial   banks,  while   commercial  

banks  depict  decaying  patterns  with  respect  to  BAAC.  For  the  village-­‐level  data,  the  results  

change   for   the   case   of   commercial   banks,   as   shown   in   figure   9(ii).   Commercial   bank  

expansion   in  a  given  village   is  not  affected  by   the  changes  of  BAAC   in  other  villages.  The  

pattern  for  the  BAAC  is  similar  to  that  obtained  with  the  amphoe-­‐level  data.  The  changes  in  

Page 24: Bank!Ownership!and!Expansion!of!the!Financial!Systemin

the  access  to  BAAC  in  a  given  village  are  affected  by  the  change  in  the  access  to  commercial  

banks  in  villages  located  at  a  wide  range  of  distances.  

 

 

The   comparison   between   these   two   patterns   reveals   interesting   features   about   the  

underlying  decision-­‐making  choices  compatible  with  the  data.  The  profitability  of  a  given  

village   in   this  environment,  conditional  on   the   location  of   the  nearby  competitors   (which  

completely  define  the  market  size),   is  independent  from  what  happen  in  the  other  (farer)  

villages.   In  this  example  we  are  closer  to  a  decaying   g   function.  For  the  flat  pattern  case,  

we  can  take  a  country-­‐level  mandate  as  an  example.  If  BAAC  (as  we  assume  in  the  model)  

maximizes   the   total   financial  access  of   the  population,   the  access  of  a  given  village  might  

become  related  to  the  access  in  all  other  villages  in  the  country.  

 

10  Conclusion  

 

This  paper  studies  location  as  another  channel  through  which  bank  ownership  may  matter.  

We  build  a  spatial  dynamic  entry  model  where  the  government  development  bank  (BAAC)  

maximizes  total  access  to  finance  irrespective  of  which  bank  provides  it,  while  the  private  

bank   (commercial   bank)  maximizes  profits.   If   the  discount   factor   is   high   enough,  we   see  

BAAC  anticipating  that  commercial  bank  has  incentive  to  attend  the  most  profitable  villages  

first  and  go  to  more  isolated  and  less  populated  ones,  as  suggested  by  our  data.  In  addition,  

we  show  that  the  behavior  of  BAAC  changes  whether  playing  against  commercial  banks  or  

not   and   that   the   geographical   location   patterns   of   BAAC   and   commercial   banks   are  

different.  

 

References  

 

Ackerberg,  Daniel  A.  and  Gautam  Gowrisankaran  (2006)  “Quantifying  equilibrium  network  

externalities  in  the  ACH  banking  industry”.  NBER  Working  Papers  12488.      

 

Page 25: Bank!Ownership!and!Expansion!of!the!Financial!Systemin

Chan,  Tat  Y.;  V.  Paddy  Padmanabhan  and  P.  B.  Seethu  Seetharaman  (2007)  “An  Econometric  

Model  of  Location  and  Pricing  in  the  Gasoline  Market”.  Journal  of  Marketing  Research,  

44(November):  622-­‐635.      

 

Chen,  X.  and  T.  G.  Conley  (2001)  “A  new  semiparametric  spatial  model  for  panel  time  

series”.  Journal  of  Econometrics,  105:  59-­‐83.      

 

Holmes,  Thomas  J.  (2008)  “The  Diffusion  of  Wal-­‐Mart  and  Economies  of  Density”.  NBER  

Working  Paper  13783.      

 

Micco,  Alejandro;  Ugo  Panizza  and  Monica  Yañez  (2007)  “Bank  ownership  and  

performance.  Does  politics  matter?”.  Journal  of  Banking  &  Finance,  31(1):  219-­‐241.      

 

Panle,  Jia  (2008)  “What  Happens  When  Wal-­‐Mart  Comes  to  Town:  An  Empirical  Analysis  of  

the  Discount  Retailing  Industry”.  Econometrica,  forthcoming.      

 

Sapienza,  Paola  (2004)  “The  effects  of  government  ownership  on  bank  lending”.  Journal  of  

Financial  Economics,  72(2):  357-­‐384.      

 

Schmidt-­‐Dengler,  Philipp  (2006)  “The  Timing  of  New  Technology  Adoption:  The  Case  of  

MRI”.  Mimeo.    

 

 

Page 26: Bank!Ownership!and!Expansion!of!the!Financial!Systemin

Figure  1  -­‐  Number  of  branches  and  Outreach  

 

 

   

198619881990199219941996

!"#$%&'(")*++,*++, ,"--./0(&%1*&)23,"--./0(&%1*&)23 4"'&%

!"#$%&'(")*/&)05.3

!"#$%&'(")6*/&)05.3 */&)05.3

!"#$%&'(")6*/&)05.3 */&)05.3

!"#$%&#'!% '( &"%#!!) *#")) %*#"!) *#'"'!&#)!'#%'% )% &%&#)&$ *#(&( *'#)'& %#+!*!&#&"(#++* *%& %''#('" %#%"' *$#%+$ %#!(*!%#&"&#$(& %*+ *$&#$(! %#$(+ *%#$!$ %#)++!%#(%)#*%$ !!% ((#*)* !#+!& *+#)$! !#!""!!#'"+#!"! $+% "'#%$% !#!(+ (#($( !#)(%

!

"#!

$!!

%#!

&!!

"'(& "'(( "''! "'') "''% "''&

!

")#*!!!

)#!*!!!

$+#*!!!

#!!*!!!!""#

,-./0123-45.67587149: 87149:6;

!

"!!!

)!!!

$!!!

%!!!

"'(& "'(( "''! "'') "''% "''&

,-./0123-45.67587149: 87149:6;

198619881990199219941996

!"#$%&'(")*++,*++, ,"--./0(&%1*&)23,"--./0(&%1*&)23 4"'&%

!"#$%&'(")*/&)05.3

!"#$%&'(")6*/&)05.3 */&)05.3

!"#$%&'(")6*/&)05.3 */&)05.3

!"#$%&#'!% '( &"%#!!) *#")) %*#"!) *#'"'!&#)!'#%'% )% &%&#)&$ *#(&( *'#)'& %#+!*!&#&"(#++* *%& %''#('" %#%"' *$#%+$ %#!(*!%#&"&#$(& %*+ *$&#$(! %#$(+ *%#$!$ %#)++!%#(%)#*%$ !!% ((#*)* !#+!& *+#)$! !#!""!!#'"+#!"! $+% "'#%$% !#!(+ (#($( !#)(%

!

"#!

$!!

%#!

&!!

"'(& "'(( "''! "'') "''% "''&

*+,-./01+23,45365/278 65/27849

!

"!!!

)!!!

$!!!

%!!!

"'(& "'(( "''! "'') "''% "''&

!

:;#!!

"#;!!!

));#!!

$!;!!!!"##$%&'()*+(,-.

*+,-./01+23,45365/278 65/27849

Page 27: Bank!Ownership!and!Expansion!of!the!Financial!Systemin

 

Figure  2  -­‐  Thai  regions  and  provinces  

 

   

Page 28: Bank!Ownership!and!Expansion!of!the!Financial!Systemin

Figure  3  -­‐  CDD  selected  villages  and  amphoe  district  centers  

 

 

Page 29: Bank!Ownership!and!Expansion!of!the!Financial!Systemin

 

Figure  4  –  Example  economy  

 

   

Page 30: Bank!Ownership!and!Expansion!of!the!Financial!Systemin

Figure  5  –  Commercial  bank  (red)  x  BAAC  (green)  

β = 1,τ = 2( )    

   

Page 31: Bank!Ownership!and!Expansion!of!the!Financial!Systemin

Figure  6  –  Commercial  bank  (red)  x  BAAC  (green)  

β = 0,τ = 2( )    

   

Page 32: Bank!Ownership!and!Expansion!of!the!Financial!Systemin

Figure  7  –  BAAC  (red)  x  BAAC  (green)  

β = 1,τ = 2( )    

   

   

Page 33: Bank!Ownership!and!Expansion!of!the!Financial!Systemin

 Figure  8  –  Geographical  pattern  of  policy  functions  (amphoe-­‐level  data)  

 

(i)  specification  with  geographical  own-­‐effects  

 (ii)  specification  with  geographical  cross-­‐effects  

 

Page 34: Bank!Ownership!and!Expansion!of!the!Financial!Systemin

 Figure  9  –  Geographical  pattern  of  policy  functions  (village-­‐level  data)  

 

(i)  specification  with  geographical  own-­‐effects  

 

Page 35: Bank!Ownership!and!Expansion!of!the!Financial!Systemin

 (ii)  specification  with  geographical  cross-­‐effects  

 

 

 

Page 36: Bank!Ownership!and!Expansion!of!the!Financial!Systemin

Table  1  -­‐  Descriptive  statistics  -­‐  CDD  Data  

 

obs.   mean   std.  dev.   min   max  

1986  

access  to  BAAC   44513   0.8   0.4   0   1  

access  to  commercial  banks   44175   0.3   0.4   0   1  

population  (thousands)   44652   603.7   420.2   11   9864  

per  capita  wealth   29235   0.5   0.3   0.0   8.9  

%  population  with  advanced  

secondary  school  35918   0.0   0.3   0.0   52.0  

time  to  the  market  (in  

minutes)  42762   35.0   30.6   1   998  

1996  

access  to  BAAC   44536   0.9   0.2   0   1  

access  to  commercial  banks   44280   0.4   0.5   0   1  

population  (thousands)   44649   606.2   421.1   3   9788  

per  capita  wealth   43515   1.2   0.5   0.0   23.2  

%  population  with  advanced  

secondary  school  44281   0.0   0.0   0.0   4.9  

 

Page 37: Bank!Ownership!and!Expansion!of!the!Financial!Systemin

Table  2  –  Branch  data  vs  CDD  Data  

 

CDD  Data 1996 1986 Change  (1986-­‐1996) BAAC Commercial  

Bank BAAC Commercial  

Bank BAAC Commercial  

Bank (1) (2) (3) (4) (5) Branch  data -­‐0.003 -­‐0.001 0.061*** 0.002 -­‐0.005 -­‐0.004** (0.530) (0.600) (3.250) (0.920) (0.540) (2.360) Constant 0.944*** 0.467*** 0.804*** 0.285*** 0.135*** 0.183*** (169.650) (45.200) (89.730) (30.140) (16.150) (19.830) observations 705 705 705 705 705 705 R-­‐squared 0.00 0.00 0.01 0.00 0.00 0.00

 

Page 38: Bank!Ownership!and!Expansion!of!the!Financial!Systemin

Table  3  –  Change  in  access  to  financial  providers  from  1986  to  1996  

Panel  (i):  Change  in  Access  to  BAAC  1986-­‐1996  

Village  Level   Amphoe  Level  

Branch  Data  (Amphoes)  

 

(1)   (2)   (3)   (4)   (5)  -­‐0.867***   -­‐0.871***   -­‐0.915***   -­‐0.666***   -­‐0.058  access  in  1986  (0.006)   (0.006)   (0.004)   (0.019)   (0.790)  

  0.052***   0.047***   0.073***   -­‐0.012  change  in  access  to  commercial  banks  in  1986-­‐96     (0.003)   (0.003)   (0.020)   (1.560)  

  0.056***   0.049***   0.028   0.001  access  to  commercial  banks  in  1986     (0.004)   (0.004)   (0.021)   (0.120)  

-­‐0.005*   -­‐0.011***   -­‐0.007***   0.001   -­‐0.030  ln(population  in  1986)   (0.003)   (0.003)   (0.002)   (0.006)   (0.830)  -­‐0.038***   -­‐0.053***   -­‐0.011**   -­‐0.084***   -­‐0.091  per  capita  wealth  in  1986   (0.005)   (0.005)   (0.005)   (0.022)   (0.670)  -­‐0.003   -­‐0.002   -­‐0.001   0.005   -­‐0.254***  %  population  with  advanced  

secondary  school  in  1986   (0.003)   (0.003)   (0.004)   (0.049)   (5.050)  0.003*   0.005**   0.006***   0.008   -­‐0.102  ln(distance  (minutes)  to  the  

market  in  1986)   (0.002)   (0.002)   (0.002)   (0.011)   (1.400)  

amphoe-­‐level  fixed  effects   no   no   yes      observations   24498   24202   24202   704   704  R-­‐squared   0.70   0.70   0.75   0.68   0.01     Panel  (ii):  Change  in  Access  to  Commercial  Banks  1986-­‐1996     (1)   (2)   (3)   (4)   (5)  

-­‐0.823***   -­‐0.832***   -­‐0.924***   -­‐0.485***   -­‐1.159***  access  in  1986   (0.007)   (0.007)   (0.007)   (0.035)   (2.580)     0.288***   0.266***   0.253***   -­‐0.18  change  in  access  to  BAAC  in  1986-­‐

96     (0.013)   (0.015)   (0.070)   (1.640)     0.294***   0.274***   0.220***   0.730***  access  to  BAAC  in  1986     (0.014)   (0.017)   (0.058)   (6.320)  0.099***   0.099***   0.095***   0.018*   0.336**  ln(population  in  1986)   (0.005)   (0.005)   (0.006)   (0.010)   (2.370)  0.214***   0.228***   0.120***   0.194***   0.05  per  capita  wealth  in  1986   (0.010)   (0.010)   (0.012)   (0.040)   (0.110)  -­‐0.011***   -­‐0.010***   -­‐0.01   -­‐0.072   0.232  %  population  with  advanced  

secondary  school  in  1986   (0.001)   (0.001)   (0.009)   (0.091)   (0.740)  -­‐0.029***   -­‐0.030***   -­‐0.031***   -­‐0.049**   0.491  ln(distance  (minutes)  to  the  

market  in  1986)   (0.004)   (0.004)   (0.004)   (0.020)   (1.240)  

amphoe-­‐level  fixed  effects   no   no   yes      observations   24221   24202   24202   704   704  R-­‐squared   0.39   0.40   0.49   0.25   0.58  

Page 39: Bank!Ownership!and!Expansion!of!the!Financial!Systemin

Table  4  –  Interaction  between  commercial  banks  and  BAAC  in  1986  

Access  to  BAAC  in  

1986

0.083*** ln(population  in  1986) (0.008)

0.017 per  capita  wealth  in  1986 (0.016)

-­‐0.096*** ln(population  in  1986)  *  %  villages  with  access  to  commercial  banks  in  the  province (0.025)

-­‐0.074 per  capita  wealth  in  1986  *  %  villages  with  access  to  commercial  banks  in  the  province (0.048)

0.123*** access  to  commercial  banks  in  1986 (0.005)

Observations 28970

R-­‐squared 0.09