FACTORS SHAPING CAPITAL STRUCTURE: EVIDENCE FROM PRIVATE COMMERCIAL BANKS IN BANGLADESH

This study investigates how various determinants shape the capital structure of commercial banks in Bangladesh, employing panel data, structured from available secondary sources, covering 22 banks as samples from 61 scheduled banks for the period of 2011 to 2020, conducting Feasible Generalized Least Squared (FGLS) Regression Model. Several diagnostic tests have been conducted to ensure the robustness and stability of the model. The study results reveal that return on assets, earnings per share, asset growth, asset structure, investment structure, cost per loan assets, and loan loss provisioning considerably influence the capital structure or the leverage of commercial banks. On the contrary, the authors find no explicit evidence that bank size, liquidity, capital adequacy, and non-performing loan ratio have significant impacts on the capital structure of the banking industry of Bangladesh. The findings of this study advocate that return on assets, earnings per share, asset structure, and cost per loan assets as the dominant explanatory factors of capital structure. Besides, asset growth, investment structure, and loan loss provisioning affect less significantly on determining the capital structure of the banking industry. This study also brings the academicians, researchers, and analysts with corroborating new routes for exploring further research in this field.

H1 = There is no significant relationship between bank size and capital structure.

Liquidity
How shorter period is recommended to convert assets into cash determines the provision of liquidity, and is one of the significant issues shaping the capital structure (Režňáková, 2010), measuring the institutional capability of redeeming short-term obligations. Liquidity and investments are on different sides of the same coin (Lalon & Naher, 2020). Trade-off theory suggests that higher liquidity navigate higher opportunities to capitalize debt instruments (Al Shubiri, 2011;Ukaegbu & Oino, 2014;and Teixeira et al., 2014). Besides, Ukaegbu and Oino (2014) administered a study based on the data of 2004 to 2008 in Nigeria also supported the previous findings. On the contrary, Amidu (2007), based on the data of 19 banks, argued that liquidity incurred financial stress, which had driven limited leverage opportunity. Prior evidence focused on significant positive influences of liquidity on leverage (Bukair, 2019); conversely, some evidence revealed a significant negative relationship (Güner, 2016;Hossain & Hossain, 2015;Režňáková, 2010;Sakunasingha et al., 2018;Shah & Jam-e-Kausar, 2012;and Ullah et al., 2017). Extracts from the studies shape the following hypothesis: H4 = There is no significant relationship between liquidity and capital structure.

Asset Growth
High growth rate conveyed high financial costs, confirmed by trade-off theory (Al-Hunnayan, 2020) by examining a study based on 12 Islamic banks in the GCC for 10 years of data. On the contrary, pecking order theory opposed to that said that high growth rate demonstrated low costs of bankruptcy, leading to constructive affiliation between growth rate and leverage (Kayo & Kimura, 2011). In the same vein, Režňáková (2010) substantiated that non-financial companies with high growth corroborated with the high option to debt financing, experienced from the study based on 6 years data in Slovakia. Sakunasingha et al. (2018) deduced that the growth rate of assets had significant impacts on leverage ratio evidenced from the Fixed Effects Model using the data of 14 local banks in Thailand. Growth had a significant positive influence on leverage (Al-Hunnayan, 2020;Cassar & Holmes, 2003;Frank & Goyal, 2009;Kayo & Kimura, 2011;Manos & Ah-Hen, 2003;Noulas & Genimakis, 2011;and Sakunasingha et al., 2018), contradicting the studies (Hossain & Hossain, 2015;Režňáková, 2010;and Sheikh & Quereshi, 2017) who found significant negative impacts on leverage. Besides, fewer studies https://www.cribfb.com/journal/index.php/ijafr International Journal of Accounting & Finance Review Vol. 9, No. 1;2021 preferred to conduct growth as explanatory variables (Güner, 2015;Kayo & Kimura, 2011;Shah & Jame-Kausar, 2012;and Ukaegbu & Oino, 2014). Based on the stated literature, the subsequent hypothesis was derived: H5 = There is no significant relationship between asset growth and capital structure.

Asset and Investment Structure
The structure of assets complemented the proportion of debt and equity against total assets. Amidu (2007) documented that asset structure played an essential character in the capital structure of banks, examining a study based on 6 years of data from 19 banks in Ghana. In the same context, Shah and Jame-Kausar (2012) appraised that investment was one of the contributing factors of capital structure. Agency costs were depressed by using assets as security against the issuance of debt instruments (Stulz and Johnson, 1985). There was a substantial constructive connection between asset structure and leverage (Ukaegbu & Oino, 2014). On the contrary, Manos and Ah-Hen (2003) highlighted a substantial negative connotation. In the case of investment structure, relevant evidence is scant; instead, some findings enumerated a significant relationship between investment and capital structure (Shah & Jam-e-Kausar, 2012) who researched from 2003 to 2008. Extracts from the reviews, the following hypotheses are developed: H6 = There is no significant relationship between total equity to total assets and capital structure. H7 = There is no significant relationship between total debt to total assets and capital structure. H8 = There is no significant relationship between total loans and advances to total assets and capital structure.

Cost per Loan Assets and Capital Adequacy
Cost per loan assets measures the cost per unit loans and advances provided to the borrowers of banks.
Operating costs are distributed among all sorts of loans and advances. Management of loans and advances upheld the operating costs of commercial banks (Berger & DeYoung, 1997). Banks contended to alleviate the operating costs to ensure performance (Gadzo et al., 2019). Basel Accords deduced minimum capital requirements to be kept against risk-weighted assets. Capital adequacy is employed to fight against credit, market, and operational risks (Ukaegbu & Oino, 2014). Instead, maintaining capital adequacy is not an alternative to mitigate risks inherent in monitoring (Kahane, 1977). It was observed negative association between capital adequacy and loan disbursement (Altman et al., 2002). Some banks were restructured through mergers due to limited capital requirements (Peek & Rosengren, 1995). Using balanced panel data, Pooled OLS and Random Effects model exposed a significant relationship between minimum regulatory capital and leverage for 2004 to 2008 in Ghana (Ukaegbu & Oino, 2014). It has tried to reveal a few related studies on cost per loan assets and capital adequacy as independent variables for determining the dependent variable. The succeeding hypotheses are: H9 = There is no significant relationship between cost per loan assets and capital structure. H10 = There is no significant relationship between capital adequacy and capital structure.

Non-performing Loan Ratio and Coverage Ratio
The non-performing loan ratio is a dominant tool to evaluate financial performance (Vatansever & Hepsen, 2013). The quality of the assets of the bank was assessed through this NPL ratio which was scrutinized to reduce the negative influence of NPL on assets (Irawati et al., 2019). Van Greuning and Bratanovic (2020) indicated that credit risk was the prime reason for failure in banks and increased nonperforming loans. These factors were negatively related to the banks' performance to a large extent. The risk-taking nature of banks was hindered by NPL volume (Lestari, 2018). To reduce the loss, every bank maintains a coverage ratio where provision on loan loss is taken as per the volume of NPL. Loan loss provision affects the bank's profitability. A well-organized bank enables to manage higher profitability to keep lower loan loss provision. The high volume of NPL deteriorates the credit policy (Bhattarai, 2018). However, this study develops hypotheses to relate NPL ratio and coverage ratio to the capital structure of banks. H11 = There is no significant relationship between non-performing loans to total loans and advances and capital structure. H12 = There is no significant relationship between loan loss provision to total non-performing loans and capital structure.

RESEARCH METHODOLOGY Sample and Data
The research is intended to analyze the factors that affect or have significant impacts on the capital structure of the banks. Bangladesh has 61 scheduled banks, out of which there are 33 conventional private banks (Bangladesh Bank, 2021). However, 22 private conventional commercial banks are taken as samples based on data availability and annual reports for conducting the research. This study deals with 10 years of data from 2011 to 2020, thereby constructing a panel dataset of 220 observations collected from available secondary sources, including annual reports, websites, academic papers, journals, and original newspapers.

Framework and Measurements of the Variables
After analyzing the literature, twelve variables are selected as independent or explanatory variables, which are assumed to have significant effects on the dependent variable. The dependent variable is total debt to total equity (TDTE), which measures capital structure or leverage or how many times total debt is compared to total equity of banks. Independent variables include bank size proxied by total assets, profitability proxied by net profit after tax to total assets (ROA) and net profit after tax to the number of outstanding shares (EPS), liquidity proxied by total loans and advances to total deposits (TLATDE), asset growth proxied by the growth rate of assets of the banks (GRTA), asset structure proxied by total equity to total assets (TETA) and total debt to total assets (TDTA), investment structure proxied by total loans and advances to total assets (TLATA), cost per loan assets proxied by operating cost to total loans and advances (CLA), capital adequacy proxied by (tier 1 capital + tier 2 capital) to risk-weighted assets (CAR), non-performing loan ratio proxied by nonperforming loan to total loans and advances (NPLR), and coverage ratio proxied by loan loss provision https://www.cribfb.com/journal/index.php/ijafr International Journal of Accounting & Finance Review Vol. 9, No. 1;2021 to non-performing loan (LLPNPL). Only bank size is in natural logarithm form to avoid any extreme value.

Model Selection and Specification
Possible estimations are shaped based on appropriate statistical models to test hypotheses. Balanced panel data are checked to derive the best model among alternative options to provide significant results with fitted values. Where α0 confers constant/intercept, β1 to β12 represents coefficients, uit denotes between-entity error term, and Ɛit represents within-entity error.

Tools for Analysis
To check the panel data and to select or specify the best statistical models, several tests are attributed, e.g., F-test, Breusch   The mean result of the total debt to total equity ratio of the selected banks is 11.693, indicating that a higher proportion of the banks' assets is financed by liabilities from deposits and non-deposits accounts. Khan et al. (2020) mentioned this scenario as the nature to take deposits more by the commercial banks. This mean value of leverage is akin to the earlier studies informed by Sakunasingha et al. (2018), Khan et al. (2020), and Anarfo (2015). This variable also exhibits the highest variation determined by standard deviation, which is 2.479. It also presents a wide gap between the minimum and maximum values ranging from 2.479 to 19.822. EPS exhibits a mean value of Tk. 2.32 and indicates a wide gap between minimum and maximum values than other variables. The mean value of ROA is 10% which reveals that at the time of sampling, about 10% of profits were earned by the banking industry of Bangladesh. On the contrary, return on assets shows a small gap between minimum and maximum value. In addition, the minimum asset growth rate is -0.266% which is the only negative growth value among all of the variables. The correlation matrix including explained and explanatory variables is conducted to scrutinize the presence of multicollinearity. The leverage ratio has the highest positive correlation with the total debt to total assets ratio, indicating a significant association. The total debt to total equity ratio has the highest negative correlation with the total equity to total assets ratio (-0.793), specifying that when the total equity to total assets ratio decreases, the leverage ratio of the sampled banks steps up. ROA, cost per loan assets, asset growth, liquidity and non-performing loan ratio are also inversely correlated to the leverage ratio, signifying trade-off theory. The leverage ratio and bank size are heading in the same direction, similar to the study reported by Bukair (2018). According to Kennedy (2008), there exists no multicollinearity between two variables, signifying that the correlation of coefficients does not exceed 80%. So, this research output does not convey multicollinearity problems between independent variables or dependent to independent variables. Multicollinearity issue arises when there is a VIF value of more than 10 (Hair et al., 2006), signifying the non-existence of multicollinearity problem in the dataset.

Regression Results
Analysis of Variance (ANOVA) discerned how these three pre-stated models are statistically fitted for further analyses and estimations of variables.
R-squared values 74.9%, 63.0%, and 65.9% for within, between, and overall, respectively, uncovered that the dependent variable (TDTE) is explained with these stated percentages by explanatory variables in Model I (Fixed Effects Model). In the same way, R-squared for Model II confirms 71.3%, 87.8%, and 77.65 for within, between, and overall, respectively. F-test for Model I is statistically significant, rejecting the null hypothesis at 1% level of significance. Wald chi2 (12) for both Model II (Random Effects GLS Regression) and Model III (Feasible GLS Regression) dispensed significant outcomes, striking model significances, predicting at p<1%. Model III (FGLS) propels no autocorrelations and no heteroskedasticity, confirming that panel data is homoskedastic.  (12) 2207.80*** 807.64*** Notes: p<0.01=***, 0.05>p≥0.01=**, and 0.10>p=* Empirical findings have been derived from stated regression models specified based on autocorrelation and heteroskedasticity issues. Feasible GLS Regression constitutes the confirmation of robustness in Random Effects GLS Regression. Standard errors and coefficients confirm the reliability and the connection between the dependent variable and explanatory variables. Notes: p<0.01=***, 0.05>p≥0.01=**, and 0.10>p=* Regression results under Fixed Effects (FE) Model reveal that bank size is statistically noteworthy at 1% level with a positive coefficient, determining that bank size positively affects the total debt to total equity ratio (leverage ratio). This estimation rejects the null hypothesis H1. In the same way, some similar studies also found that bank size had contributory positive effects on capital structure or leverage ratio (Al-Hunnayan, 2020;Amidu, 2007;Anarfo, 2015;Bukair, 2019;Heider & Gropp, 2008;Khan et al., 2020;and Sheikh & Qureshi, 2017). In contrast, bank size in some researches was found significant negative relationship (Sharif & Muhammad, 2019). In the case of Random Effects (RE) GLS Regression and Feasible GLS Regression, those results do not predict in the same way.
Model I, Model II, and Model III uncover that return on assets (ROA) predicts debt to equity ratio by rejecting the null hypothesis H2 at 1%, 5%, and 1% level of significances. The coefficient is slightly higher with negative values, stating that ROA negatively affects the capital structure of banks. Earnings per share (EPS) estimates the leverage ratio (TDTE) at 5% level of confidence for Model II and 1% level of confidence for Model III, rejecting the null hypothesis, H3. Earnings per share positively explain the dependent variable with positive coefficients. Some similar studies found that ROA significantly affects the leverage ratio negatively (Al-Hunnayan, 2020;Amidu, 2007;Anarfo, 2015;Heider & Gropp, 2008;Khan et al., 2020;Sakunasingha et al., 2018;Sharif & Muhammad, 2019;and Sheikh & Quereshi, 2017). In contrast, few studies found that profitability insignificantly affected the leverage ratio (Noulas & Genimakis, 2011;and Bukair, 2019).
The ratio of total loans and advances to total deposits (TLATDE) indicates the liquidity position of the sampled banks. Liquidity ratio (TLATDE) has positively affected leverage ratio (TDTE) with significant prediction at 1% level inducing rejection of null hypothesis, H4 under the Model I. It means that higher liquidity intends to higher capacity to take risks through the issuance of debt instruments. Bukair (2019) found that liquidity was positively affecting the leverage ratio. On the contrary, some studies also found that liquidity significantly affects the leverage ratio negatively (Sakunasingha et al., 2018;and Güner, 2016). However, few studies found liquidity insignificantly affecting the leverage ratio (Al-Hunnayan, 2017;and Sharif & Muhammad, 2019).
The growth rate of total assets (GRTA) has positive contributory explanations on leverage ratio in Fixed Effects Model and FGLS Model, estimating p<0.05, rejecting the null hypothesis, H5. The growth of assets induces positive outcomes for the banks by which the banks arrange more debt financing in capital structure. In some research, growth was found to positively affect the leverage ratio (Al-Hunnayan, 2020;and Sakunasingha et al., 2018). Similar studies also found that growth significantly affects the leverage ratio negatively (Sheikh & Qureshi, 2017), where few studies found growth insignificantly affecting the leverage ratio (Anarfo, 2015;and Bukair, 2019).
Total equity to total assets (TETA) attributes significant contributions to the capital structure of banks rejecting the null hypothesis, H6 at 1%, 5%, and 1% for Model I, Model II, and Model III, respectively. When the total equity to total assets is composed of high equity compared to debt, the leverage ratio is negatively affected and reduced by the values of coefficients. Empirical results uncover that total debt to total assets (TDTA) has positively influenced capital structure, predicting the non-acceptance of null hypothesis, H7 at 1% level of significance for all the three predefined models. High debt to assets ratio indicates a high percentage of debt in the firm size, with favorable debt issuances for financing banks' capital. The investment structure of banks in terms of total loans and advances to total assets (TLATA) constitutes dominant forces to the leverage ratio and predicts the dependent variables at 1%, 5%, and 5% levels of significance under Model I, Model II, and Model III, rejecting null hypothesis, H8. Some similar studies perceived that asset structure significantly affects the leverage ratio negatively at 1% significance level (Amidu, 2007).
Cost per loan assets (CLA) negatively influenced the leverage ratio (TDTE) because the higher the cost, the lower the chances of raising funds from debt financing. Model II and Model III represent that cost per loan assets has contributory effects on leverage of banks, estimating p<0.05 and p<0.01 respectively, rejecting null hypothesis, H9. Capital adequacy ratio (CAR) explained dependent variable at 1% significance level under Fixed Effects Model, concluding that null hypothesis, H10 is rejected. The stated coefficient is negatively affecting the leverage structure of banks.
The study result reveals no significant association between non-performing loan ratio (NPLR) and debt to equity ratio and hence fails to reject the hypothesis H11. Feasible GLS Regression propels that loan loss provision to non-performing loans has contributory attributes on the capital structure of banks. Independent variable (LLPNPL) predicts the dependent variable (TDTE) at p<0.05, rejecting the null hypothesis, H12.

CONCLUSION
The decision regarding the capital structure is a vital phenomenon to the relevant bodies of any financial institution. To ensure optimal capital structure, banks intend to understand the financing behavior, techniques, and implications. This study explores the effects of relevant factors on the capital structure of commercial banks in Bangladesh through employing FGLS Regression Model, constituted through the data of 22 commercial banks in Bangladesh for 10 years of data. Study results have uncovered that capital structure is significantly explained by profitability, asset growth, asset structure, investment structure, cost per loan assets, and loan loss provisioning. In contrast, bank size, liquidity, capital adequacy, and non-performing loan ratio do not significantly affect capital structure. Profitability, asset structure, and cost per loan assets play dominant roles in financing the capital needs of the sampled banks. Return on assets has high contributory enlightenment to the capital structure because this item negatively influences the leverage conditions in a greater aspect, fetching a slightly higher coefficient. The authors concluded that when a bank possesses more profits, it increases the chances to finance from the earnings it has generated, ensuring fewer usages of debt capital in financing capital needs.
Conversely, asset growth has positive impacts, indicating that the commercial banks with largesized have strong ground to finance more debt capital in comparison to small-sized banks. These findings provide useful indications to the concerned bodies, regulators, and stakeholders to diagnose and assess the dominant factors for exploring complementary decision supports. However, this study is not free from limitations and in this context, researchers, analysts, and academicians may have scopes to explore new avenues for further investigation covering preferred shares and retained earnings, which are also vital elements of the capital structure. This study overlooks the state-owned commercial banks, Islamic banks, and foreign commercial banks of Bangladesh. There are some scopes to conduct studies on these three stated areas and a comprehensive study on the whole banking sector of Bangladesh with applying more statistical tools and models.