1 Thesis MSc. Finance, June 2016
Bank business models, risk, return and national
regulation: An empirical analysis of European Union
banks
Rosa van de Beek a,*, supervised by Dr. M. Hernandez Tinoco a Faculty of Economics and Business, University of Groningen, The Netherlands
ARTICLE INFO JEL classifications: G01 G18 G21 G28 Keywords: Non-interest income Wholesale funding Diversification Business models Bank return Bank risk Regulation Financial crisis ABSTRACT
This paper conducts an empirical analysis on 414 listed and delisted European Union banks’ business models, return, risk and the influence of national regulation over a 2005-2015 period. We find that higher capital requirements decreases non-traditional banking. Some risk and return benefits occur from very small shares of non-interest income. High shares of non-non-interest income decrease return and increase insolvency risk significantly. Wholesale funding does not affect insolvency risk in general but decreases return. Short-term borrowings are positively associated with insolvency risk. Overall, non-traditional banking business models are not entirely efficient and can bring additional risk.
* Corresponding author. Student number: S2024462
2
1. Introduction
In 2008, the banking sector faced a global financial crisis with a massive negative impact on the global economy. The crisis unfolded when several significant banks became illiquid and distressed. Many banks have failed in appropriately managing the risks related to their activities and funding, even though these banks did meet existing capital requirements. The financial crisis has brought up a great discussion about non-traditional banking. Banks have been expanding into risky on-and-off balance sheet activities over the last decades, without adjusting their capital buffers to cover for the higher risk. The financial crisis has shown that the required capital buffers were not sufficient to overcome a period of stress caused by risk of non-traditional asset activities. Several studies point out that higher involvement in non-interest income generating activities result in higher risk (DeYoung and Roland, 2001; Lepetit et al., 2008; Demirgüç-Kunt and Huizinga, 2010). A less discussed determinant, but of upmost importance in explaining the magnitude of the financial crisis, has been the expansion into non-traditional funding structures. The financial crisis has revealed the risk of non-customer-deposit short-term funding. Money markets dried up and banks heavily dependent on wholesale funding suddenly became unable to fund their activities. The recent financial crisis has aroused new interest in the relationship between bank business models and risk.
Income and funding structures, as part of banks’ business models, bring challenges for both policy makers and banks in terms of stability, returns and risk. The spillover effect of the recent financial crisis resulting in an economic downturn proves the dependence of the economy on stability in the financial sector (Bernanke, 1983; Calomiris and Mason, 2003; Hall, 2010). Changes in regulation and supervision with the attempt to increase stability and limit future crises in the financial sector result in alterations within banks. Unfortunately, previous studies warn about changes in regulation of the banking sector that increase rather than mitigate risk-taking behaviour of banks (Calem and Rob, 1990; Leaven and Levine, 2009; Darwish et al., 2011). These studies illustrate the importance of appropriate regulation and the need for monitoring risk-taking behaviour of banks.
3 We will focus on the relationship between differences in national regulation and bank business models and on the influence of non-traditional banking on risk/return trade-offs.
This paper makes several contributions to the literature. This study uses a recent panel data set over the 2005-2015 period, which allows us to capture the most recent effects of changing regulation on bank business models and its relationship with risk. We increase accuracy of previous research by focusing solely on European Union banks. Estimated relationships and results of banking studies can be different for the worldwide, US or EU banking system (De Jong, 2007). To the best of our knowledge, we are the first to examine the influence of diversification of income and funding structures on return and insolvency risk in the European Union. As far as we are concerned, we are also the first to examine the most recent effect of national regulation of banking and shareholder protection on bank business models.
The contributions of this paper are important for European Union policy makers. Acknowledging and monitoring determinants of and developments in bank risk is essential to create and maintain a stable financial sector and limiting spillover shocks to the real economy. This research provides policymakers with better insights about banks’ business models, risk/return trade-offs and regulation. These insights can provide foundations for policymakers to improve monitoring and regulation of individual banks in the financial sector.
The purpose of this paper is as follows. First, we will discuss the developments in business models distinguishing between bank types using a data set of 414 European Union banks over the 2005-2015 period. Second, we estimate the relationship between non-interest income, diversification of earning assets, wholesale funding, return, and risk. Third, we empirically determine the impact of differences in national regulation on non-interest income shares, diversification of earning assets and wholesale funding shares. Fourth, we estimate the influence of non-traditional income and funding structures on return on assets (ROA) and insolvency risk of banks. Fifth, we estimate in-depth relationships of different sources of non-interest income and wholesale funding shares for better insights.
4 We see no correlation between non-interest income or diversification of earning assets and wholesale funding. Controlling for institutional and country specific differences we show that higher capitalized banks have higher shares of non-interest income activities and wholesale funding shares. Personnel costs are positively associated with non-interest income and negatively associated with wholesale funding shares. The level of minimum required capital has a negative relationship with both non-traditional income activities and funding structures. We use Ordinary Least Squares (OLS) regressions to show that wholesale funding shares slightly decrease return on assets, but have no significant relationship with insolvency risk. We find that the relationship between interest income and bank return and risk is non-linear. Small shares of non-interest income shares result in higher return on assets and lower insolvency risk, while high shares of non-interest income result in lower return on assets and lower insolvency risk. This relationship indicates that non-interest income is not the result of a risk/return trade-off, but rather an inefficient business model. In the robustness checks, we use alternative measures for non-traditional income activities, return, and risk. The alternative measure for non-traditional income activities shows a strong negative relationship with bank risk. Alternative measures for return and risk show that our estimations are robust. Additionally, we employ regressions with the Instrumental Variables (IV) methodology because OLS could estimate inconsistent results. The IV methodology is an econometric technique that deals with potential endogeneity issues which cannot be dealt with using OLS.
The remainder of the paper is structured as follows. Section 2 discusses the existing literature on non-traditional income activities and wholesale funding on bank return and risk. Section 3 discusses the methodology used for the empirical analysis. Section 4 describes and summarizes the data and analyses the main explanatory variables used in this study. Section 5 presents the results of the regressions on the influence of non-interest income and wholesale funding shares on bank return and risk. Section 6 concludes and discusses the estimated results.
2. Existing literature
5
2.1. Bank income structure
Traditionally, banks gain their income from interest on outstanding loans. However, over the past decades banks have extended their range of activities beyond lending into non-interest income generating activities. Subsection 2.1.1 relates to previous studies on the relationship between non-interest income and bank return. Subsection 2.1.2 discusses past research on non-interest income and risk.
2.1.1. Bank income structure and return
Previous studies are divergent in their scope, focus and findings on bank income structure and return. DeYoung and Roland (2001) observe statistically significant increases of return on assets caused by increasing fee-based income shares. The authors question the reliability of the impact on performance because a shift from interest income activities to fee based activities causes a decrease in bank assets. Lozano-Vivas and Pasiouras (2010) use a large multi-country sample to show that off-balance sheet items and non-interest income activities have a positive impact on cost efficiency of banks. The authors also find a positive effect of non-interest income activities on profit efficiency, which is even more significant under regulatory climates that restrict bank activities and enhance supervision and monitoring of banks.
Contradictory, Rajan et al. (2000) do not find a positive impact on profit efficiency and rather show that diversity in resources and opportunities lead to inefficient investment and lower profits, which the authors call discount diversification. DeYoung and Rice (2004) focus on the influence of non-interest income on return on equity and find that on average marginal increases in non-interest income is associated with lower risk-return trade-offs. Mercieca et al. (2007) identify negative effects of income diversification on average profitability and conclude that non-interest income activities do not improve operational performance of small European banks. Other studies actually find no clear relationship between diversification of income and bank returns. Stiroh (2004) finds no significant relationship between non-interest income shares and returns. Demirgüç-Kunt and Huizinga (2010) show results are contradictory and the authors are unable to identify a clear relationship between return on assets and non-interest income worldwide.
6
2.1.2. Bank income structure and risk-taking
Early studies examining the relationship between non-traditional banking activities and risk-taking are based on industry-level data and results are ambiguous. Rosen et al. (1988) analyse bank expansion into real estate activities and find that expanding into real estate provides small potential diversification of risk. However, this potential benefit is easily outweighed by the high level of real estate profit uncertainty. Boyd (1993) analyses bank mergers with non-banking firms and finds that bank holding companies mergers with insurance firms, could potentially decrease risk. However, mergers with securities or investment firms tend to increase risk of bank holding companies. Both studies find small potential diversification effects of bank risk, but the volatility of non-traditional returns from real estate and trading in comparison to interest income is evident.
More recent studies have examined the effects of activity diversification on bank risk based on firm-level data. Templeton and Severiens (1992) use market-based data to show that diversification into non-bank activities decreases risk of bank holding companies. Rogers and Sinkey (1999) confirm these results when they reveal that US commercial banks with higher shares of non-traditional activities tend to have smaller net interest margins and exhibit a lower level of risk. The authors argue that diverse banks that have better access to financial markets, mostly true for larger banks, exhibit a lower level of illiquidity risk, interest rate risk, credit risk and equity risk. It is uncertain if it is the diversification or size effect that determines risk-taking of banks.
7 income and wholesale funding can result in diversification benefits of risk. Results for the German banking sector presented by Köhler (2014) indicate that higher shares of non-interest income decreases insolvency risk of commercial banks but increases insolvency risk of investment banks. Investment banks’ experience higher average rates of non-interest income than commercial banks, which implies that only small shares of non-interest income creates diversification benefits.
Overall, most recent studies find a negative relationship between non-interest income and risk. Especially, the relationship between non-interest income and insolvency risk seems to be non-linear. We expect a small share of non-interest income to generate diversification benefits, but large shares of non-interest income to increase insolvency risk.
2.2. Bank funding structure
Banks traditionally fund their activities with customer deposits. Fragility of customer-deposit funding, caused by risk of bank runs, creates an opportunity for alternative sources of short-term funding. Over the past two decades, banks have been expanding their funding shares with sources other than customer deposits. Wholesale funding refers to all short-term funding that cannot be classified as commercial deposits and satisfies short-term funding needs that cannot fully be covered with customer deposits. Subsection 2.2.1 relates to previous studies on the relationship between wholesale funding and bank return. Subsection 2.2.2 discusses past research on wholesale funding and risk.
2.2.1. Bank funding structure and return
8 Demirgüç-Kunt and Huizinga (2010) are the first to examine the relationship between shot-term funding structures and return. The authors find that the relationship between non-deposit wholesale funding and return is negative rather than positive.
While reasoning for funding structures is not entirely clear nor straight-forward, results provided by Demirgüç-Kunt and Huizinga (2010) could imply that alternative forms of short-term funding is not used for efficient investment opportunities. We therefore expect a negative relationship between wholesale funding and bank returns.
2.2.2. Bank funding structure and risk-taking
9 Previous research point out that reliance on non-deposit funding can provide benefits but also give rise to additional bank risk. We expect wholesale funding to have an impact on insolvency risk if there is a correlation between non-interest income and wholesale funding shares of banks. In the presence of a correlation we expect non-traditional banks to use wholesale funding for riskier investment projects. As a consequence we expect a positive relationship between wholesale funding share and insolvency risk.
3. Methodology
Section 3.1 defines the model of bank return and risk, explains why the variables are included and how they are constructed. Section 3.2 elaborates on the specification of the model with fixed effects ordinary least squares and instrumental variables.
3.1. Explanation of the variables
Subsection 3.1.1 presents the estimated model of diversification on risk and return. Subsection 3.1.2 discusses the variables constructed to measure non-traditional income and funding structures. Subsection 3.1.3 describes the variables of bank return and risk. Subsection 3.1.4 deliberates on the control variables specified in the model. Subsection 3.1.5 considers other control variables used throughout this paper. An overview of all the variables can be found in Appendix A.
3.1.1. The model of diversification on bank return and risk
Formally, the model is given in Eq.(1) as follows:
𝑌𝑖,𝑡 = 𝛼 + 𝛽𝐷𝐼𝑉𝑖,𝑡+ 𝜑 ∑ 𝑋𝑖,𝑡+ 𝛿 ∑ 𝑀𝑐,𝑡 +𝜃𝐿𝑖+ 𝑢𝑖,𝑡,
Y represents several measures of risk and return as current period values over the 2005-2015 sample period. DIV is either the non-interest income, the squared term of NII, the wholesale funding ratio, the squared term of WSF, the value of earning asset diversification or its squared term, all in time-variant values. X equals a set of bank-level control variables: size, capitalization, asset growth, personnel expenses, other expenses, and ownership concentration. M defines a set of macroeconomic control variables: inflation, GDP per capita, and GDP growth.
(1)
10 L is a dummy variable for if a bank is listed or delisted. α is the constant and 𝑢 is the error term. β, 𝜑, δ, and 𝜃 are vectors of the coefficient estimates.
3.1.1. Diversification of income and funding
Operational income can be divided into interest income generated from outstanding loans and non-interest income generated from other activities. We construct a proxy for bank diversification of income activities. Our main measure of diversification of activities is an income-level variable is a ratio of non-interest income to total operating income (NII).2
Non-interest income consists of fees, commissions, trading, securities, insurance activities, financial assets and other operating income. We are interested to see if there are difference in impact between trading and non-trading non-interest income. While trading seems very volatile, some studies find that only fee-based activities increase risk, not trading (Stiroh and Rumble, 2006; Lepetit et al., 2008). We create a variable for trading income to total operating income (TRADING), which is generated from trading of derivatives, cash instruments, off-balance contracts, and mark-to-market changes in the carrying value of assets and liabilities (Stiroh, 2004). We also create a variable for non-trading non-interest income as a ratio to total operating income (NONTRADINGNII). Non-trading non-interest income includes fees, commissions, gains or losses on financial operations, and other non-interest operating income. We are aware of the fact non-interest income as a proxy for diversification can suffer from some reliability issues. The reliability issue stems from potentially large differences in yearly income affecting the proxy of diversification. A bank participating in non-traditional activities experiencing a year of negative returns on these activities is not necessarily less non-traditional during that year. We introduce a second measure of diversification to deal with this issue. We construct an asset-level variable of non-traditional income activities that measures the diversification of earning assets (DIVERS).3 Diversification of earning assets is a ratio of non-loan earning assets
to total earning assets. Total earning assets include loans, securities and investments.
On the liability side, banks traditionally fund themselves with customer deposits. Short-term funding not classified as customer deposits are organized under wholesale funding. Examples of wholesale funding are: interbank deposits, federal funds, bank borrowings, commercial papers, and certificates of deposits. We estimate the variable wholesale funding
2 This variable is also used in DeYoung and Roland (2001), Stiroh (2004) and Demirgüç-Kunt and Huizinga (2010). 3 The asset level measure of activity diversity is also included in the research of Stiroh and Rumble (2006), Leaven
11 (WSF) as a ratio of non-customer-deposit short-term funding to the sum of deposit, money market and other short-term funding. We are interested to see if there exist a difference in the effect of interbank deposits compared to other sources of wholesale funding. We create two sub-variables, namely deposits from banks (BANKDEP), other short-term borrowings (OTHERBOR), both as shares of total short-term funding.
3.1.2. Time varying bank return and risk
We use several measures of bank risk and return in our empirical analysis. Our main proxy for return is a ratio of profit before taxes to total assets (ROA).
Our focus of risk-taking will be on the distance to insolvency, generally known as the Z-score. 4 Z-score is the inverse of the probability of insolvency (Roy, 1952). A higher Z-score
indicates a larger distance to insolvency which makes it a safer bank. There are different methods to capture time-variant Z-score of banks:
𝑍𝑠𝑐𝑜𝑟𝑒𝑡=
𝐶𝐴𝑅𝑡+µ𝑅𝑂𝐴,𝑡 𝜎𝑅𝑂𝐴,𝑡 ,
In Eq. (2), mean and standard deviation of ROA are calculated over the total sample period combined with annual moving CARs (Lepetit and Strobel, 2013).
𝑍𝑠𝑐𝑜𝑟𝑒𝑡=
𝜇𝐶𝐴𝑅𝑡+µ𝑅𝑂𝐴𝑡 𝜎𝑅𝑂𝐴𝑡 ,
In Eq. (3), Z-score is the result of the sum of averages of CAR and ROA and standard deviation of ROA for periods of time (Boyd et al, 2006).
𝑍𝑠𝑐𝑜𝑟𝑒𝑡 =
𝐶𝐴𝑅𝑡+𝑅𝑂𝐴𝑡
𝜎𝑅𝑂𝐴𝑡 , 𝜎𝑅𝑂𝐴𝑡 = |𝑅𝑂𝐴𝑡− µ𝑅𝑂𝐴|,
Eq. (4) shows a Z-score estimation method based on yearly moving ROA and CAR estimations. The sum of the time variant ROA and CAR estimations are divided by the standard
4 This proxy of bank risk can also be found in the research of Stiroh (2004), Laeven and Levine (2009) and
Demirgüç-Kunt and Huizinga (2010).
(2)
(3)
12 deviation of ROA estimated by the current period values of ROA minus the mean ROA over the full sample period (Boyd et al, 2006).
𝑍𝑠𝑐𝑜𝑟𝑒𝑡=
𝐶𝐴𝑅𝑡+𝑅𝑂𝐴𝑡 𝜎𝑅𝑂𝐴,𝑡 ,
Eq. (5) constructs Z-score as the calculation of the full sample period standard deviation of ROA combined with current period estimations of ROA and CAR values (Hesse and Čihák, 2007).
We believe that accurate values of inputs are necessary to capture time-variant Z-scores that are the best representation in the specific point in time. Lepetit and Strobel (2013) empirically estimate that Eq. (2) is the most reliable estimation for capturing time-variant bank risk. The authors do state that for specific subsamples, time-varying Z-score measure based on Eq. (5) can be more appropriate and results based on Eq. (2) and Eq. (5) are similar. For this reason we will use Eq. (5) as our measure of time-varying Z-score (AZSCORE) in our model. For testing the robustness of our estimations we will use alternative measures of return and risk. As an alternative to ROA we use return on equity (ROE) as a ratio of profit before taxes to total equity. Other measures of bank risk we use are the standard deviation of the return on assets (SDROA) and the Sharpe Ratio (SRATIO). Risk-taking can be defined as choices made by banks that increase the volatility of bank profits (DeYoung and Roland, 2001; De Nicolo, Dell’Aricca, Laeven, and Valencia, 2010). We use the standard deviation of return on assets (SDROA) to measure bank profit volatility. The standard deviation of ROA is estimated as follows:
𝜎𝑅𝑂𝐴 = √(𝑅𝑂𝐴𝑡− µ𝑅𝑂𝐴)2
The higher the standard deviation of ROA, the higher the risk exhibited by the bank. The Sharpe ratio is a proxy for risk-adjusted performance. The Sharpe ratio calculated as the current period value of ROE divided by the full sample period standard deviation of ROE. A higher Sharpe ratio indicates a safer bank.
3.1.3. Control variables
We use several bank-level and country-level control variables. We use the capital to asset ratio (CAR) as a proxy for capitalization because it has a relationship with bank risk
(5)
13 (Calem and Rob, 1990). We construct a variable for size, which is the natural logarithm of assets (LNASSETS), because it can have an impact on risk-taking (Rogers and Sinkey, 1999). We include asset growth (ASSETGROWTH), which controls for operating strategies resulting in risk differences. We also incorporate personnel expenses (PERSONNEL) and other expenses (OTHEREXP) as shares of total operating expenses control for differences in cost structures and efficiency. We include an index of ownership concentration (OWNERCON). The index measures the concentration of ownership and provides an indication of a bank’s level of independence of its shareholders. The higher the index, the higher the concentration of ownership (An index of 1 is equal to a bank with no recorded shareholder with an ownership over 24,99%; an index of 2 indicates a bank with no recorded shareholder with an ownership of over 49,99%; an index of 3 is equal to a bank with a recorded shareholder that has an ownership of over 49,99% or is identified as the ultimate owner). A higher ownership concentration is found to influence bank risk (leaven, 2002; Leaven and Levine, 2009). We also control for potential relationships between delisted banks and business models, performance and risk by including a dummy variable (LISTED) that equals 1 if the bank is listed, zero otherwise. Country-level economic control variables account for country-based determinants of bank risk. INFLATION, calculated in annual consumer prices, affects bank behaviour through margins and overhead costs, possibly affecting return and risk (Demirgüç-Kunt et al., 2003). We also control for gross domestic product (GDP) per capita in constant 2011 US dollars. GDP is a measure of country level output and influences performance of banks. GDP growth (GDPGROWTH) controls for business cycle fluctuations and is estimated in percentages and is annually computed.
3.1.4. Other bank and country level control variables
14 (CAPSTRING). This index is based on questions e.g. whether the capital requirement reflects certain risk elements and deducts certain market value losses from capital before minimum of required capital is determined. Higher index values indicate higher stringency about required capital. We use ACRESTRICT, measured by an index, as a proxy for regulatory restrictions on non-traditional bank activities. Barth, Caprio and Levine (2002) find that activity restrictions are negatively associated with bank development and stability. A higher index indicates more restrictions on income generated activities in insurance, real estate or securities. We also use an index for diversification guidelines (DIVERSGUID). The higher the index the more guidelines exist for diversification. We control for supervisory power (SUPERVISION) using an index based on questions about enforcement power of supervisory institutions. The higher the index the higher the supervisory power.5
Shareholder protection can affect corporate performance and risk (Kose et al., 2008). We are interested to see if differences in shareholder protection affects bank business models. We include several country level control variables for shareholder protection. We use an index for protection of minority shareholders against self-dealing of corporate directors and board enrichment (SELFDEALNG).6 An index of anti-director rights of shareholder (RIGHTS) is
split into minority rights (MINORITYPROT) and other rights (OTHERRIGHTS) to distinguish between shareholder and minority rights. Higher country indices suggest higher shareholder rights.7
3.2. Fixed Effects Ordinary Least Squares and Instrumental Variables model of bank risk
OLS will be the main econometric technique to estimate our results throughout this paper. Our data set consists of bank observations over 11 years and 27 different countries. We expect observations to be subject to unobserved heterogeneity caused by year and country effects. We can control for unobserved heterogeneity with a Fixed Effects (FE) or Random Effects (RE) method of OLS. The RE method is appropriate if the individual-level effects are uncorrelated with the explanatory variables while the FE method is desired if the individual-level effects have a correlation with the explanatory variables. Fixed Effects OLS seems most appropriate for our sample because it allows us to encounter correlated country and time specific unobserved effects affecting coefficients of our explanatory variables. We perform a Hausman test to confirm our expectations about the heterogeneity of unobserved effects. The
5 For details on how the indices are constructed see Barth, Caprio, and Levine (2002). 6 See Djankov et al. (2008) for more information on self-dealing indices.
15 Hausman test is based on the null hypothesis that the individual-level effect is uncorrelated with the explanatory variables. We test the hypothesis based on Eq.(1) for both ROA and Z-score. The Hausman test on ROA reports a Chi2 value of 83.12 and a p-value of 0.000. The same test on Z-score reports a Chi2 value of 69.08 and a p-value of 0.000. Both tests reject the null hypothesis that the difference in coefficients is not systematic. We are now certain that the FE-OLS model is most appropriate. Because we have 11 observations per bank we errors will most likely be heteroskedastic within individual bank observations. In addition to our fixed year and country effects we will cluster the errors in our model at bank level.
Our model given in Eq.(7) is adjusted with Fixed Effects as follows:
𝑌𝑖,𝑡 = 𝛼 + 𝛽𝐷𝐼𝑉𝑖,𝑡+ 𝜑 ∑ 𝑋𝑖,𝑡+ 𝛿 ∑ 𝑀𝑐,𝑡 +𝜃𝐿𝑖+ 𝜈𝑖+ 𝜆𝑖 + 𝑢𝑖,𝑡,
In which ѵ is the cross-section fixed effect, λ is the time fixed effect and 𝑢 the error term.
One basic assumption for OLS regressions is that the explanatory variables are exogenous. In case the basic assumption cannot hold, the estimated results based on the FE-OLS model are inconsistent. Further in the paper we will use the Two Stage Least Squares (2SLS) Instrumental Variables technique to deal with potential endogeneity issues. Our explanatory variable is endogenous if it has a relationship with the error term:
𝐶𝑜𝑣[𝐷𝐼𝑉, 𝑢] ≠ 0,
The IV technique allows for a two staged estimation process to eliminate endogeneity issues by including instrumental variables in the first stage of the equation. The IV model uses exogenous instruments that have to meet two requirements: the instrument is correlated with the endogenous variable and the instrument is uncorrelated with the error term. The 2SLS IV methodology is based on two stages. The linear relationship of interest is our model defined in Eq.(7). The first stage is estimated in Eq.(9) as follows:
𝐷Î𝑉𝑖,𝑡 = 𝛼 + 𝑦 ∑ 𝑍𝑖 + ɛ𝑖,𝑡, (9)
(7)
16 In which 𝐷𝐼𝑉 is the endogenous variable and Z are the instrumental variables. The first stage eliminates the endogeneity of the explanatory variable by using predicted values based on the instrumental variables.
The second stage adjusts our model and is given in Eq.(10) as follows:
𝑌𝑖,𝑡 = 𝛼 + 𝛽𝐷Î𝑉𝑖,𝑡+ 𝜑 ∑ 𝑋𝑖,𝑡+ 𝛿 ∑ 𝑀𝑐,𝑡 +𝜃𝐿𝑖 + 𝜈𝑖+ 𝜆𝑖 + 𝑢𝑖,𝑡,
In which 𝐷Î𝑉 is the exogenous form of 𝐷𝐼𝑉. The exogenous variables eliminate the bias that would be present in OLS regressions.
4. Data
Section 4.1 presents the data sources and summarizes the final data set. Section 4.2 describes the trends in income diversification, wholesale funding, return and risk for different bank types. Section 4.3 evaluates income diversification and wholesale funding. The descriptive statistics can be found in Appendix B.
4.1. Data set and sources
Bank-level data are retrieved from Bankscope, a Bureau van Dijk database for global banking data. Country specific information about GDP, GDP growth and inflation is retrieved from The Worldbank Group. Country specific bank regulation and supervision indices are based on the database created by Barth, Caprio and Levine (2013). Barth, Caprio and Levine (2013) use surveys carried out by The Worldbank. The surveys of The Worldbank measure regulatory and supervisory approaches in 143 jurisdictions. The surveys cover data from 2011 and allow us to use the most current form of bank regulation and supervision in a wide range of countries. Information about self-dealing is retrieved from the index on shareholder protection created by the Worldbank group. The index is based on questionnaires carried out to corporate and securities lawyers on securities regulations and company laws. The survey covers data from 2015 in 189 jurisdictions. Data on legal protection of shareholders and minorities are based on an index constructed by La Porta et al. (1998).
17 banks. Listed banks are more likely to meet up to international accounting standards which increases comparability between different countries (Leaven and Levine, 2009; Demirgüç-Kunt and Huizinga, 2010). Delisted banks are added to the data set to overcome the survivorship bias. 68,9% of the banks in de sample are listed and 31,1% of the banks are delisted. Central banks and specialized governmental institutions are dropped from the sample, because it is not relevant for our research. Our final data set consists of 414 banks over the 2005-2015 period.
The banks are divided into four categories: Commercial banks, investment banks, non-credit institutions, and other banks.8 Bank Holding Companies, Commercial Banks and Savings
Banks are classified as “commercial banks”. Investment Banks, Investment and Trust Corporations, Securities Firms, Private Banking and Asset Management Companies are categorized in “investment banks”. Finance Companies (e.g. credit card, factoring, and leasing companies), Group Finance Companies and other non-banking credit institutions are classified as “non-bank credit institutions”. Cooperative Banks, Islamic Banks, Real Estate and Mortgage Banks are grouped in “other banks”. 62.3% of the banks in the data set are commercial banks, 20.5% are investment banks, 5.6% are non-bank credit institutions and 11.4% of the data set consists of other type of banks. We have adjusted very influential outliers to the 1th and 99th
percentile of the sample values.
4.2. Trends for different bank types
The sample captures 11 years of information allowing us to identify trends in wholesale funding and non-interest income shares and developments in bank risk and return. Fig. 1 presents the developments in non-interest income shares for the different bank categories since 2005. It shows that investment banks and bank credit institutions gain much higher non-traditional income shares than commercial banks and other banks. We see that the non-interest income seems much more volatile for investment banks and non-credit institutions than for commercial banks and other banks. Non-interest income is known to be more volatile than net interest income (DeYoung and Roland, 2001; Stiroh, 2004). Fig.1 suggests that the volatility of non-interest income marginally increases with the participation rate of non-traditional income activities. All banks experienced a steep decrease in non-interest income during the 2008 financial crisis. Additionally, we are interested in the differences and trends in non-traditional earning assets, because it is less subject to income volatility. Trends in non-interest earning assets shares can be found in Fig. 2. We see that the slope of diversification of earning assets
18 is more stable but similar to the non-interest income shares, except for non-bank credit institutions. Non-bank credit institutions have non-interest income shares that are much higher compared to their relative rate of diversification. While investment banks and non-bank credit institutions have higher diversification rates in 2015, commercial banks have lower levels of activity diversification compared to 2005. Other banks have equal diversification levels in 2015 compared to 2005.
Fig.1. Trends in non-interest income for different bank categories from 2005 to 2015. Bank categories are
commercial banks, investment banks, non-bank credit institutions and other banks.
Fig.2. Trends in diversification of earning assets for different bank categories from 2005 to 2015. Bank categories
are commercial banks, investment banks, non-bank credit institutions and other banks.
.2 .4 .6 .8 N o n -i n te re st i n co me 2005 2010 2015 year
Commercial banks Investment banks
Non-bank credit institutions Other banks
.3 .4 .5 .6 .7 .8 D ive rs if ica ti o n o f e a rn in g a ss e ts 2005 2010 2015 year
Commercial banks Investment banks
19 Fig. 3 shows the trends in wholesale funding since 2005. Wholesale funding shares have declined for all bank types except for non-bank credit institutions. The recent financial crisis has uncovered a negative side of wholesale funding, which could explain the downward trend especially from 2008 onwards. Fig. 3 shows that non-bank credit institutions and other banks rely most heavily on wholesale funding. Commercial banks and investment banks show the biggest change in wholesale funding rates.
Fig.3. Trends in wholesale funding for different bank categories from 2005 to 2015. Bank categories are
commercial banks, investment banks, non-bank credit institutions and other banks.
Fig. 4 presents the developments in return on assets for the different bank categories since 2005. We see an overall decline in returns since 2005 for all bank categories. Investment banks and non-bank credit institutions have experienced the steepest decrease in returns. Fig. 5 shows the development of Z-scores for the different bank categories. We see an overall increase in Z-score for all categories of banks, except for other banks. Other banks do have an overall Z-score that is much higher than the Z-scores of the other banks. Summarizing Fig.1-4 we see an overall decline in wholesale funding, non-interest income, returns, and risk.
.1 .2 .3 .4 .5 W h o le sa le f u n d in g 2005 2010 2015 year
Commercial banks Investment banks
20
Fig. 4. Developments of return on assets for different bank categories from 2005 to 2015. Bank categories are
commercial banks, investment banks, non-bank credit institutions and other banks.
Fig. 5. Developments of Z-score between four bank categories from 2005 to 2015. Bank categories are commercial
banks, investment banks, non-bank credit institutions and other banks.
-. 1 5 -. 1 -. 0 5 0 .0 5 .1 R O A 2005 2010 2015 year
Commercial banks Investment banks
Non-bank credit institutions Other banks
10 20 30 40 50 Z -s co re 2005 2010 2015 year
Commercial banks Investment banks
21
4.3. Evaluation of income diversification and wholesale funding
Next, we estimate a correlation table for further evaluation of income diversification and wholesale funding. The correlation matrix is shown in table 1. Table 1 reports a large and significant correlation coefficient between diversification of earning assets and non-interest income shares. Surprisingly, wholesale funding has no statistically significant correlation with non-interest income or diversification of earning assets. It could imply that wholesale funding is not necessarily used by banks to expand into non-traditional banking activities. Furthermore, Z-score and return on assets have a statistically significant correlation of 0.04 at a 5% significance level. As expected, non-interest income and diversification of earning assets are both negatively correlated with Z-score and the relationships are both statistically significant. The correlation coefficient for diversification of earning assets and Z-score of banks is much larger and stronger than of non-interest income and Z-score. Both non-interest income and diversification of earning assets have a positive and statistically significant relationship with return on assets. Wholesale funding has a positive correlation that is statistically significant with a bank’s Z-score. This correlation could indicate that there exists a relationship between wholesale funding and Z-score even tough wholesale funding shares are not related to income diversification. Wholesale funding has no statistically significant correlation with return on assets.
Table 1
Correlation matrix.
The correlation matrix reports relationships between Z-score, return on assets, wholesale funding, non-interest income and diversification of earning assets. * , ** and *** indicate significance at the 10%, 5% and 1% level respectively.
Variable Z-score ROA Wholesale funding Non-interest income Diversification
azscore 1 roa 0.0416** 1 wsf 0.218*** 0.005 1 nii -0.0362*** 0.0528*** -0.00293 1 divers -0.164*** 0.0454*** 0.0106 0.307*** 1
22 concentration does not affect non-interest income shares in any of the regressions. Also, country specific economic factors (GDP, GDP growth, and inflation) have no effect on non-interest income shares of bank in any of the regressions. Regression 1 and 2 reveal that investment banks have statistically and significantly higher shares of non-interest income. This can be explained by the fact that the core business of investment banks is centred on non-interest income activities. Other banks are statistically and significantly associated with lower non-interest income shares. Regression 3 and 4 show that higher capital requirements result in lower non-interest income shares of banks. Interestingly, activity restrictions show no relationship with non-interest income shares of banks. Regression 6 reports a small significant relationship between shareholder protection laws and non-interest income at the 10% level. We see that protection of minority shareholders has a positive effect on non-interest income shares at the 5% level. Countries with more shareholder rights will have more banks with higher non-interest income shares, while minority protection has a negative impact on non-interest income shares of banks.
Table 2
Model of non-interest income shares.
Regressions 1-2 include bank dummies to control for bank specific characteristics influencing non-interest income shares. Regressions 3 and 4 control for country specific regulation and supervision. Regressions 5- 6 control for country specific level of shareholder protection and rights. Regressions 2, 4, and 6 include country level economic control variables. Standard errors are reported in parentheses. * , ** and *** indicate significance at the 10%, 5% and 1% level respectively.
23 (0.006) (0.006) (0.012) investment 0.126** 0.154*** (0.050) (0.047) nonbank 0.043 0.055 (0.102) (0.104) otherbank -0.118*** -0.124*** (0.031) (0.031) supervision -0.006 -0.001 (0.014) (0.014) acrestrict 0.009 0.010 (0.010) (0.009) capreq -4.141*** -2.573** (1.013) (1.061) capstring 0.022 0.021 (0.014) (0.013) diversguid 0.018 (0.037) rights 0.045 0.043* (0.025) (0.021) minorityprot -0.103 -0.114** (0.060) (0.050) otherrights 0.000 0.000 (.) (.) selfdealing -0.650 -0.507 (0.607) (0.519) N 2902 2706 2394 2257 1820 1712 R2 0.113 0.136 0.188 0.254 0.180 0.254 adj. R2 0.099 0.121 0.181 0.246 0.172 0.245
Country FE Yes Yes No No No No
Time FE Yes Yes Yes Yes Yes Yes
Clustering Bank Bank Country Country Country Country
24
Table 3
Model of diversification of earning assets.
Regressions 1-2 include bank dummies to control for bank specific characteristics influencing non-interest income shares. Regressions 3 and 4 control for country specific regulation and supervision. Regressions 5-6 control for country specific level of shareholder protection and rights. Regressions 2, 4, and 6 country level economic control variables. Standard errors are reported in parentheses. * , ** and *** indicate significance at the 10%, 5% and 1% level respectively.
25 selfdealing -0.257 -0.795 (0.707) (0.536) N 2879 2683 2387 2249 1809 1701 R2 0.493 0.495 0.319 0.370 0.292 0.324 adj. R2 0.485 0.486 0.313 0.363 0.285 0.315
Country FE Yes Yes No No No No
Time FE Yes Yes Yes Yes Yes Yes
Clustering Bank Bank Country Country Country Country
Table 4 reports the model of wholesale funding. Large and higher capitalized banks tend to rely more heavily on wholesale funding. Banks located in countries with a higher growth of GDP have lower shares of wholesale funding than banks located in countries with lower GDP growth. Regressions 1 and 2 show that non-bank credit institutions and other banks rely more heavily on wholesale funding. Regression 3 and 4 illustrate that power of supervisory institutions and a higher level of required capital decrease wholesale funding shares. Capital structures might not be fully determined by capital regulation as argued by Gropp and Heider (2010), but at least it does partly influence the choice of funding. Especially the coefficient of capital requirements is very large and significant. Shareholder protection has no influence on wholesale funding shares of banks.
Table 4
Model of wholesale funding.
Regressions 1-2 include bank dummies to control for bank specific characteristics influencing non-interest income shares. Regressions 3-4 control for country specific regulation and supervision. Regressions 5-6 control for country specific level of shareholder protection and rights. Regressions 2, 4, and 6 country level economic control variables. Standard errors are reported in parentheses. * , ** and *** indicate significance at the 10%, 5% and 1% level respectively.
26 (0.181) (0.423) (1.055) inflation -0.001 -0.010 -0.024* (0.004) (0.006) (0.012) investment 0.004 0.006 (0.046) (0.047) nonbank 0.232** 0.240** (0.113) (0.114) otherbank 0.156*** 0.154*** (0.040) (0.040) supervision -0.013** -0.009 (0.006) (0.006) acrestrict 0.001 0.004 (0.018) (0.014) capreq -2.965** -2.795** (1.165) (1.028) capstring 0.010 0.016 (0.015) (0.016) diversguid 0.044 0.058 (0.043) (0.043) minorityprot -0.157 -0.170* (0.104) (0.090) rights 0.062 0.056* (0.037) (0.028) otherrights 0.000 0.000 (.) (.) selfdealing 0.059 0.560 (0.717) (0.658) N 2454 2281 2190 2058 1755 1511 R2 0.369 0.368 0.184 0.205 0.172 0.170 adj. R2 0.357 0.355 0.176 0.196 0.164 0.158
Country FE Yes Yes No No No No
Time FE Yes Yes Yes Yes Yes Yes
Clustering Bank Bank Country Country Country Country
5. Results
Section 5.1 presents the FE-OLS regressions of wholesale funding and non-interest income on bank return and risk. Section 5.2 provides an in-depth analysis on different sources of wholesale funding and non-interest income in their relationships with return and risk. Section 5.3 estimates robustness checks by employing alternative measures of income diversification, bank risk and return. Section 5.4 deals with and corrects potential endogeneity issues by performing Instrumental Variables (IV) regressions.
5.1. Bank return and risk
27 growth is statistically significant in all regressions and positively affects ROA. Regressions 1 and 2 show that wholesale funding has a statistically negative effect on ROA at a 5% significance level. Non-interest income is only significant in regression 6. In regressions 4 and 5 we take into account the squared values, but they have no explanatory power. Regression 6 combines the linear and squared terms of wholesale funding and non-interest income. Wholesale funding is still negatively affecting return on assets at a 10% significance level. More interestingly, non-interest income has become highly significant in positively affecting return on assets, but the relationship seems to be non-linear. This suggests that only a marginal increase in non-interest income would increase return on assets, but banks participating heavily in non-interest income activities would have lower returns instead.
Table 5
Model of Return on Assets.
28 niisq 0.000 -0.026*** (0.001) (0.007) N 2281 2706 2277 2281 2706 2277 R2 0.190 0.109 0.191 0.191 0.109 0.216 adj. R2 0.174 0.094 0.174 0.174 0.094 0.199
Country FE Yes Yes Yes Yes Yes Yes
Time FE Yes Yes Yes Yes Yes Yes
Clustering Bank Bank Bank Bank Bank Bank
Table 6 presents the model of Z-score. Table 6 reveals that asset growth has a negative and statistically significant impact on bank risk. Personnel expenses are significantly increasing risk. Ownership concentration also reduces Z-scores of banks at a 10% significance level in almost every regression. This outcome confirms the results of Leaven and Levine (2009).9
Table 1 reported a positive and significant correlation between wholesale funding and Z-score. Wholesale funding has no explanatory power in any of the regressions in table 6. Regressions 5 and 6 show that non-interest income does affect bank risk, but the sign differs between the regressions. The coefficients of non-interest income are large and negatively and statistically related to the Z-score in regressions 3 and 5. In regression 6, the coefficient turns positive and the quadratic interest income term is negative and very large. This suggests that non-interest income has a non-linear relationship with bank risk.10 Small percentages of non-interest
income activities will result in a higher Z-score while a high level of non-interest income increases risk.11
9 Leaven and Levine (2009) find that ownership concentration increases shareholder power and results in higher
bank risk.
10 This result is consistent with outcomes provided by Baele and De Jonghe (2007) and Demirgüç-Kunt and
Huizinga (2010).
11 Leaven and Levine (2009) find that there exists an interaction effect between national banking regulation,
29
Table 6
Model of Z-score.
Regressions 1-2 report the impact of wholesale funding and non-interest income share. Regressions 4-5 include squared terms of wholesale funding and non-interest income to detect potential nonlinear relationships. Regression 6 includes wholesale funding and non-interest income together with the squared terms. Standard errors are reported in parentheses. * , ** and *** indicate significance at the 10%, 5% and 1% level respectively.
Variable (1) (2) (3) (4) (5) (6) lnassets 0.360 0.404 0.318 0.229 0.329 0.110 (0.487) (0.404) (0.491) (0.546) (0.404) (0.538) assetgrowth -1.643** -1.219*** -1.554** -1.639** -1.175*** -1.569** (0.690) (0.301) (0.649) (0.709) (0.289) (0.646) car 11.534 2.631 14.770* 11.861 4.506 18.226** (8.036) (5.843) (7.818) (7.939) (6.381) (7.688) personnel -55.656*** -42.620*** -50.474*** -56.138*** -39.500*** -47.891*** (15.755) (12.667) (15.325) (16.121) (12.129) (14.681) otherexp -21.823 -2.211 -21.607 -23.004 -2.101 -25.451 (22.348) (1.378) (2.1614) (22.822) (1.446) (21.764) ownercon -2.448* -1.269 -2.411* -2.450* -1.244 -2.395* (1.423) (1.326) (1.425) (1.423) (1.329) (1.401) gdp 0.000** 0.000 0.000** 0.000** 0.000 0.000** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) gdpgrowth 3.600 0.997 4.540 4.251 1.665 1.962 (7.465) (8.746) (7.545) (7.412) (8.931) (7.511) inflation 0.298*** 0.296** 0.254** 0.301** 0.274** 0.179 (0.115) (0.128) (0.120) (0.116) (0.128) (0.135) listed 5.067* 3.029 5.182* 5.080* 3.222 4.959* (2.932) (2.797) (2.943) (2.923) (2.773) (2.873) wsf 6.172 5.453 14.315 13.385 (4.372) (4.330) (11.001) (10.807) nii -2.255 -5.518** -5.776** 7.806** (1.984) (2.787) (2.913) (3.631) wsfsq -9.362 -8.024 (12.929) (12.735) niisq -0.340** -13.257*** (0.170) (3.495) N 2281 2706 2277 2281 2706 2277 R2 0.337 0.255 0.337 0.338 0.260 0.351 adj. R2 0.324 0.242 0.323 0.325 0.248 0.337
Country FE Yes Yes Yes Yes Yes Yes
Time FE Yes Yes Yes Yes Yes Yes
Clustering Bank Bank Bank Bank Bank Bank
5.2. Trading income and short-term borrowings on return and risk
30 Regression 2 shows that the effect of trading income on return on assets is slightly negative. Once non-trading non-interest income enters the equation in regression 3 trading income loses its significance. Regression 6 reports the combined effect of trading and other non-interest income on Z-score. Both trading income and other non-interest income are negatively affecting Z-scores at a 5% significance level.12
Table 7
Model of trading income and other non-interest income on ROA and Z-score.
Regressions 1-3 present the effect of trading income and other non-interest income on ROA. Regressions 4-6 present the effect of trading income and other non-interest income on Z-score. Standard errors are reported in parentheses. * , ** and *** indicate significance at the 10%, 5% and 1% level respectively.
Variable ROA Z-score
(1) (2) (3) (4) (5) (6) lnassets 0.003** 0.003** 0.003** 0.319 0.449 0.298 (0.001) (0.001) (0.002) (0.404) (0.404) (0.408) assetgrowth 0.002 0.002 0.002 -1.155*** -1.262*** -1.125*** (0.002) (0.003) (0.002) (0.289) (0.311) (0.279) car 0.071*** 0.078*** 0.073*** 3.762 2.201 5.185 (0.016) (0.017) (0.016) (6.108) (5.682) (6.459) personnel 0.036 0.057 0.040 -39.295*** -45.010*** -34.849*** (0.120) (0.116) (0.122) (11.745) (13.095) (11.710) otherexp -0.052 -0.052 -0.052 -1.882 -2.358* -1.852 (0.092) (0.092) (0.092) (1.393) (1.384) (1.480) ownercon -0.004 -0.004 -0.003 -1.396 -1.261 -1.318 (0.003) (0.003) (0.003) (1.318) (1.326) (1.324) gdp -0.000 -0.000 -0.000 0.000 0.000 0.000 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) gdpgrowth 0.233*** 0.233*** 0.232*** 2.796 1.163 1.840 (0.072) (0.070) (0.072) (9.181) (8.818) (9.016) inflation 0.003** 0.002** 0.003** 0.256** 0.317** 0.231* (0.001) (0.001) (0.001) (0.122) (0.132) (0.125) listed -0.002 -0.001 -0.001 3.087 2.937 3.229 (0.003) (0.003) (0.003) (2.791) (2.808) (2.775) Nontradingnii 0.018 0.015 -6.432** -9.269** (0.014) (0.018) (3.059) (4.025) trading -0.019* -0.009 -2.172 -8.467** (0.010) (0.016) (1.620) (3.761) N 2706 2706 2706 2706 2706 2706 R2 0.113 0.111 0.113 0.260 0.253 0.265 adj. R2 0.098 0.096 0.098 0.248 0.240 0.253
Country FE Yes Yes Yes Yes Yes Yes
Time FE Yes Yes Yes Yes Yes Yes
Clustering Bank Bank Bank Bank Bank Bank
31 Furthermore, we distinguish between interbank deposits and other short-term borrowings. Table 8 reports the effect of interbank deposits and short-term borrowings on ROA and Z-score. Regressions 1-3 show that interbank deposits have a negative effect on return on assets, while other short-term borrowings are not statistically significantly related to ROA. Regressions 4-6 estimate the impact on Z-score. The coefficient of short-term borrowings is large, negative and statistically significant. While interbank deposits have a positive impact on Z-score at a 10% significance level in regression 4, it loses its explanatory power when other borrowings enters the regression in regression 6. Regression 6 indicates that, with the magnitude and significance of the coefficient of other borrowings, wholesale funding can actually be associated with lower Z-score of banks.13
Table 8
Model of interbank deposits and short-term borrowings on ROA and Z-score.
Regressions 1-3 present the effect of interbank deposits and other short-term borrowings on ROA. Regressions 4-6 present the effect of interbank deposits and other short-term borrowings on Z-score. Standard errors are reported in parentheses. * , ** and *** indicate significance at the 10%, 5% and 1% level respectively.
Variable ROA Z-score
(1) (2) (3) (4) (5) (6) lnassets 0.001 0.001 0.000 0.162 -0.169 -0.912 (0.001) (0.001) (0.000) (0.462) (0.617) (0.759) assetgrowt h -0.000 0.005 0.000 -1.419 *** -1.066* -1.826*** (0.001) (0.004) (0.001) (0.354) (0.560) (0.452) car 0.072** 0.105*** 0.146*** -0.963 9.258 66.290** (0.035) (0.020) (0.031) (5.431) (7.687) (31.692) personnel -0.103 0.249*** 0.649*** -63.045*** -31.660 -300.524* (0.239) (0.085) (0.235) (19.974) (22.170) (165.114) otherexp 0.039*** -0.119 -1.096*** -3.899** 20.645* -155.664 (0.007) (0.091) (0.317) (1.975) (11.606) (145.406) ownercon -0.001 -0.001 -0.002* -2.412* -2.231 -3.251** (0.002) (0.002) (0.001) (1.375) (1.517) (1.616) gdp 0.000 0.000*** 0.000*** 0.000*** 0.001** 0.001** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) gdpgrowth 0.108*** 0.325*** 0.152*** 3.611 -14.339 -18.821 (0.039) (0.108) (0.047) (7.716) (14.263) (17.051) inflation 0.001 -0.001 -0.001* 0.317** 0.922** 1.180*** (0.001) (0.002) (0.001) (0.124) (0.371) (0.418) listed 0.001 -0.001 -0.002 5.009* 3.324 6.131 (0.003) (0.003) (0.002) (2.816) (3.679) (4.191) bankdep -0.018* -0.009*** 6.920* 11.618 (0.009) (0.003) (3.675) (7.163) otherbor -0.011 -0.008 -17.170*** -16.320** (0.008) (0.005) (5.260) (7.287)
13 The result is consistent to the result of non-deposit wholesale funding and bank risk in the sample of
32
N 2358 1112 973 2358 1112 973
R2 0.119 0.263 0.468 0.340 0.321 0.400
adj. R2 0.102 0.232 0.442 0.327 0.293 0.370
Country FE Yes Yes Yes Yes Yes Yes
Time FE Yes Yes Yes Yes Yes Yes
Clustering Bank Bank Bank Bank Bank Bank
5.3. Robustness checks
We will now consider alternative measures of activity diversification, bank return and risk to check the robustness of earlier results. We start with diversification of earning assets as a robustness measure for interest income. The results are presented in table 9. While non-interest income has a significant positive effect on return in regression 8 of table 5, diversification of earning assets has no explanatory power in any of the regressions (1-3) in table 9. Moreover, the relationship between diversification of earning assets and bank risk is linear, in contrast to non-interest income, and its relationship is highly negative. It confirms the increase in risk related to non-traditional activities.
Table 9
Model of diversification of earning assets.
Regressions 1-3 present the effect of earning assets diversification and the quadratic term on ROA. Regressions 4-6 present the effect of earning assets diversification and the quadratic term on Z-score. Standard errors are reported in parentheses. * , ** and *** indicate significance at the 10%, 5% and 1% level respectively.
Variable ROA Z-score
33 (0.014) (10.736) divers -0.007 -0.027 -0.005 -16.597*** -23.152* -28.094* (0.008) (0.031) (0.021) (3.867) (12.227) (15.109) wsfsq 0.011 -10.060 (0.017) (12.354) diverssq 0.020 0.005 6.345 10.869 (0.031) (0.023) (10.632) (13.653) N 2683 2683 2269 2683 2683 2269 R2 0.150 0.151 0.191 0.283 0.283 0.364 adj. R2 0.136 0.136 0.173 0.271 0.271 0.350
Country FE Yes Yes Yes Yes Yes Yes
Time FE Yes Yes Yes Yes Yes Yes
Clustering Bank Bank Bank Bank Bank Bank
Next, we evaluate alternative measures for bank return and risk. We use ROE as the measure of return and the Sharpe ratio as the measure of risk. The regression outputs are reported in table 10. Wholesale funding has a small negative influence on return on equity in regression 1 and 3 at a 10% significance level, but loses its explanatory power in regression 4. The relationship between non-interest income and return is similar to that in table 5, small shares increase bank return but high shares of non-interest income will decrease return rates of banks. Wholesale funding is again not significant in any of the regressions for bank risk (regression 5-8). Regression 8 illustrates the non-linear relationship of non-interest income on bank risk. Small shares of non-interest income can increase a bank’s Sharpe ratio but high shares of non-interest result in a declining Sharpe ratio.
Table 10
Model of Return on Equity and Sharpe Ratio.
Regressions 1-4 present the effect of wholesale funding, non-interest income and the quadratic terms on ROE. Regression 5-8 presents effect of wholesale funding, non-interest income and the quadratic terms on the Sharpe Ratio. Standard errors are reported in parentheses. * , ** and *** indicate significance at the 10%, 5% and 1% level respectively.
Variable ROE Sharpe Ratio
34 (0.013) (0.069) (0.013) (0.013) (0.155) (0.146) (0.154) (0.151) gdp 0.000*** 0.000 0.000*** 0.000*** 0.000** 0.000** 0.000** 0.000** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) gdpgrowth 1.515** 0.057 1.511** 1.437** 6.334*** 7.000*** 6.350*** 5.921*** (0.713) (1.876) (0.714) (0.718) (1.514) (1.446) (1.521) (1.557) inflation 0.011* 0.001 0.011* 0.009 0.034* 0.048** 0.038* 0.029 (0.006) (0.018) (0.006) (0.006) (0.020) (0.020) (0.020) (0.022) listed 0.035 -0.084 0.035 0.030 0.526 0.419 0.499 0.467 (0.025) (0.136) (0.025) (0.025) (0.345) (0.319) (0.346) (0.339) wsf -0.073* -0.072* -0.175 -0.289 -0.265 -0.400 (0.041) (0.040) (0.116) (0.480) (0.471) (1.434) nii -0.063 0.015 0.292*** 0.029 0.312 2.066*** (0.113) (0.051) (0.111) (0.096) (0.386) (0.508) niisq -0.269*** -1.714*** (0.095) (0.352) wsfsq 0.142 0.301 (0.127) (1.587) N 2281 2706 2277 2277 2281 2706 2277 2277 R2 0.105 0.019 0.105 0.128 0.315 0.284 0.312 0.328 adj. R2 0.087 0.002 0.086 0.110 0.301 0.272 0.298 0.313
Country FE Yes Yes Yes Yes Yes Yes Yes Yes
Time FE Yes Yes Yes Yes Yes Yes Yes Yes
Clustering Bank Bank Bank Bank Bank Bank Bank Bank
Next, we estimate the relationship with earnings volatility as a robustness measure of risk. In addition, the effect of non-interest income and wholesale funding on earnings volatility provides an in-depth analysis of the influence on the Z-score. Table 11 reports the effect of wholesale funding and non-interest income on earnings volatility. Regression 4 shows that the relationship between non-interest income and earnings volatility is non-linear. Earnings volatility decreases with higher non-interest income shares but at a decreasing rate. As a consequence, high shares of non-interest income will result in higher earnings volatility than low shares of non-interest income. This is consistent with previous studies that find income from non-traditional banking activities is more volatile than interest income (Rosen et al, 1988; Boyd, 1993; Stiroh, 2004; Stiroh and Rumble, 2006). Wholesale funding is not statistically related with earnings volatility in any of the regressions. The results strengthen earlier implications that wholesale funding is not necessarily used by banks to expand into non-traditional banking activities.
35
Table 11
Model of earnings volatility.
Regressions 1-4 present the effect of wholesale funding, non-interest income, and the quadratic terms on earnings volatility. Regressions 1-4 include country fixed effects, year fixed effects and clustering at bank level. Standard errors are reported in parentheses. * , ** and *** indicate significance at the 10%, 5% and 1% level respectively. Variable (1) (2) (3) (4) (0.001) (0.001) (0.001) (0.001) assetgrowth 0.000 -0.002 0.000 0.000 (0.001) (0.001) (0.001) (0.001) car 0.040* 0.085*** 0.038* 0.034* (0.021) (0.014) (0.020) (0.019) personnel 0.026 0.095 0.023 0.019 (0.047) (0.062) (0.047) (0.049) otherexp 0.181*** 0.073 0.181*** 0.184*** (0.062) (0.053) (0.062) (0.062) ownercon 0.000 -0.000 0.000 0.000 (0.001) (0.003) (0.001) (0.001) gdp -0.000 -0.000 -0.000 -0.000 (0.000) (0.000) (0.000) (0.000) gdpgrowth -0.083*** -0.134*** -0.083*** -0.078*** (0.027) (0.042) (0.027) (0.026) inflation 0.001 -0.000 0.001 0.001* (0.000) (0.001) (0.000) (0.000) listed -0.000 -0.001 -0.000 0.000 (0.002) (0.003) (0.002) (0.001) wsf 0.008 0.008 0.017 (0.005) (0.005) (0.011) nii 0.000 0.003 -0.018** (0.005) (0.005) (0.008) niisq 0.020*** (0.006) wsfsq -0.012 (0.009) N 2281 2706 2277 2277 R2 0.287 0.306 0.287 0.308 adj. R2 0.272 0.295 0.272 0.293
Country FE Yes Yes Yes Yes
Time FE Yes Yes Yes Yes
Clustering Bank Bank Bank Bank
5.4. Endogenous variables
36 eliminates potential causality issues because it is less likely for current bank risk values to influence lagged non-interest income and wholesale funding values. Also, lagged values are less likely to be correlated with the error term because they are affected less by current shocks. Table 12 reports the regressions of the lagged values of wholesale funding and non-interest income on return on assets and Z-score. The lagged value of wholesale funding still has a negative impact on return on assets at a 10% significance level. Lagged non-interest income value and its squared term have no explanatory power on return on assets anymore. The lagged value of non-interest income squared has a large negative effect on Z-score in regression 8. Lagged wholesale funding has a positive relationship with Z-score at a 10% significance level, which loses explanatory power once lagged non-interest income and its quadratic term enters the equation in regression 8.
Table 12
Model of Return on Assets and Z-score based on lagged values of non-interest income and wholesale funding. Regressions 1-4 present the effect of lagged values of wholesale funding, non-interest income and the quadratic terms on ROA. Regressions 5-8 present effect of the lagged values of wholesale funding, non-interest income and the quadratic terms on Z-score. Standard errors are reported in parentheses. * , ** and *** indicate significance at the 10%, 5% and 1% level respectively.
Variable ROA Z-score
