Do the diversification of income and assets spur bank profitability in Bangladesh? A dynamic panel data analysis

Purpose – This paper aims to explore the effects of bank diversification (i.e. diversification of income and diversification of assets) on Bangladeshi banks’ profitability. Design/methodology/approach – Using a dynamic panel data model with system generalized methods of moments, the authors examine an unbalanced panel data from 32 banks spanning 318 bank-year observations from 2007 to 2016. Findings – The findings indicate a significant positive association of income diversification and asset diversification on bank profitability. Therefore, the results show that banks can generate profit from diversification of income and diversification of assets. Originality/value – One of the rare attempts to investigate the relationship between diversification and profitability in Bangladesh’s banking sector is this report. The authors anticipate the results to have major consequences for Bangladeshi bank regulators and other related economies.


Introduction
The financial sector, especially the banking sector in developing countries as well as in developed countries, has undergone major changes over the past few decades. Because of globalization, growing competitiveness and deregulation, banks have strengthened their attention to diversified business lines in addition to conventional bank interest-based operations. Diversification is seen as part of the extreme gap in bank behavior in the postcrisis era, as a strategic choice dictated by the macroeconomic climate shifts (Hidayat et al., 2012;Nguyen et al., 2012a;Lin et al., 2012). Nowadays, owing to the evolving business scenario, noninterest-based activities are more popular among banks in earning a profit. By incorporating new revenue streams or new assets, such as investment provision, brokerage services, stock trading and underwriting services, the bank will diversify its asset portfolio (Meslier et al., 2014). Most previous research shows that noninterest income (NII) not only yields a higher return; it also causes danger because of its volatility (Moudud-Ul-Huq et al., 2018). By investigating South Asian banks, Nguyen et al. (2012b) suggest that if banks diversify their revenue from all interest and noninterest streams, they will become more resilient. Assets diversification reduces profits and increases costs (Berger et al., 2010). Edirisuriya et al. (2015) suggest that banks' efficiency because of asset diversification is not enhanced. Previous research has shown that the correlation between diversification and bank profitability is inconclusive. Therefore, we have implemented two research questions on the Bangladeshi banking industry: What are the impacts on bank profitability of income diversification? Is there any significant relationship between the diversification of assets and the profitability of banks?
The earlier diversification research concentrated primarily on developing countries (i.e. the USA and Europe). Very few studies have found on emerging economy and provide different results (Moudud-Ul-Huq et al., 2018). Deesomsak et al. (2004) contend that Asian banks are a critical source of financing for private sector business activity. In addition, the Asian region's banks have been hit by the financial crisis as well, and consolidation projects are underway in several countries. In addition, it has been found that there is a shortage of study on Asian banks. Lee and Hsieh (2013) suggest that in the Asian regions, the banking sector is an exciting and critical research laboratory. Thus, by taking into account the lack of analysis on the Asian economy, this study considered Bangladesh as a key laboratory to explore the effect of income diversification and asset diversification on bank profitability. The latest literature indicates that several scholars have sought to study the determinants of bank profitability in Bangladesh (Rahman et al., 2015;Sufian and Habibullah, 2009;Robin et al., 2018;Sufian and Kamarudin, 2012;Hossain and Ahamed, 2015;Noman et al., 2015;Majumder and Rahman, 2017;Majumder and Li, 2018;Akter, 2017). However, to the best of our understanding, this is a specific study that considers bank diversification as an independent variable and a significant profitability determinant for Bangladesh's banking sector. For the duration 2007-2016, we use an unbalanced panel data collection of 32 Bangladeshi commercial banks, which provides a total of 318 bank-year observations. For the study, we used a system generalized methods of moments (GMM) regression method with a dynamic panel data model. According to the findings of this empirical research, income and asset diversifications in Bangladesh's commercial banks are significant positive factors in improving profitability of the banks.
In many aspects, the present research adds to the current literature. First, we are analyzing a sample of 32 Bangladeshi banks over the period 2007-2016. To the best of our comprehension, this is the unique research on this subject that uses such a rich collection of data on Bangladesh. Second, most recent research has concentrated primarily on the relationship between NII and risk. Still, very little attention has been given to the relationship between NII and profitability, as well as noninterest-bearing assets and profitability. Third, we use four bank profitability proxies to ensure the robustness of the outcomes that are unusual in methodological literature. Fourth, the analysis used a system GMM regression method with a dynamic panel data model. In contrast, the relevant research focusing on Asian countries are largely centered on a static panel data model. The research would eventually supplement other studies in Asian countries (Nisar et al., 2018;Lee et al., 2014b;Ahamed, 2017;Berger et al., 2010;Nguyen et al., 2012b).
This report is structured as follows. The study of literature is illustrated in Section 2. The research methodology is described in Section 3, whereas the empirical findings are addressed in Section 4. Conclusion and policy implications are given in Section 5.

Literature review 2.1 Theoretical underpinnings
The most significant theory of this study is the "portfolio theory." Harry Markowitz is the pioneer of the portfolio theory. The theory originated from Markowitz's (1952) write-up regarding "portfolio selection" and "foundations of portfolio theory" (Markowitz, 1991). The main concept of "portfolio selection" is that the investors should diversify their funds among those securities which give maximum anticipated return and minimum variance. Under this assumption, the law of large numbers to a portfolio of securities cannot be accepted. The yields from securities are highly intercorrelated. Diversification cannot mitigate all variance (Markowitz, 1952). The modern portfolio theory is a philosophy for risk-averse investors to build their portfolio to increase or optimize their anticipated rate of return with a specified amount of market risk. To improve or increase the anticipated benefit of a portfolio, it is only important to invest in one security (Markowitz, 1991). The portfolio theory says, diversified banks avail from economies of confine which refine the performance and reduce risk simultaneously (Nisar et al., 2018). By referring portfolio theory, Elsas et al. (2010) argue that banks will benefit from economies of scale, which gradually decreases the risk of banks and increases their profitability. Another group of researchers say banks can reduce their risk by diversifying their assets into different geographic areas or diversifying their revenue sources . According to the portfolio theory, banks may face risk diversification rewards if NII runnels are uncorrelated with interest income. Further, banks may get a higher risk if NII brooks are riskier and have a high correlation with interest income (Moudud-Ul-Huq et al., 2018).

Empirical review
In this section, this study reviews the existing literature on the impacts of diversification on bank profitability. There is a puzzle about whether diversification improves bank profitability. For example, Boyd et al. (1993) claim that the diversification cost may outweigh the advantages if the diversified activities are riskier than the traditional banking activities. Another research using the sample of developed and developing countries, Doumpos et al. (2016), concludes that developing countries can gain more advantages from revenue diversification than developed countries. The prior literature indicates two types of bank diversification: income or revenue diversification and asset diversification.
Concerning to the impacts of income diversification on bank profitability, the existing literature shows mixed results. For example, Nisar et al. (2018) examine the impact of revenue diversification on bank profitability on eight South Asian countries and conclude that NII positively impacts profitability. Another recent study by Moudud-Ul-Huq et al. (2018) investigates association of southeast asian nations (ASEAN's) emerging economies and finds that the diversified banks have lower risks and higher performance. Lee et al. (2014b)

examine 22
Asian countries that show that NII reduces risk and increases bank profitability in the middle and low-income countries while raising the risk for the high-income countries. Baele et al. (2007) highlight the positive impacts of income diversity on the firm's long-term value by examining European banks. By investigating Italian banks, Chiorazzo et al. (2008) find that income diversity increases bank risk-adjusted returns. Using data from Canada, Australia, Italy, France, Germany, Switzerland, Spain, the USA and the UK banks, Elsas et al. (2010) find that diversification improves bank profitability. Applying 310 yearly observations from a total sample size of 31 Kenyan commercial banks using panel data for the period of 2008-2017 Githaiga (2020) found a significantly positive relationship impact on income diversification and market power with bank performance. Using a sample size of 31 commercial banks and panel data for the period 2008-2017; Githaiga et al. (2019) found the positive significant of income diversification on financial performance. The French 412 financial institutions over 10 years ' (2002-2012) effect of income diversification on profitability is seen as significantly positive (Jouida, 2018). The effect of diversification on profit efficiency and cost of six commercial banks from ASEAN countries over the period 2007-2014 is a justified positive relationship (Nguyen, 2018). The higher portion of net interest income (NII) generates higher profits and risk-adjusted profits and NII diversification tasks improves the bank profitability (Ahamed, 2017). An empirical study on nationalized banks in Bangladesh for period of 2010-2014 found that the effect of income diversification on bank profitability is found significantly positive (Majumder and Uddin, 2017). The study on unbalanced panel data over the period of 2007-2016 of listed commercial banks in Vietnam stock market investigated that income diversification reduces the insolvency risk and simultaneously enhance the bank performance (Nguyen and Hong, 2017).
Another group of studies reports negative impacts of bank income diversification on profitability. For example, a recent study showed that the relationship among revenue diversification, risk and bank performance among 26 commercial banks in Vietnam over the period 2010-2018 is negative (Githaiga, 2020). The period of 2000-2012, the research on G7 and BRICS (Brazil, Russia, India, China, and South Africa) country's result of revenue diversification, decreases the bank's efficiency (Hu, 2018). Another analysis conducted in European banks from 2002-2012 showed that share of NII had a detrimental impact on bank profitability (Maudos, 2017). Borroni and Rossi (2017) have found using 110 extensive sample size data on eight European Monetary Union (EMU) countries throughout 2005-2013 commercial, saving and cooperative banks, revenue diversification reduces the bank performance. Stiroh (2004) investigates the US community bank and finds a negative association between diversification and bank profitability. Using unbalanced panel data on 88 Chinese banks, Berger et al. (2010) argue that diversification increases costs and deteriorates profitability.
Furthermore, some studies use two types of diversification measures, such as income diversification and asset diversification. For example, the study of Moudud-Ul-Huq et al.
(2018) differentiates between revenue and asset diversification. Their study results conclude that the effects of asset diversification vary from one country to another country. Most of the studies based on the US economy find that assets diversification increases bank performance, such as Hughes et al. (1999) and Deng et al. (2007). The outcome of asset diversification on bank performance in three Asian countries (Indonesia, Malaysia and Pakistan) with a dual banking system from 2006 to 2012 is positive (Chen et al., 2018). Other studies find negative impacts of asset diversification on profitability (Acharya et al., 2006;Hughes et al., 1996;Rose, 1996). Using listed public banks in south Asian countries, Edirisuriya et al. (2015) explore that diversification of assets does not improve bank performance. The income or asset diversifications do not increase the bank stability in Gulf Cooperation Council (GCC) countries is investigated on data period 2001-2014 (Abuzayed et al., 2018). XIMB Thus, the previous study findings were inconclusive. Different diversification measures are researched, e.g. revenue and assets diversification, and different diversification has different impact on bank profitability. Table 1 describes the prior literature more scientifically: 3. Research methodology 3.1 Data and sample For this analysis, the sample size consists of 32 commercial banks listed on Bangladesh's Dhaka Stock Exchange (DSE). Now, in Bangladesh, a total of 57 scheduled banks are operating. But, because of data unavailability, the analysis omitted 25 banks. Some banks are still recent, hence being omitted from the report. Our final data set involves unbalanced panel data representing a total of 318 observations from 2007 to 2016. Data were extracted from each bank's audited financial statements, accessible on both the bank's website and the DSE. The study also uses macroeconomic control variables that are available in the World Bank database [1]. In addition, we use numerous books, papers and Web resources.

Model specification, variables and estimation techniques
The empirical model specification is as follows: [ it indicates the idiosyncratic error term. l and b j represent the coefficients to be estimated. The definitions of all the variables and their references are represented in Table 2.
The above dynamic panel model is estimated by using GMM estimator developed by Arellano and Bover (1995) and Blundell and Bond (2000). We use a one-step system GMM rather that two-step system GMM because the former one produces a smaller standard deviation of the estimation and smaller bias (Judson and Owen, 1999). The advantage of using GMM is it avoids autocorrelation and heteroskedasticity problems in the model, and it gives better results compared to ordinary least squares (OLS).

Descriptive statistics
The descriptive statistics of all the study variables are presented in Table 3. The table shows that the minimum value of bank profitability (ROA, ROE, SHROA and SHROE) indicates a negative figure, which is due to negative earnings of some state-owned banks (see Rupali Bank Ltd.  -2007, 2008 and 2009). Those state-owned banks are still operating because of support from the Government of Bangladesh. Interestingly, we find that the average ROA of Bangladeshi banks is 1.1%, which is higher than 0.93% as   Table 4 displays the Pearson correlation coefficient matrix, which shows that the maximum correlation between the independent variables is 0.67 between liquidity (LIQD) and diversification of assets (ADIV). Gujarati (2009) indicates that when the two independent variables' association value exceeds 0.80, multicollinearity is a serious concern. The analysis, therefore, indicates no serious multicollinearity issue in the interpretation of the effects of regression.

Regression analysis 4.3.1 Effects of bank diversification on profitability.
The empirical results of bank diversification's effect on profitability are listed in Table 5. Here, as a dependent variable, we use four proxies of bank profitability such as ROA, ROE, SHROA, SHROE in Models 1 to 4 and 5 to 8, respectively. As independent variables, the analysis uses two proxies of bank diversification. Income diversification is seen in Models 1 to 4 and Models 5 to 8 for asset diversification. The significant positive coefficient of lagging dependent variables is expressed by all models, confirming the degree of persistence in all models and the model's dynamic character.
The findings of the analysis indicate that income diversification (IDIV) has a positive and significant effect on the profitability of Bangladeshi commercial banks in all models, suggesting that more diversified non-income generating activities contribute to higher profitability and the results supported by the research by Jiang et al. (2003). We also find that diversification of assets has significant positive effects on profitability, which means that more value is produced from more noninterest-bearing assets. This is aligned with that of Edirisuriya et al. (2015).
We see a positive correlation between liquidity and bank profitability with respect to the control variables, except in Models 5 and 8. The high ratio means that a large amount of loan sanction to the customers produces more interest revenue, thereby growing profitability. In Models 1, 5, 6 and 7, the Equity-to-Total Assets Ratio (EQTA) reveals a strong positive correlation with profitability, suggesting that the higher the bank's equity capital produces Table 4. Pearson correlation matrix Notes: Total number of observations 318; ***correlation is significant at 1% level (two-tailed); **correlation is significant at 5% level (two-tailed); *correlation is significant at 10% level (two-tailed); all variables are winsorized at the 1% level   Notes: The estimation technique is a one-step system GMM dynamic panel estimators. The dependent variable is the profitability measured by return on assets (ROA), return on equity (ROE), risk-adjusted return on assets (SHROA) and risk-adjusted return on equity (SHROE) in Models 1 to 4 as well as in Models 5 to 8, respectively. Income diversification (INDIV) is the independent variable in Models 1 to 4, whereas asset diversification (ADIV) is the independent variable in Models 5 to 8. *, ** and *** denote significance at 10%, 5% and 1% levels, respectively. Robust standard errors are presented in parenthesis. a Test of overidentifying restrictions (Ho: over-identifying restrictions are valid). The tests accept the null hypothesis that over-identifying restrictions are valid. b

Arellano-
Bond test for the first-order autocorrelation (Ho: no autocorrelation). c Arellano-Bond test for the second-order autocorrelation (Ho: no autocorrelation). The test results of AR (1) and AR(2) indicate there is autocorrelation exists in the first-order but not in the second-order. All variables are winsorized at the 1% level Table 5. Zheng et al. (2017), Casu et al. (2016) and others. Bank size (BSIZE) is significantly negatively related to profitability, suggesting that big banks have management problems relative to smaller banks, thus reducing profitability. This findings in line with Tan (2016) and Majumder and Uddin (2017). The study showed that credit risk (NPLTL) negatively impacts bank profitability; it decreases profitability by suggesting low credit quality. In this case, the results of Almekhlafi et al. (2016), Zhang et al. (2013) and Lin et al. (2005) found identical findings, although not compatible with Naceur and Omran (2011). Cost control (the overall cost-to-total asset ratio [CMGT]) reveals a strong negative profitability correlation. The low ratio implies cost accounting that is efficient; this proof is in line with Rahman et al. (2015). The analysis findings indicate that concentration (Herfindahl-Hirschman Index [HHI]) is important and positively related to bank profitability, which is consistent with this research (Tan and Floros, 2012a). The analysis uses two variables of macroeconomic influence, such as GDP and inflation. The significant positive relationship between GDP and bank profitability shows that the higher the GDP growth in Bangladesh, the higher the profitability. The proof is compatible with this study (Tan, 2016). With respect to inflation, the findings show that all models except Models 1 to 3 have a substantial positive effect on profitability, which is confirmed by these analyses (Tan and Floros, 2012b;Majumder and Uddin, 2017). 4.3.2 Robustness of results. By flipping the regression approach from GMM to OLS, we have conducted a robustness examination. As can be seen in Table 6, the expected significant positive association between income diversification (INDIV) and asset diversification (ADIV) and various bank profitability measures has been shown. We noticed a similar association with profitability as in previous approaches in Table 5 with respect to the control variables, which confirm the robustness of our findings.

Summary and policy implications
This article explores the effect on bank profitability in Bangladesh of income diversification and asset diversification. The research uses a one-step GMM framework approach with a dynamic panel model for 32 Bangladeshi commercial banks over the period 2007-2016 for empirical review, producing a total of 318 bank-year observations. The analysis also uses OLS regression for the verification of robustness. As a measure of the dependent variable, we use four bank profitability proxies (ROA, ROE, SHROA and SHROE) and two diversification proxies (income and asset diversification) as a measure of the independent variable. The other control variables such as liquidity, capitalization, bank size, credit risk, cost management, concentration, GDP and inflation are also considered for the analysis.
The study results show that the diversification of income and assets has a significant positive effect on the profitability of commercial banks in Bangladesh. Liquidity, capitalization, concentration and GDP provide a substantial positive effect on profitability in control variables. Bank size, credit risk, cost management and inflation, on the other side, have major negative ties to bank profitability.
Our studies have essential policy consequences for Bangladeshi banking sector regulators and administrators, as well as for developing and emerging economies. The study indicates that banks still generate the majority of income from traditional interestgenerating outlets. Banks should also search at other revenue streams, such as fee income and noninterest-bearing assets, as these can be significant and helpful to the banks' profitability.

Limitations and directions for future research
This analysis is focused on secondary data obtained from 32 Bangladeshi commercial banks and excludes other financial and nonfinancial nonbank organizations. In addition to the banking industry, the prospective researcher may identify other industries or undertake cross-country study to analyze diversification's success impacts. The report uses two forms of diversification, i.e. diversification of income and diversification of assets. The potential researcher should then suggest other diversification forms, such as diversification of bank financing, product diversification, regional diversification and functional diversification. The current study aims to guide the interaction between the dependent and independent variables only; therefore, a mediator or moderator can be identified by future researchers on the correlation. Notes: The estimation technique is OLS estimators. The dependent variable is the profitability measured by return on assets (ROA), return on equity (ROE), risk-adjusted return on assets (SHROA) and risk-adjusted return on equity (SHROE) in Models 1 to 4 as well as in Models 5 to 8, respectively. Income diversification (INDIV) is the independent variable in Models 1 to 4, whereas asset diversification (ADIV) is the independent variable in Models 5 to 8. *, ** and *** denote significance at 10%, 5% and 1% levels, respectively. Robust standard errors are presented in parenthesis. All variables are winsorized at the 1% level XIMB Note