Results  of  the  estimation

Table  4:  Estimation  of  firms’  switching  costs  dependency  on  firms’  characteristics.  

Table  4  shows  the  results  of  the  regression.  The  first  column  shows  the  name  of  the  variable,  the   third  and  fourth  show  the  actual  parameter  estimate,  of  which  only  the  sign  is  of  interest.  The  last   two  columns  show  the  t  value  and  corresponding  P-­‐value.  Below  will  individual  interpretations  of   each  parameter  estimate  be  presented.  

The  intercept  of  the  estimation  is  not  of  interest,  both  because  the  level  of  switching  costs  is  not   interesting  in  this  estimation,  but  also  because  it  does  not  make  economic  sense  to  consider  the   switching  costs  of  a  firm  where  independent  variables  are  zero.  Generally  most  of  the  variables   appear  to  be  significant  and  the  standard  errors  are  quite  low,  which  is  most  likely  because  of  the   large  dataset  used  for  the  estimation.  

The  parameter  estimates  of  the  balance  and  capital  are  both  positive  and  very  significant  compared   to  the  other  variables.  Firms  with  a  larger  balance  or  capital  are  thus  predicted  to  have  larger   switching  costs.  So  the  size  of  the  firms  and  their  switching  costs  are,  according  to  the  estimation,   positively  correlated.  It  seems  reasonable  that  both  parameters  have  the  same  sign,  as  both  variables   are  expressions  of  size  of  the  consumers.  The  positive  relationship  is  also  what  would  be  expected   from  theory.  The  parameter  of  capital  is  larger  and  more  significant  than  balance.  This  could  be   because  capital  is  a  better  measure  for  firm  size  than  the  balance.  The  balance  includes  all  the  firm’s   operations  and  can  easily  be  affected  by  both  the  firm’s  accounting  and  financial  policies.  The  capital   is  more  narrowly  defined,  and  therefore  better  at  predicting  the  level  of  switching  costs  of  the  firm.  

The  reason  for  this  positive  correlation  can  be  found  on  both  the  supply  and  demand  side.  Larger   firms  are  likely  to  require  more  banking  services  and  are  therefore  relatively  more  important  for  the   banks.  Therefore  banks  put  more  effort  into  keeping  the  larger  customers.  The  high  volume  of  

business  probably  also  makes  it  more  inconvenient  to  switch,  but  may  also  increase  the  firms   incentive  to  choose  the  best  current  contract.  The  last  effect  seems  to  be  offset  by  the  others.  All  in   all,  these  effects  and  maybe  other  effects,  reduce  the  likelihood  of  firms  switching  bank.    

The  variable  profit  is  a  dummy  variable  that  is  one  if  a  firm  made  a  profit  in  their  latest  financial   report.  The  parameter  is  negative  and  significant.  According  to  the  estimation,  profitable  firms  have   lower  switching  costs  than  firms  that  did  not  make  a  profit.  This  result  is  in  line  with  the  theoretical   prediction.  Unprofitable  firms  are  generally  not  desirable  customers  in  banks,  as  they  have  a  higher   risk  of  defaulting  on  their  loans.  The  most  obvious  reason  for  the  relationship  between  switching   costs  and  firms’  profitability  is  that  the  firms  and  the  banks  have  obligations  to  each  other  through  a   contract,  and  that  there  is  very  low  supply  of  credit  for  unprofitable  firms.  Competing  banks  are  not   likely  to  offer  loans  to  unprofitable  firms,  let  alone  attractive  contracts,  so  the  firms  often  have  to   stay  with  their  current  bank  connection.  The  current  bank  often  have  an  interest  in  continuing  the   relationship  with  the  firm  if  it  is  unprofitable,  to  lower  the  risk  of  the  firm  defaulting  on  its  loans.  If   the  bank  stop  supplying  credit  to  the  firm,  and  the  firm  is  unable  to  get  financing  elsewhere,  the  firm   may  default  and  be  unable  to  pay  back  its  debt  in  full.  

The  debt  of  firms  is  correlated  with  the  other  variables  through  the  size  of  the  firm,  but  it  is  also  the   product  being  traded  between  the  firm  and  the  bank.  This  makes  it  hard  to  predict  the  variables’  

relationship  with  the  firms’  switching  costs.  The  estimation  reveals  that  the  parameter  is  negative   and  very  significant.  Thus  the  more  debt  a  firm  has,  the  lower  is  its  switching  costs.    

There  are  at  least  two  effects  that  contribute  to  the  parameter  estimation.  The  first  is  the  size  effect   of  the  firm.  The  larger  a  firm  is,  the  larger  is  their  debt  generally.  Other  variables,  such  as  balance  and   capital  that  are  also  linked  to  the  size  of  the  firm,  revealed  a  positive  correlation  between  size  of  the   firm  and  switching  costs  of  firm.  The  other  effect  is  a  consequence  of  the  increased  size  of  the   contract  between  the  bank  and  the  firm.  The  larger  a  potential  contract  is,  the  more  incentive  does   the  competing  banks  have  to  offer  a  good  contract  to  the  firm,  which  can  make  firms’  incentive  to   switch  larger.  The  firms  also  have  a  larger  monetary  incentive  to  get  the  best  possible  contract,  due   to  the  large  volume.  The  current  bank  connection  of  the  firm  obviously  has  the  same  incentive  to   offer  the  firm  a  good  contract,  but  the  increased  competition  between  banks  and  the  strong  incentive   for  the  firm  to  choose  the  best  current  contract  seems  to  exceed  this  effect.  The  size  effect,  that   definitely  is  present  in  the  estimation,  is  offset  either  by  the  before  mentioned  effect  or  by  other   unknown  effects.  

The  equity  of  the  firm  is  estimated  to  be  positively  correlated  with  the  switching  costs  of  the  firm.  

The  larger  a  firm’s  equity  is,  the  larger  is  the  estimated  switching  costs  of  the  firm.  Equity  is  also   positively  correlated  with  the  size  of  the  firm,  which  is  positively  correlated  with  switching  costs.  But   equity  is  also  an  indication  of  firms’  ability  to  absorb  losses.  A  high  equity  is  therefore  also  a  sign  of  a   healthy  firm,  which  is  negatively  correlated  with  switching  costs  as  with  the  profit  variable.  

According  to  the  sign  of  the  parameter,  the  size  effect  is  the  dominating  factor.  This  is  likely  because   the  variable  does  not  compare  the  equity  level  to  other  factors,  so  it  is  not  possible  to  infer  if  the   equity  for  a  given  firm  is  high  or  low.  The  solvency  ratio  does  exactly  that.  It  relates  equity  to  the   assets  and  therefore  excludes  the  effect  of  the  size  of  the  firm  from  the  variable.  

The  solvency  ratio  is  a  continuous  logarithmically  transformed  variable.  The  estimated  parameter  is   negative  and  significant.  It  is  one  of  the  least  significant  variables  with  a  standardized  P-­‐value  around   0.7%,  which  is  still  well  below  the  conventional  cutoff  value  of  5%.  According  to  the  estimation  the   higher  a  firm’s  solvency  ratio  is,  the  lower  is  the  firm’s  switching  costs.  The  solvency  ratio  is  an   expression  for  a  firm’s  ability  to  meet  its  obligations.  It  is  related  to  the  profitability  measure,  as  both   variables  measure  the  firms’  health.  It  is  therefore  not  surprising  that  both  have  the  same  sign.  

Generally  the  estimation  reveals  that  healthier  firms  have  lower  switching  costs,  most  likely  because   they  are  more  attractive  customers  to  competing  banks.  The  reason  for  the  low  significance  of  the   variable  is  most  likely  due  to  high  variation  in  solvency  ratios  of  otherwise  similar  firms.  This   variation  is  caused  by  sector  and  industry  differences,  as  well  as  the  nature  of  the  ratio.  If  a  firm  has   few  assets,  which  is  the  denominator  of  the  fraction,  the  equity  has  a  high  impact  on  the  total  ratio,   and  vice  versa.        

The  number  of  bank  connections  that  each  firm  has  in  the  dataset  is  not  a  financial  figure  like  the   other  variables.  It  is  mainly  included  to  reduce  the  effect  of  some  firms  weighting  more  than  others   in  the  estimation,  if  they  have  more  than  one  bank  connection.  The  parameter  is  negative,  but  not   very  significant.  The  adjusted  P-­‐value  is  slightly  above  10%,  so  the  variable  would  usually  be  

considered  insignificant.  The  standard  error  does  however  reveal  that  the  parameter  estimate  with  a   high  probability  is  negative,  which  is  an  interesting  result.  The  number  of  bank  connections  a  firm   has  is  possibly  correlated  to  the  size  of  the  firm,  but  it  is  not  clear  how  strong  this  effect  is.  Most  firms   can  get  their  demand  fulfilled  by  one  bank,  but  larger  firms  may  want  more  bank  connections,  to   spread  their  risk  and  dependency  across  several  banks.  The  logical  relationship  between  the  number   of  bank  connections  that  a  firm  has,  and  the  switching  costs  of  the  firm,  is  negative.  Multiple  bank   connections  indicates  that  the  firm  is  not  loyal  to  one  bank  and  therefore  have  lower  costs  of   switching.  If  a  firm  has  more  than  one  bank  connection,  then  it  will  be  easier  to  close  a  bank   connection  with  a  bank,  which  in  the  dataset  will  be  interpreted  as  a  switch.  This  is  not  a  desired   feature,  but  it  is  the  disadvantage  of  including  firms  with  more  than  one  bank  connection.  The  last   mentioned  effect  dominates  the  size  effect  in  the  dataset,  according  to  the  estimation.  

The  last  parameter  estimates  are  those  of  the  dummy  variables  containing  information  about  the   year  of  establishment.  The  default  category  is  the  group  consisting  of  firms  established  before  1950.  

All  parameter  estimates  are  therefore  in  relation  to  this  group.  All  the  parameter  estimates  related  to   the  year  of  establishment  are  negative,  meaning  that  the  other  groups  of  firms  has  lower  switching   costs  than  the  default  group.  It  is  expected  that,  if  there  is  a  significant  relationship,  then  it  is  a   positive  correlation  between  the  number  of  years  a  firm  has  existed  and  the  switching  costs  of  the   firm.  The  parameter  estimations  are  thus  in  line  with  the  expectations.  The  group  of  firms  

established  between  1950  and  1959  has  a  P-­‐value  of  0.23%,  which  yields  a  standardized  P-­‐value  

slightly  above  5%.  The  group  of  firms  established  between  1960  and  1969  also  has  a  standardized  P-­‐

value  above  5%,  which  suggest  that  the  parameter  estimates  are  not  significant,  and  consequently   that  the  switching  costs  of  the  firms  established  between  1950  and  1969  are  not  significantly  

different  from  the  switching  costs  of  firms  established  before  1950.  The  three  groups  are  arbitrarily   composed  and  they  all  contain  firms  that  are  very  mature,  it  is  therefore  reasonable  to  accept  that   they  do  not  have  different  switching  costs.  The  other  groups  have  parameter  estimates  that  decrease   as  the  ages  of  the  firms’  decreases.  According  to  the  estimation,  the  more  recently  a  firm  was  

established,  the  lower  switching  costs  does  it  have,  which  is  also  in  line  with  the  expectations  of  the   relationship.  The  relationship  can  be  partly  explained  by  the  likely  correlation  between  the  amount   of  time  a  firm  has  been  a  customer  at  a  bank  and  the  year  a  firm  were  established,  caused  by  the  fact   that  firms  generally  do  not  switch  banks  very  often.  The  longer  a  firm  has  been  a  customer  at  a  bank,   the  higher  switching  costs  are  the  firm  expected  to  have.    

The  variables  of  the  estimation  are  generally  all  relatively  highly  correlated,  as  firms’  characteristics   are  all  related  to  size  of  the  firms,  which  may  lead  to  multicollinearity.  The  issue  of  multicollinearity   would  be  an  important  issue  if  a  smaller  sample  size  were  used  for  the  estimation.  Multicollinearity   increases  the  standard  errors,  because  it  is  not  clear  which  of  the  independent  variables  that  are   responsible  for  the  variation  in  the  dependent  variable.  This  effect  is  mitigated  by  the  large  sample   size,  as  can  be  seen  from  the  standard  errors  of  the  estimation.  

The  coefficient  of  determination,  𝑅^!,  is  slightly  above  2%.  It  is  very  low,  which  means  that  the  model   does  not  do  a  very  good  job  at  predicting  the  observed  values.  Thus  the  characteristics  of  firms,  used   in  the  estimation,  are  not  good  predictors  of  the  firms’  switching  costs.  It  was  not  expected  that  the   firm  characteristics  could  explain  very  much  of  the  variation  in  the  consumer  switching  cost,  so  a  low   coefficient  of  the  determination  was  expected.  The  dataset  include  all  firms,  without  controlling  for   outliers,  other  than  transforming  the  variables.  Looking  at  the  data,  it  is  evident  that  there  are  many   large  outliers,  as  well  as  significant  variation  across  industries.  This  may  also  contribute  to  a  low  the   coefficient  of  the  determination.  The  focus  of  the  estimation  was  not  to  set  up  a  model  to  predict  the   consumer  switching  costs  on  the  basis  on  consumers’  characteristics,  but  to  examine  if  there  were   some  relationship  between  the  characteristics  and  the  switching  costs.  If  the  model  was  constructed   to  predict  the  switching  costs  of  consumers,  the  coefficient  of  determination  could  most  likely  be  

increased  if  outliers  were  removed  from  the  dataset,  and  the  industry  of  the  firms  were  controlled   for.  

Generally  the  parameter  estimates  are  all  very  reasonable,  and  are  consistent  with  both  theory  and   expectations.  The  most  important  result  of  the  estimation  is  that  almost  all  of  the  variables  are   significant.  While  the  variables  does  not  explain  very  much  of  the  variation  in  the  switching  costs,  it   is  still  notable  that  they  have  an  influence  on  the  switching  costs  of  the  firm.