Potential  bias

The  dataset  used  for  the  estimation  contains  all  available  bank  connections  of  firms  that  satisfy  the   requirements.  If  firms  have  more  than  one  bank  connection,  then  they  appear  more  than  once  in  the   dataset.  It  is  interesting  to  examine  if  it  has  an  effect  on  the  firms’  switching  costs.  A  simple  variable   that  contains  the  number  of  bank  connections  each  firm  has  in  the  dataset  is  therefore  included  in   the  estimation.  

The  last  variable  that  is  included  in  the  dataset  is  the  year  of  establishment.  It  is  not  a  continuous   variable  and  there  are  too  many  years  to  include  each  year  as  a  dummy.  Therefore  the  seven  groups   used  in  Figure  5  are  used.  The  groups  are  included  in  the  estimation  as  six  dummy  variables,  with  the   group  of  firms  established  before  1950  as  the  default  group.  Each  dummy  thus  represents  the  

switching  costs  of  the  firms  that  were  established  in  the  time  period,  compared  to  firms  that  were   established  before  1950.  

year  a  firm  were  established,  but  rather  on  other  characteristics  of  the  firms.  In  the  following  size  of   firms  will  be  examined,  as  it  should  be  highly  correlated  with  the  other  characteristics.  The  size  of   the  firm  is  thus  used  a  proxy  of  the  firms’  demand.  The  relevancy  of  the  examination  depends  on  how   well  the  size  of  the  firm  explain  the  demand  of  the  firms.  

The  supply  of  credit  is  based  on  the  firms’  ability  to  fulfill  their  credit  commitments.  This  is  mostly   depending  on  future  factors  and  therefore  hard  to  measure.  Usually  a  subjective  assessment  by  each   bank  decides  the  terms  of  the  contract.  It  is  not  possible  to  observe  the  individual  contracts,  so  a   formal  examination  of  this  is  not  possible.  The  subjective  assessment  is  however  partly  based  on   financial  figures,  so  it  is  likely  to  be  correlated  to  the  observed  figures  in  the  estimation.  

If  the  newly  established  firms  have  different  preferences  than  the  existing  firms,  observed  

differences  in  the  market  shares  can  be  caused  by  consumer  heterogeneity  and  not  the  lock-­‐in  effect.  

The  preferences  of  consumers  are  likely  to  depend  on  the  size  of  the  consumer,  it  is  therefore   important  that  the  two  distributions  of  markets  shares  that  are  compared  are  based  on  a  

homogenous  distribution  of  firms.  The  first  examination  looks  at  the  distribution  of  both  the  median   balance  and  median  capital  of  each  group  depending  on  year  the  firms  were  established.  The  median   is  presented  instead  of  the  average,  because  the  switching  cost  estimates  are  based  on  the  absolute   number  of  firms  that  each  bank  serve  in  each  period.  Thus  each  firm  has  identical  weights,  

independent  of  their  characteristics,  when  estimating  the  switching  costs⁴.  The  amount  of  firms  in   each  category  is  also  added  to  the  graph,  to  illustrate  the  weight  each  group  has  in  the  estimations.  

The  financial  figures  approximate  the  size  of  the  firm.  If  the  sizes  of  firms  are  independent  of  the  year   the  firms  were  established,  the  median  financial  figures  would  be  the  same  for  each  group  of  firms.  

Below  are  the  two  aforementioned  graphs.    

4  The  averages  are  similar  but  influed  by  the  large  firms,  most  notably  in  the  groups  consisting  of  the  oldest  firms.  

  Figure  6:  Median  capital  of  firms  distributed  according  to  year  of  establishment.  

  Figure  7:  Median  balance  of  firms  distributed  according  to  year  of  establishment.  

It  is  clear  that  firms  are  not  homogenously  distributed  according  to  size  over  the  year  they  were   established.  Firms  established  a  longer  time  ago  are  generally  larger  than  newly  established  firms,   which  can  be  seen  from  the  larger  median  capital  and  median  balance.  The  median  capital  of  firms,   illustrated  by  Figure  6,  is  largest  for  the  two  groups  containing  the  oldest  firms.  It  is  constant  for  the   group  of  firms  established  in  1960-­‐1989,  but  considerably  smaller  for  the  group  of  firms  established   in  2011  and  after.  The  median  balance  of  firms,  illustrated  by  Figure  7,  is  continuously  decreasing   from  the  group  of  firms  established  the  longest  time  ago,  to  the  group  of  firms  established  in  2011   and  after.  These  graphs  are  not  surprising,  as  firms  are  likely  to  either  grow  or  go  out  of  business.  

This  can  unfortunately  potentially  bias  the  observations,  if  the  sizes  of  the  firms  are  also  correlated   with  the  choice  of  bank.  As  mentioned  before  firms  are  not  weighted  according  to  their  size  in  the   switching  cost  estimations,  but  all  count  equally  towards  the  market  shares.  In  Figure  7  and  Figure  8   the  number  of  firms  in  each  group  are  also  graphed.  The  number  of  firms  displays  how  much  each   group  contribute  to  the  estimations.  The  graph  of  the  number  of  firms  is  in  line  with  the  tendency  for  

firms  to  either  grow  or  go  out  of  business.  While  the  medians  indicate  that  the  estimation  is  

potentially  biased,  the  number  of  firms  in  each  group  reduces  this  effect.  The  groups  with  the  largest   medians  that  potentially  bias  the  estimation  are  very  small  compared  to  the  groups  with  medians   most  similar  to  those  of  the  new  firms.  The  groups  most  similar  to  the  newest  firms  are  the  largest   groups  and  because  firms  are  not  weighted,  this  suggest  that  while  the  estimation  does  have  

potential  to  be  biased,  the  issue  is  not  as  severe  as  the  medians  suggest  without  taking  the  size  of  the   groups  into  account.  

One  way  to  reduce  the  potential  bias  that  arise  because  of  firm  heterogeneity,  could  be  to  use   another  group  of  firms  that  consists  of  firms  that  has  been  established  after  1990,  or  some  other   year,  and  compare  that  group  to  the  group  of  new  firms.  The  same  method  for  estimating  the  

consumer  switching  costs  could  then  be  applied  to  the  two  distributions  of  market  shares.  As  the  two   groups  of  firms  are  more  similar  than  the  total  population  of  firms,  it  would  reduce  the  potential  bias.  

The  group  is  not  more  correct  than  the  original  but  will  contain  firms  that  are  more  similar  to  the   newly  established  firms  and  still  have  a  significant  number  of  observations.  The  disadvantage  of  this   approach  is  that  many  firms  that  could  potentially  have  large  switching  costs,  due  to  a  longer  lasting   relationship  with  a  bank,  will  be  excluded.  The  alternative  group  definition  yields  very  similar   results,  but  with  lower  statistical  power  –  as  would  be  expected.  The  results  are  not  reported   independently.  

It  is  evident  that  the  size  of  the  firms  are  correlated  with  the  year  the  firms  were  established.  It  will   bias  the  estimations  if  the  sizes  of  the  firms  have  an  affect  on  the  firms’  choice  of  bank.  If  all  banks   serve  consumers  of  the  same  size,  the  market  share  differences  will  not  be  biased  as  a  consequence   of  size  differces  of  firms.  To  examine  if  the  size  of  the  firms  are  correlated  with  the  choice  of  bank,  the   median  capital  and  median  balance  are  graphed  for  each  bank.  The  interesting  aspect  is  the  

difference  of  the  medians  across  banks.  If  some  banks  banks  serve  customers  with  low  medians,  ie   small  firms,  the  market  share  of  new  firms  will  be  to  high  compared  to  the  situation  with  

homogenous  firms,  and  likewise  will  banks  that  serve  firms  with  high  medians  have  a  high  market   share  among  the  existing  firms.  This  means  that  the  switching  cost  estimates  of  banks  that  serve   small  firms  will  be  biased  such  that  their  switching  cost  estimates  are  too  low,  and  similarly  will  the   switching  cost  estimations  of  banks  that  serve  large  firms  will  be  too  high.  Below  are  the  median  

capital  and  median  balance  of  firms  that  are  customers  at  the  ten  largest  banks  graphed  on  separate   graphs.  The  ten  largest  banks  accounts  for  the  majority  of  the  market,  so  for  a  better  overview  the   smallest  banks  are  left  out.  Graphs  including  all  banks  can  be  found  in  appendix  A-­‐1.  

  Figure  8:  Median  capital  of  firms  served  by  the  ten  largest  banks.  

Figure  8  illustrates  that  there  are  differences  in  the  amount  of  capital  that  the  customers  of  the  banks   have.  Bank  number  three  and  bank  number  five  to  ten  serve  customers  with  the  same  median  

capital,  while  other  banks  serve  customers  with  a  higher  median  capital.  The  fourth  largest  bank   serves  customers  with  much  larger  capital  than  the  other  banks,  while  the  two  largest  banks  are  also   quite  a  bit  above  the  other  banks.  These  banks,  especially  bank  number  four,  will  have  a  tendency  to   have  a  too  high  market  share  among  existing  customers,  compared  to  new  customers,  as  they  

generally  serve  larger  customers.  This  will  bias  their  switching  costs  upwards.  The  amount  of  firms   these  banks  serve  accounts  for  a  large  part  of  the  market,  which  suggest  that  the  switching  costs  of   those  banks  could  be  upwards  biased.  Generally  however  it  seems  that  there  is  not  much  difference  

between  the  sizes  of  customers  that  most  of  the  banks  serve.  So  while  Figure  6  and  Figure  7  suggests   that  there  are  significant  size  differences  between  the  groups,  Figure  8  suggests  that  there  is  not  a   general  tendency  for  banks  to  serve  customers  of  different  size.    

  Figure  9:  Median  balance  of  firms  served  by  the  ten  largest  banks.  

The  median  balance  of  firms  served  by  each  bank  has  more  variance  than  the  median  capital.  The   four  largest  banks  serve  firms  with  comparable  balance,  but  there  seems  to  be  a  general  tendency  for   smaller  banks  to  serve  firms  with  a  smaller  balance.  This  problem  is  enhanced  as  the  large  banks   serve  most  of  the  market,  so  a  large  part  of  the  market  is  affected  by  potentially  biased  switching   costs.  Figure  7  shows  that  the  existing  firms  are  generally  larger  and  Figure  9  shows  that  the  larger   banks  generally  serve  larger  customers.  As  before  this  has  the  consequence  that  the  switching  costs   of  the  larger  banks  are  potentially  upwards  biased.  

The  potential  bias  ascending  because  of  heterogeneity  of  firms  cannot  be  controlled  for  in  the  

estimations  conducted  in  this  thesis,  and  is  therefore  an  unavoidable  downside  to  using  this  method   to  estimate  consumer  switching  costs.  The  potential  bias  is  of  less  importance  when  doing  the   regression  of  firm  characteristics  on  switching  costs,  as  the  switching  cost  estimates  are  all  

potentially  upwards  biased,  but  the  concern  of  the  estimation,  is  only  the  sign  and  significance  of  the   parameters.  The  sign  of  the  parameters  will  not  change  if  the  switching  cost  estimates  are  increased,   but  their  individual  significance  may  be  overestimated.  Chen  and  Hitt  estimate  consumer  switching   costs  with  and  without  controlling  for  consumer  heterogeneity  and  find  no  substantial  differences   between  the  two  estimations  (Chen  &  Hitt  2002).  This  suggests  that  their  estimation  of  consumer   switching  costs  is  not  very  sensitive  to  consumer  heterogeneity.  If  the  estimation  in  this  thesis  yields   the  same  traits,  the  estimates  should  be  practically  unbiased.