Bank Performance and Noninterest Income: Evidence from Countries in the Asian Region

Noninterest income is understood as income generated by banks from sources other than interest payments, the most common examples of which are service fees such as ATM fees and loan origination fees (Haubrich & Young, 2019). There has been a substantial expansion of off-balance sheet activities in many banks globally, indicating that banks have been diversifying, thus making non-interest income an important source of bank revenue. As interest margins have been observed to be narrowing in banks of numerous countries, there has been a tendency towards diversifying into non-interest sources of income (Heffernan, 2005).

Numerous studies have been conducted on the relationship between non-interest income, and bank profitability and/or risk. However, much of this literature is focused on Europe and the USA with relatively little research on the Asian sector (Lee et al., 2014). Hsieh et al. (2013) also state that studies in Asian banks are sparse and claim that theirs is one of the first studies using Asian data. European studies by Smith et al. (2003), Baele et al. (2007) and Lepitit et al. (2008) have generally shown a positive effect of non-interest income on profitability though there is a possibility that greater diversity in bank income may reduce it. Kohler (2014) in a detailed study of German banks finds evidence that the impact of non-interest income on bank stability depends on banks' overall business models. Specifically, it was shown that banks that focus on traditional services become more stable with an increasing share of non-interest income. Hahm (2008) finds for OECD countries that non-interest income does not have a significant effect on risk-adjusted returns. Studies on the US banking system (DeYoung and Rice (2004), DeYoung and Roland (2001), Stiroh (2004, 2006), Stiroh and Rumble (2006)) have found that an increase in non-interest income led to a greater profit variability, which worsened the banks’ risk-return tradeoff. It was also noted that bank risk was found to be strongly negatively correlated to non-interest income sources, specifically trading income.

The view put forward in this paper is that experiences of the USA and Europe may not necessarily hold for Asian countries. The main reason for this is that the characteristics of the banking system in Asia and the role played by it in their respective economies may be quite different than what we observe in the USA or Europe. We discuss these details below.

Research in this area has sought to explain profitability of banks or bank risk by relating it to a variety of factors, some of which are financial factors related to the specific bank and others are macroeconomic and policy variables which affect all banks within a country. The equation that we specify is stated as follows:We define bank risk in terms of the ZSCORE which evaluates the risk exposure of banks (Baek et al., 2018; Stiroh & Rumble, 2006) and is indicative of a bank’s “distance-to-default” (Chiaramonte et al., 2016). A higher value for ZSCORE indicates lower risk and higher stability as the bank has higher levels of profitability and equity with less variability in profitability (Engle et al., 2014). Following Kohler (2014), we define:where ROA_it = Return on Asset for bank i in time period t. TE_it = Total equity of bank i in time period t. TA_it = Total assets of bank i in time period t. SDROA_i = standard deviation of ROA of bank i, computed over the sample period.

It may be pointed out that ZSCORE can be computed in different ways depending on how the three components (ROA, TE/TA and SDROA) in its definition are measured. For instance, Houston et al. (2010) measure all three components as averages over their period of analysis while Berger et al. (2014) compute rolling averages averages of ROA and (TE/TA) as well as rolling standard deviation of ROA (SDROA) over previous 12 quarters. It should also be noted that other measures of bank stability/risk have been used in the literature, such as, bank capitalization, non-performing loans and risk-weighted capital-asset ratio among others (Jayakumar et al., 2019; Pradhan et al., 2019; Baek et al., 2018; Skala & Weill, 2018). While there can, indeed, be a debate regarding which measure of bank risk or bank stability is to be used, we make use of the ZSCORE in our exercises. Several advantages of using the ZSCORE have been listed. Boyd and Graham (1986) and Strobel (2011) point out that ZSCORE does not require strong assumptions about the distribution of assets. Chiaramonte et al. (2016) also note the simplicity in the computation of the measure as an important advantage but, more importantly, report on the ability of ZSCORE to forecast bank failures with an accuracy of 76% as also the ability to predict bank defaults three years in advance. The ZSCORE has become a very commonly used measure of bank risks. Morgan and Pontines (2014) have used it in their study of 193 countries listed in the Global Financial Development Database of the World Bank (World Bank, 2020) which covers countries in all continents. Finally, many studies (Chen & Lin, 2010; Bian et al., 2015; Baek et al., 2018; Ahamed, 2017; Pham et al., 2021) related to banks in Asia make use of the ZSCORE to measure bank risk when studying the effects of income diversification, which makes it possible to compare our results with these studies. Clearly, the ZSCORE has been used as a measure of bank risk/stability in diverse countries with widely differing institutional settings.

Table 1 lists out all the independent variables used in our exercises. The first group defines the financial variables, the second defines macroeconomic variables and the third defines policy variables.

Our variable of interest in the determination of ZSCORE is NII. Engle et al. (2014) note that nontraditional business activities could permit banks to circumvent capital regulations and allow increased risk-taking that traditional activities do not permit. This is likely to create agency problems that have been studied extensively in the literature. The reason why risk-taking increases is that non-interest income is often more volatile than interest income but not necessarily more profitable. (Stiroh, 2004; Stiroh & Rumble, 2006). If this is true, the coefficient of NII will likely be negative and make the banks more risky in the presence of rising NII.

Total assets (TA) is included as an independent variable to proxy bank size (Nguyen, 2012). Bank size allows availing of advantages of economies of scale and earn higher incomes and also provide protection from negative shocks during recessions (Salike & Ao, 2018) The annual growth rate of assets (TAGR) is expected to have a positive impact on the risks faced by banks (Ahamed, 2017). In order to control for the asset structures in each bank, the ratio of loans to total assets (LOANRATIO) is included since it detects differences in the asset portfolios of banks (Sanya & Wolfe, 2011; Stiroh & Rumble, 2006). Equity capital to total assets ratio (EQUITYRATIO) is expected to affect risk by allowing banks with higher capital amounts to absorb any negative shocks. Loan loss provisions to total assets ratio (LLPRATIO) is a measure for the quality of loans in a bank. Banks could use loan loss provisions as a means to smooth their revenues (Ahamed, 2017). The ratio of deposits to total assets (DEPOSITRATIO) is also included as control variable following Lee et al. (2014). Finally, ratio of non-performing loans to total loans (NPL_TL) represents asset quality following Salike and Ao (2018).

In addition to bank specific financial variables, we also include a few macroeconomic variables in our equations. As noted by Salike and Ao (2018), macroeconomic variables are factors that are not within the control of the banks’ management. We include two of the macroeconomic factors mentioned by Engle et al. (2014) and by Salike and Ao (2018), namely, rate of growth of GDP (GROWTH) and rate of inflation (INFLATION). High GDP growth tends to increase the demand for bank loans and will likely improve bank profitability and reduce bank risks. However, the effect of inflation is ambiguous (Salike & Ao, 2018). This is further elaborated by Dhal et al. (2011) who term the relationship between financial stability and inflation to be a very “contentious issue” and point out that financial stability may be affected only when inflation rises above a threshold level. In our analysis, we take the average rate of inflation during a given time period as the threshold level and compute the gap between actual inflation for each year and the average inflation for the relevant time period. Since we will be estimating separate regression equations for two time periods, one for before the GFC and one after the GFC, we compute the average rate of inflation for each country separately for each time period and then compute the gap for each year.

We also include two more macroeconomics variables which we call policy variables. Monetary policy is operated by the central bank of a country and a variety of instruments are used to operationalise it dependening on the objectives (Filardo & Genberg, 2010; Morgan, 2013). Given that the monetary policy instruments differ across countries, it is difficult to capture measures of the policy in one representative variable. However,it is possible to consider a proxy measure. As stated by Mathai (2020), monetary policy is understood as the central bank seeking to change the size of the money supply. Albertazzi et al. (2021) further note that the traditional transmission channel of monetary policy depends on the central bank’s control over the level of deposits via reserve requirements and the money multiplier. Hence, we take the ratio of broad money to GDP (MONEY) as a proxy measure of the monetary policy variable. Fiscal policy is the other means by which governments seek to manage the economy and involves the use of government spending and taxation (Horton and ElGanainy, 2020). We include in our analysis the ratio of government expenditure to income (GOVT_EXP) as a fiscal policy variable.

Our data on financial variables have been extracted from CapitalIQ (2020), a comprehensive database for banks while data on macroeconomic variables have been sourced from World Development Indicators of the World Bank (undated). The distribution of banks and the number of observations across the 24 countries is given in Table 2. It will be be noticed that we have defined the first group of countries as the broad Asia–Pacific region. The Asian Development Bank (1965) include Australia and New Zealand as regional members in its Charter. However, when we estimate equations for this group we do so with and without Australia and New Zealand. As in Salike and Ao (2018), the banks used in the study were a mix of commercial banks, savings banks, Islamic banks, housing and mortgage banks, rural banks and holding finance companies.

It can be seen from Table 2 that, for each country, we have a panel dataset and the entire dataset is made up of a panel of panels of 1122 banks with data spread over twenty-five years from 1996 to 2020. It must be mentioned that we do face the problem of missing data for some of the variables that we employ in our models. Hence, we are forced to work with an unbalanced panel and the total number of effective observations is much less than 28,050. It may be noted that we have divided the countries in our dataset into three sub-groups based on the categorization followed by Capital IQ.

Our empirical strategy involves estimating dynamic panel data models employing the two-step system GMM estimator. We estimate two equations-one for 1996–2007 and the other for 2008–2020-for (a) all countries together (b) group of Asia/Pacific developed countries excluding Australia and New Zealand (c) group of Asia/Pacific developed countries including Australia and New Zealand (d) group of East and South-east Asian countries and (e) South Asian countries.

Having estimated equations for all countries in our dataset as well as for sub-groups of these countries, we now bring together all our results for the purpose of comparison. We present in Fig. 1 a map of all the countries included in our data set (created using the tool at mapchart.net) which also colour codes the sub-groups of countries that we have created. Embedded in Fig. 1 are the summarised results of Eqs. (1) and (2) from Table 4 for all countries.

In Table 11, we report on the results of the sub-groups of countries. In this table, we include the results of Eqs. (3) and (4) from Table 6, Eqs. (5) and (6) from Table 8, Eqs. (7) and (8) from Table 10 and equations (A1) and (A2) from Appendix Table 12. Since detailed results are available in earlier tables, in Table 11 we report only the significance or otherwise of the coefficients and, if signficant, their signs are also reported.

The differences in results reported in Table 11 justifies our approach of estimating equations for various sub-groups of countries. This diversity would have been lost had we confined ourselves to the overall group of 24 countries. One result that carries across all sub-groups is the one related to NII. Our results clearly show that in the post-GFC period, diversifying income (as captured by NII) reduces the stability of the banks. For South Asia, this results holds in the both the time periods. Size of banks as reflected by the variable TA has an ambiguous relationship with bank stability. IMF (2014) distinguishes between individual bank risk and systemic risk. The perception that large banks are “too big to fail” and will be bailed out leads to a moral hazard problem leading the banks to take more risks. Our results seem to bear this out for the group of All Countries and Asia/Pacific Developed Countries (A and B) in the post-GFC period. This is not surprising since in the Asia/Pacific Developed Countries, top four banks account for a very large market share in terms of Assets: in almost all countries, the top four banks account for more than 50% market share with Australia the highest at 77% (Dahl et al., 2019). IMF (2014) also states that large banks are able to diversify risks better which reduces risks and hence TA will be positively associated with ZSCORE.  This appears to be true for South Asian banks in both the time periods and Asia/Pacific Developed (A) in the pre-GFC period. Further, rate of growth of TA (TAGR) plays a role only in the South Asia group.

Among the other bank-specific variables, LOANRATIO plays a negative role in Asia/Pacific Developed (A) Countries and East and South-east Asian Countries but a positive role in South Asian countries. Kohler (2012) has reported that higher credit growth (which will be reflected in the LOANRATIO) increases bank risk. However, in the largest economy in South Asia, namely, India, the there is a history of bailout of banks by the government when loans turn bad (Parkin & Kazmin, 2020). This safety net results in a contrary relationship between LOANRATIO and bank risk in South Asian countries. LLPRATIO is non-significant in most country sub-groups while DEPOSITRATIO is negative in the post-GFC time period when all countries are considered but positive for East and South-east Asian countries. EQUITYRATIO is positive for many sub-groups of countries in the post-GFC period. Finally, NPL_TL has a negative impact on bank stability except for South Asian countries in the pre-GFC period.

Both the macroeconomic variables, GROWTH and INFLATION, are signficant for the All Countries group with GROWTH improving bank stability and INFLATION lowering it. It is important to note that when all countries are considered together, both the macroeconomic variables show consistent results before and after the GFC but the effect of these two variables is not seen when we consider sub-groups of countries. A possible explanation for this is that, when all countries are considered, there is sufficient variablity among the countries as far as GROWTH and INFLATION is concerned which is not the case when sub-groups of countries are considered. For instance, the standard deviation of GROWTH when all countries are taken together is higher than for each of the sub-groups and likewise for INFLATION barring East and South-east Asian countries which show higher variability. Among the policy variables, MONEY is seen to be positive at the All Countries level in both the time periods but is negative for Asia/Pacific Developed (A) countries and South Asian countries in the post-GFC period. GOVT_EXP is marginally significant in the countries of East and South-east Asia and South Asia.

The role of noninterest income in determining bank risk has been an active area of research especially for the USA and for Europe. By and large, studies have found that diversification away from traditional sources earnings has worsened bank risk (this has generally been true of studies related to the USA) while others have found the reverse (this has been true of studies related to European banks). This paper was concerned with examining bank risk for Asian banks. We have provided arguments that banking in Asia needs to be explored more deeply in view of significant differences between American/European banks and Asian banks.

Studies on Asian banks have not yet provided a coherent view on the impact of non-interest incomes on bank risk. While some studies have reported a worsening of bank risk due to income diversification towards non-interest incomes others have reported a lowering of risk. The latter studies have called for a further diversification of bank incomes as a way of improving bank stability. Our results cast doubt on this policy prescription. Further, our exercises have demonstrated the usefulness of adopting a granular approach to estimating bank stability. Confining our estimation to the aggregate level would have likely prevented us from examining differences across regions. We have also been able to demonstrate that we need to take account of the global GFC of 2008 in studying the impact of NII on bank risk given that this impact certainly changed as compared to the time period before the GFC.

To summarise the main result of our paper, we can say that there is reasonable unanimity across groups of countries that diversification of income hurts bank stability. It is interesting that, despite differences among these groups and among the countries within the groups, there is a clear message that an emphasis on NII does not render banks more stable. This result has become stronger in the post-GFC time period. It is also seen from the summary statistics that in all the groups the level of NII has come down in the post-GFC period which would suggest that banks would have gained in stability.

There are some important directions in which this research needs to be extended further. The role of macroeconomic variables needs to be explored further. We have stated that inflation above a certain threshold is likely to impact bank stability but the exact threshold for every group of countries or, indeed, every country needs to be determined. The role of policy variables, especially monetary policy, needs further explication especially since there is a great diversity among countries regarding the conduct of this policy. Finally, the most signficant direction in which research needs to extended is to explore the implications of the COVID-19 pandemic on bank stability. Since the full effects of the pandemic are still being played out in many countries, it might need some time for reliable data to be generated for such analysis to be carried out.