Analysis of Factors Influencing Bank Profitability: Evidence from the West African Economic and Monetary Union Banking Sector

A function of a financial system is to intermediate between lenders and borrowers so that transaction and information costs for both parties can be reduced. Financial institutions known as financial intermediaries perform brokerage and asset transformation functions. Considered as the important financial intermediary, banks permit credit and liquidity provisions through credit channels and protect companies and households against unexpected needs for cash, they permit rapid economic development through the financing of different sectors of the economy such as agriculture, industry, and trade, and help the promotion of entrepreneurship that leads the private sector to participate effectıvely to economic growth (Aziakpono,2005, Rosengren, 2008). They cannot play this important role if they are not profitable and well capitalized. A profitable banking sector is better able to withstand negative shocks and contribute to the stability of the financial system (Jiang et al., 2003). The profit is important for all participants in the economy and depends on such factors as received and paid interest on bank transactions, the share of non-interest income, current expenditures and the structure of assets and liabilities (Myktybekovich, 2013). The main aim of the analysis on profitability is to uncover the main center of bank performance and factors that affect the increase in profits and profitability of the bank based on the effective management of revenues and expenditures.

Moreover, in countries with no strong financial markets or with a weak financial system the banking sector constitutes the key to the economic growth.

What is the situation about the banking sector in the West African Economic and Monetary Union?

Eight developing countries, including Benin, Burkina Faso, Ivory Coast Guinea Bissau, Mali, Niger, Senegal and Togo comprise the WAEMU banking sector, where banks mostly constitute the backbone of the financial system. Despite the fact that the regional stock market (BRVM) exists, the regional financial system remains weak Banks in the WAEMU countries and most other African countries are for the most part private businesses that have to attract capital from the public to fund their operations. If profits are inadequate or if the risk is excessive, they will have greater difficulty in obtaining capital, and their funding costs will grow, which will erode profitability.

What is the situation of banks from under-developed countries with weak financial system?

WAEMU countries are developing countries that have a weak financial system dominated by banks. Bank performance (commercial banks mostly) has been relatively poor, and it is characterized by low levels of economy financing (it represents on average 20 % of GPD in 2012), and private credit (high growth potential sector (industry), high levels of non- performing loans, poor asset quality, operational inefficiencies, among others (Berger et al., 2005; Bonaccorsi and Hardy, 2005; Imam and Kolerus, 2013; and Panayiotis et al., 2005). Also, there is a lack of information about African banks in general and about WAEMU banks in particular (Bourke, 1989; Demerguc-Kunt and Huizinga, 1999; Short, 1979; and Molyneux and Thornton, 1992).

In line with the aforementioned problems, the objectives of this study are to highlight the factors that influence bank profitability between the period 2006 and 2014 with a view to proposing some recommendations to decision makers and regulatory authorities. In the pursuance of these objectives, the following research questions were administered: What are the factors that influence the profitability of the West African Economic and Monetary Union Banking sector during the period 2006-2014? Do small and medium size and large size affect bank profitability in WAEMU?

These questions will lead this study to provide the appropriate answers, capture and highlight the importance of banks and gauge their role in WAEMU’s zone in these recent years.

The available empirical evidence tends to show that studies on banking have extensively been concentrated more on developed and a few developing countries and limitedly on WAEMU. There is thus insufficient information on the determinants of bank performance in Africa in general, and in WAEMU, in particular, that would require further investigation (Short, 1979; Bourke, 1989; Molyneux and Thornton, 1992; Demerguc-Kunt and Huizinga, 1999). Determinants of bank profitability can be divided between those that are internal and those that are external. Mercia, et al. (2002), Toddard, et al. (2004), and Panayiotis et al. (2005) showed that bank profitability is a function of internal and external factors. Internal factors include bank-specific; while external factors include both industry-specific and macroeconomic factors. Internal determinants of bank profitability can be defined as those factors that are influenced by the bank management decisions and policy objectives. The key factors of bank profitability performance are as following:

(i) Bank-Specific Factors.

In view to understanding bank performance in the global context, studies on profitability have largely used returns on bank assets (ROA), net interest margin (NIM) and return on equity (ROE), as common measures. Bank-specific determinants include financial statement ratios in four areas: (1) capital; (2) earnings, profitability, and efficiency; (3) liquidity; and (4) asset quality (Golin, 2001). The details of these variables are presented in Table 1 and Table 2.

(ii) Industry-Specific Determinants

Bank industry determinants are external factors that may relate to bank profitability, such as the extent of industry concentration and the size of the banking system in relation to the size of the economy as a whole. Industry concentration is the degree to which the industry in a market is served by just a few or by many banks. When a banking market is more concentrated, customers have fewer choices, competition is less, and the market power of individual banks is greater. The common variables include Market concentration, Stock market capitalization, bank assets, Herfindahl-Hirschman index and others. The details of the latter can be found in Table 1 and Table 2.

(iii) Macroeconomic Determinants

The last group of profitability determinants deals with macroeconomic control variables. Economic growth is thought to impact bank profitability favorably, by increasing loan demand, decreasing loan default rates, and allowing banks to charge more for their services. This may be offset by increasing the supply of banking services, as expansions and new entrants are encouraged by perceived favorable conditions. This variable is assessed by the year’s real change in gross domestic product (GDP) for the nation the bank is located in, sometimes on a per capita basis. The common variables include inflation rate, the long- term interest rate, rate of economic growth (Panayiotis et al., 2005) other variables. The detail of these variables are found in Table 1 and Table 2. Factors influencing bank profitability is summarized in Table 1.

The overall sample consists of 86 active banks in the WAEMU’zone. Out of the 86 banks, 92% are commercial banks, and 8% are quasi-commercial 1banks where attention was focused on the commercial bank in order to avoid problems of comparison between different types of banks and to provide homogeneity in the comparison between countries. Also, the study covers the time period of 2006-2014 with a balanced panel in order to have an acceptable, relevant and recent sample in line with the study’s objectives and also to be able to get enough data to carry out the econometric analysis. The data are taken from the West African Central Bank (BCEAO) website www.bceao.int where banks’ annual financial statements are published, so are macroeconomic indicators Furthermore, all the bank financial ratios were calculated because the financial statements are raw data.

Return on assets (ROA), Return on equity (ROE), and Net interest margin (NIM) are the three measures of profitability supported by the literature review (see Growe et al. 2014, Alberttazzi and Gambacorta, 2009, Angbazo, 1997, Demirguc-kunt and Huizinga, 2008, Golin and Delhaise (2013). Return on assets (ROA) is used as a primary measure of bank profitability. In calculating ROA, and ROE, the average assets and equity in the denominator respectively were considered. ROA is preferred for some reasons that assets directly reflect both income and expense levels (Olson & Zoubi, 2011), and also ROA does not vary according to the amount of leverage employed, as does ROE (Golin & Delhaise, 2013). The drawback of ROA is that it takes into account the off-balance assets. But, these off-balanced assets are negligible in WAEMU bank balance sheet.

1 Most of non-commercial bank act like commercial bank. Because in WAEMU, commerce activities are dominant when comparison is made with the other sectors. Due to this; some noncommercial bank use to combine their first domain intervention activity with the commerce in order to survive.

Panel data analysis is a method of predicting relationships using time series with cross- sectional series (Greene, 2003, p.612. The general model of balanced panel estimation is written as follows:

where, 𝑌_i𝑡 is the dependent variable of bank i at time t, with i = 1,…,N; t = 1,…, T, 𝑐 is a constant term, X_i𝑡 are k explanatory variables and s_i𝑡is the disturbance with 𝑣i the unobserved bank-specific effect or random effect and 𝑢_i𝑡 the idiosyncratic error. This is a one-way error component regression model, where 𝑣_i ∼ 𝐼𝐼𝑁 (0, 𝜎²) and independent of 𝑢_i𝑡 ∼ IIN (0,𝜎²).

The development and application of Generalized Methods of Moments (GMM) estimation for panel data have been extremely fruitful in the last decade. In the empirical growth literature, GMM estimation has become particularly popular. The Arellano and Bond (1991) estimator in particular initially benefited from widespread use in different topics related to growth. The Arellano and Bond model is as follows: The original estimator is often entitled difference GMM

The first difference transformation removes both the constant term and the individual effect:

∆𝑦it = 𝜌∆yi,t−1 + ∆Xit𝛽2 + ∆sit

The DPD (Dynamic Panel Data) approach is usually considered by the work of Arellano and Bond (Rev. Ec. Stud., 1991), who have popularized the work of Holtz-Eakin, Newey, and Rosen (Econometrica, 1988). Generalized Method of Moments (GMM) is more efficient estimates of the dynamic panel data model. As the DPD estimators are instrumental variables methods, it is particularly important to evaluate the Sargan test results when they are applied. Also, another important diagnostic in DPD estimation is the AR test for autocorrelation of the residuals.

^₃ With $E\left(\mu_{\mathrm{i}}\right)=0, E\left(\mathrm{v}_{\mathrm{it}}\right)=0$, and $\mathrm{E}\left(\mu_{\mathrm{i}} v_{\mathrm{it}}\right)=0$ for $i=1,2, \ldots \ldots, N$ and $t=1,2, \ldots, T$

Sargan-Hansen Test

The standard test for testing the validity of moment conditions used in the GMM estimation procedure is the Sargan test of overidentifying restrictions (Sargan 1958) and the development of Hansen (1982). For the GMM estimator in the first-differenced model, this test is given as follows by:

where, WN is the optimal weight matrix and ^~$\widehat{\Delta \mu}$ are two step in the differenced model. When the number of moment conditions is greater than the dimension of the parameter vector, the model is said to be over-identified. Over-identification allows the study to check whether the model's moment conditions match the data well or not. The hypotheses for Sargan test are as follows:

H0: Overidentifying restrictions are valid

When the null hypothesis is accepted, then it is concluded the instruments are valid. In other words, the higher the p-value (p>0.05) of the Sargan statistic, the better. In robust estimation, Stata reports the Hansen J statistic instead of the Sargan with the same null hypothesis.

Arellano-Bond Test for Autocorrelation

The hypotheses for Arellano-Bond test are as follow:

𝐻o: No Autocorrelation: The Arellano – Bond test for autocorrelation has a null hypothesis of no autocorrelation and is applied to the differenced residuals.

a) The test for AR (1) process in first differences usually rejects the null hypothesis

b) The test for AR (2) in first differences is more important because it will detect autocorrelation in levels.

For profitability model, the study uses cost-efficiency frontier model (Battese and Coelli,1992; Marko ,2006; Munyambonera, 2013) which is a technical efficiency concept based on a production function that is used to measure bank cost efficiency. Cost efficiency is derived from the cost function and is a modified form of Cobb-Douglas function. Cost efficiency reflects the position of particular bank relative to the cost frontier. In fact, after the transformation of the stochastic cost frontier function into a log-linear generalized production function framework which is destined to estimate bank profitability. Concerning bank profitability estimation, this study adopts almost a similar framework as applied by Wilson et.al. (2004) on European banks; Naceur (2003) on Tunisian banks; and Panayiotis et al. (2005) on Greece banks. The only differences are that balanced panel and more explanatory variables than the latter were used. So the final general linear4 model is written as following.

where, П_i𝑐,𝑡 is the profit of bank i in country 𝑐 for period t; α is the regression constant; X^j and X^𝑚 denote vectors of bank-specific and bank - industry determinants, respectively; X^𝑛 refers to macroeconomic factors specific to each country; and εit is the idiosyncratic error. s_i𝑡= 𝑣_i + 𝑢_i𝑡𝑠 and s_i𝑡is the disturbance with 𝑣i the unobserved bank-specific effect or random effect and 𝑢_i𝑡 the idiosyncratic error. So models used are as follows:

$\begin{aligned} & \mathrm{ROA}_{\mathrm{t}}= \\ & \alpha+\left[\beta_1 S H E R_{\mathrm{ic}, \mathrm{t}}+\beta_2 L i D R_{\mathrm{ic}, \mathrm{t}}+\beta_3 L i A R_{\mathrm{ic}, \mathrm{t}}+\beta_4 L D R_{\mathrm{ic}, \mathrm{t}}+\beta_5 L A R_{\mathrm{ic}, \mathrm{t}}+\beta_6 N I R_{\mathrm{ic}, \mathrm{t}}+\beta_7 N A R_{\mathrm{ic}, \mathrm{t}}+\beta_8\right.\end{aligned}$

$\begin{aligned} & \left.N I I R_{\mathrm{ic}, \mathrm{t}}+\beta_9 \quad N P A R_{\mathrm{ic}, \mathrm{t}}+\beta_{10} \quad \text { LogSize }_{\mathrm{ic}, \mathrm{t},}\right]+\left[\begin{array}{llll}\boldsymbol{\beta}_{\mathbf{1 1}} & \boldsymbol{B C R}_{\mathrm{c}, \mathrm{t}}+\boldsymbol{\beta}_{\mathbf{1 2}} & \boldsymbol{S M R}_{\mathrm{t}}\end{array}\right]+\beta_{13} \quad \text { CPI }_{\mathrm{c}, \mathrm{t}}+\beta_{14} \\ & P R G D P_{\mathrm{c}, \mathrm{t}}+\varepsilon_{\mathrm{it}} \\ & \text { (Model 1) } \\ & \end{aligned}$

In the line of the objective to check the effect of bank size on the profitability, the sample is divided (see table 3.) into two part such large bank (LogsizeLarBK) and Small and Medium bank (LogsizeSmBk) based on the criteria used in the WAEMU banking sector concerning the bank size classification5. The first factor will lead to negative coefficients if increased diversification leads to lower risk and thus, lower the expected returns, and the second factor will lead to the positive coefficient for profitability if there are significant economies of scale. In the models, all the other variables have been kept intact. The models are here after

$\begin{aligned} & \mathrm{ROA}_{\mathrm{ic}, \mathrm{t}}= \\ & \alpha+\left[\beta_1 S H E R_{\mathrm{ic}, \mathrm{t}}+\beta_2 L i D R_{\mathrm{ic}, \mathrm{t}}+\beta_3 L i A R_{\mathrm{ic}, \mathrm{t}}+\beta_4 L D R_{\mathrm{ic}, \mathrm{t}}+\beta_5 L A R_{\mathrm{ic}, \mathrm{t}}+\beta_6 N I R_{\mathrm{ic}, \mathrm{t}}+\beta_7 N A R_{\mathrm{ic}, \mathrm{t}}+\beta_8\right.\end{aligned}$

$\begin{aligned} & \left.N I I R_{\mathrm{ic}, \mathrm{t}}+\beta_9 N P A R_{\mathrm{ic}, \mathrm{t}}+\beta_{10} \operatorname{LogSizeLarBk} \boldsymbol{k}_{\mathrm{ic}, \mathrm{t},]}\right]+\left[\beta_{11} B C R_{\mathrm{c}, \mathrm{t}}+\beta_{12} S M R_{\mathrm{t}}\right]+\beta_{13} C P I_{\mathrm{c}, \mathrm{t}}+\beta_{14} \\ & P R G D P_{\mathrm{c}, \mathrm{t}}+\varepsilon_{\mathrm{it}} \\ & \text { (Model 2) }\end{aligned}$

$\begin{aligned} & \mathrm{ROA}_{\mathrm{ic}, \mathrm{t}}= \\ & \alpha+\left[\beta_1 S H E R_{\mathrm{ic}, \mathrm{t}}+\beta_2 L i D R_{\mathrm{ic}, \mathrm{t}}+\beta_3 L i A R_{\mathrm{ic}, \mathrm{t}}+\beta_4 L D R_{\mathrm{ic}, \mathrm{t}}+\beta_5 L A R_{\mathrm{ic}, \mathrm{t}}+\beta_6 N I R_{\mathrm{ic}, \mathrm{t}}+\beta_7 N A R_{\mathrm{ic}, \mathrm{t}}+\beta_8\right.\end{aligned}$

$\begin{aligned} & \left.N I I R_{\mathrm{ic}, \mathrm{t}}+\beta_9 N P A R_{\mathrm{ic}, \mathrm{t}}+\beta_{10} \operatorname{LogSizeSmB} \boldsymbol{k}_{\mathrm{ic}, \mathrm{t},}\right]+\left[\begin{array}{lll}\beta_{11} & B C R_{\mathrm{c}, \mathrm{t}}+\beta_{12} & S M R_{\mathrm{t}}\end{array}\right]+\beta_{13} C P I_{\mathrm{c}, \mathrm{t}}+\beta_{14} \\ & P R G D P_{\mathrm{c}, \mathrm{t}}+\varepsilon_{\mathrm{it}} \\ & \text { (Model 3) }\end{aligned}$

Moreover, in order to capture the tendency of profits to be persistent over time, (due to market structure imperfections or high sensitivity to auto correlated regional or macroeconomic factors) the general model was reestimated using a dynamic panel GMM model by lagging the dependent variable among the regressors. The Arellano- Bover/Blundell -Bond (1995, 1998) generalized method of moments (GMM) which includes additional moment conditions and shows the absence of autocorrelation in the idiosyncratic errors was used. The dynamic model can be written when based on equation (5) as follows:

$\varepsilon_{\mathrm{it}}=v_{\mathrm{i}}+u_{\mathrm{it}}$

where, Π_{i𝑐,𝑡−1} is one-period lagged dependent variable and ð measures the speed of mean reversion. Also, it is possible that, given the relative large time frame of the dataset and the reforms that took place in the WAEMU banking sector during the sample period, it is very crucial to include dummy time variables in the model. Failing to account for these may bias the estimates in unknown magnitudes and directions. So time effects exist in the error component of the model, as follows:

εit= 𝑣i +λt +𝑢it

where, 𝜆𝑡 is the unobservable time- effect. The joint significance of the unobservable time- effects is tested by the H0 hypothesis: 𝐻0: 𝜆2=𝜆3 …=𝜆𝑇=0.

The relevant LM test (Table 5) indicates that H0 is rejected at the 95% confidence level for model 4, implying that it is necessary to include year-specific dummy variables to account for λt. All the coefficient of dummy variables are significant for all the years for model 4, Therefore, the researcher extend expand equation (7) as following:

εit = 𝑣i +𝑢it

where, Dt are the dummy variables for the years (2006 2014).

In fact, the LM test for model (4) doesn’t reject 𝐻0 (see Table 5) and thus the study proceeds with the estimation of this model. This yields the following models specification:

$\mathrm{ROA}_{\mathrm{ic}, \mathrm{t}}=$

$\alpha+\delta R O A_{\mathrm{ic}, \mathrm{t}-1}+\left[\beta_1 S H E R_{\mathrm{ic}, \mathrm{t}}+\beta_2 L i D R_{\mathrm{ic}, \mathrm{t}}+\beta_3 L i A R_{\mathrm{ic}, \mathrm{t}}+\beta_4 L D R_{\mathrm{ic}, \mathrm{t}}+\beta_5 L A R_{\mathrm{ic}, \mathrm{t}}+\beta_6 N I R_{\mathrm{ic}, \mathrm{t}}+\beta_7\right.$

$\begin{aligned} & \left.N A R_{\mathrm{ic}, \mathrm{t}}+\beta_8 \quad N_{I I R_{\mathrm{ic}, \mathrm{t}}}+\beta_9 \quad N P A R_{\mathrm{ic}, \mathrm{t}}+\beta_{10} \quad \text { LogSize }_{\mathrm{ic}, \mathrm{t}}\right]+\left[\begin{array}{lll}\beta_{11} & B C R_{\mathrm{c}, \mathrm{t}}+\beta_{12} & S M R_{\mathrm{t}}\end{array}\right] \gamma D_t+\left[\begin{array}{l}\beta_{13} \\ \hline\end{array}\right. \\ & \left.C P I_{\mathrm{c}, \mathrm{t}}+\beta_{14} P R G D P_{\mathrm{c}, \mathrm{t}}\right]+\gamma D_t+\varepsilon_{\mathrm{it}} \quad \text { (Model 4) } \\ & \end{aligned}$

The analysis of the descriptive statistics results (Appendix 1.) shows that the three indicators of profitability, return on assets (ROA), return on equity (ROE) and net interest margin (NIM) all have an average positive mean of 0.7%, 2.8%, and 6.8% respectively. These results show low profitability index for WAEMU banks, particularly for ROA and ROE. In sum, the low level of profitability of banks can be explained by the excessive operating expenses occurred by banks that swallowed an important part or all their profits. Also, the shareholders ‘equity ratio stood on average for 7.6% during the period covered by the study. A Shareholder’s equity Ratio (SHER) indicates an average value of 7.8%. In regards to capital adequacy, range from a maximum value of 36.13% to a minimum value of -14.72 % shows that most banks in WAEMU do not respect the minimum capital adequacy during the period covered by this study.

Concerning liquidity variables, a liquid asset to deposit ratio (LiDR), a liquid asset to asset ratio (LiAR), loan to deposit ratio (LDR), loans to asset ratio (LAR) indicate a mean value of 108.5%, 89.4%, 64%, 59.5%. The union's bank sector recorded a high liquidity ratio during 2006-2014.

When based on the efficiency factors, it points that Cost to income ratio (NIR), cost to asset ratio (NAR) have an average value of 80.5% and 11.8% respectively. NIR is ranged from a maximum value of 852% to a minimum value of 26.5%. This explains the high level of banks ‘operating expenses in WAEMU. In regards to asset quality factors, nonperforming asset ratio (NAR) has a mean value of 13.3% which seems to be high. This ratio with a maximum value of 59.4% and a minimum value of 0.1%, indicates that most banks have a higher value than the mean of 13.3%. While the bank size variable (logsize) records an average value of 13.31.

Relatively to bank-industry variables, bank concentration ratio (BCR), stock market capitalization ratio (SMR), stands respectively for an average value of 60.8% and 9.7%.

Furthermore, when focused on macroeconomic factors, inflation (CPI), and real GDP growth per capita have an average value of 0.1%, and 0.13% respectively. The level of inflation stands at its lowest level in WAEMU during 2006-2014. This proves the effectiveness of monetary policy implemented in the zone.

In this work, panel data regression models have been applied. These models are some static panels and others dynamic panel models. In order to prevent the occurrence of false associations among variables, panel unit root tests such as common unit root process test of Levin, Lin and Chu (2002); and individual process unit root test of Im, Pesaran and Shin (2003) and Augmented Dickey-Fuller(ADF) panel unit root test for all series have been performed. The analysis of results of the unit root tests (Appendix 2) shows that all the series do not have unit root (p<0.05). The results show the calculated p -values are lesser than the critical value of 0.05 for all the variables. For this, the null hypothesis that the series have unit root is rejected for all the variables for the three-unit root tests. The analysis of the results are discussed in the following subsections.

A) Analysis of Results Based on Bank ‘Size Effect (Panel Static)

Before performing the analysis, some precondition tests have been performed. For the model 1 (original model), model 2 (large size bank) and 3 (small and medium-size bank), Chow test with a p-value < 0.05 reports that the fixed effect model is the appropriate model. After the Breusch Pagan test has been performed and with the calculated p-value < 0.05, it indicates that the random effect model is more appropriate than the pooled regression. With p-value <0.05, Hausman test indicates that, between fixed and random model, the fixed effect model is the efficient and consistent model for the model 1 and 3. Concerning the model 2 (large bank size), Hausman test indicates that random effect model is the appropriate one with p-value>0.05. Also concerning Model 1, 3, Panel Corrected Standard Error (PCSE) is performed because of the presence of the autocorrelation and variance problems, while PCSE is not performed for model 2 because of the absence of the autocorrelation and variance problem after performed modified Wald variance test and Wooldridge test of autocorrelation. The results of model 1, 2 and 3 are found in the Table 4. The model 1, show that the profitability is positively and significantly affected by the bank size. The model 2 indicates that the profitability has been negatively and significantly affected by the coefficient of large size of banks (LogsizelargBk) at the level of 1%. In theory, increased size is presumed to confer benefits which can enhance profitability. However, increasing size beyond a certain point may conduct to scale inefficiencies as the organization’s bureaucracy impedes communication. Larger size may allow banks to diversify, affecting both risk and profitability and decision making. This is called the theory of diseconomies of scale. So, the findings are in line with Staikouras and Wood (2004) and others scholars (Chronopoulos et al., 2012; Flamini et al., 2009. This confirms the non- linearity between size and profitability and the ambiguity of size. The model 3 indicates that the profitability is significantly and positively affected by the small and medium size (LogsizeSmBK) of banks at the level of 10%. According to the theory of economies scale, increased size is presumed to confer benefits which can enhance profitability. So in accordance with this theory and some studies (Staikouras and Wood (2004 and others,) Small and medium size of banks in WAEMU zone are positively related to profitability.

a) Analysis of results based on panel dynamic model

In order to find out whether profits are persistent or not in WAEMU banking sector by following Flamini V, McDonald C.and Schumacher L., (2009), Athanasoglou, et al. (2006b,) and others)), the generalized method of moments (GMM) approach was applied. In this study the dependent variables ROA was lagged one period and, the instrument (endogenous) variables are lagged 2 through 2 to override the endogeneity problems. When related to the literature and following Growe et al. (2014), Roodman (2009), all the bank –specific variables were used as endogenous variables among the explanatory variables which are first differenced, while bank-industry and macroeconomic variables are used as exogenous variables but are not first differenced and considered as control variables. Among control variables, Stock market capitalization ratio was considered as external instrument variable. That is why it is not figured in the outcomes. Also, a robust test was conducted by using two-step GMM estimator. More so, time dummies were used to control time invariant for Models. Table 6 reports the results for Model 4. The GMM estimator is consistent only if the lagged values of the explanatory variables are valid instruments. The Sargan (Johansen) test presented evidence that the underlying overidentifying restrictions are valid with Pr (J) >0.05 and the Arellano-Bond test for serial correlation in the first-differenced residuals presents no evidence of model misspecification. The test has rejected the null of zero autocorrelation in the first differenced errors at order one (AR (1) <0.05. However, the value test for the second order autocorrelation (AR (2) >0.05) indicated that the moment conditions of the model are valid. Model 4 reported in Table 6 indicates that in addition to the significance of lag dependent variable ROA, eleven variables included bank-specific, bank- industry and macroeconomic variables and all the time dummy variables have significantly affected the dependent variable. These variables are Shareholder’s equity ratio (SHER), Liquid asset / Total Asset Ratio (LiAR), Loans/total deposit ratio (LDR) , Loans/ total asset ratio(LAR), Cost/ Income Ratio(NIR), Cost / Asset Ratio (NAR) Non -interest income /revenues ratio(NIIR), Natural logarithm of total assets (Logsize), Bank Concentration ratio measured by Total bank assets/GDP (BCR), Inflation rate (Change in Price Index (CPI)) and Real GDP growth per capita (PRGDP). While only two bank-specific variables such as the ratio of a liquid asset to deposit (LiDR) and ratio of Nonperforming asset/Total Asset (NPAR), do not have significant effects on profitability. Moreso, dummy variables are positive and significant over the period except those of the year 2006 and 2007 which have been removed in the model. This indicates that most reforms performed in the banking sector during the period of study have a significant effect on bank profitability. Mostly those of 2014, have positively affected bank profitability.

When based the analysis on the direction of the relationship, it indicates that the lag dependent variable ROA has positively and significantly impacted the dependent variable at the 5% level. The coefficient of 0.021 indicated that profits of banks in WAEMU are persistent but with weak magnitude. The dynamic nature of the model is confirmed by the significance of the coefficient of the lagged ROA. This estimated coefficient of 0.021 has tended to reveal the existence of market power in the WAEMU banking sector pointing out a relatively competitive market structure. This shows that profits tend to adjust fast to their equilibrium level in WAEMU. Furthermore, other Studies of bank profitability have found lagged profitability to be significantly predictive of current profitability (Athanasoglou et al. (2005 Ayaydin & Karakaya, 2014; Chronopoulos et al., 2012; Dietrich & Wanzenried, 2011; Garcia-Herrero et al.). The low value of this coefficient reflects a high degree of adjustment or adaptability of the banking industry to changes in its macroeconomic and financial environment (Athanasoglou et al., 2008). Moreover, the finding is also consistent with Gammadigbe (2012) who found the persistence of profit in the WAEMU banking sector. The analysis of bank-specific factors shows the results as follows:

Shareholder’s equity ratio considered as a key proxy of capital has positively impacted the return on an asset at the level of 1% significance. This result is in line with theory and empirical evidence. Studies showing a positive relationship between capital levels and profitability include Alp et al. (2010), Ameur and Mhiri (2013), Athanasoglou et al. (2005, 2006); Bourke (1989), Demirguc-Kunt and Huizinga (1999), Flamini et al. (2009). However, the positive sign of this coefficient does not mean that WAEMU financial system is a perfect market. In other words, in spite of the relative perfect market shown by the one-period lagged ROA model, WAEMU financial system is far from being characterized as a perfect capital market with symmetric information, under which the impact of increased capital on profitability will be negative (Berger, 1995b, Athanasoglou, et al, 2006, Flamini et al., 2009).

The coefficient of Liquid asset to Total deposit ratio have the expected positive effect on profitability, but it stands insignificant. The higher liquid is the bank and the less vulnerable to a run. Higher liquidity may act to increase profitability by reducing insolvency risk costs. The results are in line with some empirical evidence (Almumani, 2013; Ayaydin & Karakaya, 2014) that found this ratio to be no significantly related to profitability.

Despite the negative expected effect of the coefficient of Liquid asset/Total Asset, it stands positive and highly significant at 1% level. According to portfolio theory higher risk leads to higher profitability. Lower liquidity means higher risk. This theory is in the line of some authors (Alp et al., 2010; Goddard et al., 2004;) who find a negative effect on profitability. They point out that Liquid assets typically earn a lower rate of return than the longer term loans banks make. However, some authors ((Bourke, 1989; Kosmidou et al., 2005; Shen et al., 2009) have found a positive effect on profitability. They have argued that the rationale offered is that more liquid banks have less need to resort to costly external funding. In other words, less expensive funding meant increased income. The findings are consistent with most of the previous empirical evidence in banking literature. So in line with some authors, banks are highly liquid6 in WAEMU banking sector.

Regarding coefficient of (Loans/total deposit ratio) credit risk, it shows an expected positive association with profitability at the 1% level significance. Theoretically, a high figure means lower liquidity. Flamini et al. (2009) have pointed out that credit risk is the main source of bank-specific risk in SSA. According to Golin & Delhaise, (2013), a figure in the 70_90% range is seen as optimal (with higher numbers being on the risky side, and below this range is conservative). In this thesis, the average value of this ratio stands at 64.5% in WAEMU banking sector. So this ratio stands below 70%. This can be explained that on average, WAEMU banks are conservatives in their credit risk levels. The results are consistent with studies of Flamini et al. 2009; Sohail et al., (2013) who find positive and significant relationships between this ratio and profitability. Both of these studies have reported averages on this ratio below 70%. This is also in line with financial models where risk-averse banks require larger earnings to compensate for higher credit risk.

⁶ The average ratio of liquid asset/total asset is 105.8% as the descriptive statistic has indicated.

Basing on the coefficient of Loans/ total asset ratio (widely used to measure liquidity), there is evidence of expected positive and significant relationship between this ratio and profitability at 5% level. Loans are less liquid than the other main component of a bank’s asset portfolio _ investment securities. Some authors (Francis (2013), Gul et al. (2011) find that the larger the share of loans on the balance sheet the greater will be the bank’s profitability. So Higher liquidity may act to increase profitability by reducing insolvency risk costs. Thus, the results i are in line with most of the latter previous empirical banking literature.

In regards to Cost/ Income Ratio considered as a key of efficiency ratio, its coefficient is highly negative and significant at 1% level. Theoretically, higher numbers indicate that a bank is operating less efficiently. Its relationship with profitability is almost uniformly negative. The findings are consistent with theory and most empirical studies (Alexiou & Sofoklis, 2009; Almumani, 2013; Bourke, 1989; Francis, 2013; Heffernan & Fu, 2008; Kosmidou et al., 2005). The results can be interpreted that banks in WAEMU are relatively incapable of passing on their costs to their customers completely.

Moreover, there is also evidence of a strong negative effect of Cost to Asset Ratio on profitability at 1% level significance. This ratio is less important than cost to income ratio considered as the most efficiency ratio. Also, the relative weak magnitude of this coefficient also points to potential large gains in profitability if banks manage to reduce their operating expenses better. The results are consistent with several studies which have found that high values of this ratio lower profitability (Alp et al., 2010; Athanasoglou et al., 2005, 2006; Demirguc-Kunt & Huizinga, 1999; Staikouras & Wood, 2004; Sufian, 2011.

Meanwhile, the coefficient of Ratio of Non-interest income /revenues (NIIR) has a positive and significant effect on profitability at 1% level. The expected positivity and significance of this ratio can be interpreted that in WAEMU, banks have been given through technological changes or innovations to enlarge their activities into nonbanking activities. But this cannot last like can do the interest income. Therefore, in WAEMU zone, banks’ most important revenues are from interest income. This is line with Golin & Delhaise, (2013) who underlined that this ratio is seen to be less sustainable and of lower quality than interest income. Therefore, the results of this study are consistent with these authors (Alper & Anbar, 2011; van, 2011) who find this ratio to be positively significant.

The coefficient of bank size (Logsize) was highly positive and significant on profitability at 1% level. Theoretically increased size is presumed to confer benefits which can enhance profitability. The fact that large banks carry out activities in the very less competitive market permit them to make efficiency gains that can be captured as higher earnings. This coefficient had the expected sign according to the economies of scale market power hypothesis. The results were in line with the followings studies (Flamini et al.2009; Athanasoglou et al., 2006;) that found size to be positively significant related to profitability.

The ratio of Nonperforming asset/Total Asset has no direct effect on profitability. Despite the negative sign of this coefficient, it has not significantly impacted profitability. This ratio provides indications of difficulties with a bank’s loan portfolio. Problems with a bank’s asset quality are usually assumed to decrease profitability. Many studies show that this ratio is negatively and significantly related to profitability ((Lee, 2012; Macit, 2012).

When based the analysis on the bank-industry factors, it shows that the coefficient of Bank concentration ratio (BCR) has a positive relationship with profitability as expected. This coefficient is weak and highly significant at 1% level. Contrary to the theory which states that when the banking sector is larger, it can be expected to be more competitive, which lowers the profitability of individual banks. But this is linked to developed economies. However, when based on less developed economies with developed banking system this ratio used to be positively significant to profitability. This is the case in this study. WAEMU remains less developed economies which have fairly developed banking system. The findings are consistent with Raza et al., 2013; Tan & Floros, 2014 who have pointed out that in less developed countries greater financial system development can enhance efficiency and profitability. The small magnitude of the coefficient can be interpreted that, despite that WAEMU zone is from less developed economies, its banking system remains relatively developed when making a comparison with that of some developing countries.

In regard to macroeconomic factors, there is evidence of positive effect of inflation (CPI) on bank profitability as expected. The coefficient of 5.89 is highly strong and significant. This can be interpreted that banks in WAEMU zone foresee changes in inflation successfully and promptly enough to adjust interest rates and margins. Furthermore through this result, one can find mathematical evidence when assuming that the Fisher (1911) equation holds. Let denote by rL and rD the real interest rate on loans and deposits, respectively, and π the inflation, bank spreads can be written in nominal terms as follows:

$\left(1+r_L\right)(1+\pi)-\left(1+r_D\right)(1+\pi)$

This can be rearranged as follows

$\left(r_L-r_D\right)(1+\pi)$

Assuming that net interest margins are a major component of bank profits. Therefore, this translates into a positive effect of inflation on bank profitability even when there is no attempt by banks to adjust interest rates in order to counter the impact of inflation shocks. The results are consistent with Flamini et al., 2009 who found a positive and significant relationship between inflation and profitability on SSA. Also the outcomes are in line with others studies (Athanasoglou et al., 2005, 2006; Demirguc-Kunt & Huizinga, 1999; Flamini et al) which found inflation to be positively and significantly related to profitability. Moreover, these outcomes is also the result of the lowest level of inflation kept in WAEMU zone during the period that covered this study. This is also the consequence an effective monetary7 policy implemented in the union.

Meanwhile, the coefficient of Real GDP Growth per capita (PRGDP) has negatively affected profitability. Despite an unexpected negative sign of this coefficient, it stands highly significant. The negative relationship can be interpreted that when ease of entry and competition increase, profitability is reduced along with GDP growth. Studies (Ayaydin & Karakaya, 2014; Staikouras & Wood, 2004) found negative relationship.

^₇ According to The monetary policy, the standard level of inflation should not exceed 3% in the union. Each country have the duty to respect it.

This table reports results of model 4 for which Dynamic panel GMM first difference method was applied. Sargan test of over identified instruments is valid (p-value >0.05) and Arellano and Bond autocorrelation test shows that the model is not misspecified (R(1) <0.05 and R(2) >0.05)) . all the bank –specific variables were used as endogenous variables among the explanatory variables which are first differenced, while bank-industry and macroeconomic variables are used as exogenous variables but are not first differenced.Also, the instrument variables are lagged 2 to 2 while the dependent variable is lagged 1 to 1 Also .For robustness of the model Two step GMM estimator was applied.

The topic of this study which is sorting to highlight factors that influence bank profitability in WAEMU has permit to used robust and sophisticated econometric models which lead to getting efficient outcomes. In regards to empirical evidence, the outcomes from the analysis of determinants of bank profitability based on cost efficiency theory, show that the relevant factors that influence the bank profitability in WAEMU are bank-specific factors ( capital (SHER), liquidity (LiAR, LAR, LDR), efficiency (NIR, NAR, NIIR), Asset quality (Bank size), bank sector- factors such as bank concentration ratio (BCR) and macroeconomic factors (inflation (CPI) and real GDP per capita)). Out of these factors, the most important and most indispensable factors that need to be addressed by regulatory, authorities; bank executives; professionals and researchers revealed in this dissertation are capital (SHER) which without a required level and without being under the control of regulators will not be able to allow banks to carry out their important role dedicated to them, the loan-to-deposit ratio (LDR) which constitute the main sources of revenue for banks; the cost income ratio which is the factor degrading banks profitability by its excessive level; the inflation which is the main macroeconomic factor that positively affects bank profitability.

The limitation of this study is that it does not take into account all the banks in the regional sector. This because some banks’ data do not cover the period (2006-2014) of the study, and some of them are newly created banks and do not possess any data. More so, some banks that cover the period have lacked data or have been merged with other banks or went into bankruptcy.

Based on the findings of the research, the following recommendations were made. It is recommended for.

Bank executives

• To focus on reducing operating expenses, which killed banking profitability in order to be more efficient source of improving banking effectiveness;

• To allocate and direct a significant share of medium and long-term credits to sectors with high potential for development, such as agricultural processing industries; and others; Inter-state transport projects.

Regulatory and decision-making authorities (BCEAO);

• To promote and strengthen bank penetration within the Union by subsidizing banks to enable the population to open free bank accounts as in the Western countries; the issuance of free debit and credit cards. Thus; debit cards will greatly solve the problem of the fluidity in the banking operations which makes customers wait for a long time. While credit cards will allow banks to have more revenues since the interest rates applied to these cards after being used remain high (between 16 and 18%).

• To ensure that the individual bank data (financial statements: balance sheet and income statements) published on its (BCEAO) database must be produced in Excel format and not in PDF format in order to facilitate the work of the researchers and encourage research on the sector.

For the future research, it may focus on the Analysis of the effect of bank sector performance on the economic growth.