Ogura “the objective function of government controlled banks in a financial crisis”, j banking and finance (2018)

We find that, in response to the increased loan demand, the welfare-maximizing GCB increases its lending more for firmswith a weak relationship with its main bank, in the sense that the

To appear in: Journal of Banking and Finance

Received date: 21 October 2016

Revised date: 15 January 2018

Accepted date: 27 January 2018

Please cite this article as: Yoshiaki Ogura, The Objective Function of Government-Controlled Banks in

a Financial Crisis, Journal of Banking and Finance (2018), doi:10.1016/j.jbankfin.2018.01.015

This is a PDF file of an unedited manuscript that has been accepted for publication As a service

to our customers we are providing this early version of the manuscript The manuscript will undergocopyediting, typesetting, and review of the resulting proof before it is published in its final form Pleasenote that during the production process errors may be discovered which could affect the content, andall legal disclaimers that apply to the journal pertain

ACCEPTED MANUSCRIPT

Highlights

• GCBs increased lending to SMEs with a weaker main-bank relationship

in the financial crisis

• This is consistent with the welfare maximization by GCBs rather than the

profit maximization

1

JEL Classification: G21; H44

Keywords: government-controlled banks, mixed oligopoly, relationship banking, small businessfinancing

ACCEPTED MANUSCRIPT

1 Introduction

The existing literature on government-controlled banks (GCBs) has presented mixed judgments

on the banks’ contribution to economic efficiency The seminal empirical study by La Porta et al.(2002) shows international evidence of the underperformance of GCBs Several subsequent studiesshow evidence that such inefficiency mainly comes from the political constraint or the politicalcapture (e.g., Sapienza, 2004; Din¸c, 2005),1 and that a privatization significantly improves theefficiency (Bertrand et al., 2007) On the other hand, recent studies show evidence of the benefits

of GCBs, such as mitigating the credit constraint against SMEs (Behr et al., 2013; Lin et al.,2014; Sekino and Watanabe, 2014) and the less procyclicality of their lending (Micco and Panizza,2006; Brei and Schclarek, 2013; Cull and Per´ıa, 2013; Coleman and Feler, 2015; Behr et al., 2017),2especially in countries with good governance (Bertay et al., 2015) Moreover, macroeconomicanalyses theoretically predict the possibility of welfare improvement via counter-cyclical policylending to firms in a model with a financial friction (e.g., Gertler and Karadi, 2011; Martin andVentura, 2016) However, it remains an open empirical question whether the lending behavior ofGCBs improves welfare

This interesting and important issue boils down to the question of which is the actual objectivefunction of GCBs among various alternatives, such as their own profits, the social welfare, or someother political interests To figure out an empirical strategy to detect their objective function, first

we applied a mixed Cournot oligopoly model (Fraja and Delbono, 1989; Ide and Hayashi, 1992;Matsumura, 1998) to the loan market for a firm The standard mixed-oligopoly model assumes

a public firm, which maximizes the social welfare, and multiple profit-maximizing private firms

We introduce an additional twist of the asymmetry among profit-maximizing private banks totake into account relationship banking, which is a widely accepted phenomenon in small businessfinancing (for the list of the existing studies, see, e.g., Degryse et al., 2009) Namely, we assume acredit market with a GCB, a main bank providing a differentiated service based on its information

1 More recently, Pereira and Maia-Filho (2015) find a slower transmission of the monetary policy to the interest rates of GCBs Illueca et al (2014) and Iannotta et al (2013) provide evidence of excess risk-taking by government- controlled banks.

2 Brei and Schclarek (2015) theoretically explain that this phenomenon is due to the differences between private banks and public banks in terms of the objective functions and the funding sources.

ACCEPTED MANUSCRIPT

advantage, and another private bank without such an advantage

We consider two cases; first, the case where the GCB maximizes the social welfare, which isdefined by the sum of the profits of all banks and the surplus for the borrowing firm; second, thecase where the GCB is a profit maximizer, like a private non-main bank We find that, in response

to the increased loan demand, the welfare-maximizing GCB increases its lending more for firmswith a weak relationship with its main bank, in the sense that the extent of differentiation of themain bank is lower and that the main-bank loan demand is more price-elastic This is becausethe GCB is less willing to interrupt a lending relationship between a firm and its main bank if

it provides a differentiated service that is more valuable for the firm and contributes more to thesocial welfare In contrast, a profit-maximizing GCB never adjusts its lending according to thestrength of the main-bank relationship This result suggests that we can detect whether a GCB is

a profit maximizer or a welfare maximizer by examining whether it increases lending more to firmswith a weaker main-bank relationship in response to a surge in loan demand

The microdata provided by the Small and Medium Enterprise Unit of the Japan Finance poration (JFC), one of the major GCBs for SMEs, enables us to conduct this empirical test Thedataset contains information on the annual financial statements and other basic characteristics ofeach past and current borrower, as well as outstanding loan amounts from the SME Unit of JFCand private banks up to the four largest lenders The most desirable feature of the data is that itcontains the identifier of these private banks so that we can match the bank information

Cor-We focus on the dataset from 2007 to 2011 before and after the 2008-09 financial crisis severelyaffected the Japanese economy through the sharp reduction of exports to the USA and Europe in theaccounting period ending in 2009 The benefits of using the Japanese dataset are threefold First,the financial crisis was an exogenous shock to Japanese banking and industrial sectors The bankingsector was barely affected by the shock, while the shock had a deep impact on the performance andfinancing behavior of the industrial sectors, especially the exporting sectors Second, we observed

a clear surge in the demand for bank loans in the accounting period ending in 2009 (for typicalJapanese firms, the end of the accounting year is in March) The survey of large banks conducted

by the Bank of Japan clearly shows this (Figure 1) This is because of the temporary shutdown

of the commercial paper and bond markets (Uchino, 2013) and the precautionary motivations in

ACCEPTED MANUSCRIPT

response to the disastrous drop in corporate earnings in the exporting sector, as shown in Section 3.These points ensure the theoretical assumption of the exogenous loan demand shift for our empiricalhypothesis Third, the dataset enables us to construct a three-way panel dataset by firm, year,and lender This three-way panel data enables us to fully control for the unobservable time-varyingfirm characteristics, such as the magnitude of a demand shock and other credit characteristics byintroducing the firm-by-year cross fixed effect In the context of this study, both the intensity

of a main-bank relationship and the lending attitude of GCBs are correlated with unobservablefirm characteristics The estimated correlation between the lending attitude of GCBs and theintensity of main-bank relationship can be biased due to these unobservables, if we cannot controlfor them perfectly The firm-by-year cross fixed effect minimize this problem and provide a clearidentification, as proposed by Gan (2007) and subsequent studies

From the regression using the three-way panel data to control for the firm-by-year cross fixedeffect, we find that the GCBs increased their lending to SMEs whose main bank is a large bank,which operates nationwide and internationally, in the crisis period of two years after September

2008, while they decreased lending for other SMEs On the other hand, we also find that a mainbank decreased lending in the crisis period if it is a large bank, whereas it increased if it is a regionalbank Thus, GCBs filled in the loan supply shortage of large main banks

Since we control for unobservable time-varying firm characteristics as mentioned above, thisresult is less likely to be driven by unobservable firm characteristics, such that a large main bankobtains negative private information and reduces lending to an SME while a GCB without it keeplending, or that a GCB increases its lending because of positive private information while a mainbank without it reduces lending

Further analyses with more explicit relationship measures, such as the main-bank loan or depositshare, a dummy variable indicating that a firm switched main banks before the crisis, or the number

of lenders before the crisis, show that this result is driven by the weak relationship between largebanks and SMEs, which has been recognized in the existing literature (e.g., Cole et al., 2004;Berger et al., 2005; Uchida et al., 2008; Ogura and Uchida, 2014) This is consistent with thewelfare-maximizing behavior in the above theoretical prediction

The remaining part of this paper is organized as follows We describe the source of our dataset in

ACCEPTED MANUSCRIPT

Section 2 The financial condition and financing behavior of Japanese SMEs in the 2008-09 financialcrisis are described in Sections 3 and 4 A theoretical model to derive an empirical strategy to detectthe objective function of the GCB is presented in Section 5 The hypothesis for the statistical test,the data description, the specification for the estimation, and the result of the test are presented

in Section 6 Section 7 presents the conclusion and the limitation of our analysis

The dataset for this study is the internal credit information on borrowers at the Small and MediumEnterprise (SME) Unit of the Japan Finance Corporation (JFC) JFC is a 100% government-owned and government-controlled lending institution that provides subsidized long-term loans toSMEs; microcorporations including start-up firms and farmers; and individuals It also providesreinsurance for the public guarantee system for SME loans JFC does not take deposits and isfinanced mostly by borrowing from the Japanese government and partially by issuing bonds with

or without government guarantees It has a nationwide branch network of 152 branches (March2009) The SME Unit is the business unit focusing on loans to SMEs The total outstanding loanamount of this unit was about 5.2 trillion JPY in March 2009 The asset size is close to that oflarger regional banks The unit was called the Japan Finance Corporation for Small and MediumEnterprise (JASME) before its merger with other units in October 2008

The internal credit information of the SME unit includes the annual financial statement mation and other basic characteristics of each borrowing firm, such as the industrial classification,the year of establishment, and the location of the JFC branch that transacts with the firm, aswell as the internal credit rating The most notable feature of the dataset is that it contains theoutstanding loan amount provided by JFC and other private and government-owned institutions.The names of lenders can be identified for the largest four lenders to match the financial and otherinformation of each lender JFC identifies a main bank of each firm based on information such asdeposit share and loan share This information enables us to examine what types of firms becamemore dependent on GCB lending in the crisis and evaluate the economic efficiency of GCB lending

infor-We use the observations from calendar years 2007 to 2011, from right before the outbreak of thecrisis to several years after The dataset covers not only firms with a current positive amount of loan

ACCEPTED MANUSCRIPT

outstanding from JFC but also those without this for several years before starting a transaction orafter closing a transaction with JFC The number of firm-year observations is 230,587 From theoriginal sample, we drop firms whose main bank is a JA bank (agricultural cooperative), a JF marinebank (fishery cooperative), a GCB (739 observations), or a Shinkumi bank (3,298 observations),which is a smaller credit cooperative, since the data of their characteristics are not fully available

We drop 14,454 observations whose borrow/asset is greater than one to avoid the effect of firmsunder a bankruptcy procedure in all of the estimations Finally, we drop 64,514 observations forwhich any item required for the preliminary regression in Section 4 is not available The remaining147,582 observations are the baseline sample for our analysis

The industrial composition of the borrowers at the SME Unit of JFC tilted more toward themanufacturing sector than did the population, which was measured by the 2009 Economic Census(Table 1) Table 3 shows the descriptive statistics of the variables to be used for the regressionslater The definition of each variable is listed in Table 2 The median of the main-bank share ofloans is about 38% and that of deposits are about 66% The median of the loan share of GCBs isabout 35%, somewhat lower than that of the main bank The number of lenders other than JFC

is three on average The minimum is one, i.e., each firm has a relationship with at least one bankother than JFC This is because JFC does not provide checking, savings, or settlement services.The median of the asset size is 780 million JPY The Credit Risk Database (CRD), which is closer

to the population of the SMEs with access to the loan market, indicates that the median asset sizewas 85 million JPY in 2003 (Table 1.4 on p.21 in Shikano, 2008) Thus, our dataset focuses onlarger firms among the SMEs

More than 70% of our sample firms chose regional banks3as their main bank (Table 4) Regionalbanks operate within a single or a couple of adjacent prefectures The remaining 30% chose largebanks, which have a nationwide branch network and operate nationwide or internationally.4 Largebanks have features clearly different from those of other types of banks First, the main-bank shares

3 Regional banks include both the member banks of the Regional Banks Association of Japan and the Second Association of Regional Banks, and cooperative banks, such as Shinkin and Shinkumi banks The asset size of the banks in the two regional bank associations ranges from 0.2 to 11.6 trillion JPY as of March 2009 The asset size of cooperative banks is smaller, and ranges from 0.004 to 3.9 trillion JPY as of March 2009.

4 Large banks include city banks (Mitsubishi UFJ, Sumitomo-Mitsui, Mizuho, Mizuho Corporate, Risona, Saitama Risona, Shinsei, and Aozora) and trust banks (Mitsubishi UFJ Trust, Mizuho Trust, Chuo-Mitsui Trust, and Sumit- omo Trust) The asset size ranges from 6.1 to 149 trillion JPY as of March 2009.

ACCEPTED MANUSCRIPT

of deposits and loans are significantly lower when the main bank is a large bank than otherwise(Panels (a) and (b), Figure 2) Second, firms switch their main banks more frequently when theirmain bank is a large bank than otherwise (Table 5) The probability that a firm had switchedmain banks from the previous year is higher by at least 1% for larger banks The difference was

at a maximum in 2010, the later stage of the financial crisis Third, the ratio of SME loans overtotal loans of large banks is significantly lower than that for other types of banks (Panel (c), Figure2).5 The difference is about 10-17% The gap significantly widened in 2009, in the midst of thecrisis, and has remained wide since then, as large banks decreased the SME ratio considerably,while regional banks slightly increased the SME ratio In contrast to the decline of large banks inSME lending, the share of the GCBs for firms whose main bank was a large bank kept increasing

in 2008 (Panel (d), Figure 2) These figures and table suggest that large banks maintain a weakerrelationship with SMEs than regional banks do, even if they are recognized as a main bank by thefirm or other lenders

Table 6 shows a comparison of the characteristics of those firms whose main bank was a largebank and others in the crisis period of two years from September 2008 The main-bank share ofloans and deposits decreased significantly more for firms whose main bank was a large bank, andthe loan share of GCBs increased more for them In terms of creditworthiness, the firms whosemain bank is a large bank had assets twice as large as the other firms The JFC credit rating forthem was significantly higher than that for others, whereas the damage to the credit rating, sales,interest coverage ratio (∆credit rating, ∆ln(sales), and ∆int.cover) were more severe for the formergroup of firms This is because the weight of exporters such as the manufacturing sector is largerfor the clientele of large banks than that of regional banks or cooperative banks In short, thosewhose main bank was a large bank were larger and more creditworthy, but they were affected moreseverely by the temporary shock of the global financial crisis

5 Cooperative banks are allowed to lend to individuals and SMEs only by regulation.

ACCEPTED MANUSCRIPT

3 Corporate Finance of Japanese SMEs in the 2008-09 Financial Crisis

3.1 Loan demand increased sharply in 2009

The dataset shows that the 2008-09 financial crisis severely affected the Japanese SME loan marketwith a short lag through the plummeting export to the USA and Europe Panel (a) of Figure 3 isthe plot of the sector average of the ratio of EBITDA over total assets, which is calculated from themicrodata provided by JFC and is normalized to 100 in 2007 for all sectors Clearly, the earnings

of Japanese SME exporters in the electronics, transportation equipment (including auto makersand their suppliers), and other manufacturing sectors dropped by more than 50% from 2008 to

2009 These exporters increased their cash holdings in response to this serious crisis, probably with

a precautionary motivation (Panel (b) in Figure 3) despite the fact that the cash flow from theirusual operation had dramatically contracted The increased cash holdings were mainly financed

by bank loans as is indicated by the sharp increase in the ratio of loans over assets in the exportsectors (Panel (c) in Figure 3)

3.2 Banks responded differently by type

The response of each individual bank varies by bank type Figure 4 shows the average annualchange in loans from each lender to each firm The values are normalized by the total asset of eachfirm in the previous year GCBs for SMEs including all units of JFC and the Shoko Chukin Bank,6another GCB for SMEs, increased their lending sharply in 2009 and kept increasing it until 2011 as

a result of the reinforcement of the safety-net lending for SMEs by the government through theseGCBs Regional banks also increased their lending in 2009 but decreased it in 2010 In contrast,large banks never increased their lending even in 2009, although the speed of reduction slowed in

2009 This stark contrast between regional banks and large banks is likely to stem from the factthat the relationships of a large bank with SMEs are weaker than those of a regional bank, as shown

6 The JFC SME Unit accounts for about 70%, JFC other units account for about 10%, and the Shoko Chukin Bank accounts for about 20% of observations in each year in our sample (Table 8) The influence of the government

is somewhat smaller for the Shoko Chukin Bank than for JFC The government holds 46.46% (March 2009) of the share of the Shoko Chukin Bank, and the remainders are widely held by various private entities, including financial institutions The bank is mostly financed by deposits and bank debentures (the latter is until 2012) Its board includes several members sent from the government It has a nationwide branch network of 93 branches The amount

of outstanding loans is 9.2 trillion JPY, which is larger than that for the SME Unit of JFC (March 2009).

ACCEPTED MANUSCRIPT

in extant empirical studies (e.g., Uchida et al., 2008; Ogura and Uchida, 2014) and the tables andfigures in the previous section As if to fill in this gap, GCBs increased lending more extensivelyfor firms whose main bank was a large bank after 2009

A quick guess suggests that this difference between large main banks and regional main banks

is due to exposure to the negative shock of the financial crisis However, the financial indicators ofthese banks in the period show that the Japanese banking sector was not seriously affected by theoverseas shock For example, Figure 5 is the plot of the average risk-adjusted capital ratio of eachtype of bank The figure indicates that the damage to the capital ratio in 2009 was quite limited,even for large banks that had more exposure to foreign assets although the size of the reductionwas larger for large banks than that for the other types of banks The capital ratio of large banksquickly recovered in 2010 The movement of large banks is mainly driven by the denominator ofthe capital ratio, the amount of total loans The total loans of city banks, which consists of themajor part of large banks, increased by 4.6% from March 2008 to March 2009, whereas it decreased3.7% in the next year (Source: Bank of Japan) The sharp increase in the capital ratio of largebanks in 2010 is also due to the capital increases by massive seasoned equity offerings (SEO) inresponse to the expected reinforcement of the regulatory requirement for capital.7

4 Preliminary Regression Analysis: Share of the Main Bank and GCBs

To check whether the casual observation that the share of GCBs increased for firms whose mainbank was a large bank after 2009 was neither driven by the deteriorated financial soundness of mainbanks nor the characteristics of the clientele of each type of bank, we regress the GCBs’ share tobank-type dummies and various firm and main-bank characteristics

The precise specification for this preliminary regression is

ln(GCB f or SM Es shareit) = β0+ β1· MB largeit+ β2· MB largeit· crisist

+β3· MB largeit· post − crisist+ δ0Xit+ µi+ it, (1)

7 For example, Mizuho Financial Group raised capital of 516 billon JPY on July 24, 2009, and 729.17 billion JPY

on July 22, 2010 by SEO Mitsubishi UFJ Financial Group raised capital of 1 trillion JPY on December 22, 2009, and Sumitomo Mitsui Banking Corporation raised 827.4 billion JPY on June 22, 2009, by SEO.

ACCEPTED MANUSCRIPT

where i is the index for each firm, t = {2007, · · · , 2011}, and βs, γs and a column vector δ arecoefficients to be estimated Xit is a column vector of control variables including main-bank char-acteristics, firm characteristics, crisis dummies, and the interaction term of the year dummies andsector dummies Main-bank characteristics consist of the financial soundness or risk-taking capac-ity of each bank, such as capital ratio, ROA, and non-performing loan ratio Firm characteristicsinclude those related to the creditworthiness of each firm The details of the variable definition andthe descriptive statistics are listed in Tables 2 and 3, respectively θt is the year fixed effect µi

is the firm fixed effect it is the error term ln(GCB f or SM Es shareit) is a logit-transformedshare of GCBs for SMEs in the loan for firm i in year t (see Table 2 for details) M B largeit is

a dummy variable, which equals one if the main bank of firm i in year t is a large bank crisis

is a dummy variable, which is equal to one if the observation is reported in the crisis period fromSeptember 2008 to August 2010 post-crisis is a dummy variable to indicate the post-crisis periodfrom September 2010 to the end of the sample period, December 2011

We also estimate a model in which the dependent variable is replaced with the logit-transformedloan share of the main bank, ln(M B loan share), or the ratio of total borrowing over total asset

of each firm, borrow/asset to look at the change in the main-bank share and the change in totalborrowing in the crisis

The regression result is listed in Table 7 Column (1) is the list of estimated coefficients andthe firm cluster robust standard errors when we regress the logit-transformed GCB share Thebase category includes firms whose main bank is a regional bank including a cooperative bank.The estimated coefficient for the dummy variable crisis is negative and significant This indicatesthat the GCB dependence of those whose main bank is a regional bank or a cooperative bank keptdecreasing in the crisis period In contrast, the estimated coefficients of the interaction terms of

M B large and the crisis-phase dummies are positive and significant, i.e., those whose main bank

is a large bank increased its dependence on GCBs in the crisis period, relative to those whose mainbank is a regional bank

Column (2) in Table 7 shows the result when we regress the logit-transformed main-bank share.The estimated coefficient of M B large is deeply negative and significant This point shows that themain-bank share is smaller when the main bank is a large bank, given the same firm characteristics

a cross fixed effect of firm× year by making use of the three-way panel data of the firm, year, andbank level to address this problem sharply in the full analysis.

The control variables also show interesting results Main-bank characteristics in Columns (2)and (3) show that the main-bank share and total borrowing reduce when the main bank suffersfrom a higher non-performing loan ratio However, the effect of the capital adequacy ratio of themain bank, which is adjusted by subtracting the minimum requirement for the credit risk, on themain-bank share is opposite: the main-bank share is smaller for those with a main bank with ahigher capital ratio As for the firm characteristics, firms with a higher credit rating have a highershare of GCBs, a lower share of main banks, and lower borrowing Perhaps this captures thereverse causality, i.e., those with less debt are rated higher On the other hand, the improvement incredit rating increases the main-bank share and total borrowing, while it reduces the dependence

on GCBs Larger firms, which are presumably more creditworthy, depend less on GCBs Thosewith more tangible assets that are pledgeable as collateral are more dependent on their main bankand have a higher dependence on borrowing

5 Model for the Welfare Evaluation of GCBs

To understand the welfare effect of such lending behavior by GCBs theoretically, and to clarify arelevant hypothesis, we construct a model based on the mixed Cournot oligopoly model (e.g., Fraja

We consider the case where a main bank, a non-main bank, and a GCB are potential lenders for afirm We assume that the firm prefers loans from its main bank to those from others This sort ofbrand loyalty would be generated by the borrower’s expectation that the main bank is willing toprovide additional loans in a flexible manner when the firm is under temporary financial distress(Chemmanur and Fulghieri, 1994; Din¸c, 2000; Bolton et al., 2016) or the expectation of additionalservices, such as more effective advising and monitoring, based on proprietary information at themain bank generated from a long-term relationship (Boot and Thakor, 2000; Yafeh and Yosha,2001) We assume that the loans from non-main banks and from the GCB are homogeneous services.Firms expect that these additional benefits will improve their corporate value To formulate thisassumption into an analytical model, we assume the following loan demand function of a firm,which can be derived from the profit maximization problem of each firm, which is not explicitlymodeled here.

where Lm is the amount of a loan from the main bank and L−m is the amount of loans from anon-main bank or a GCB, i.e., L−m≡ Lo+ Lg, where Lo is a loan amount from an outside bankand Lgis a loan amount from a GCB Rm is the gross interest rate of the loan from the main bank,and R−m is that of the loan from other banks The interest rate is identical for loans from a non-

We consider a mixed-oligopoly loan market, where a GCB, which may maximize the socialwelfare, and private banks, which engage in asymmetric Cournot competition,8 are operating Bysolving the above simultaneous equations of demand functions with respect to each interest rate,

we obtain the following inverse demand functions,

We assume that the firm defaults on all loans with a probability of 1− p ∈ (0, 1) to take intoaccount the credit risk The firm does not yield anything, and so the profit of the firm and therepayment from the firm are zero in the default case The funding cost or an opportunity cost for

8 An empirical study about the Japanese banking sector by Uchida and Tsutsui (2005) does not reject the possibility that large banks are competing in a Cournot manner There is no clear consensus whether the Cournot model is a good approximation for lending competition, but we use this setup for the tractability of the model.

of the firm profit and the total profit of all banks.

The profit maximization problem for the main bank is

Z L o +L g

0(a− bl − cLm)dl− r(Lm+ Lo+ Lg) (11)The first term is the firm profit from the loan from a main bank, the second term is the firm profitfrom the loans from a non-main bank and the GCB, and the last term is the total funding cost forthe banking sector

The FOCs for each of these problems are

ACCEPTED MANUSCRIPT

Solving this system of equations with respect to Lm, Lo, and Lg gives the equilibrium supply

of loans by each type of banks as follows:

p{2b2− (1 + K)c2}

b(apK− r)(ap− r)K − c

(b− cK)(2b2− (1 + K)c2) < 0. (19)This is negative under assumptions (7) and (8) It means that the GCB supplies larger amounts to afirm whose main-bank relationship is weak and supplies less to a firm whose main-bank relationship

is strong in response to the increase in the loan demand of the firm This is because the maximizing government bank takes into account the fact that a unit of loans from the main bankgenerates more benefits for the firm than does a unit of loans from the other banks, including thegovernment bank, because of the additional benefit from relationship banking by the main bank.This effect is captured by the first term in the LHS of the FOC for the GCB (14) As K gets largerand the loans from the main bank Lmget larger due to the increase in demand, the marginal socialwelfare of a unit of loans by the GCB decreases Thus, the GCB is less willing to provide a loan tothose with higher K, i.e., those with a strong relationship with their main bank

welfare-5.2.2 Profit-maximizing GCB

Now we consider the case where the GCB behaves in the same way as the non-main bank as aCournot competitor The GCB solves the same maximization problem as that of the non-main

Namely, the GCB does not adjust its supply according to the intensity of the main-bank relationship

of the borrowing firm

The following proposition is a summary of the results in this section

Proposition (Welfare Maximizing vs Profit Maximizing) In response to a demand increase, awelfare-maximizing government-controlled bank (GCB) increases lending more for a firm with aweaker relationship with its main bank Lending by a profit-maximizing GCB is independent of thestrength of the relationship between a firm and its main bank

This proposition suggests that we can identify whether the government bank is trying to imize the social welfare by examining the negative correlation between the supply of GCB lendingand the strength of the main-bank relationship of the borrower under a surge of loan demand likethat in the Japanese loan market in the crisis period from September 2008 to August 2010, asmentioned in Section 3.1 Thus, the hypothesis to be tested is as follows

6.1 Measure for the strength of the bank-firm relationship

To test the above proposition, we need a good proxy measure for the strength of the main-bankrelationship

Our primary proxy measure is a dummy variable to indicate a main bank is a large bank,

M B large Many existing studies suggest that the information on whether the main bank is a largebank can work as a proxy for the weakness of a main-bank relationship The theory suggests that

a large bank with a more centralized lending-decision mechanism is not competent in utilizing thesoft information that is required for relationship banking (e.g., Stein, 2002), and several empiricalstudies show supportive evidence of this (e.g., Cole et al., 2004; Berger et al., 2005; Uchida et al.,2008; Ogura and Uchida, 2014) Typical large banks in Japan are city banks and trust banks,which have a nationwide branch network and operate nationwide or internationally The weakness

of the relationships of large banks with SMEs is also consistent with the significantly lower bank share when the main bank is a large bank in Figure 2 and with the higher probability of amain-bank switch at large banks in Table 5

main-We also use the measures, which is more explicit and focused The first ones are the loan shareand the deposit share of the main bank as of 2007, M B loan share and M B deposit share Thehigher values of them indicate the stronger main-bank relationship The second one is the dummyvariable, which equals one if a firm switched main banks in the pre-crisis period or zero otherwise,

M B switch It indicates a stronger relationship if this variable equals zero The last one is themaximum number of lenders except for GCBs in the pre-crisis period, #lenders The larger value

of it indicates a weaker main-bank relationship

ACCEPTED MANUSCRIPT

6.2 Identification strategy

Our test strategy is to look at the difference in differences (DID) To put it more precisely, wetested the statistical significance of the difference between the response of GCB lending to firmswith a strong main bank relationship and to those without one after the loan demand surge in the2008-09 financial crisis We argue that the loan demand surge in the crisis period was an exogenousshock to the Japanese credit market, given that it was mainly propagated from the USA and the

EU to Japan through the exporting sector, as discussed in Section 3.1

To conduct this DID analysis, we use the three-way panel data of the annual change in theamount of outstanding loans from each bank at the end of each accounting period of each firm,which was normalized by the total asset of the firm in the previous year (∆loan/asset) A notablebenefit of estimating the model with the three-way panel data is that we can introduce the firm×yearcross fixed effect to control for the heterogeneous magnitude of a loan demand shock and othertime-varying unobservable individual firm factors, such as profitability, growth opportunity, andcreditworthiness, since we have multiple observations for each firm-year cell Thanks to this crossfixed effect, we can eliminate the endogeneity problem due to the potential correlation betweenthe choice of a main bank and the lending attitude of a main bank through an time-varying andunobservable firm characteristics For example, an under-performing firm might be willing tocontact an under-performing bank with an inferior monitoring ability This correlation resultingfrom an unobservable firm factor makes difficult to tell apart the effect of the bank-firm relationshipfrom the firm factor as pointed out by Gan (2007) The cross fixed effect enables us to control forthe latter factor more sharply and to extract the former effect In our context, the elimination ofthe time-varying unobservable firm factor by the cross fixed effect minimizes the possibility thatthe difference in the lending behaviors of GCBs and large banks are driven by the unobservablefirm factors Likewise, we can introduce the lender×year fixed effect to control for the unobservabletime-varying characteristics of each lender, such as the financial soundness of each lender

The three-way panel data is obtained by transforming the original dataset described in Section

2 from the firm-year level panel to the firm-year-lender level panel data After this transformation,

we obtain 900,635 firm-year-lender observations for which ∆loan/asset is available We drop the

ACCEPTED MANUSCRIPT

entire observation of a firm whose ∆loan/asset is in the top 1% or the bottom 1% in any of theyears from 2007 to 2011 (69,154 observations) We also drop 112,369 observations which are theonly observation within a firm-year cell and cannot control for the firm-year fixed effect Theresulting 831,481 observations are our baseline sample for the three-way panel regression.9

As noted in the data description, our dataset includes the amount of outstanding loans from abank to a firm in each year for the largest four lenders and the SME Unit of JFC If the number oflenders other than the SME Unit of JFC exceeds four, the outstanding loans of the fourth largestlender and other smaller lenders including the unknown lenders, are summed up and classified asloans from miscellaneous “Other institutions.” The composition of the types of lenders in each year

is listed in Table 8 Regional banks, including cooperative banks, account for the largest part ofthe dataset, about 44% GCBs for SMEs, including the SME Unit and the Micro Corporation andIndividual Unit of JFC, and the Shoko Chukin Bank account for the next largest part, about 35%.Large banks account for about 19% The class “Other institutions,” which is the mixture of thefourth largest and smaller lenders, and the other GCBs, including the Development Bank of Japan,account for a very small part of our observations

Table 9 shows the descriptive statistics of the annual change in loan outstanding from a bank to

a firm in a year, which is normalized by the total asset of the firm in the previous year, ∆loan/asset.The mean and median of ∆loan/asset are around zero Figure 4 is the plot of the sample mean of

∆loan/asset by each class of lenders Clearly, the loans by GCBs to SMEs kept increasing from

2009 to 2011 The increase in GCB loans to SMEs increased more precipitously when the mainbank of a firm was a large bank than otherwise The figure also shows that the loan growth of largebanks is always lower than that of regional banks

9 We adjust ∆loan/asset right after each bank merger in the following way (1,291 observations) We treat the value for each acquired bank, which loses its official bank ID (Kin’yu Kikan code) upon merger, as missing after the merger For each acquiring bank, which maintains its official bank ID after merger, we calculate ∆loan/asset on the consolidated basis, i.e., we set the numerator equals to the difference between the loan from the post-merger bank and the sum of the loans of pre-merger banks.

ACCEPTED MANUSCRIPT

We estimate the following firm×year fixed-effect model by using the three-way panel data

∆loan/assetijt = β0+ (β1+ β2· crisist+ β3· post-crisist)· GCB for SMEjt

+ (β4+ β5· crisist+ β6· post-crisist)· GCB for SMEjt· MB largeit+ (β7+ β8· crisist+ β9· post-crisist)· large bankjt+ δ0Xijt+ ρit+ ijt, (25)where i is the index of a firm, j is the index of a bank, and t (= 2007,· · · , 2011) is the year βs and

a column vector δ are the coefficients to be estimated Xijt is a column vector of control variables

ρit is the cross fixed effect of firm i and year t ijt is the error term The definitions of the othervariables are listed in Table 2

The most important coefficients for our hypothesis test is β5 If Japanese GCBs behave as awelfare maximizer rather than a profit maximizer, they should increase their lending more for firmswhose main-bank relationship is weaker, typically firms whose main bank is a large bank, i.e., β5

is positive and significant We also tested the hypothesis by replacing M B large with alternativemeasures of the strength of a main-bank relationship; i.e., a dummy variable to indicate whether

a firm has switched main banks in the pre-crisis period, M B switch; the pre-crisis loan share ofthe main bank, M B loan share; the pre-crisis deposit share of the main bank, M B deposit share;and the number of non-GCB lenders in the pre-crisis period, #lenders We need to note that theinterpretation of the signs of coefficients is opposite for M B loan share and M B deposit shares,since these variables are increasing in the strength of the relationship as opposed to the othermeasures

The theoretical model suggests that the response to a demand surge differs between a mainbank and a non-main bank, and this difference depends on the strength of the relationship Thus,the most important control variable is a dummy variable to indicate whether a lender is a mainbank, main bank, and its interaction with a dummy variable to indicate whether a lender is alarge bank, large bank The relationship with non-main banks, which are supposed to be weakerthan that with main banks, also can affect the lending by GCBs, although it is not explicitlymodeled in the theoretical model We control for this effect by introducing the interaction terms

of GCB f or SM E, N M B large, which is a dummy indicating whether a firm borrowed from anon-main large bank in the pre-crisis period, and the crisis-phase dummies Dummies of other

The coefficient of GCB f or SM E× crisis × MB large (the row of GCB for SME × crisis ×relation in Column (1)) is positive and significant at a 1% significance level, i.e., GCB loans inthe crisis increased more for firms whose main bank is a large bank The coefficient is 0.491 Sincethe mean of ∆loan/asset for regional banks in 2009 is 0.44 (Figure 4), the impact is economicallysignificant This result supports that GCBs for SMEs behaved as welfare-miximizers rather thanprofit-maximizers This effect is still significant in the post-crisis period although the magnitude

is smaller than that in the crisis period This extended effect is mainly because GCBs providelong-term loans only by the regulation

The coefficient of GCB f or SM E× crisis × NMB large is also positive and significant Thisresult means that GCB loans increased more for firms who have borrowed from a non-main largebank in the pre-crisis period This result is also consistent with welfare-maximizing GCBs since theywill increase loans to firms who receive less differentiated services from existing lenders includingnon-main banks

To clarify the change in the lending behavior of each type of main banks and GCBs, we marize the marginal effects in Table 11 and Figure 6, which are calculated from the estimates inColumn (1), Table 10 The estimated marginal effects listed in Table 11 indicate the mean change

sum-in ∆loan/asset for each type of banks relative to the mean of regional banks The first row showsthat GCBs for SMEs reduced its lending in the pre-crisis period relative to regional banks, but theyincreased in the crisis and post-crisis periods for firms whose main bank is a large bank (Columni) while they did not for firms whose main bank is a regional bank or cooperative bank (Column

ACCEPTED MANUSCRIPT

ii), which supposedly maintains stronger relationship with SMEs The second row shows that largemain banks keep reducing their lending (Column iii), whereas regional main banks increased itslending in the crisis period (Column iv) The difference between them is statistically significant(Column v) These points are consistent with the lending behavior of a welfare-maximizing GCB,which increases lending more to firms whose main bank relationship is weaker and less likely toobtain additional loans from the main bank

Columns (2)-(5) show the estimates with alternative measures of main-bank relationships Allthese results are consistent with welfare-maximizing GCBs rather than profit-maximizing Columns(2) and (3) are the results when we replace M B large with the loan or deposit share of the mainbank, M B loan share or M B deposit share Here, we posit that a main bank with a highershare have stronger relationship with a firm GCBs increased lending less for firms whose mainbank loan/deposit share is higher and has a stronger relationship with a main bank This result

is consistent with the welfare-maximization by GCBs again The coefficients of the interactionterms of the main bank dummy (the latter half of the table) indicate a counter-intuitive result thatthose with stronger relationship obtain smaller amount of loans from their main bank in the crisisperiod This is probably because a part of the effect of the main bank share is captured by thelarge-bank dummy in the same way as in the baseline regression in Column (1), which is negativeand significant in the crisis period, since the main bank share negatively correlated with the largebank dummy as shown in Table 2

Columns (4) is the result when we use the dummy indicating whether a firm switched mainbanks in the pre-crisis period as a measure of the strength of the main-bank relationship We assumethat those has switched main banks should have weaker relationship with their main banks Thecoefficient of GCB f or SM E× crisis × MB switch (the row of GCB for SME × crisis × relation

in Column (4)) is positive and significant in the same magnitude as in Column (1) This is alsoconsistent with welfare-maximizing GCBs Among the control variables, main bank× MB switchhas a highly positive coefficient This is because the loan growth of a main bank has to be higherfor switchers in the pre-crisis period due to the definition of M B switch, i.e., a switch to a newmain bank implies the increase of loans from the new main bank

Column (5) is the result when we use the number of lenders except for GCBs in the pre-crisis