Financial Inclusion And Bank Profitability- Evidence From A Developed Market2021.Pdf

Financial inclusion and bank profitability Evidence from a developed market Global Finance Journal xxx (xxxx) xxx Please cite this article as Vijay Kumar, Global Finance Journal, https //doi org/10 10[.]

Available online 21 January 2021

1044-0283/© 2021 Elsevier Inc All rights reserved

Financial inclusion and bank profitability: Evidence from a

developed market

Vijay Kumara,*, Sujani Thrikawalab, Sanjeev Acharyac

aSenior Lecturer in Finance, School of Accounting, Finance and Economics, Waikato Management School, The University of Waikato, Private Bag

3105, Hamilton 3240, New Zealand

bSenior Academic Staff Member, Centre for Business and Enterprise, Waikato Institute of Technology (WINTEC), Private Bag 3036, Waikato Mail

Centre, Hamilton 3240, New Zealand

cFaculty of New Media Arts and Business, Southern Institute of Technology, Invercargill, New Zealand

A R T I C L E I N F O

Keywords:

Financial inclusion

Bank profitability

ATMs

Japanese banks

A B S T R A C T Previous literature supports the view that financial inclusion leads to economic growth and helps alleviate poverty; however, it is still unclear whether financial inclusion increases bank profit-ability Using a sample of 122 Japanese banks from 2004 to 2018, we investigate this question

We find that financial inclusion is important even in a developed economy; branch contraction reduces the profitability of Japanese banks, although the numbers of loan accounts and auto-mated teller machines (ATMs) do not affect bank profitability Among bank-specific variables, cost management, credit risk management, and bank size are the key drivers of profitability

1 Introduction

Financial inclusion refers to the access to and use of a range of financial products and services by all of a society’s adult members at

a price that is affordable even to deprived and lower income groups (Demirguc-Kunt, Klapper, Singer, Ansar, & Hess, 2018) Since this concept was introduced, in 2005, it has received significant attention from researchers and policymakers Making financial services such as savings, loans, insurance, and payment systems accessible to all sectors of the population nurtures their financial autonomy and amplifies the country’s economic growth (Lal, 2017) It allows individuals and firms to invest in education, save for retirement, take advantage of business opportunities, and insure against risks (Demirgüç-Kunt, 2008) It also enhances the efficiency and accessibility

of financial services in a secure, convenient, safe and cost-effective manner (Ikram & Lohdi, 2015) Financial inclusion is recognised as

a continuous process of improving the quality, quantity and efficiency of financial intermediary services (Babajide, Adegboye, & Omankhanlen, 2015)

It is evident that financial inclusion plays an important role in economic development and the stability of the financial system When a country is financially inclusive, its economic activities rely more on banking transactions and it has greater financial sus-tainability and more effective monetary policy (Mehrotra & Yetman, 2015) Most empirical studies of financial inclusion have focused mainly on developing economies (Andrianaivo & Kpodar, 2012; Kim, Yu, & Hassan, 2018; Raman, 2012), but financial inclusion is also essential for developed economies Even in developed markets with sound financial systems and advanced technologies, closure of banks and post offices and bank indebtedness have helped exclude some portions of populations from obtaining financial services—for example, people living in remote areas (Godinho & Singh, 2013) In some developed nations, one in five adults does not have a bank

* Corresponding author

E-mail addresses: vijay.kumar@waikato.ac.nz (V Kumar), Sujani.thrikawala@wintec.ac.nz (S Thrikawala)

Contents lists available at ScienceDirect Global Finance Journal journal homepage: www.elsevier.com/locate/gfj

https://doi.org/10.1016/j.gfj.2021.100609

Received 5 June 2020; Received in revised form 16 December 2020; Accepted 19 December 2020

account or other access to the formal financial sector (Demirgüç-Kunt & Klapper, 2012)

Japan has a well-developed financial system compared to many other Asian countries Sarma (2016), using the Index of Financial Inclusion (IFI) for the period 2003–2013, found that Japan, Malaysia, the Republic of Korea, and Brunei Darussalam achieved high levels of financial inclusion Fung´aˇcov´a and Weill (2016) compared the level of financial inclusion in Japan with the world average and found that Japan has a high level of financial inclusion in terms of formal accounts and savings In terms of account ownership, 98% of Japanese adults have a bank account In addition, there is no rich vs poor gap in account ownership in Japan (Demirguç-Kunt, Klapper, Singer, Ansar, & Hess, 2018) In comparison, in the United States, among the poorest 20% of households, only 79% of adults own accounts And it remains unknown whether Japan’s high level of financial inclusion translates into bank profitability and financial stability

Most studies of financial inclusion have focused on defining it (Hunter, dela Cruz, & Dole, 2016; Patwardhan, 2018; Raddatz,

2003), identifying its determinants (Allen, Demirguc-Kunt, Klapper, & Martinez Peria, 2016; Fung´aˇcov´a & Weill, 2016), and measuring it (Demirgüç-Kunt et al., 2018) Only a few studies have investigated its impact on bank profitability Yet a profitable banking sector is necessary for economic development (Athanasoglou, Brissimis, & Delis, 2008) and financial stability (Klein & Weill,

2017) Banks foster economic growth (Levine & Zervos, 1998) by funding productive projects (Levine, Loayza, & Beck, 2000) Several bank-specific variables, such as bank capital, nonperforming loans, bank size, liquidity, cost management, and bank efficiency, are known to affect the profitability; however, the impact of financial inclusion of the profitability of banks is still unclear There is limited research in this area, especially for Japan Kondo’s (2010) study is not comprehensive; it used only one measure of financial inclu-sion—the number of automated teller machines (ATMs)—and employed a very small dataset for a period of five years from 2000 to

2004 In addition, Kondo’s study does not address issues related to endogeneity In contrast, the present study uses three measures of financial inclusion (number of loan accounts, number of ATMs, and number of commercial bank branches) and an index combining five variables (deposit accounts with commercial banks per 100,000 adults, loan accounts with commercial banks per 1000 adults, number of life insurance policies per 1000 adults, number of ATMs per 100,000 adults, and number of commercial bank branches per 100,000 adults) created using principal component analysis (PCA) We also analyze a large dataset, with 1604 observations from 122 banks over a period of 15 years from 2004 to 2018 Finally, we use robust methods (system GMM estimation and fixed effect models) to address the persistence and dynamic nature of bank profitability (see Athanasoglou et al., 2008; Delis & Kouretas, 2011) while controlling simultaneity and unobserved heterogeneity across banks

The remainder of the paper is structured as follows Section 2 reviews the existing literature Section 3 focuses on data and methods Section 4 presents and discusses the results Section 5 discusses the robustness of results Section 6 provides conclusions

2 Literature review and hypothesis development

A number of studies have investigated the determinants of bank profitability using the structure-conduct-performance (SCP) hy-pothesis and the efficient-structure (ES) hyhy-pothesis (Demirgüç-Kunt & Huizinga, 1999; Dietrich & Wanzenried, 2011; Iannotta, Nocera, & Sironi, 2007; Micco, Panizza, & Yanez, 2007; Molyneux & Thornton, 1992; Short, 1979; To & Tripe, 2002) The SCP hy-pothesis suggests that firms with large market power are more profitable than their counterparts The ES hyhy-pothesis suggests that efficient firms capture a large market share through comparative advantage, which results in an increase in market concentration and higher profitability (Peltzman, 1977) Bank-specific variables that have been found to significantly affect bank profitability include overheads, capital ratios, liquidity, bank size, and bad loans Athanasoglou et al (2008) suggested that capital ratio and overheads significantly affect the profitability of banks in Greece, while Nguyen (2018) suggested that asset diversification and income diver-sification are key drivers of bank efficiency in ASEAN countries Mirzaei, Moore, and Liu (2013) investigated the factors influencing the profitability of banks in emerging and developed markets They found that bank size, overheads, and bank loans are the key de-terminants, and they suggest that the impacts of some of the variables differ between emerging and advanced economies Pasiouras and Kosmidou (2007) suggested that capital ratio, cost-to-income ratio, and bank size are the factors influencing bank profitability in the European Union Demirgüç-Kunt and Huizinga (1999), using data from 80 countries, found that capital ratio and bank liquidity are the major factors behind bank profits Similarly, Bitar, Pukthuanthong, and Walker (2019) suggested that capital and liquidity ratios are the key drivers of bank profitability in Arab countries Kumar, Acharya, and Ho (2020) suggest that capital adequacy ratio, nonperforming loans, and cost efficiency are the major factors influencing bank profitability in New Zealand

Only a few studies have focused on the relationship between financial inclusion and bank profitability The Global Partnership for Financial Inclusion (GPFI) 2016 report1 on G20 financial inclusion indicators suggested that financial inclusion has three dimensions: use of financial services, access to financial services, and quality of products and service delivery Indicators of the use of financial services include the percentage of adults having a bank account and the percentage of adults having outstanding loans Indicators of access to financial services include the number of branches and the number of ATMs per 100,000 adults Quality indicators include the use of savings for emergency funding and the percentage of SMEs required to provide collateral on their bank loans

Different researchers have used different measures of financial inclusion Kondo (2010) used the number of ATMs and suggested that the number of ATMs does not affect the profitability of banks in Japan On the other hand, Holden and El-Bannany (2004) revealed

a positive relationship between the number of ATMs and the profitability of banks in UK Shihadeh and Liu (2019) investigated the impact of financial inclusion on banks’ risk and performance in 189 countries, using the number of branches as a measure of financial

1 https://www.gpfi.org/news/g20-financial-inclusion-indicators

inclusion Their results suggest that an increase in branch networks leads to an increase in bank profitability Shihadeh, Hannon, Guan,

Ul Haq, and Wang (2018) investigated the relationship between financial inclusion and bank performance in Jordan, and found that number of ATMs and number of credit cards increased bank profits, while number of ATM services and SME deposits decreased profits There are also mixed views on how financial inclusion may influence bank profitability As financial inclusion has the potential to alleviate poverty, it also has the potential to increase bank profits Extending services to a larger pool of customers can increase de-posits and loans, and hence profitability Han and Melecky (2013) suggested that an increase in customer deposits mitigates the risk of deposit withdrawals when banks are under financial stress; and Boot and Schmeits (2000) suggested that financial inclusion helps banks diversify and reduce risk On the other hand, providing financial services to individuals and small businesses may increase transaction costs and other overhead costs Investing in resources to increase financial inclusion requires capital expenditure, which may reduce bank profitability in the short term (Shihadeh & Liu, 2019) Further, lending to individuals and small businesses is risky (Burgess, Pande, & Wong, 2005)

Han and Melecky (2013) suggested that banks with higher loans are more vulnerable to credit risk Credit risk is a major risk for banks as it affects the quality of loans, and banks with higher credit risks are likely to have more nonperforming loans Nonperforming loans erode bank profitability (Athanasoglou et al., 2008; Tan, Floros, & Anchor, 2017) Shihadeh and Liu (2019) found a positive relationship between financial inclusion and nonperforming loans in their study covering 189 countries However, for China, Chen, Feng, and Wang (2018) argued that increased financial inclusion reduces nonperforming loans; their results suggest that financial inclusion increases the number of customers and diversifies risk It is important to note that these two studies used different measures

of financial inclusion: Shihadeh and Liu used number of branches, while Chen and colleagues created an index using ten dimensions of financial inclusion The results clearly show that different measures of financial inclusion produce different results

In sum, both theoretical and empirical research suggest mixed impacts of financial inclusion on bank profitability, and there is little evidence concerning developed markets For Japan, we adopt the following hypotheses:

Hypothesis 1a There is a positive relationship between the number of loan accounts and the profitability of banks

Hypothesis 1b There is a positive relationship between the number of ATMs and the profitability of banks

Hypothesis 1c There is a positive relationship between the number of bank branches and the profitability of banks

3 Data and methods

3.1 Data

The panel dataset for this study contains 1604 bank-year observations of 122 Japanese banks for the period 2004–2018 We used the number of loan accounts with commercial banks per 1000 adults as our financial use variable and the number of ATMs per 100,000 adults and number of commercial bank branches per 100,000 adults as financial access variables For an additional analysis we developed a composite financial inclusion index (FIN_INDEX) using principal component analysis (see Le, Chuc, & Taghizadeh-Hesary,

2019).2 The index combines five variables extracted from the International Monetary Fund (IMF) database: deposit accounts with commercial banks per 100,000 adults, loan accounts with commercial banks per 1000 adults, number of life insurance policies per

1000 adults, number of ATMs per 100,000 adults, and number of commercial bank branches per 100,000 adults The data on indi-vidual banks come from the BankFocus database, and other country-specific data such as GDP, interest rates, and inflation come from the World Bank database Some observations in BankFocus provide bank statements that cover only a part of a year, such as three months or six months; we considered only banks that provided data for a 12-month financial year We winsorized all data at the 1st and 99th percentiles to remove the impact of outliers

3.2 Dependent and independent variables

3.2.1 Dependent variables

In line with the literature (see Athanasoglou et al., 2008; Mirzaei et al., 2013; Shihadeh & Liu, 2019), we use return on assets (ROA) and return on equity (ROE) as measures of profitability ROA is calculated by dividing profit before tax by total assets, and ROE by dividing profit before tax by total equity Both variables are expressed as percentages

3.2.2 Independent variables

3.2.2.1 Financial inclusion variables Researchers have used a number of variables to measure financial inclusion, but for some of

these variables only triennial data are available We used three variables for which annual data are available: the number of loan accounts with commercial banks per 1000 adults (FIN_LOAN), our variable for use of financial services; and the numbers of ATMs per 100,000 adults (FIN_ATM) and of commercial bank branches per 100,000 adults (FIN_BRANCH), our variables for access to financial services For additional analyses we used the composite index explained above in section 3.1

2 The authors would like thank an anonymous reviewer for this suggestion

3.2.2.2 Control variables We used a number of bank-specific and macroeconomic control variables The bank-specific variables are

cost-to-income ratio, capital adequacy ratio, nonperforming loan ratio, bank size, and loan-to-deposit ratio; the macroeconomic variables are inflation rate, interest rate, and GDP growth rate

Cost-to-income ratio (COST) is often used as a measure of operating efficiency A number of studies have used COST as a deter-minant of bank profitability (Athanasoglou et al., 2008; Dietrich & Wanzenried, 2011; Mirzaei et al., 2013); most studies suggest that it has a negative impact

Capital adequacy ratio (CAR) is often used as a measure of credit risk management CAR is the ratio of tier 1 and tier 2 capital to risk-weighted assets and is expressed as a percentage Though research has found that CAR affects bank profitability, the direction of the relationship is uncertain: Athanasoglou et al (2008) found a positive impact on profitability measured with ROA, but Dietrich and Wanzenried (2014) suggested a negative effect on profitability measured with ROE Accordingly, we expect opposite signs for ROA and ROE

Nonperforming loan ratio (NPLR) is also often used as a measure of credit risk management, on the assumption that banks with high nonperforming loans manage credit risk poorly and so are less profitable than their counterparts Athanasoglou et al (2008), Dietrich and Wanzenried (2014), and Tan et al (2017) all found that NPLR reduced bank profitability

Total Assets (logSize): We measured bank size with the natural log of the bank’s total assets There are mixed findings on the impact

of bank size on bank profitability: Smirlock (1985) suggested that size has a positive impact, Tan and Floros (2012a) that it has a negative impact, and Shehzad, De Haan, and Scholtens (2013) that there is no relationship

Loan-to-deposit ratio (LDR) is often used to measure liquidity Again, there are mixed views: Tan and Floros (2012a) argued that liquid banks are less profitable, Heffernan and Fu (2010) that they are more profitable

Inflation rate (INF): Most studies suggest that banks perform better during inflationary periods For example, Athanasoglou et al (2008) and Tan (2016) found a positive relationship between inflation rate and bank profitability; however, Mirzaei et al (2013) suggest that banks do worse during inflationary periods

Interest rate (INT) is a monetary policy tool used by central banks to expand or contract economic activities The literature suggests that banks are more profitable when interest rates are high (Bourke, 1989; Short, 1979)

Gross domestic product (GDP) growth: The literature suggests that business growth and demand for loans increase during boom periods Both Athanasoglou et al (2008) and Mirzaei et al (2013) found that GDP growth increased bank profits

Table 1 summarizes the variables and their expected impacts on bank profitability, as determined from the literature

3.3 Descriptive statistics

The descriptive statistics in Table 2 provide the mean, standard deviation, and minimum and maximum values for each variable The dependent variables are ROA and ROE; the others are independent variables, broadly classified as financial inclusion variables and control variables Japanese banks vary in profitability, with ROA ranging between − 16 and 6.03 and ROE between − 21 and 29.6 Albertazzi and Gambacorta (2009) found similar variation of ROE in their study of European and Anglo-Saxon countries’ banks from

1981 to 2003

Loan accounts average 187.7 for every 1000 adults, with a minimum of 170.5 and a maximum of 210.6 ATMs average 127 for every 100,000 adults, with less variability Similarly, there are 34 bank branches for every 100,000 adults, with a standard deviation of only 0.21 FIN_INDEX has a minimum value of − 2.32 and a maximum of 3.82, with a standard deviation of 1.63

Table 1

Definition, notation and expected effect of the variables

Dependent Variables

Independent Variables

Financial inclusion variables

Number of loan accounts FIN_LOAN Number of loan accounts with commercial banks per 1000 adults +

Financial inclusion index FIN_INDEX Composite index of five financial inclusion variables +

Control variables

Notes: “+” refers to a positive relationship between the dependent variable and the explanatory variable “-” indicates a negative relationship between the dependent variable and explanatory variable “+/− ” shows either a positive or a negative relationship between the dependent variable and the explanatory variable

The cost-to-income ratio (COST) and GDP both vary substantially during the study period The average capital adequacy ratio is 11.4%, with a standard deviation of 3.63 In contrast, the log of total assets has an average of 5.3 with a standard deviation of 1.45 The mean loan-to-deposit ratio is 70.5%, with a minimum of 22.1% and a maximum of 90% Inflation varies significantly, with a minimum

of − 1.35% and a maximum of 2.75% The mean interest rate is 1.85%, with a standard deviation of 0.88

3.4 Model specification

We estimated the following two models:

ROA it=β0+β1FIN LOAN t+β2FIN ATM t+β3FIN BRANCH t+β4COST it+β5CAR it+β6NPLR it+β7logSIZE it+β8LDR it+β9INF t

+β10INT t+β11GDP t+u it

(1)

ROE it=β0+β1FIN LOAN t+β2FIN ATM t+β3FIN BRANCH t+β4COST it+β5CAR it+β6NPLR it+β7logSIZE it+β8LDR it+β9INF t

+β10INT t+β11GDP t+u it

(2)

where the subscript i refers to the bank and t refers to the time period; ROA and ROE are the dependent variables and refer to the

profitability of the bank; FIN_LOAN, FIN_ATM, and FIN_BRANCH are the independent variables and refer to the financial inclusion of the country; COST, CAR, NPLR, logSize, and LDR are bank-specific control variables; INF, INT, and GDP are country-specific

mac-roeconomic control variables; and u it is the error term We also performed additional analysis by using the composite financial in-clusion index described in section 3.1

3.4.1 Data analysis techniques

3.4.1.1 Ordinary least-square regression (OLS) analysis Studies using ordinary least-square regression (OLS) analysis to find the

relationship between financial inclusion and bank profitability have applied either fixed effects or random effects to deal with simultaneous causality and unobserved heterogeneity across banks The fixed-effect procedure addresses estimation issues related to unobserved heterogeneity (individual firm characteristics) across firms, and it is appropriate when the unobserved heterogeneity is constant over time for an individual firm (Schultz, Tan, & Walsh, 2010) Unlike the fixed-effect model, the random-effect model re-duces variability within the sample by partially pooling the data, and explores the differences in error terms across individual firms and time periods

For studies using panel data, the most widely used econometrics test for choosing between fixed-effect and random-effect models is the Hausman test, as it tests for orthogonality of common effects and regressors (Greene, 2012; Hausman, 1978) It examines whether the individual effects (unique errors) are correlated with other regressors in the model Given the results of the Hausman test, we used fixed-effect methods We also used pooled OLS to determine whether or not the results obtained with these methods were consistent

3.4.1.2 Generalized method of moments (GMM) estimation In general, panel regression analyses are subject to endogeneity problems,

Table 2

Descriptive statistics of variables over the period 2004–2018 (number of observations: 1604)

Dependent Variables

Independent Variables

Financial Inclusion Variables

Control Variables

Notes: ROA = Return on assets; ROE = Return on equity; FIN_LOAN = Number of loan accounts; FIN_ATM = Number of ATMs; FIN_BRANCH = Number of bank branches; FIN_INDEX = Financial inclusion index; COST = Cost-to-income ratio; CAR = Capital adequacy ratio; NPLR = Nonper-forming loan ratio; logSIZE = Total assets; LDR = Loan-to-deposit ratio; INF = Inflation rate; INT = Interest rate; and GDP = Gross domestic product

including dynamic endogeneity, simultaneity, and time-invariant unobserved heterogeneity across banks To mitigate the endogeneity problem in our banking dataset, we applied a generalized method of moments (GMM) estimation, which was initially used by Arellano and Bond (1991) This is the method most commonly used to investigate the determinants of bank profitability (Dietrich & Wan-zenried, 2011; Liu & Wilson, 2009) and is considered an appropriate estimation method to explore the dynamic nature of relationships (Flannery & Hankins, 2013)

This study follows the method proposed by Schultz et al (2010), working on the basis that “selected lags have the desirable instrumental variable properties of being correlated to the regressors, yet uncorrelated with contemporaneous errors” (p 149) However, in a GMM approach, a large number of generated instruments may be suspicious Therefore, we used the Hansen-J test or overidentification test to diagnose the validity of the instruments used in the system GMM estimator (Roodman, 2009) The null

hypothesis that the instruments are valid was not rejected, as the p-value was greater than 0.05, confirming that all the instruments

employed in the model were appropriate We used one-year lagged dependent variables to capture the influence of past performance and added explanatory variables in the right-hand side of the research model to capture the unobserved factors that interact with the link between financial inclusion and bank profitability Though Wintoki, Linck, and Netter (2012) proposed two-year lags of the dependent variable to capture the dynamic nature of a relationship, we found that a one-year lag was sufficient, as the coefficients for a two-year lag were not statistically significant at the 5% level

4 Empirical results and discussion

Table 3 shows the regression results for Eq 1, with ROA as the dependent variable, in Panel A, and the results for eq 2, with ROE as the dependent variable, in Panel B Each panel shows three sets of results obtained using three different methods In model 1, we used the OLS method In models 2 and 3, we applied the fixed-effect (FE) and GMM estimation methods, respectively

The results indicate that the number of branches (FIN_BRANCH) has a positive and significant impact on bank profitability in Japan This finding is consistent for both profitability measures (ROA and ROE) and for all the estimation methods (OLS, FE, and GMM), with the exception of model 2 in Panel A, where the coefficient is not statistically significant Hence, we accept Hypothesis 1a:

“There is a positive relationship between the number of bank branches and the profitability of banks.” Contraction of branch networks reduces the profitability of Japanese banks Our results are consistent with those of Chen et al (2018) and Shihadeh and Liu (2019), which suggest that an increase in the number of branches leads to an increase in the number of customers, which in turn increases deposits and loan portfolio and diversifies risk Boot and Schmeits (2000) also suggested that financial inclusion allows banks to diversify and reduces risk Berger, Leusner, and Mingo (1997) and Bernini and Brighi (2018) indicated that branch networks play an important role in increasing bank revenue Nguyen (2014) suggested that bank branches are important for underprivileged sectors of society, and closing branches reduces the loans granted to small firms

Before the global financial crisis, the number of bank branches in Japan was declining It then increased somewhat until 2015, since when it has been decreasing year by year.3 The decline is due to population shrinkage in the countryside,4 increasing bad debts, growing pressure to reduce cost, the increasing number of self-service branches,5 and a drastic shift of customers towards internet banking.6 However, because there is little empirical research, it is not yet clear whether reducing the number of branches leads to cost efficiency in Japan Harimaya and Kondo (2016) suggested that increasing the number of branches leads to cost inefficiency up to a certain point and after that point to cost efficiency Kondo (2015) suggested that Japanese regional banks that expand their network into other prefectures have higher lending income; accordingly, our results may indicate that new branches can help banks to increase profitability and diversify risk if those branches are opened in financially excluded areas or regions We performed Granger causality tests (see Table B1 in Appendix B) to explore whether the relationship between bank branches and bank profitability is bi-directional; however, our results suggest that bank branches cause bank profitability, not the reverse.7

The coefficients on the number of loan accounts (FIN_Loan) and the number of ATMs (FIN_ATM) are not statistically significant for either of the profitability measures (ROA and ROE) in any of the models, suggesting that these variables do not affect the profitability

of banks in Japan Therefore, we reject Hypothesis 1a: “There is a positive relationship between number of loan accounts and the profitability of banks,” and Hypothesis 1b: “There is a positive relationship between the number of ATMs and the profitability of

banks.” An increase in the number of loan accounts (FIN_LOAN) does not necessarily increase bank profitability because, among other

factors, transaction costs and overheads may offset the additional income from additional loan accounts Our results related to FIN_ATM are in line with the findings of Kondo (2010), who also found no relationship between the number of ATMs and return on assets for banks in Japan He argued that ATMs offer only a certain type of service and reduce customer waiting time

Among the bank-specific control variables, our results suggest that cost management, credit risk management, and bank size are the key factors underlying bank profitability There is strong evidence that banks with a high cost-to-income ratio (COST) are less prof-itable than their counterparts The coefficient of COST is negative and statistically significant for both profitability measures (ROA and ROE) in all models This finding is in line with our expectations and previous studies (Athanasoglou et al., 2008; Dietrich & Wan-zenried, 2011; Mirzaei et al., 2013)

3 https://fred.stlouisfed.org/series/DDAI02JPA643NWDB

4 https://asia.nikkei.com/Business/Companies/Japan-to-crack-down-on-struggling-regional-banks

5 https://www.reuters.com/article/us-japan-banks-results/japans-biggest-banks-revamp-close-branches-in-efficiency-drive-idUSKCN1IG1E5

6 https://www.reuters.com/article/mufg-strategy/japans-mufg-to-close-up-to-40-of-domestic-bank-branches-idUSL4N2D21B0

7 The authors would like to thank an anonymous reviewer for suggesting this possibility

We used the nonperforming loan ratio (NPLR) and capital adequacy ratio (CAR) to measure credit risk management There is weak evidence to suggest that CAR increases bank profitability The coefficient of CAR is positive and significant only in one model (model 1, Panel A), in which we used ROA as a measure of profitability and used the OLS estimation method There is ambiguity about the relationship between CAR and bank profitability Some studies have found a positive relationship (Bourke, 1989; Lee & Hsieh, 2013), others a negative one (Athanasoglou et al., 2008; Dietrich & Wanzenried, 2011) Dietrich and Wanzenried (2011) suggested that banks with more capital are considered financially stable It appears that more capital helps Japanese banks to attract low-cost deposits and increase profitability The coefficient of NPLR is negative and statistically significant for both profitability measures (ROA and ROE) For ROE the coefficient is significant in all three models, while for ROA it is significant only in model 1 A 1% increase in NPLR can reduce bank profitability by 0.873% (β = 0.0873 in model 2, Panel B) This result is consistent with our expectations and the previous literature It appears that some banks in Japan are compromising on lending quality The banks involved in aggressive lending practices are more vulnerable to credit risk and likely to have more nonperforming loans (Han & Melecky, 2013) This interpretation also gains some support from the positive correlation between loan-to-deposit ratio (LDR) and NPLR

There is strong evidence that larger bank size (logSIZE) reduces bank profitability, especially when profitability is measured using ROE In the literature, the results related to the impact of logSIZE on bank profitability are mixed Akhavein, Berger, and Humphrey (1997) argued that large banks are more profitable because of economies of scale and scope, while Tan and Floros (2012b) and Tan (2016) suggested that there is a negative relationship between logSIZE and bank profitability Our results are in line with the latter claim It appears that in Japan, small size allows banks to concentrate on profitable segments and lowers their agency costs Finally, our results suggest that the loan-to-deposit ratio does not affect the profitability of Japanese banks

Turning to the country-level control variables, we found that the inflation rate (INF) increases bank profitability This is not surprising; Demirgüç-Kunt and Huizinga (1999) and Athanasoglou et al (2008) also found a positive relationship between INF and

Table 3

Regression results for full sample

Notes: ROA = Return on assets; ROE = Return on equity; FIN_LOAN = Number of loan accounts; FIN_ATM = Number of ATMs; FIN_BRANCH = Number of bank branches; COST = Cost-to-income ratio; CAR = Capital adequacy ratio; NPLR = Nonperforming loan ratio; logSIZE = Total assets; LDR = Loan-to-deposit ratio; INF = Inflation rate; INT = Interest rate; and GDP = Gross domestic product t-values are given in brackets The Hansen-J over-identification test confirmed that all the instruments employed in the GMM model are appropriate The absence of second-order autocorrelation

shows that the estimates are consistent *** p < 0.01, ** p < 0.05, and * p < 0.1

bank profitability Perry (1992) suggested that if banks can predict the inflation rate, they tend to increase their lending rate and thereby their profitability It appears that banks in Japan are able to predict the inflation rate The coefficient of the interest rate (INT)

is insignificant As we expected and in line with previous studies (Athanasoglou et al., 2008; Mirzaei et al., 2013), we found a positive relationship between GDP growth and bank profitability

4.1 Additional analysis

4.1.1 Regression results with a composite financial inclusion index

We performed additional analysis employing the composite financial inclusion index described above in section 3.1 Table 4 shows the regression results in Panel A, with ROA as the dependent variable, and Panel B, with ROE as the dependent variable Each panel shows three sets of results, obtained using the OLS, fixed-effect (FE), and GMM estimation methods, respectively The results are largely consistent with our results in Table 3 They show that FIN_INDEX increases both ROA and ROE, in line with the arguments of Boot and Schmeits (2000) and Chen et al (2018) that financial inclusion allows banks to diversify, reduce risk, and increase profitability

4.1.2 Unit root tests

We performed an augmented Dickey-Fuller (ADF) test to determine whether or not the variables are stationary The results, re-ported in Table 5, suggest that, except FIN_LOAN, FIN_ATM, COST, CAR, logSIZE, and LDR, all the variables are stationary However, after the first lag, all variables become stationary, suggesting that all variables are integrated of order one I (1)

4.1.3 Cointegration test

We also performed a Kao test for cointegration The results are reported in Table 6 P-values are less than 0.01, which suggests that

all panels are not cointegrated

Table 4

Regression results with composite financial inclusion index (FIN_INDEX)

Notes: ROA = Return on assets; ROE = Return on equity; FIN_INDEX = Index of 5 financial inclusion variables; COST = Cost-to-income ratio; CAR = Capital adequacy ratio; NPLR = Nonperforming loan ratio; logSIZE = Total assets; LDR = Loan-to-deposit ratio; INF = Inflation rate; INT = Interest rate; and GDP = Gross domestic product t-values are given in brackets The Hansen-J over-identification test confirmed that all the instruments

employed in the GMM model are appropriate The absence of second-order autocorrelation shows that the estimates are consistent *** p < 0.01, ** p

< 0.05, and * p < 0.1

5 Robustness analysis

The Japanese banking sector has three mega banks—large banks with significant market share and more branches than other banks In order to determine whether our results still hold, we conducted robustness tests by dividing the sample into two groups: mega and other banks.8 Table 7 reports the results without mega banks, using the same estimation techniques (OLS, FE, and GMM) that we used for the full sample Again, Panel A shows the results for ROA, and Panel B shows the results for ROE Our results without mega banks are largely consistent with our results for the full sample,9 but some of the results are stronger when mega banks are excluded, perhaps because the mega banks have more significant investments outside Japan than other banks do.10

We also conducted tests to detect possible multicollinearity among the variables Table A1 in Appendix A shows the correlations between explanatory variables and the results of a VIF test It indicates the strength and the direction of the relationships between variables Multicollinearity is deemed to be a serious concern when the correlation coefficients between variables are above 0.80 (Gujarati & Porter, 2009); our results show that there is no multicollinearity among the variables The VIF results for all variables were less than 2.4, again confirming that there is no multicollinearity among the variables in this study

6 Conclusion and policy implications

Our results suggest that financial inclusion is important even for developed markets We found that the number of branches had a positive and significant impact on the profitability of banks, a result consistent with those of Chen et al (2018) and Shihadeh and Liu (2019) However, we did not find any significant impact on bank profits of the numbers of loan accounts and ATMs Among the control variables related to individual banks, our results suggest that cost management, credit risk management, and bank size are the key drivers behind bank profitability Cost-efficient banks, and banks with prudent credit policies and fewer nonperforming loans, are more profitable than their counterparts It appears that large banks in Japan do not take advantage of economies of scale Our results

Table 5

Unit root tests

Notes: ROA = Return on assets; ROE = Return on equity; FIN_LOAN = Number of loan accounts; FIN_ATM = Number of ATMs; FIN_BRANCH = Number of bank branches; COST = Cost-to-income ratio; CAR = Capital adequacy ratio; NPLR = Nonperforming loan ratio; logSIZE = Total assets;

LDR = Loan-to-deposit ratio; INF = Inflation rate; INT = Interest rate; and GDP = Gross domestic product t-values are given in brackets *** p < 0.01,

** p < 0.05, and * p < 0.1

Table 6

Cointegration test

***Significant at 1% level

8 We also conducted another robustness test by using another profitability measure, net interest margin (NIM), and the results were largely consistent with those for ROA and ROE Results are not reported but are available on request from the authors

9 We also produced results (not reported here) for mega banks only, but found no significant relationship between financial inclusion and bank profitability of mega banks

10 https://asia.nikkei.com/Business/Finance/Japanese-banks-maintain-global-lead-in-overseas-investments

suggest that small banks are more profitable than large banks Perhaps small size allows banks to focus on profitable segments Among the country-level control variables, we found that banks in Japan perform better during inflationary and high-growth periods The findings of this study have important implications for policymakers in Japan Yoshino and Taghizadeh-Hesary (2017) suggest that a vertical investment-saving curve has caused the Japanese economy to stagnate The economy has structural problems instead of temporary downturns Some of the major problems that cause this situation are an aging population, ineffective fiscal policy, overtight monetary policy, local government dependency on central government funds, misallocation of public and private investments, and Basel capital requirements Overtight monetary policy and strict Basel capital requirements reduce banks’ lending capacity and make it difficult for small and medium enterprises (SMEs) and startup businesses to obtain loans Policymakers in Japan need to focus on improving financial inclusion by developing policies that support lending to SMEs and startup businesses In addition, it is important for banks to encourage households to diversify their assets rather than depending on cash and deposits Further, banks and other financial institutions should reduce their asset management fees to encourage people to use banking services

This study does not investigate the impact of financial inclusion on bank risk One of the major risks for banks is credit risk Future research could be undertaken to confirm the link between financial inclusion and credit risk The findings could be applied to countries

or regions with similar settings

Funding information

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper

Table 7

Regression results without mega banks

Notes: ROA = Return on assets; ROE = Return on equity; FIN_INDEX = Index of 5 financial inclusion variables; COST = Cost-to-income ratio; CAR = Capital adequacy ratio; NPLR = Nonperforming loan ratio; logSIZE = Total assets; LDR = Loan-to-deposit ratio; INF = Inflation rate; INT = Interest rate; and GDP = Gross domestic product t-values are given in brackets The Hansen-J over-identification test confirmed that all the instruments

employed in the GMM model are appropriate The absence of second-order autocorrelation shows that the estimates are consistent *** p < 0.01, ** p

< 0.05, and * p < 0.1