2.1 Effect of market concentration on banking performance.

According to various studies, banking industry structure is a relevant factor influencing banks’ performance. The seminal works of ^{Klein (1971)} and ^{Monti (1972)} formulated the theoretical relationships that associate improved banking performance with a more concentrated banking industry.

There are several studies and arguments that support the positive relationship between banking performance and market concentration. Some studies based on structure-conduct-performance hypothesis (SCP) indicate that when banks develop monopolistic behavior they obtain advantages that allow them to control interest rates and industry prices. These conditions lead to improved performance. As a result, several researchers have supported this approach for developed markets such as the United States and Europe (^{Bourke, 1989}; ^{Molyneux & Thornton, 1992)} as well as in emerging markets (^{Garza-García, 2012}; ^{Guillén, Rengifo, & Ozsoz, 2014}). Other studies suggest that the positive relationship between market concentration and banking profitability is due to the banks can reduce their production costs by distributing them through higher production of financial services (^{Shepherd, 1983}). ^{Berger (1995)} adds that larger banks take advantage of scale factors that allow them to obtain greater advantages and returns. This is the “efficient structure” hypothesis, also known as the “relative-market-power hypothesis” (RMP). ^{Chortareas, Garza-Garcia, and Girardone (2011(}, in an empirical study based on Latin American banks between 1997 and 2005, found evidence in favor of this hypothesis. Their results show that market concentration is associated with a broader financial product portfolio that allows banks to reduce their operational activity costs.

Through empirical analyses of both hypotheses, some scholars have concluded that market concentration increases banking performance, since banks benefit from developing monopolistic behaviors (^{Martínez-Peria & Mody, 2004}; ^{Athanasoglou, Brissimis, & Delis, 2008}; ^{Gelos, 2009}; ^{Jara, Arias, & Rodriguez, 2014}). This phenomenon can be observed in a banking industry with few banks (^{Tregenna, 2009}), and where the large banks obtain the greatest profits in the sector due to a broader financial products portfolio (^{Maudos & Fernandez de Guevara, 2004}).

However, there are those who remain skeptical, arguing that the aforementioned effect on banking performance is marginally reversed when market concentration is high. This negatively impacts on banking performance and has collateral effects on financial stability and the creation of non-banking institutions. ^{Goddard, Molyneux, and Wilson (2004)} have shown through a dynamic panel data model applied to European banks that the effect of market share on banking performance is not significant. They suggest that, in comparison to banks with greater market power, banks with small market share can achieve greater returns because they operate within specific and riskier niches. Similar results have emerged from other studies (^{Smirlock, 1985}; ^{Demirgüç-Kunt & Huizinga, 1999}; ^{Staikouras & Wood, 2004}; ^{Carbó & Rodríguez, 2007}; ^{Mirzaei, Moore, & Liu, 2013}). ^{Chortareas et al. (2011)} and ^{Behname (2012)} warn that if the banking market tends towards a monopoly, profitability can decrease as institutions become less efficient. Other scholars add that this negative relationship could be due to banks with greater market power becoming inefficient as they neglect the goal of maximizing profits (^{Berger & Hannan, 1998}). ^{Zhang, Jiang, Baozhi, and Wang (2013)} in their work on BRICS countries indicate that the negative effect of market concentration on banking performance is explained by a pursuit of a “quiet life” whereby banks tend to diminish their efforts to maximize profits.

Finally, a possible non-linear relationship between market concentration and banking performance is not an illogical idea. Although there is no concrete evidence, some studies may help to support this hypothesis. ^{Boyd and De Nicoló (2005)} conclude in their research that higher interest rates (as a result of greater concentration in the credit market) increase the debt burden for borrowers, the incentive to participate in riskier projects and the probability of defaults. These aspects negatively affect banks’ returns. In this way, the positive effect of market concentration on banking performance is not persistent, due to this adverse selection problem and the creation of an environment conducive to a “quiet life”. According to these arguments we propose the following hypothesis:

H1: Market concentration has a non-linear effect on banking performance.

2.2 Effect of income diversification on banking performance.

Income diversification is another relevant determinant for banking performance. Although some empirical evidence has shown that income diversification is beneficial for banking performance, there is no consensus.

Banking products have similar productive technology. Banks can take advantage of economies of scope to distribute their production costs across various financial products (^{Gregoire & Mendoza, 1990}). For this reason, several empirical studies have argued that income diversification increases profitability. ^{Elsas, Hackethal, and Holzhäuser (2010)} analyzed 380 banks from Australia, Canada, France, Germany, Italy, the UK, the US, Spain and Switzerland between 1996 and 2008. Their results corroborated the positive effect of income diversification on banking performance. The authors add that this impact generates a greater market value for banks. ^{Lee, Hsieh, and Yang (2014)} analyzed 2372 banks from 29 countries in the Asia-Pacific region between 1995 and 2009, and found similar results. These empirical findings have been corroborated by other studies for both developed and emerging economies (^{Deng & Elyasiani, 2008}; ^{Demirgüç-Kunt & Huizinga, 2010}; ^{Fang, Hasan, & Marton, 2011}; ^{Sanya & Wolfe, 2011}; ^{Meslier et al., 2012}; ^{Amediku, 2012}; ^{Jara et al., 2014}). Even the positive effects of income diversification are consistent with the RMP hypothesis as banking efficiency is achieved by distributing production costs across a diversified financial services portfolio and concentrated banking industry.

Other studies provide a different vision and results. For instance, they indicate that income diversification also has collateral effects, such as performance instability and higher bank risk (^{Stiroh, 2004a}, ^2004b, ²⁰⁰⁶; ^{Stiroh & Rumble, 2006}; ^{Demirgüç-Kunt & Huizinga, 2010}). Accordingly, ^{Chiorazzo, Milani, and Salvini (2008)} analyzed Italian banks and showed that income diversification generated a trade-off between risk and return, which conditioned its impact on performance. In fact, the higher risk from income diversification strategies reduced performance. ^{Berger, Hasan, and Zhou (2010)}, in a study of Chinese banks, showed that income diversification effectively reduces banking performance and increases costs. ^{Ben and Plihon (2011)} analyzed 714 banks from East Asia and Latin America and found evidence that corroborates these findings. Such conclusions can also be found in the work of other researchers (^{DeYoung & Rice, 2004}; ^{Acharya, Hasan, & Saunders, 2006}; ^{Baele, De Jonghe, & Vander-Vennet, 2007}; ^{Lepetit, Nys, Rous, & Tarazi, 2008}; ^{De Jonghe, 2010}; ^{Fiordelisi, Marques-Ibanez, & Molyneux, 2011}). It would seem that greater income diversification is a source of adverse selection that increases risk and reduces performance. We believe that the impact of income diversification is non-linear, and although there may initially be a positive relationship between diversification and performance, this relationship is reversed at high levels of diversification as a result of higher risk. Accordingly, we propose the following hypothesis:

H2: Income diversification has a non-linear effect on banking performance.

3.1. Data

The data used in this research was extracted from Global Financial Development (GFDD) and World Development Indicators (WDI), both World Bank databases. The data covers the period between 1994 and 2011. Based on this information, we constructed panel data for 134 countries.

Table 1 presents the geographical distribution of those countries. As shown, 30.60% (41 countries) of the sample are developed countries, while emerging countries represent 69.40% (93 countries). Europe concentrates the highest proportion of developed countries (80.49%), while Latin America and the Asia-Pacific region are the main emerging markets. The geographical distribution of the sample reveals a somewhat more equitable distribution, except for counties in Sub-Saharan Africa and North America.

Table 2 provides a description of the variables. The dependent variable is banking performance (BPER), measured by return on assets (ROA). According to several studies, this ratio compares the net profit generated by banks of a particular country to their total assets. Therefore, this measure is a global performance indicator for these institutions (^{Bourke, 1989}; ^{Mercieca, Schaeck, & Wolfe, 2007}; ^{Athanasoglou et al., 2008}; ^{Goddard, McKillop, & Wilson, 2008}; ^{Jara et al., 2014}; ^{Yahya, Akhtar, & Tabash, 2017}). We also use net interest margin (NIM) as an alternative measure for banking performance in the robustness analysis.

Some variables were also used as measures of market concentration. First, we used the five largest banks’ asset concentration in each country. ^{Williams (2003)} and ^{Chen and Liao (2011)} suggest that the concentration ratio indicates the possible barrier to entry for other financial intermediaries. Therefore, a higher concentration ratio is related to a less-competitive banking industry and improved performance for banks (^{Molyneux & Thornton, 1992}; ^{Beck, Demirgüç-Kunt, & Levine, 2006}; ^{Athanasoglou et al., 2008}; ^{Jara et al., 2014}). Second, we used the Lerner and Boone indices as competition measures. According to ^{Berger, Klapper, and Rima (2009)} and ^{Chen and Liao (2011)} an increase in the Lerner index indicates a higher degree of monopoly in the banking industry, and therefore a deterioration of the competitive conduct of financial intermediaries. Any value within this range indicates a noncompetitive market structure for a country’s banking sector (^{Klein, 1971}; ^{Monti, 1972}). Similar arguments correspond to the Boone Index. Both the Lerner and Boone Indices are used as market power measures associated with less-competitive banking markets and a potentially greater degree of industry concentration. The empirical data of this research reveals that the correlation between the asset concentration ratio and the Lerner Index was 0.12, while between the concentration ratio and the Boone Index it was 0.11. In both cases, the correlations were statistically significant.

We measured income diversification through the percentage of nontraditional or non-operating activity income to total income (^{Baele et al., 2007}; ^{Stiroh, 2004a}, 2004b; Stiroh & Rumble, 2006; ^{Chiorazzo et al., 2008}; ^{Deng & Elyasiani, 2008}; ^{Fang et al., 2011}; ^{Meslier et al., 2012}; ^{Jara et al., 2014}).

The analysis also incorporates other control variables suggested by empirical studies. At the banking industry level, we used the capitalization ratio to quantify the impact of banks’ financing on performance (^{Goddard et al., 2008}; ^{Gul, Irshad, & Zaman, 2011}); the credits to deposits ratio as a proxy for liquidity and growth opportunities (^{Maudos & Solis, 2009}; ^{Gul et al., 2011}; ^{Yahya et al., 2017}); the non-performing loans to total bank loans ratio as a measure of operational risk exposure faced by banks and the quality of their assets (^{Jara et al., 2014}); and the bank credits to GDP ratio as a proxy for banks’ degree of penetration (^{King & Levine, 1993}). We also used the gross margin ratio to quantify the role of operational efficiency (^{Martínez-Peria & Mody, 2004}; ^{Gelos, 2009)} and the Z-Score indicator as a proxy for banking system stability.

At the macroeconomic level, we used economic growth, annual inflation rates and the dummy variable denoted by crisis as control variables. These variables were suggested by ^{Demirgüç-Kunt, Laeven, and Levine (2004)}, ^{Gul et al. (2011)} and ^{Yahya et al. (2017)} amongst other authors. At the institutional level, we used the political stability and rule of law indices. These indices vary between -2.58 and 2.58, where positive (negative) values indicate higher (lower) political stability and respect for investors’ rights (^{Girma & Shortland, 2008}; ^{Roe & Siegel, 2011}; ^{Marcelin & Mathur, 2014}; ^{Montes, Mendonça, & Oliveira 2016}).

3.2. Econometric methodology

To determine the effect of market concentration and income diversification on banking performance we estimated a dynamic panel data regression using the GMM estimator proposed by ^{Arellano and Bover (1995)}. The empirical model is:

(1)

Where BPER _it is the dependent variable that measures the banking performance of the country i in the period t. The variable MC _it measures banking market concentration, which was measured through the five largest banks’ asset concentration, Lerner index and Boone Index. Income diversification (DIV _it ) was measured through non-traditional income to total bank income ratio. Note that model (1) incorporates the variables MC ² it and DIV ² it to control the possible non-linear effects of market concentration and income diversification on banking performance. Model (1) incorporates other control variables, at industry, macroeconomic and institutional system levels, grouped in the matrices X _bit , X _mit and X _eit respectively. Finally, Ʃ _it is the remaining random disturbance.

Model (1) includes individual fixed-effects ƞ _i related to country i and temporary effects ƞ _t linked to year t. The model also includes dummy variables to measure unobservable heterogeneity across geographic zones and countries’ income levels. The dynamic panel data model proposed by ^{Arellano and Bover (1995)} is widely used to control the endogeneity problem. In this case, endogeneity lies in the lag of banking performance (BPER _it-1 ) and banks’ operational risk (NPL _it ). According to ^{Stiglitz and Weiss (1981)}, bank performance and operational risk are endogenous variables. When banks seek higher returns, interest rates will increase and therefore their assets portfolio becomes more risky. The risk increase is consistent with the adverse selection problem. Therefore, if banks reduce their risk, a reduction in performance would be expected. These variables were instrumentalized through the t-2 and t-3 lags. For the model to be correctly specified, ^{Arellano and Bover (1995)} point out that these estimators must be consistent and the model must be instrumentally overidentified. To ensure the consistency of GMM estimators the presence of first-order autocorrelation was required (but not of a higher order); while the instrumental overidentification of the model was verified through the Sargan Test. Model (1) was estimated with robust variance to control heteroskedasticity patterns.

4.1 Descriptive analysis

Table 3 shows the descriptive and correlational statistics. At an international level, banking industry performance shows an average return on assets of 1.37%. Within this figure there is a contrast between the inferior performance of banks in developed countries (0.54%) and the higher return of banking in emerging countries (1.65%). The NIM has a similar pattern. This disparity is significant, showing that the structural characteristics of emerging markets offer the conditions for superior performance.

Banking markets are not competitive. The five largest banks have 79.58% of banking assets on average, while the Lerner index has an average value of 0.24. The results also show significant differences between developed and emerging markets. Emerging economies have a more concentrated banking structure, even above the world average. In any case, it is observed that both market concentration measures correlate positively and significantly with banking performance. Similar results are shown in the Boone index.

Income diversification indicates that 38.40% of bank income comes from non-operational (nontraditional) activities, a figure that rises to 39.82% in developed countries and falls to 37.92% in emerging markets. The difference between these markets is significant. It is observed that income diversification is positively related to banking performance, an idea that supports the view that banks obtain higher benefits from economies of scope than economies of scale.

At the banking industry level, the variables also show interesting results. The Z-Score index indicates an average of 15.33, which demonstrates the high financial stability of banking worldwide. On average, 8.48% of bank assets are financed with capital, which indicates a heavy dependence on external financing, while banks obtain an average gross margin of 43.32%. Regarding these figures, we observe that emerging markets have a more stable banking system, with higher capital requirements and greater costs control efficiency. It should be noted that financial stability, bank capital requirements and operational efficiency correlate positively and significantly with banking performance.

Other banking system characteristics, such as operational risk, bank liquidity and the degree of banking development, are negatively correlated with performance. On average, 6.57% of bank loans are delinquent, 97.86% of bank deposits are converted into credits, and bank credit penetration amounts to 59.84% of GDP. It should be noted that developed countries show the lowest levels of non-performing loans (4.46%), as well as more advanced bank development (112.88%) and greater liquidity (115.43%) than emerging countries.

Regarding macroeconomic characteristics, an annual growth of 3.99% is observed. The low growth of the developed countries stands out, with a figure of 3.05%, contrasts with the greater activity of emerging countries, which is above average. The variables that measure political stability and rule of law show obvious differences between developed and emerging countries. Descriptive results show that institutional development in developed countries is significantly higher than in emerging ones. Furthermore, it appears that institutional characteristics are both negatively and significantly correlated with bank performance. Preliminarily, this result would indicate that greater bank performance is related to national institutional weaknesses.

4.2 Effect of market concentration and income diversification on banking performance

Table 4 shows the results of model (1). According to ^{Arellano and Bover (1995)}, the GMM estimators are consistent because the z-test denoted as AR1 reveals the presence of first-order autocorrelation, but the AR2 test discards the second order autocorrelation. In addition, the Sargan test indicates that the model is instrumentally overidentified, while the Wald test supports global significance for all models.

Some control variables report the expected results according to empirical evidence. At the banking industry level, factors such as capital requirements, operational efficiency and financial stability have a positive and significant effect on banking performance. These results suggest that banks’ returns increase due to greater external financing autonomy (^{Jara et al., 2014}), greater control of costs (^{Martínez-Peria & Mody, 2004}; ^{Gelos, 2009)} and operating in a financially sound banking system (^{Yahya et al., 2017}). Contrary to these results, non-performing loans and banking development have a negative and significant impact on performance. This indicates that an increase in operational risk reduces profitability, while financial development leads to stricter regulations that could limit banks’ behavior and their returns (^{Athanasoglou et al., 2008}; ^{Yahya et al., 2017}). At the macroeconomic level, GDP growth has a positive and significant impact, which shows that banking performance is directly linked to economic activity (^{Revell, 1979}; ^{Demirgüç-Kunt & Huizinga, 2000}; ^{Bikker & Hu, 2002}). Finally, inflation and periods of economic crisis have the expected negative effect (^{Jara et al., 2014}).

Market structure is a relevant factor. Our results indicate that the five largest banks’ asset concentration, Lerner index and Boone index have a positive and significant effect on performance. This result is not a novelty, since an extensive research literature has demonstrated that market concentration increases returns (^{Bourke, 1989}; ^{Molyneux & Thornton, 1992}; ^{Demirgüç-Kunt et al., 2004}; ^{Martínez-Peria & Mody, 2004}; ^{Athanasoglou et al., 2008}; ^{Jara et al., 2014}). However, the effect of market concentration is nonlinear, validating hypothesis H1. The positive effect of market concentration on performance is observed at low concentration levels (^{Goddard et al., 2004}). However, when the five largest banks’ asset concentration exceeds a threshold of 75.26% (the average of critical values indicated in Table 4), the Lerner index exceeds the critical value of 0.46 (the average of critical values indicated in Table 4 and the Boone index exceeds the critical value of 0.16, the effect reverses and becomes negative. Thus, above these threshold values, the phenomenon of banks seeking a “quiet life” impacts on performance and reduces profitability (^{Carbó & Rodríguez, 2007}; ^{Zhang et al., 2013}).

Income diversification strategies are another relevant factor acting on performance. It is noted that the diversification variable has a positive and significant effect on banks’ ROA. Income diversification generates higher returns as a result of economies of scope, which promote a multi-product banking industry, being more profitable than economies of scale. This result is in line with several studies (^{Deng & Elyasiani, 2008}; ^{Elsas et al., 2010}; ^{Demirgüç-Kunt & Huizinga, 2010}; ^{Fang et al., 2011}; ^{Sanya & Wolfe, 2011}; ^{Meslier et al., 2012}; ^{Jara et al., 2014}; ^{Lee et al., 2014}). However, the effect of income diversification on banking performance is not linear, a result that validates hypothesis H2. When the non-traditional income to total bank income ratio exceeds the average threshold value of 44.24%, the positive effect of diversification is diluted and becomes negative. This second effect of income diversification shows that risk predominates over risk-return trade-off (^{Berger et al., 2010}; ^{DeYoung & Rice, 2004}; ^{Acharya et al., 2006}; ^{Lepetit et al., 2008}; ^{De Jonghe, 2010}; ^{Fiordelisi et al., 2011}). Therefore, economies of scope are less predominant in comparison to economies of scale. The interactive variable (DIV × CRISIS) is negative and significant for all models. Periods of economic crisis and contraction impose higher costs on banks which offer a more diverse portfolio of services. This leads banks to cut back on nontraditional activities, and therefore, to become more specialized. This result contradicts the findings of ^{Jara et al. (2014)}.

4.3 Robustness analysis.

In this section we analyze the robustness of our estimations. To do this, we used the net interest margin as an alternative measure of bank performance. Model (1) was estimated through the GMM estimator proposed by ^{Arellano and Bover (1995)}. This regression satisfies the necessary conditions for consistency and specification according to the autocorrelation and overidentification tests, respectively. To correct the endogeneity problem caused by non-performing loans and bank performance in t-1, these regressors were instrumentalized through the t-2 and t-3 lags

Table 5 shows the results. As in the case of the results described in Table 4, the control variables have the same effects on banking performance, with the exception of the negative and significant effect of liquidity.

Regarding interest variables, similar results are observed that support the robustness of the estimation. On the one hand, market structure has a positive and non-linear effect on banking performance. The five largest banks’ asset concentration, the Lerner index and Boone index have threshold values of 73.48%, 0.43 and 0.19, respectively. These figures are similar to those obtained previously which separate the effects of market concentration on bank performance.

On the other hand, income diversification also has a negative and non-linear effect on net interest margin. This result is consistent with the previous positive effect of income diversification on ROA. The net interest margin is an specialized banking performance measure related to purely operational activity. Therefore, an increase in income diversification reduces the proportion of operating income in relation to total bank income, but increases global performance (^{Jara et al., 2014}). A similar argument applies to the impact of income diversification on net interest margin in periods of economic crisis.