Financial  data

The  various  financial  data  chosen  for  the  estimation  reflects  characteristics  of  consumers  that  the   switching  costs  could  be  affected  by.  Rather  than  just  financial  figures,  the  variables  should  be  seen   as  characteristics  of  the  firms.  Each  variable  is  reviewed  below.  The  database  includes  financial   reports  of  firms  that  are  not  active  anymore,  as  well  as  historical  financial  reports.  The  bank   connections  are  not  registered  with  a  date,  so  they  are  all  assumed  to  be  the  current  bank  

connections.  It  is  not  clear  how  the  bank  connections  are  registered,  but  it  is  unlikely  that  they  are  all   collected  on  the  same  day  and  updated  daily.  The  financial  data  used,  has  to  be  the  latest  available  to   reflect  the  current  bank  connections,  and  the  time  the  switching  costs  are  estimated.  The  financial   report  also  has  to  be  for  2011  or  later,  to  ensure  that  the  characteristics  of  the  firm  are  current.  

The  dependent  variable  of  the  estimation  is  the  consumer  switching  costs  estimated  empirically  from   the  firms’  bank  connections.  All  customers  at  each  bank  thus  have  the  same  switching  costs.  The   variable  will  be  extensively  reviewed  in  section  5.2.  

The  balance  and  capital  of  the  firms  are  included  in  the  estimation  to  model  the  size  of  the  firms.  

Both  figures  are  non-­‐negative  and  have  a  high  amount  of  variance  across  individual  firms,  and  is   often  highly  dependent  on  the  industry  of  the  firm  as  well  as  the  firms’  financial  policy.  The  balance   of  the  firm  is  sensitive  to  accounting  methods,  and  items  that  may  not  be  relevant  for  the  size  of  the   firm  can  also  be  included  in  the  balance.  Capital  of  the  firm  can  also  vary  between  otherwise  similar   firms.  One  can  be  undercapitalized  while  the  other  can  be  overcapitalized.    The  high  variance  across   otherwise  identical  firms  can  decrease  the  predictive  power  of  the  variables.  The  variance  across   time  is  on  the  other  hand  not  very  large  compared  to  other  financial  figures  that  may  change  

significantly  from  one  year  to  another.  Both  of  the  variables  are  skewed  across  firms,  due  a  small   amount  of  very  large  firms  with  very  high  balance  and  capital.  To  reduce  this  skewedness,  both   variables  are  logarithmically  transformed.  Alternative  measures  of  market  size  could  be  output  of  the   firm,  market  capitalization  or  revenue,  which  may  be  more  suitable  for  some  subgroups  of  firms.  

Output  and  market  capitalization  is  however  not  available  for  all  firms  in  the  dataset.  Revenue  is   registered  for  around  50%  of  firms,  but  it  is  optional  for  firms  if  they  want  it  published,  so  it  is  not   included  in  the  estimation.  

Firms’  profit  are  available  through  the  financial  reports  and  possibly  has  influence  on  the  switching   cost  of  the  firm.  Firms’  profit  provides  some  indication  of  the  current  financial  situation  of  the  firm,   and  thus  its  ability  to  pay  back  its  loans.  Banks  are  generally  not  interested  in  acquiring  new  

customers  with  high  credit  risk.  The  profit  itself  is  not  particularly  interesting,  when  it  is  not  related   to  another  financial  figure  that  contains  information  about  the  base  upon  which  the  profit  is  earned.  

The  hypothesis  being  examined  is  concerned  with  the  switching  costs  of  good  and  bad  customers,  so   it  is  not  interesting  to  estimate  exactly  how  much  switching  costs  change  when  the  profit  change.  

Therefore  a  dummy  variable  that  is  1  if  the  firm  earned  a  positive  profit  and  0  otherwise  is  used   instead  of  the  original  variable.  The  profit  of  firms  also  varies  considerably  across  firms,  most  likely   more  than  the  balance  and  capital,  but  also  over  time.  A  firm’s  profit,  especially  in  a  distressed   economic  environment  as  the  one  the  firms  in  this  thesis  operate  in,  can  vary  significantly  from  year   to  year.  Using  the  dummy  variable,  instead  of  the  actual  profit,  should  reduce  this  effect.    

The  equity  of  firms  is  included  in  the  estimation  because  it  is  a  central  financial  figure.  It  is  

dependent  on  the  assets  and  liabilities,  as  well  as  the  size  of  the  firm  and  the  profit  over  time.  It  is   therefore  possible  that  the  amount  of  equity  a  firm  has  is  correlated  with  some  of  the  other  variables   in  the  estimation.  While  it  is  related  to  many  factors,  the  amount  of  equity  a  firm  has  is  a  financing   decision,  and  can  vary  from  firm  to  firm.  Large  outliers  characterize  the  data,  as  the  balance  and   capital  figure,  so  a  logarithmic  transformation  is  applied  to  reduce  the  effect  the  of  outliers.  Equity   can  be  negative,  and  there  are  both  positive  and  negative  outliers.  Since  the  logarithm  of  negative   numbers  or  zero  is  not  defined,  the  lowest  equity  in  the  dataset  plus  one  is  added  to  all  equity   figures.  It  obviously  does  not  make  sense  to  interpret  the  parameter  value,  but  since  the  concern  of  

this  estimation  is  only  the  dependencies  and  not  the  actual  parameters,  there  is  no  loss  of  generality   due  to  the  logarithmic  transformation.  

The  total  amount  of  debt  each  firm  has  is  also  included  in  the  estimation.  It  is  calculated  as  the  sum  of   the  short-­‐term  and  long-­‐term  debt  for  each  firm.  The  debt  is  related  to  equity  and  balance  by  the   accounting  identity  that  says  debt  and  equity  must  equal  assets.  Including  debt  could  thus  possibly   cause  multicollinearity,  which  could  affect  the  parameters  estimations  of  individual  variables.  

Multicollinearity  cannot  reduce  the  predictive  power  or  reliability  of  the  estimation,  but  should  be   avoided  to  reduce  the  standard  errors  of  the  parameter  estimates.  In  the  real  world,  the  accounting   identity  is  not  that  simple.  There  are  other  instruments  such  as  subordinated  loan  capital  and  other   instruments  that  are  not  short-­‐  or  long-­‐term  debt  and  not  equity.  These  instruments  are  therefore   not  included  in  the  dataset  variables  used  for  the  estimation.  Including  the  debt  in  the  estimation   should  therefore  not  cause  multicollinearity.  Like  many  of  the  other  variables,  the  calculated  variable   debt  is  skewed  to  the  right  with  many  large  outliers.  A  logarithmic  transformation  of  the  variable  is   therefore  used  in  the  estimation.  

The  solvency  ratio  is  the  only  financial  ratio  included  in  the  estimation.  It  is  defined  as  equity  divided   by  total  assets.  The  solvency  ratio  is  an  expression  for  a  firm’s  ability  to  incur  losses.  It  measures  the   percentage  of  capital  the  firm  can  lose  before  the  more  senior  financing  is  affected.  The  solvency   ratio  is  negative  if  the  equity  is  negative,  which  is  a  clear  sign  of  financial  distress.  The  solvency  ratio   has  a  tendency  to  be  very  negative  when  it  is  negative.  In  the  database  the  solvency  ratio  is  capped  at   -­‐999  and  999,  which  is  relevant  for  some  firms  in  the  dataset.  The  solvency  ratio  is  very  sensitive  in   its  nature,  with  many  outliers  and  high  variation,  so  the  solvency  ratio  is  logarithmically  transformed   in  the  estimation  to  reduce  these  effects.  The  solvency  ratio  can  be  negative,  so  the  database  cap  plus   one  is  added  to  all  solvency  ratios.  It  could  be  preferred  to  use  a  dummy  variable  to  proxy  firms  with   a  good  solvency  ratio  and  a  bad  solvency  ratio.  It  is  however  not  obvious  what  a  high  and  low  

solvency  ratio  is  for  the  entire  population  of  firms,  so  a  continuous  scale  seems  like  the  better   choice³.    

3  Using  a  dummy  variable  yield  equivalent  results.    

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.