The paper applies the generalized method of moments (GMM) estimator to estimate our dynamic panel model. This estimator is robust to possible multicollinearity, heteroskedasticity, and autocorrelation issues (Greene, 2012 ). Notably, the endogeneity problem (caused by the reciprocal relationship between explanatory and dependent variables or the omission of variables) could be well tackled by the GMM dynamic estimator. There are two different versions of the dynamic panel GMM estimator: (i) the difference GMM estimator performs regressions on the converted equation in the first-difference form, using variables in levels as instruments (Arellano Bond, 1991 ), and (ii) the system GMM estimator conducts regressions on both primitive and transformed equations (level and first-difference equations), employing variables in levels and differences as instruments (Arellano Bover, 1995 ; Blundell Bond, 1998 ). This study prefers the system GMM estimator since it has been improved on the basis of the difference GMM version to yield better estimates (Roodman, 2009 ). The two-step estimator is also selected because it is more efficient than the one-step version, especially for the system GMM estimator (Windmeijer, 2005 ). To avoid the problem of “too many instruments” in the GMM estimator that could destroy the power of regressions, we limit the proliferation of instruments using the procedure proposed by Roodman ( 2009 ), based on the rule that the number of created instruments should not exceed the number of banks. The usage of the dynamic system GMM estimator is justified by mandatory diagnostic tests, namely, the Hansen test and the Arellano-Bond tests. The null hypothesis of the Hansen test is that the instruments employed are not correlated with the residuals, meaning that the set of instruments is valid. The null hypothesis of the Arellano-Bond tests is that the errors in regressions display no first- and second-order autocorrelation.
4 online installment loans GA.1. Descriptive statistics and correlations
Table 1 exhibits summary statistics for all variables employed. The means of ROA, ROE, and NIM are 0.904%, 9.816%, and 3.866%, respectively. The statistical distribution of these variables indicates the considerable variations in banks’ performance (reflected in the large values of the standard deviations and the differences between extreme values).
Second, almost all of the existing studies have been only concerned about the overall average impact of loan portfolio diversification on bank performance, but not considering the heterogeneous effects across different bank groups. Many prior authors have shown that the lack of expertise, information resources, and competitiveness capacity could explain the costs of loan portfolio diversification, but they do not provide additional evidence to shed further light on such reasons (e.g., Acharya et al., 2006 ; Behr et al., 2007 ; Hayden et al., 2007 ; Jahn et al., 2013 ; Tabak et al., 2011 ). If these aspects are at the center of the functioning of loan portfolio diversification, it should be expected that banks’ response to portfolio changes would differ according to heterogeneous bank characteristics.
This paper empirically investigates whether and how loan portfolio diversification affects bank returns, particularly highlighting the conditioning roles of business models and market power in this nexus. We employ the banking data of Vietnamese commercial banks in the period of 2008–2019 to achieve the research objectives. To ensure the robustness of findings obtained from the dynamic panel models, we use multiple alternative variables of primary interest in the regression treatment. For loan portfolio diversification, we consider two traditional diversification measures, including the Herfindahl-Hirschman index (HHI) and the Shannon Entropy index (SE). Regarding bank return indicators, we analyze a rich set of accounting ratios, namely, return on assets (ROA), return on equity (ROE), and net interest margin (NIM). The non-interest income share and the Lerner index are the appropriate indicators that account for business models and bank-level market power in this study.
For example, in 2008, the SBV warned 21 private commercial banks with an excessive proportion of outstanding loans in the real estate sector by the end of 2007. In response to the SBV’s tightening lending policies to potentially risky sectors, Vietnamese banks reported reduced real estate loans since 2009. Besides, the SBV also actively issued guidelines to encourage banks to increase lending in some priority areas, in line with the government’s economic development orientation.
where n is the number of economic sectors in each loan portfolio. The greater value of the HHI measure implies that the bank has a higher level of loan portfolio diversification. In some detail, the value 0 of the HHI measure represents an extremely specialized bank, in which all loans are assigned to only one economic sector. In contrast, the value of (1–1/n) describes a bank with a perfectly diversified loan portfolio, suggesting all equal exposures to economic sectors.
3.2.3. Business models and market power
Once both the marginal cost and the output price are available, we employ the following equation to compute the Lerner index for each bank and thereby derives the direct value of bank market power: (6) L e r n e r i t = P i t ? ? M C i t P i t (6)
The presence of the lagged dependent variable is to adopt the dynamic nature or, in other words, to highlight the persistence in bank earning-making behavior. We take one-period lags of all independent variables with the purposes to (i) reduce the endogeneity problem caused by the causality between the explanatory variables and the dependent variables, and (ii) record that banks take time to absorb the changes in their balance sheet structure and the macroeconomic situation, before translating them into business outcomes.
