BANKING COMPETITION AND PROFITABILITY IN GHANA:

Transcription

1 BANKING COMPETITION AND PROFITABILITY IN GHANA: AN EMPIRICAL INVESTIGATION OF MARKET POWER, BANK-SPECIFIC AND MACROECONOMIC DETERMINANTS USING SYSTEM GMM. MSc. Banking and Finance Student ID: Supervised By: Dimitrios Vavarigos 1

2 Acknowledgement I would like to thank my supervisor Dr Dimitrios Vavarigos for his helpful comments and suggestions throughout the entire thesis period. May the Lord bless and favour you at all times. Also, I would like to thank my spiritual fathers, Rev. Samuel Commey, Charles Emmanuel Odum and Edem Fianu for their support and encouragement. Above all, the praise and honour goes to God Almighty for seeing this study to a successful end. Without him, it would have been in vain. 2

3 TABLE OF CONTENTS Acknowledgement 2 Abstract 5 1. Introduction 6 2. Literature Review Persistence of Profit as a measure of Competition Market Power and Bank Profitability Structural Conduct Performance Hypothesis Relative Market Power (RMP) Hypothesis Bank Specific Determinants of Profitability Efficient-Structure Hypothesis Capital Adequacy Credit Risk Macroeconomic Factors and bank profitability Aggregate GDP Inflation Empirical Studies on the Ghanaian Banking Sector Overview of the Ghanaian Banking Sector The Ghanaian Banking Sector over the past 6 year Methodology Hypothesis Model Specification Determinants of bank performance Data Econometric Estimation Data Envelopment Analysis Dynamic Panel Data Generalised Method of Moment Justification of System GMM xtabond Results and Discussion Result and Analysis Descriptive Statistics 35 3

4 4.1.2 Analysis and Interpretation of Empirical Results Discussion of Results Diagnostic Tests Heteroskedasticity Autocorrelation Hansen J Statistics F-Test 46 5 Conclusion Summary of Findings Policy Recommendations Limitations Further Study 50 References 51 Appendices 56 Appendix A Names of Banks 56 Appendix B Mathematical Notation of Variables 57 Appendix C Tables 58 Appendix D Figures 61 Appendix E Regression output from Stata 63 4

5 Abstract This study assesses banking competition and profitability in Ghana from 2007 to To measure banking competition, lags of the dependent variable is used to derive the level of profit persistence within the banking sector. Market power, bank- specific and macroeconomic determinants are also used to measure bank profitability in the study. The market power determinants include Structural Conduct Performance (SCP) Hypothesis and Relative Market Power (RMP) hypothesis. The Bank-Specific Determinants encapsulates, Efficient-Structure Hypothesis, Credit risk and Capital Adequacy whiles the macroeconomic determinants incorporates Annual growth of GDP and Inflation. System Generalized Method of Moment (GMM) is the estimator employed to determine profit persistence and control for endogeneity in the empirical framework. The empirical results show that, the banking sector is quite competitive with a moderate level of profit persistence. The two market power determinants did not affect profitability significantly. Credit Risk (ratio of Non-Performing Loans) is the only Bank-Specific determinants that affect bank profitability significantly. The two macroeconomic determinants, Annual growth of GDP and Inflation are reported as the main determinants of Bank profitability in Ghana showing strong significance in the model. The results generally, show a departure from previous studies that observed the Ghana s banking sector to be uncompetitive and characterized by financial repression. Keywords: Banking competition, Persistence of Profit, Bank Profitability, Market Power Hypothesis, Efficient-Structure Hypothesis, Data Envelopment Analysis, and Generalized Method of Moment. 5

6 0 mean of hhi ,000 CHAPTER 1 1. INTRODUCTION The Ghanaian Banking Sector has experienced series of financial reforms in the past two decades to resuscitate it from the financial repressive regimes that characterised it in the 1980 s. As a result, the banking sector has become more responsive and competitive (Brownbridge and Gockel, 1996). The Ghanaian Banking Industry is now one of the most vibrant financial sectors within the West African Sub-region attracting a lot of foreign banks. Over the past decade, the banking sector has experienced some structural changes such as a reduction in market concentration (Figure 1 below). Profitability and efficiency were impressive, whiles capitalisation increased over the period as well 1. Figure 1 shows the trend of market concentration (HHI) over the past 10 years. Market concentration reduced consistently from 2004 until This reduction coincided with the introduction of the Universal Banking License in Concentration however, increased slightly in 2011 and can be attributable to the Recapitalization Policy in 2007, which led to a couple of Mergers and Acquisitions (M&A s). Figure 1 Market Concentration (HHI) 1 Figure 2, 3, 4, and 5 in Appendix D show the trend of important Indicators like Return on Asset, Return on Equity, Market Share, Capitalization and Efficiency indicating the competitiveness of the banking sector. 6

7 Despite these positive outlooks, the banking sector, in recent years has been confronted with problems such as excessively high interest rate and increasing ratio of non-performing loans (Bawumia, Belnye and Ofori, 2005). These problems have raised concerns about the soundness and competitiveness of the banking sector once more with stakeholders like the Association of Ghana Industries (AGI), advocating for further reforms to curb these problems that are at the heart of the banking sector. Therefore, it is important to determine, the level of competitiveness of the banking sector and what drives their profitability. This should help provide a good insight into the health of the financial sector. The present thesis s objectives are to find out, 1) The level of competitiveness of Ghana s banking sector. 2) Whether market power, bank-specific or macroeconomic variables determines bank profitability in Ghana. Answers to the above objectives, should provide a good basis upon which reforms can be undertaken. This thesis investigates banking competition and profitability using a dynamic panel estimator (system GMM). Banking competitiveness is measured using the Persistence of Profit (POP). POP involves the use of lags of the profitability (dependent variables) to determine how quickly profits adjust to equilibrium (Athanasoglou et al, 2005). POP was used by Berger et al (2000), Athanasoglou et al (2005), and Garza-Garcia (2012), to measure the competiveness of various banking sectors 2. To examine the determinants of bank profitability, Market power, Bank-specific and Macroeconomic variables are used. Market Power variables incorporate Structural Conduct Performance (SCP) and Relative Market 2 Athanasoglou et al (2005) and Garza-Garcia (2012) used the lagged profitability variables to test Persistence of profit (competiveness) for Greek and Mexican Banks respectively. 7

8 Power (RMP). The bank-specific variables are Efficiency, Capital ratio and ratio of Nonperforming loans whiles the macroeconomic variables are Annual growth of Aggregate GDP and inflation rates. Data on all the 23 commercial banks in the sector from the period 2007 to 2012 is used in the study. 3 The empirical results indicate that the banking sector is moderately competitive with profit persistence coefficient ranging between 0.27 and 0.36 in the four models. The study found all the market power variables to be insignificant in explaining bank profitability. Non-performing loans is the only banks-specific variable that is significant in the model. Lastly, the macroeconomic variables are reported as the main determinants of bank profitability in Ghana. The main limitation of the study has to do with the sample period employed in the study. Though the study intended to use data covering 10 year period, it only end up using 6 year period due to restrictions imposed by the econometric estimation technique used 4. Thus the generalisation of these findings must be done albeit cautions. The remaining study is organised as follows. Chapter 2 reviews the relevant literature on bank competition and performance as well as reforms within the Ghanaian Banking Sector. Chapter 3 touches on the methodology and the justification for the estimation method. Chapter 4 entails a presentation and analysis of the empirical results and the necessary diagnostic tests. The last chapter summarises the main findings of the study and some policy implications from the findings. The limitations of the study are reported in this chapter as well. 3 Refer to Appendix A for a list of all the commercial banks used in the present study. 4 Refer to the section labelled limitation in Chapter five, for explanation on the limitations of system GMM on the sample period. 8

9 CHAPTER TWO 2. LITERATURE REVIEW Banking competition and profitability is an area in applied finance that has received a lot of academic attention. Generally, two different methods have been used by previous studies to determine the competitiveness of banks. The first is the Persistence of Profit (POP) hypothesis which entails the use of lags of the dependent variable (Berger et al, 2000). The second method is the Panzar-Rose Model derived under the condition that profit is maximised in equilibrium (Bikker Shaffer and Spierdijk, 2012). This study however, used the first approach (POP) in measuring the competitiveness of the Ghanaian Banking Sector. Bank profitability have been measured using varied determinants believe to influence their performance. These determinants fall in three broad categories namely, Market Power, Bank-Specific and Macroeconomic determinants (Athanasoglou et al, 2005). The market power determinants include Structural Conduct Performance (SCP) and Relative Market Power (RMP) hypotheses whiles Bank-Specific determinants entailed efficiency, capitalisation, and risk management practices. The macroeconomic determinants consider variables like GDP and Inflation and how they impact bank performance. In terms of coverage, previous studies often employed either cross-country or individual country as their sample. The present paper used the Ghanaian Banking Sector, falling under the latter (country specific) approach. 2.1 Persistence of Profit (POP) as a Measure of Competition Profit persistence is the likelihood of firm(s) staying in the same domain of industry performance induced by obstacle to competition and information opacity (Berger, 2000). In a very competitive banking sector a period of high profitability will induce the influx of other bank into the sector (Garza-Garcia, 2011). This behaviour is expected to wipe out excess 9

10 profit bringing the market to perfect competition. Athanasoglou et al (2005) describe this phenomenon as the speed of profit adjustment to equilibrium. According to them a positive coefficient is to be expected for the lagged dependent variable and lies between 0 and 1 showing the dynamics of adjustment. Thus a value close to 0 represents a very competitive banking sector whiles value closer to 1 indicates an uncompetitive sector. 2.2 Market Power and Bank Profitability Structural Conduct Performance (SCP) Hypothesis: Structural Conduct Performance (SCP) Hypothesis is a market power phenomenon that allows existing banks to set prices above the market clearing equilibrium due to market concentration and entry barriers (McWilliams and Smart, 1993). In this hypothesis, banks find it easier to collude and enjoy higher Interest rate spreads 5 because they charge a higher interest rate on loans and lower interest rates on deposits (Shepherd, 1982; Goddard, 2001). Concentration of the banking sector occurs when there are only a small number of banks in the industry or when a relatively small number of banks control greater proportion of the market share (Berger, 1995; Goddard et al, 2001). Banks play a crucial role in every economy; and evidence of SCP paradigm often leads to reforms to reduce market concentration and restricting larger banks from taking over smaller banks (Berger, 1995). Recent studies on SCP hypothesis usually tested for a positive relationship between profitability and market concentration using the Hirfindahl-Hirschman Index 6 (HHI). Where a positive relationship exists, the hypothesis is not rejected indicating an acceptance of the SCP hypothesis. Molyneux and Forbes (1995) tested the SCP among European banks covering the period 1986 to The findings of their study showed that profitability was 5 Interest rate spread in the study refers to the difference between the rate received on loans and the rate rates they pay on deposits. 6 Herfindahl-Hirschman Index is measured by first squaring the market share of each bank and adding all of them up to derive the market concentration. 10

11 mainly driven by market concentration measured by HHI. They pointed out that in most of the countries in their sample; there exist a small number of banks that commanded a greater proportion of the market share. Goldberg and Rai (1996) also found some evidence in support of the SCP hypothesis albeit some limitations. The study covered 11 top European banks over the period 1988 to 1991 and the results showed a mild positive relationship between market concentration and bank profitability. They however, conceded that the results were not robust enough. Most recent studies by Athanasoglou et al (2005) and Garza-Garcia Garcia (2012) also used the HHI to measure concentration and found no significant relationship between market concentration and bank performance in Greece and Mexico respectively Relative Market Power (RMP) Hypothesis Profitability according to this hypothesis is driven by market share (Garza- Garcia, 2012). Firms with comparatively larger market share and/or unique products are more likely to be profitable than their counterparts (Berger, 1995). The RMP hypothesis is traced to a study by Shepherd in Shepherd (1982) showed that during the period , market share was the main determinant of the profitability of Fortune 500 firms in the United State after controlling for other determinants like barrier to entry and concentration. Using market share together with other variables such as size of dominant firms, advertising power, and growth, the market share was found to be significantly correlated with profit, even after controlling for the effect of other variables in the study (Nissan, 2003). Rhoades (1985) added that market share is an important determinant of profitability due to Inherent Product Differentiation for the market leader. This is because if the market is oligopolistic, product differentiation by the market followers will only benefit the market leader ultimately. For instance, when a market follower differentiates his product, other followers together with the market leader differentiate theirs as well. This action leads to cancelling of each market followers strategies. This situation returns the market leader to its original market position of 11

12 dominance (Rhoades (1985). Using a sample of 6492 banks in USA, and controlling for other variable such as concentration, economies of scale and explicit product differentiation, Rhoades (1985) then showed that, market share was a significant determinant of banks profitability in the 1970 s. More recent studies by Berger (1995), Garza Garcia (2011), Mirzaei, Liu, Moore (2011), supported the RMP hypothesis especially in advanced economies. These studies followed similar procedure in measuring RMP using market share as a proxy as well as controlling for the effect of concentration and other key variables. Berger (1995) in a very exhaustive study used dataset comprising of 1300 to 2000 observation covering ten years period in The findings supported the RMP with the coefficient for the measure of market share positive and significant even after controlling for market concentration and efficiency. Berger (1995) further noted that the inclusion of market share changes the positive coefficient of the concentration estimate (SCP) confirming that profitability was influenced by greater market share and not concentration. Garza-Garcia (2011) tested the RMP among the Mexican banks over the period, 2001 to The coefficient for market share was highly significant in the study confirming market share as a main determinants of bank profitability in Mexico. Mirzaei et al (2011) tested the RMP among advanced and emerging countries. Their study covered 1621 and 308 banks in advanced and emerging economies respectively. For emerging economies, there was no evidence for RMP hypothesis unlike the advanced economies where the coefficient was positive and significant. They further noted that market share influences other variables after interacting it with them in their study. 12

13 2.3 Bank-Specific Determinants of Profitability Part of the literature also attributed bank profitability to certain internal factors. Most of these studies employ internal factors such as efficiency, capital, risk management among others (Athanasoglou, 2005) Efficient-Structure Hypothesis Earlier works on banking efficiency used market share as a proxy for efficiency. The intuition behind this approach was that, efficient banks control greater market shares. Demsetz (1973) the pioneer of this idea argued that a positive relationship between market power (concentration or market share) and profitability could be due to efficiency. To prove this proposition, Demsetz (1973) conducted a study on two groups of firm (Small and Large). According to Demsetz (1973), a positive relationship for both small and large firms should indicate the presence of market power. If the result shows otherwise, then profitability was driven by efficiency which is captured by either market concentration or market share (Bond and Greenberg, 1976). Demsetz theory on using market share to measure efficiency was not convincing to some scholars leading to widespread criticism. Berger (1995) and Berger and Hannan (1997) argued that market share captures more than just efficiency. They instead favoured using explicit variables that measure efficiency. This proposition led to two measures of efficiency known as X-Efficiency and Scale Efficiency proposed by Berger (1995). X-Efficiency occurs when a bank operates at a lower cost because their management and technology expertise are superior to that of their competitors (Berger, 1995). X-efficient banks often tend to control a larger market share because of this superiority (Zouari and Mensi, 2010; Garza-Garcia, 2012). Scale efficiency on the other hand, espoused that a bank is more efficient because it produces at a lower unit cost compared to its competitors resulting in higher unit profitability. Therefore it s possible for two or more banks to have the 13

14 same management and technological expertise but one is more efficient because, it scale of production results in lower cost and higher profitability (Berger 1995). Berger (1995) further conceded that, efficiency can increase market concentration or market share and the positive relationship must not be taken on the face value. Berger (1995) proposed that the efficiency coefficients must be positive together with any of the market-power hypotheses (SCP and RMP) to prevent misinterpretation. Empirical studies on X- and Scale efficiency includes, Berger, 1995; Goldberg and Rai 1996; Maudos 1998; Zouari and Mensi, 2010; Garza-Garcia, Most of these studies found efficiency to be less important in explaining bank profitability. Berger (1995) was the first to test the two types of efficiency explicitly and reported that, X- efficiency only supported profitability partially, whiles scale efficiency was insignificant in the model. The condition that X-efficiency needed to be positively related with the market structure hypotheses was weak in his study. Zouari and Mensi (2010) tested for bank efficiency in the Tunisia and reported similar results to that of Berger, (1995). Zouari and Mensi (2010) even reported a negative relationship for X-efficiency which was counterintuitive since a positive relationship was expected. Garza- Garcia (2012) tested efficiency of Mexican banks and reported insignificant results for both scale and managerial efficiency. Studies that found positive significant relationship between efficiency and bank performance includes Goldberg and Rai, (1996) and Maudos (1998). Goldberg and Rai (1996) reported that for banking industries where market concentration is low, a positively significant relationship was found between efficiency and bank performance. Maudos (2008) tested for efficiency of Spanish banks over the period 1990 to 1993 and found efficiency to be positively related to profitability and market share. Maudos (1998) further observed that the introduction of efficiency variables increases the explanatory power of the model by increasing R 2 significantly. 14

15 2.3.2 Capital Adequacy Capital adequacy for banks is important for financial stability. It justify why the Basel Committee has devoted much effort to ensure that banks are adequately capitalised. Even though the Basel accords emphasised on financial stability and soundness, it has an important implication for bank profitability as well. There is a general consensus that a well-capitalised bank or sector is likely to be profitable than their less capitalised counterparts. Most of the empirical studies showed that capitalisation should have a positive relationship with bank profitability. Garcia-Herrero, Gavila and Santabarbara (2009) showed how capitalisation could affect bank profitability. According to them, capitalisation influences profitability through two channels; 1. A well-capitalised bank can easily increase its risk level by advancing more loans, to increase profitability without necessarily worsening its credit risk exposure. 2. A well-capitalised bank signals to the market it creditworthiness. These signals attract deposits at a lower rate compared to less creditworthy banks that will be considered risky. Most of the empirical studies report a positive relationship between capital adequacy and bank profitability. Athanasoglou, (2005), Garcia- Herrero et al, (2009), Garza-Garcia, (2012) all found capitalisation as a determinant of bank profitability in their studies. Athanasoglou (2005) measured capitalisation of Greek banks using Equity to Asset ratio as a proxy. The coefficient for Equity to Asset ratio in their study was positive and significant. Garcia- Herrero (2009) measured the capital adequacy of Chinese banks over the period 1997 to 2004 and reported interesting results. First, banks that were adequately capitalised were comparatively profitable. When capitalization variable was interacted with efficiency variables in the model, the results showed a higher significance for the sampled banks. Garza- Garcia (2012) also found positive relationship between capitalisations and profitability of 15

16 Mexican banks over the period 2001 to However, the results were only robust with ROA as the measure of profitability and not ROE, which was the other dependent variable Credit Risk Credit risk management is an integral function for financial institutions (especially banks). Commercial banks main business is lending and this poses the risk of non-payment on these loans. Widespread credit default can cause the failure of banks and undermine the stability of the financial system. Bernanke (2003) argued that customers defaulting on loans they took from the banks at the time propagated the Great Depression of the 1930 s. Higher default on loans reduces the profit margin of banks and negative relationship is expected with profitability (Miller and Noulas, 1997 cited in Athanasoglou, 2005). Miller and Noulas (1997) examined the credit risk situation of American banks in the 1980s and came to the conclusion that bank performance over this period was poor because their portfolio of loans were of low quality. Their study further discovered a strong negative effect of loan loss provision on bank profitability in the 1980 s. Recent studies by Athanasoglou et al (2005), found credit risk being a significant determinant of bank profitability in Greece. Garza-Garcia (2012) also reported similar findings for Mexican banks but the results were not robust in all the models that were employed in the study. 2.4 Macroeconomic factors and Bank Performance Aggregate GDP Most scholars posit that bank performance is related to flow of goods in an economy. The emphasis of these studies is on whether a period of strong economic growth will increase bank profitability or not. There were plausible reasons to believe that bank performance and economic growth are linked. This is because; banks act as facilitators (suppliers of credit) for 16

17 economic agents. In periods where output increases, demand for banking services increase as well Conversely a slowdown in economic growth should reduce the demand for banking services. Bikker and Hu (2000) argued various means by which a downturn in business cycle will affect bank profitability. Among them are 1) reduction in demand for loans; 2) Increase in the ratio of non- performing loans and; 3) less favourable interest rate structure. Empirical studies by Demirguc-Kunt and Huizinga (1998) Arpa et al (2001), Bikker and Hu (2002), Athanasoglou et al (2005) showed ample support for the cyclical movement of bank profit with the business cycle. Demirguc-Kunt and Huizinga (1998) studied on 80 countries over the period 1988 to They found that interest margin and bank profitability move together with the business cycle. Arpa et al (2001) focused on how macroeconomic development affects banks income and provision for loan loss in Austria. The findings indicated that bank profit and provision for loan loss are pro-cyclical with the business cycle. Likewise, Bikker and Hu (2002) reinforced the earlier findings when they looked at bank profit and business cycle of 20 OECD countries spanning the period, 1979 to Using macroeconomic variables such as real GDP growth, inflation, unemployment, share price index, they found a strong significant relationship between these variables and bank profitability. Additionally they found out that credit loss in these countries move together with the business cycle Inflation Periods where there is persistent increase in the prices of goods and services (inflation) often leads to reduction in the value of money. Boyd, Levine and Smith (2001) argued that increasing rate inflation does not only reduce the real return on money but assets in general. This reduction in real return has repercussions on credit intermediation by worsening credit market friction because customers will be willing to borrow, but banks unwilling to lend which leads to credit rationing. These credit market frictions reduce bank profitability in the 17

18 long run (Boyd, et al 2001). Revels, 1979, (cited in Athanasoglou et al, 2005) also noted that the effect of inflation on bank profitability depends on the rate at which bank expenses and cost increases with inflation. If banks operating expenses and wages increases higher than the inflationary rate, then bank profitability will decrease and vice versa (Athanasolglou et al, 2005). Perry (1992) in a related paper to that of Revel (1979) argued that the effect inflation on bank performance depends on whether it is anticipated or not. If inflation is fully anticipated, then banks can adjust their interest rates appropriately to be able to cover the increased cost due to inflation. However, the problem with Perry (1992) paper is how anticipated and unanticipated inflation can be determined. Empirical works on how inflation affects bank profitability and the direction of the effect often reported mixed results. Boyd et al, (2001) studied how inflation affected the performance of the financial sectors around the world. Their study covered 97 countries from 1960 to They found a negative significant relationship between inflation and financial sector development. Garza- Garcia, (2012) in a study on Mexican banks found no relationship between bank profitability and inflation whiles Bourke (1989), Molyneux and Thornton (1992) and Athanasoglou et al (2005) found a positive relationship between inflation and bank performance. 2.5 Empirical Studies on the Ghanaian Banking Sector Studies on the determinant of bank performance in Ghana are few and most of them are not directly aligned with this study. However their review gives an insight on the dynamics on the Ghanaian Banking sector. Busch and Mathisen (2005) measured the level of competition and efficiency of banks in Ghana from 1998 to They employed the Panzar- Rosse Model to measure bank competition. The Panzar- Rosse model, which gives a coefficient between 0 and 1, was 0.56 in their study, which meant the Ghanaian Banking sector was 18

19 uncompetitive. Busch and Mathisen (2005) further found high interest rate, bank size, expense ratio and inflation to be the main determinants of profitability. Bank size, Busch and Mathisen (2005) argued was a major determinant of profitability due to the low saving rates in Ghana which makes it difficult for smaller banks to penetrate the market. Ratio of nonperforming loans in their paper was positively related to profitability which seemed counterintuitive while nominal exchange rate and banking system credit to government was insignificant in explaining bank profitability. Another important finding of their paper was the excessive investment by banks in government securities. They found out that, a large amount of the bank s assets were channelled into government treasury bills. This they alluded to, as a source of inefficiency since the funds could have been easily been extended to more profitable ventures. Ametefe, Aboagye and Sarpong-Kumankoma (2011) analysed the performance of banks in Ghana from the year 2001 to Deposit ratio of banks was found not to impact profitability of bank in Ghana. They argued that because a larger proportion of these deposits are demand deposits it becomes difficult for managers to commit them into long-term projects that yield higher profits. Equity to asset ratio was positively related to profitability in their model. Ametefe et al (2011) further pointed out that banks with higher equity ratio could easily increase their portfolio of investment to increase their profitability. Other determinants like loan ratio was positive and significant in their study as expected. 2.6 Overview of the Ghanaian Banking Sector The Ghanaian banking sector, like in any emerging economy has gone through phases of financial deregulations with the aim of making the sector very competitive. These reforms, 19

20 rather than competition shaped the market structure, and the kind of financial products and services that are offered (Bawumia, Belnye and Ofori, 2005). Ghana s banking industry before the 1980 s was characterised by a lot of government intervention. Most of the banks in this era were government owned and restrictions were imposed on foreign entry as well as interest rate control by the Bank of Ghana; Central Bank (Brownbridge and Gockel, 1998). As with many developing countries at the time, the banking system was characterised by financial repression, negative interest rate and massive public sector borrowing which are often unproductive (Brownbridge and Gockel, 1998). A landmark reform to address these market imperfections known as Financial Sector Adjustment Programme (FINSAP) was implemented in the 1989 with the assistance of the World Bank and IMF. The aim of the FINSAP was to resuscitate the financial sector to make it more responsive. Most of the obstacles especially the interest rate ceiling was removed to allow the market to determine the equilibrium interest rate (Brownbridge and Gockel, 1998). By 1994, the second phase of FINSAP was introduced allowing most publicly owned banks to be privatised, and the setting up of Non-Bank Financial Institutions. These reforms were successful albeit some difficulties. Parallel with these reforms, a New Banking Act was introduced in The Banking Act spelt out the Capital requirement for local and foreign banks, and gave supervisory and legal power to the Bank of Ghana (BOG). The BOG among its duties has the power to review the minimum capital requirement, issuing new license and revoking license of existing banks when they violate the laws (Brownbridge and Gockel, 1998). These reforms were enough to stabilize and stimulate the banking sector. For instance, as at 1994, there were only 14 commercial banks operating in the sector but by the end 2012 the number almost doubled to 24 (Ghana Banking Survey, 2012). The increased number of banks was due to the introduction of the Universal Banking License in 2003 which led to series of 20

21 Merger and Acquisitions (M&A) as well as new entrants in to the industry (Ghana Banking Survey, 2008). In 2007, the central bank (BOG) introduced a new Minimum Capital Requirement aimed at ensuring that the banking sector is adequately capitalised to serve the economy. This saw the Minimum Capital raised to GHc 60 million ($30 million) with foreign banks given up to 2009 whiles their local counterpart were given up to 2012 to meet this new requirement (Ghana Banking Survey 2008). More recent reforms include the introduction of Credit Reference Bureaus to help in collection and relaying of credit information for efficient credit risk analysis, a re-establishment of the Collateral Registry and introduction of Antimoney Laundering Law (Ghana Banking Survey, 2008) The Ghanaian Banking Sector over the past 6 Years The introduction of the Universal Banking License in 2003 and the re-enactment of Banking Act in 2004 led to a lot of investor interest over the period. The sector over the period has seen a lot of M&A and the asset of the industry have more than doubled over the period (Ghana banking Survey, 2008). Figure 1, 2, 3, 4 and 5 in Appendix D shows the trend of the industry over the past decade from 2003 to Figure 1, for instance shows a reduction in market concentration measured using the Herfindahl-Hirschman Index (HHI). Market share from Figure 2 also declined over the past decade whiles the capitalization in Figure 4 went up. Banking efficiency shown by figure 5, also increased over the past decade. 21

22 CHAPTER THREE 3. METHODOLOGY 3.1 Hypothesis The objectives of the present study is to investigate the level of banking competition and find out whether market power, bank-specific and macroeconomic factors are significant determinants of bank performance in Ghana. In line with this the hypotheses for the study are formulated as follows. The first hypothesis is that the Ghanaian Banking Sector is competitive and the second hypothesis is that Market Power, Bank-Specific and Macroeconomic factors are statistically significant in determining the profitability of banks in Ghana. 3.2 Model Specification The general model below is in a linear form and expressed in the form used by Athasoglou et al (2005). k Π it =α + k X it + ε it (1) ε it= µ it + v it. Where Π it is the measure of profitability (Return on Asset and Return on Equity) of bank i at k time t, α is the constant term whiles X it s are K independent variables whiles ε it is the error term composed of µ it, the unobserved bank specific effect and v it, a random error term independent across all the banks. k The independent variables X it are categorized into three groups based on the second objective into market power, bank specific and macroeconomic determinants. The general model (1) is modified to give us. l m p Π it =α + l X it + m X it + p X it +ε it (2) 22

23 Where, X it s with l, m, p are superscripts to represent, Market power, Bank-specific and Macroeconomic determinants. Additionally, Berger et al (2000) postulated that bank profit most of the time tends to persist. Profit persistence is the tendency of this period s profit being dependent on that of the last period. This profit persistence is attributed to the existence of barrier to competition, informational opacity and bank sensitivity to macroeconomic shocks (Athanasoglou, 2005). Berger et al, (2000), proposed the inclusion of the lagged dependent variable as one of the explanatory variables to measure the level of profit persistence (competition). The inclusion of the lagged dependent variable as an explanatory variable, changes the model into a dynamic one (Athanasoglou et al, 2005). Appropriately, equation (3) is modified to reflect the dynamic nature of the model. Augmenting equation (2) with the lagged dependent variable gives us; Π it =α + Π it-1 + l X it l + m X it m + p X it p +ε it (3) Where, Π it-1 is the dependent variable lagged by one period and shows the level of profit persistence. The present study employed four (4) models in line with that of Garza-Garcia (2012). For the two dependent variables (ROA and ROE), the models were formulated with and without the macroeconomic variables. The intuition behind such an approach was to find out if the presence macroeconomic factors (GDP and Inflation) in the model influence the other determinants or not. Model 1: ROA it = α it + β 1 ROA it-1 + β 2 HHI t + β 3 MS it + β 4 NPL it + β 5 CAP it + β 6 EXS it + β 7 ESS it + µ it + v it. Model 2 ROA it = α it + β 1 ROA it-1 + β 2 HHI t + β 3 MS it + β 4 NPL it + β 5 CAP it + β 6 EXS it + β 7 ESS it + β 8 GDP t + β9cpi t + µ it + v it. 23

24 Model 3 ROE it = α it + β 1 ROE it-1 + β 2 HHI t + β 3 MS it + β 4 NPL it + β 5 CAP it + β 6 EXS it + β 7 ESS it + µ it + v it. Model 4 ROE it = α it + β 1 ROE it-1 + β 2 HHI t + β 3 MS it + β 4 NPL it + β 5 CAP it + β 6 EXS it + β 7 ESS it + β 8 GDP t + β9cpi t + µ it + v it. where, ROA it = Return on Average Asset of bank i at time t, where i = 1, N; t =1, T. ROA i,t-1 = Lagged Return on Average Asset for bank i at time t. ROE it = Return on Average Equity of bank i at time t. ROEi, t-1 = Lagged Return on Average Equity of bank I at time t. α it = the constant term of bank i at time t. HHI t = Market concentration of the industry at time t. MS it = Market Share of bank i at time t. NPL it = Ratio of non-performing loans of Bank i time t. CAP it = Capital Adequacy of bank i at time t. EXS it = Managerial Efficiency of bank i at time t. ESS it = Scale Efficiency of bank i at time t. GDP t = Annual growth of Aggregate GDP at time t. CPI= Consumer Price Index at time t. µ= Bank specific fixed (Time-Invariant Effect) disturbance term v= random disturbance term 24

25 3.3 Determinant of Bank performance Dependent Variables (Profitability Measures) Following Garza-Garcia (2011); Goddard (2004); Sufian and Noor (2012), two alternative measures of profitability, that is, Return on Average Asset (ROA) and Return on Average Equity (ROE) are employed for the present study Return on Average Asset (ROA) Return on Average Asset (ROA) shows how much revenue a unit of asset generates and an indication of how bank managers efficiently use the assets at their disposal (Athanasoglou et al, 2005). ROA is mathematically expressed as a ratio of total revenue (net profit) over total average asset 7. Athanasoglou et al (2005) argued that ROA could be diluted by off-balanced sheet activities, which results to misleading interpretations. This is because banks carry out a number of off-balance sheet transactions such as asset management, brokerage, financial advisory services and derivative contracts. These off-balance sheet transactions do not necessarily appear on the face of the balance sheet but impact profitability. As a result, reporting ROA using only assets listed on the balance sheet can lead to misleading interpretations Return on Average Equity (ROE) Return on Average Equity (ROE) shows how much returns are made on a unit of shareholders fund (equity) (Athanasoglou et al, 2005). It is expressed as a ratio of total revenue (net profit) over total average equity. The difference between ROE and ROA is that the ROE does not take into account the financial leverage (debt) whiles the latter does. Thus highly leveraged firm will have a higher ROA but lower ROE and vice versa (Athanasoglou 7 Refer to Appendix 2 for mathematical notations of all the variables used in the present study. 25

26 et al, 2005; IMF, 2002). However, synonymous to previous studies, both ratios are employed in the present study. The use of average asset and equity in computing the ratios takes care of variation within a financial year. Figures on a financial statement show the values at a specified date. Using these values will imply that throughout the financial year, the same values were observed. To cater for these dynamics, an average of last period s value (Assets and Equity) and that of this year is used rather than the balance sheet figure (Athanasoglou et al, 2005) Explanatory Variables: Lagged profitability Ratio (Π it-1 ): The lag of the dependent variables ROA and ROE are included in the model to measure the level of banking industry competitiveness and captures the dynamics of banks profitability (Garza-Garcia, 2011). The expected relationship with profitability is positive with a value between 0 and 1. In a non-competitive banking industry, profit this is expected to persist till the next period leading to a coefficient closer to 1. However, in a very competitive industry, the value will be closer to 0 (Berger et al, 2000; Athanasouglou et al, 2005; Garza-Garcia, 2011). The remaining independent variables used in the present study are grouped into three categories. These are market power determinants, bank specific determinants and macroeconomic determinant variables. Market Power Determinants: i. Market Concentration (HHI): Market concentration in the study is measured using the Herfirndahl-Hirschman Index (HHI). The HHI is expressed as the sum squares of market share. It measures the depth of market concentration and a positive coefficient will offer credence to the Structural Conduct Performance (SCP) hypothesis (Garza-Garcia, 2011) 26

27 ii. Market Share (MS): This is a measure of the proportional market share of the banks. It is expressed as a ratio of a bank s asset to that of the industry and a positive relationship will support the Relative Market Power Hypothesis (RMP), which suggests that profitability is driven by market share and inherent product differentiation (Berger, 2005). Bank Specific Determinants: i. X-Efficiency (ESX): The X-efficiency is a measure of efficiency attributable to managerial expertise and/or technology (Berger, 2005; Garza-Garcia, 2011). Two inputs and outputs were used in deriving the X-Efficiency. The inputs were total deposits and total cost whiles the outputs were total loans and other earning Assets. In estimating, the efficiency score, Data envelopment Analysis was used. The relationship with profitability is expected to be positive (Garza-Garcia, 2011). ii. Scale Efficiency (ESS): This is efficiency attributable to a firm s ability to operate at a lower cost compared to competitors. Following Berger (1995) and Garza-Garcia (2012) the scale efficiency was computed by deriving the X-efficiency under Constant Return to Scale (CRS) and Variable Return to Scale respectively. The scale efficiency is then the CRS over VRS (CRS/VRS). The relationship with profitability is expected to be positive (Berger, 2005). iii. Credit Risk (NPL): Credit risk is the risk of default by customers who take out loans from the bank. It is expressed as the ratio of Non-Performing Loans to Total Asset. It is expected have a negative relationship with profitability since a large number of defaults reduce profitability (Garcia- Herrero, 2009; Garza-Garcia, 2011). 27

28 iv. Capital Adequacy (CAP): This variable measures how adequately banks are capitalized and the relationship with profitability is expected to be positive. It is computed as the ratio of total capital over Total Asset Garza-Garcia, 2011). Macroeconomic Determinants i. Growth of Annual GDP (GPD): Gross Domestic Product is a measure of the volume of economic activities in a fiscal year. It gives an idea about the flow of goods and services in an economic cycle. This has a relationship with the level of loans and deposit rate in an economy. The expected relationship with inflation is positive (Sufian and Noor, 2012). For the study, Annual growth of Aggregate GDP was obtained from the World Bank database. ii. Inflation (INF): Inflation affects both the earnings and costs (Sufian and Noor, 2012). Inflation also affects important variables like labour cost and asset prices (Staikouras and Wood, 2003). The Consumer Price Index (CPI) is used to proxy for annual inflation rate. The sign of this inflationary determinant is often is often ambiguous (Garza-Garcia, 2011). 3.4 Data To ensure a comprehensive coverage of the Ghanaian Banking Sector, the dataset for the present study is constructed using an unbalanced panel of all the 23 commercial banks in the industry (see Appendix A for list of banks). The dataset includes both bank-level data (financial ratios) and macroeconomic variables spanning the period 2007 to 2012 (Appendix B shows the mathematical expression of the variables). The bank-level data are obtained from three main sources, that is, Bankscope Database, PricewaterhouseCoopers Ghana and the banks Annual Report. The macroeconomic variables Annual growth of GDP and Inflation, are obtained from the World Bank and the Ghana Statistical Service database respectively. The dataset is unbalanced because over the sample period, there were series of Mergers and 28

29 Acquisitions 8. This result in some banks exiting the market whiles new ones entered. Therefore it becomes impossible to construct a balanced panel where all the cross-sectional observations have the same number of time series. 3.5 Econometric Estimation The estimation was done in two stages. The first part dealt with the estimation of the X- efficiency and Scale efficiency using the Data Envelopment Analysis (DEA). The second part entails the use of the dynamic panel estimator (Generalized Method of Moment) to derive the coefficients of the four models Data Envelopment Analysis (DEA) Data Envelopment Analysis (DEA) is a non-parametric method of determining the efficiency of a group of similar entities referred to as Decision Making Units (DMUs) (Cooper, Seiford and Zhu, 2004). It uses linear programming approach to first of all determine an efficient frontier. This efficiency frontier represents the optimum level of efficiency that each firm (DMU) can attain. Thereafter, it measures the distance of each DMU from this frontier to determine their level of inefficiency. DMU s that lie exactly on this frontier are referred to as fully efficient whiles those who are not are inefficient based on their distance from this frontier (Garza-Garcia, 2012). The attraction of this method is that, it measures the efficiency frontier based on the data available (Cooper et al, 2004). Also there is no need for any strong assumptions about the availability of the inputs to every DMU, or whether the market is in equilibrium or not (Cooper et al, 2004). This often makes DEA a preferred method to other parametric methods such as the Stochastic Frontier Analysis. As long as the inputs and output of the DMU s are quantifiable, efficiency of each DMU (Bank) can be determine. For an industry, that is input-oriented, efficiency will represent a reduction in the amount of inputs it 8 Appendix 1 shows how many yearly observations of each bank were used in constructing the unbalanced panel for the present study. The year of incorporation for each bank was stated as well. 29

30 uses whiles output remain constant or increase (Cooper et al, 2004). Studies on Banks efficiency (Garza-Garcia, 2011; Casu and Molyneux, 2003) mostly treat banks as inputoriented entities. This is because bank profitability is determined mostly by how efficient they control their cost (Garza-Garcia, 2011; Casu and Molyneux, 2003). Charnel et al, (1978) first proposed a Constant Return to Scale (CRS) when computing the efficiency scores of the DMU s. Banker et al, (1984) thereafter, proposed a variable return to scale (VRS) when determining the efficiencies of the DMU s since Constant return to scale (CRS) will imply that all the DMU s are operating at an optimal scale. Banker et al, (1984), argued that firms do not always operate at an optimal scale. Casu and Molyneux (2003) further explained that imperfect competition and financial constraints restraints banks from operating at optimal scale. Following Garza-Garcia (2012) and Casu & Molyneux (2003) the inputs and outputs that were used for the efficiency estimates are listed below. Inputs: 1. Total Cost (Personnel and administrative) 2. Total Deposit Output: 1. Total loans 2. Other earning assets. The efficiency scores for each bank were computed using the Efficiency Measurement System (EMS) 9 software. The EMS computed the X-efficiency and the Scale efficiency, which are part of the bank- Specific determinants. 9 Efficiency Measurement System (EMS) is software used to calculate X and Scale Efficiency of the peer entities. This can be downloaded from 30

31 Dynamic panel Data Panel data is mostly preferred for estimating dynamic econometric models because it mitigates some of the shortcomings associated with pure cross sectional and time series regressions. Bond (2002) pointed out that cross-sectional regressions are inappropriate for dynamic models because it only considers observations over a single point in time. Therefore it is impossible to establish any long run dynamics (changes) in the data. Whiles a pure time series will capture these dynamics, it does not control unobserved heterogeneity associated with data on different entities. Therefore using a pure time series will mean that certain bankspecific effects in the dataset will not be dealt with leading to biased estimates (Baltagi, 2001, Boyd et al 2001; Bond, 2002). The use of panel data helps to overcome the shortcomings above. A dynamic panel data is a panel data model that includes lags of the dependent variables as one of the explanatory variables (Bond, 2002). In the present study, lagged ROA and ROE were used as explanatory variables making the model a dynamic one. The use of dynamic panel data models violates one the classical assumptions in econometrics known as strict exogeneity (Arelanno and Bond, 1991; Bond 2002). Strict exogeneity rules out any feedback from current or past shocks to current values of the variable, which is often not a natural restriction in the context of economics models relating several jointly determined outcomes, such as consumption and income or investment and Tobin s q (Bond 2002, pp. 2). Endogeneity therefore arises when there is dependence between the explanatory terms and the error term Generalized Method of Moment Generalized Method of Moment (GMM) is used to estimate dynamic models when endogeneity is present, even after controlling for unobserved fixed effect by first differencing (Wooldridge, 2001). Roodman (2009) touched on the following advantages of GMM estimator such as, 31

32 (1) Appropriate for panels with large number of cross-sectional observation N over a short time period T 2) Where the model is a linear one. 3) When the model contains a lag of the dependent variable. 4) Where strictly exogenous cannot be assumed for some of the explanatory variables. 5) Presence of fixed effect as well as heteroskadasticity and autocorrelation. GMM estimators are of two types, that is Differenced GMM and System GMM. (i) Difference GMM (Arellano and Bond Estimator): Difference GMM estimators operate by using the immediate lag values to first-difference (transform) the model (Roodman, 2009). This eliminates the unobserved fixed effect (µ it ) in the model but does not entirely, resolve endogeneity because, the idiosyncratic disturbance term (v it ) will still be correlated with the lagged dependent variable (Wooldridge, 2001; Bond 2002). Difference GMM then uses the second lag and beyond as the instrumental variables (IV) 10 in the model to deal with this endogeneity problem (Wooldridge, 2001). Anderson Hsiao originally proposed GMM in 1982 and espoused the use of only the second lag as instruments (Athanasoglou, 2005). However, Arrelano and Bond (1991) argued that the use of additional lags aside the second lag, greatly improves the efficiency of the estimates (Athanasoglou, 2005). Therefore Arrelano and Bond (1991,) proposed that to gain efficiency, all available lags of the dependent and exogenous explanatory variables must be used as instruments. Even though Arellano and Bond estimator (Differenced GMM) became very popular, it also came under a lot of criticism especially when applied to panels with small time period. (Athanasoglou et al, 2005). Arellano and Bover (1995) and Blundell and Bover (1998) pointed out that, where the 10 Instrumental variables are correlated with the other explanatory variables but uncorrelated with the error term. The second lag and beyond are valid instruments because they are uncorrelated with the idiosyncratic error term (V it ). 32

33 time period T is small, the estimates tend to be inefficient if the instruments are weak (Cited in Athanasoglou et al, 2005). ii. System GMM: The weakness of the difference GMM, to fully exploit all the moment conditions to improve efficiency of the estimates led to the development of an augmented estimator known as System GMM, by Arellano and Bover (1995) and Blundell and Bover (1998). System GMM compliments differenced GMM, by making an additional assumption about the instrumental variable (IV) (Roodman, 2009). System GMM assumes that the first difference of the instrument variables is uncorrelated with the fixed effect (Roodman, 2009). The implication of this assumption is that, the first difference of the instrument can also be used as valid instruments in the model. This increment in the number of instrumental variables greatly improves the efficiency of the estimates (Roodman, 2009). According to Roodman (2009), system GMM basically works by building two systems of equation. The first is the original equation of the model while the second equation is the transformed one. System GMM is often preferred to difference GMM, where the panel is unbalanced and contains missing values (Roodman, 2009). 3.6 Justification of System GMM. The present study used the System GMM to estimate the parameters of the model. The justifications are that, 1) The model is a dynamic one containing lags of the dependent variable. Strict exogeneity cannot be assumed. 2) The other explanatory variables in the model are not strictly exogenous specifically; there will be some correlation between variables like market share (MS) and the efficiency variables (X-efficiency and Scale efficiency) 33

34 3) The sample period in the dataset is not long enough (2007 to 2012). Using, difference GMM can lead to inefficient estimates if the instruments are not strong enough. Arellano and Bover (1996) showed that in such an instance system GMM performs better. 4) The panel was an unbalanced one and contains some missing values. Difference GMM will reduce the number of observation making it smaller. 3.7 xtabond2 The present studying performed the System GMM using the xtabond2 11 command developed by Roodman (2009). Xtabond2 is widely used in Stata because of its attractive features among which is its diagnostic testing. It reports the necessary diagnostic test and indicates whether the model is valid or not. 11 Refer to How to do xtabond2: An introduction to Difference and System GMM in Stata for full explanation of xtabond2 command. 34

35 CHAPTER FOUR 4. RESULTS AND DISCUSSION: The present study employs an unbalanced panel of 23 commercial banks in Ghana covering the period 2007 to Table 1 (Appendix C) shows a descriptive statistic of all the variables of interest in the study whiles Tables 2 and 3 (also in the Appendix C) provides a correlation matrix of the dependent and explanatory variables. The study used a system GMM to estimate banking competition and profitability. Two profitability measures, (ROA and ROE), were used to as the dependent variables in the present study. Following Garza- Garcia (2012) four models are used in the estimation process. Model 1 and 2 use ROA as the dependent variables whiles Model 3 and 4 use ROE. Model 1 and 3 are estimated without the macroeconomic explanatory variables whiles Model 3 and 4 has the macroeconomic variables included in the explanatory variables. The section begins with the descriptive statistics of the study. This is followed by analysis of the empirical estimates of the four models. Afterwards, a brief discussion of the results is done in relation to the Ghanaian Banking Sector. Lastly, the necessary diagnostic tests are reported to determine the robustness and reliability of the results. 4.1 Results and Analysis Descriptive Statistics. Table 1(in Appendix C) gives a summary description of the data that is used for the study and revealed some interesting insights into the Ghanaian Banking sector over the past six years. From Table 1, the average ROA is 1.86% with a standard deviation of 3.5%. The standard deviation is probably wide because there were mergers and acquisitions of banks over the sample period and this led to instances where banks reported negative returns in the first few years of operation. The ROE has a mean value of 14.59% with a maximum value of 60.33% 35

36 indicating a very high turnover within the banking sector. The market share over the period has a mean value of 4.32%, which possibly reflects the competitive nature of the sector. The highest market share recorded over the period is 15.67%. The minimum level of market concentration represented by the Herfindahl-Hirschman Index (HHI) is and increase to over the period as well. The efficiency measures show a 12% and 28% level of inefficiencies for X-efficiency and Scale efficiency respectively. Non-Performing loans average 7.13% and went up as high as 34.41% over the sample period. Capital adequacy shows an average value of 14.55% and a maximum level of 83.21%. This high ratio could be a sign of inefficiency and probably confirms the long held perception that loan penetration of banks in developing economies is low. This could also be a good sign of capital sufficiency of the banks as a result of the Recapitalisation Policy introduced in Lastly macroeconomic variables like GDP and CPI shows a mean value of 8.37% and 12.49% respectively. The annual GDP growth rate is quite high over the period and went up as high as 15% representing a fast growing economy. However just like the Philips curve predicted, the inflationary rate is high as well with a maximum value of 19.3% and a standard deviation of 4.07% Analysis and Interpretation of Empirical Results From Table 4 (Appendix C), the lagged profitability measures (ROA and ROE) are positive and significant in all the four models. Model 1 for instance, has a coefficient of 0.27 but increases to 0.3 when GDP and CPI are introduced in Model 2. Similarly, Model 3 had a coefficient of 0.33 but increases to 0.36 with the inclusion of GDP and CPI in Model 4. Using Model 2, it can be interpreted that a unit increase in last year s ROA will increase this year s ROA by 0.3 units. These coefficients indicate the level of profit persistence (Goddard et al, 2004; Garza Garcia, 2011). According to Athanasoglou et al (2005) a positive coefficient is expected for the lagged dependent variable and ranges between 0 and 1 showing the 36

37 dynamics of profit adjustment. Thus a value close to 0 represents a very competitive banking sector whiles values closer to 1 indicates an uncompetitive sector. Overall the result shows profit persistence within the Ghanaian banking sector is a moderate one. This finding is similar to work done by Athanasoglou et al (2005) who reported a coefficient of 0.35 for Greek banks. Garza-Garcia (2011) on the other hand reported profit persistence for Mexican Banks to be between and Other papers like Goddard et al (2004) found a weak insignificant coefficient for profit persistence among European banks reflecting the competitive nature of the European banking sector. The estimate for market concentration (HHI) in Table 4 shows a negative relationship with bank profitability without the macroeconomic variables in Model 1 and 3, but changes to positive with GDP and CPI introduced in Model 2 and 4. However, the results in the four models are insignificant which implies a rejection of the SCP hypothesis. This possibly indicates that the Ghanaian banking sector is not in control of a few banks leading to abnormal profits. This finding contradicts earlier work by Busch and Mathisen (2005) who found the Ghanaian Banking Sector to be a monopolistic one over the period 1998 to Since the present study used recent data covering 2007 to 2012, it could be an indication that the sector has become more competitive. Also there were a lot of reforms that led to series of mergers and acquisitions and could possibly explain the varied result from Busch and Mathisen (2005). The findings are however, consistent with that of Athanasoglou (2005) and Garza- Garcia (2011) who found no significance relationship between market concentration (measured by HHI) and bank profitability in Greece and Mexico respectively. The empirical results also show that bank profitability is not driven by market share as postulated by Relative Market Power (RMP) hypothesis. From Table 4 the coefficients for market share (MS) in the four models are positive. For model 1 and 2, the coefficient remained the same at It increases from to 0.43 in Model 3 and 4 respectively. 37

38 However, these estimates are insignificant and thus we reject the conjecture that profitability is driven by market share (Relative market power). This result contradicts work done by Garza-Garcia (2011) and Berger (1995) who found positive significant relationship between profitability and market share. None of the efficiency measures (EXS & ESS) used in the models shows significant relationship with profitability. Table 4 shows that X efficiency (EXS) are positively related to profitability while the scale efficiency (ESS) shows a mixed relationship with profitability. Therefore, the hypothesis that bank profitability is determined by managerial and/or scale efficiency is rejected. Berger (2004) and Garza- Garcia (2012) who found similar results concluded that efficiency generally does not drive profitability. Banks may remain efficient but the insignificant result possibly indicates that this efficiency is not geared toward improving on profitability. Turning to other bank-specific determinants of profitability, Table 4 showed that credit risk is a significant determinant of bank profitability in Ghana. Ratio of Non-Performing Loans (NPL), used to measure credit risk, shows a negative significant relationship in all but the 3rd model. The introduction of GDP and CPI in Model 2 increases the coefficient slightly from to and are both significant at 5% significance level. NPL is insignificant in Model 3, but gained significance at 10% in Model 4, with in the introduction of macroeconomic factors. Using Model 2, the result indicates that a unit increase in the ratio of non- performing loans will decrease bank profitability by The negative relationship is expected since an increase in NPL reduces bank profitability. The finding corroborates that of Athanasoglou et al, (2005) and Miller & Noulas, (2007) but contradicts that of Garza-Garcia (2011) who found no relationship between NPL and profitability for Mexican banks. 38

39 Capital ratio (CAP) is positive in the entire four models but only significant in the 1st from Table 4. The coefficient of the 1st Model indicates that a unit increase in capitalisation will lead to unit increase in the ROA. CAP is a measure of capital adequacy and a positive relationship with profitability is expected. For instance a well-capitalised bank is able to take advantage of new opportunity and increase profitability (Garcia-Herrero et al, 2007 cited in Garza-Garcia et al, 2011). The findings are not robust enough because the significance was only observed in the first model. The Recapitalisation Policy implemented in 2007 could possibly be a plausible reason. Banks over this period increased their capital because regulators demanded them to do so and not to take advantage of new investment opportunities. This result is similar to Garza-Garcia (2011) who reported significant relationship between capitalisation and ROA as the dependent variable and not ROE. For the macroeconomic variables (GDP and Inflation) the coefficients are significant. From Table 4 GDP is positively significant at 1% and 10% in Model 2 and 4 respectively. Taking Model 2, the coefficient indicates that one unit increase GDP will increase banks profitability (ROA) by 0.26 units. In the same way Model 4 also indicates that a unit increase in GDP will increase banks profitability by 2.3 units. The result is consistent with that of Demirguc-Kunt and Huzinga (1998) Bikker and Hu (2000) and Athanasoglou et al (2005). GDP growth is associated with demand for financial activities and a positive significant relationship as the model indicated is not a surprising result. The result however contradicts that of Garza-Garcia (2011) who found no relationship between bank performance and GDP growth in Mexico. Inflation coefficient, measured using the Consumer Price Index (CPI) shows a negative significant relationship with profitability. Model 2 in Table 4 indicate that a unit increase in inflation will decrease bank profitability by units. This result is not surprising because inflation rate in developing countries is volatile and usually high. Periods of high inflation can result in lower savings, which in the long run affect bank profitability. 39

40 4.2 Discussion of Results The analysis above revealed some interesting trends into Ghana s Banking Sector. The discussion is grouped into three phases namely Competiveness, Bank Specific and Macroeconomic determinants of bank profitability. Competitiveness and Market power The banking sector seems moderately competitive with profit persistence within the range of 0.27 to 0.36 in the four models. There are many reasons why we would not expect profit persistence to be 0. Financial institutions are one of the most regulated sectors of any economy, because their failure can be contagious to the entire economy. These regulations impose restrictions such as strict rules on entry and high capital requirement. These in a way lead to some level of profit persistence for the banking sector (Berger, et al 2000). Also, retained profit is the cheapest source of capital for any organisation. The more profitable a firm is the higher the capital accumulation and its flexibility in diversifying its portfolio to increase profitability. The efficiency of the banking sector was quite high as well from the descriptive statistics in Table 1 but not a significant determinant of their profitability in the empirical models. The SCP and RMP hypothesis, which relates profitability to market power and not competitiveness, were both insignificant in the models. A rejection of these two hypotheses meant that the banking sector is not concentrated, as SCP would suggest and market share does not drive banks profitability as RMP posited. These findings clearly show that the Ghanaian banking sector is moderately competitive and a departure from the repressive sector in the 1980 s. Bank Specific Determinants Financial markets in developing economies are highly imperfect making credit risk management an important consideration for banks. Problems of loan default are quite high in 40

41 developing economies mainly due to information asymmetry. Therefore it is not surprising to realise that the ratio of non- performing loans was significant in the model. Periods in which this ratio is high can lead to huge losses for the banks and the challenge for managers will be to keep this ratio as low as possible. The other bank-specific variable, capital adequacy was only significant in Model 1 so we do not generalise it as a determinant of bank profitability. Macroeconomic Determinants The macroeconomic variables are reported as the main findings of the study. The results confirm that GDP is the main determinant of bank profitability in the Ghanaian Banking Sector. The coefficients in for GDP in Model 2 and 4 are significant at 1% and 10% respectively. GDP is measures the volume of transactions that are undertaken within a fiscal year. An increment in the GDP shows an increased level of economic activities within that fiscal period. These increased economic activities will stimulate demand for banking and financial services. Therefore bank profitability will likely increase in periods of economic boom and shrink when economic activities decline. However, it could also bring about the issue of pro-cyclicality of banks performance with the general economy (Bikker and Hu, 2002). When the economy experience periods of low demand, bank profitability will likely plummet further worsening the economic situations. Subsequently, default on loans due to economic decline will increase the ratio of non- performing loans for the banks reducing profit further (Bikker and Hu, 2002). Inflation, in the model was also significant and negatively related to bank profitability, which implies that high levels of inflation will likely reduce bank profitability, and vice versa. Banks makes profit mainly through the spread between borrowing and lending rate. When there is high level of inflation, the interest rates are often adjusted upward. This increases the cost of borrowing, which leads to reduction in demand for loanable funds and reduces bank 41

42 profitability. Borrowers who are charged variable interest rates or lines of credit will experience difficulty in repaying back loans or taking out further loans in the future. 4.3 Diagnostic test In any econometric test, there are various assumptions that must hold for the model to be robust and make the findings acceptable. These specification tests help improve the validity of the results. Below are the necessary specification tests that have been carried out for the study. It is important to point out that, the model is robust for all the specification tests Heteroskedasticity Classical regression models assume that the error term (µ it ) has a constant variance over time (Brooks, 2008). Baltagi (2005) pointed out that, this assumption is unrealistic for panel data models since it comprises data on different entities (groups). For this reason, it will be impossible to expect that the variance of the error term for all the banks in the present study will be constant. The presence of heteroskedasticity in panel data model make the standard error inefficient and Baltagi (2005) recommended a robust standard error to correct for heteroskedasticity. This study reported the robust standard error together with the estimates in Table 4 in parenthesis () Autocorrelation The error term of a panel data has two components namely the individual-specific error term (µ it ) and a random error term (v it ) (Brooks, 2008). First differencing the model with the immediate lag removes this individual-specific error term leaving the model with only the random error term (v it ) (Baltagi, 2001). Additionally dynamic panel estimators (GMM) work by using the second lags and beyond of the dependent variable as instrumental variables, which makes the model susceptible to serial correlation if there is white noise in the error term (Arellano and Bond, 1991). Arellano and Bond (1991) argued that the presence of 42

43 autocorrelation in the residual term does not lead to inconsistency of the estimates if it is of first order. i. First order Autocorrelation: AR (1) First differencing dynamic panel regression removes the individual specific error term (µ it ). This leaves the first differenced equation with only the random error term (v it ) (Baltagi, 2001). Arellano and Bond (1991) showed that the random error term in the first differenced equation does not need to be necessarily serially uncorrelated with its lag (E (v it v it-1 ) 0). They argued that the presence of first order autocorrelation AR (1) does not make the GMM estimates inconsistent. H 0: E (v it v it-1 ) 0 H 1: E (v it v it-1 ) =0 ii. Second Order Autocorrelation: AR (2) The AR (2) test for lack of second-order serial correlation in the first differenced residuals. If the lagged dependent variables used as instrument has a second order autocorrelation with the differenced equation, then it must be dropped since the GMM estimates will be inconsistent (Baum, 2013). Arellano and Bond (1991) further pointed out that the absence of second order autocorrelation means that the error components follow a random walk process. H 0: E (v it v it-2 ) 0 H 1: E (v it v it-2 ) =0 As shown in Table 4, we reject the null hypothesis that there is second order autocorrelation. Table 4 reports the t-stats and the p- value for the AR (1) and AR (2) process. As expected the AR (1) shows that there is autocorrelation of first order in all the four models at 5% significance level. This means that we are 95% confident that there is first serial autocorrelation between the error terms. 43

44 Similarly the AR (2) process shows that there is no second order autocorrelation in the model with high p-values justifying that the first differenced residuals are not second-order serially correlated. This test result is consistent with that of Athanasoglou et al (2005) and Garza- Garcia, (2011) where there was first order autocorrelation but no second order autocorrelation Hansen J-Statistic (Over-identification) GMM estimators work very well in models when strict exogeneity of the variables cannot be assumed (Wooldridge, 2010; Arellano and Bond 1991; Roodman 2009). An explanatory variable is said to be exogenous when it is not correlated with the disturbance term (Baltagi, 2001). That is, E (X µ it ) = 0, where X is the explanatory variable and µ it the disturbance term. On the other hand, a variable is endogenous if it is correlated with the disturbance term (Wooldridge, 2010). This implies that the expectation of the variable and the error term is not equal to zero (E (X µ) 0)). Since the residual term (µ) is related with the dependent variable (Y), the endogenous variable is correlated with the dependent variable as well (Bond, 2002). Instrumental Variables (IV) are required to deal with endogeneity in econometric models. An Instrumental Variable can be thought of as an explanatory variable that is correlated with the explanatory variables but not with the error term (Baltagi, 2001). Their presence in the model helps to deal with the endogeneity in dynamic models. The use of instrumental variables in the model requires a test specification known as the Hansen J- Statistic. This test requires that the number of excluded instruments is greater than the number of endogenous variables to make the model valid 12 (Baum Schaffer and Stillman, 2007). Baum et al, (2007) assumed a model where, Y = X i β + µ i. (1) 12 Refer to Baum et al, (2007) for exposition on the J-Hansen Statistic. 44

45 Where: X i is the explanatory variable and β, the coefficient. The explanatory variable, Xi s matrix is n K, where n represents the number of observation. The explanatory variables are grouped into two that is, X 1 and X 2 Where; X 1 explanatory variables K 1 are assumed to be endogenous and X 2 explanatory variables K 2 are assumed to be exogenous. Equation (1) is rewritten to become, Y = ((X 1 X 2 ) (β 1 β 2 )) + µ i. (2) A matrix of instrumental variables Z is n P, assumed to be exogenous is introduced into the model. These instrumental variables are also separated into two, that is, Z 1 Z 2 with their corresponding instruments P 1 and P 2. P 1 instruments are referred to as excluded instruments and P 2 instruments Z 2 as included instruments. The Z 2 instruments (Included) are exogenous instruments and added to the original explanatory variables. The combination of the instrumental variables Z 2 with the original explanatory variables leads to X= (X 1 X 2 ) = (X 1 Z 2 ) (3) The explanatory variables now have two terms; endogenous explanatory variables (X 1 ) and exogenous instruments (Z 2 ). The Instrumental variables can be written as, Z = (Z 1 Z 2 ) (4) Observing equation (3) and (4) one realises the Z 2 term is common to both of them. The Hansen test for over-identification requires that the number of instruments is greater than the number of endogenous explanatory variable (P K). This intuitively, means, the number of excluded instruments P 1 should be more than the number of endogenous explanatory variables, K 1 for the model to be over-identified. (Baum et al, 2007). Where this condition 45

46 does not hold, the model is said to be under-identified. The Hansen test has a null hypothesis that the model is not over identified. H 0: The model is not over identified. H 1 : The model is over identified For the model to be over to be over-identified, the p-value must not be significant at any level of significance employed in the study. Table 4 shows very high p-value for J-Hansen Statistics at all the three levels of significance (1%, 5% and 10%). Therefore we reject the null hypothesis and conclude that the model is strongly overidentified F Test: The F-test result shows the overall significance of the regression equation. It tests for the significance of all the variables used in the model unlike the t-statistics that test the significance of individual explanatory variables (Brooks, 2008). In all the four models, the p- value shows that the model is strongly significant at 1% in all but the 1st Model, which is significant at 5%. However the coefficients for the F-test are very low overall. 46

47 CHAPTER 5 5 CONCLUSIONS 5.1 Summary of Main Findings The thesis investigates banking competition in Ghana and the effect of market power, bankspecific and macroeconomic factors on bank profitability in Ghana. The estimation process explicitly, modelled for all the parameters especially efficiency using the non- parametric Data Envelopment Analysis (DEA). Then system GMM estimator was then used to estimate the parameters of the model. The main findings of this study are summarised as follows. Firstly, the Ghanaian Banking Sector is found to be moderately competitive. The lagged profitability coefficients are low (between 0.27 and 0.36) indicating a low level of profit persistence. The Structural Conduct Performance (SCP) and Relative Market Power (RMP) Hypotheses are both rejected in the study indicating that the sector is not concentrated and banks that control larger market share do not necessarily determine competition and profitability. This finding is a departure from earlier study by Bushs & Mathisen (2005) who found the Ghanaian Banking Sector to be uncompetitive. Macroeconomic factors, Annual growth of GDP and Inflation are found to be the main determinants of bank profitability. Theories suggest that bank profitability tend to be procyclical with economic performance. This is because banks act as intermediaries for economic agents. Therefore, periods of economic transactions will increase banks profitability (Bikker and Hu, 2002). Inflation rate in Ghana has been quite high over the period of the study and the thesis confirmed a negatively impacts on bank profitability. Subsequently the study also found credit risk to be the only bank specific determinants of bank profitability in Ghana while capital adequacy was not robust enough to be considered as a main determinant of profitability. Though the efficiency scores show that on the average, 47

48 the sector is efficient at 88% and 72% for X-efficiency and Scale efficiency respectively, they weren t significant determinants of bank profitability. 5.2 Policy Recommendations The main policy recommendation from the study is that, the central bank of Ghana should put in place measures to manage the pro-cyclicality of the banking sector with the economy. From the study, annual growth of GDP is found to be the main determinant of bank performance. Pro-cyclicality of the Banking sector with the economy is harmful in periods of recessions or financial crisis (Bikker and Hu, 2002). This is because, the economy in these periods, experience low demand for goods and service. This also trickles down to the financial sector because demands for banking services reduce well. Moreover, during recessions and financial crisis, there is a lot of credit default increasing the ratio of Non- Performing Loans. This will adversely impact on the profitability of banks. Also it has been established by scholars that in periods of low economic growth, bank tend to reduce, the amount of credit they give out. Government can manage this pro-cyclicality by designing stimulus plans early enough during recession to mitigate some of the adverse impact of recessions on the banking sector and the economy as a whole. The central bank also needs to re-evaluate their Recapitalization Policy introduced in 2007 to find out how it can be enhanced to boost bank profitability. The present study found capitalisation to be insignificant in determining bank profitability. Coincidentally, the sample period of the study encapsulate the period when the Recapitalisation Policy was introduced. It may be possible that banks are well capitalised, but leave most of these funds idle which is a source of inefficiency for them and the entire economy. Specifically, regulators should set different thresholds of capital requirement. This will ensure that banks who do not have 48

49 enough investment opportunities in a particular period will keep a lower capital and vice versa. Lastly, the study also showed that, the sector has responded quite well to the reforms over the period. The findings indicated that the sector is not concentrated, or being dictated by banks with large market share. Thus the competitiveness of the sector has improved over the sample period. This competitiveness can be accredited to the series of reform over the years such as the introduction of the Universal Banking License in 2003 and Banking Act The mergers and acquisitions over the sample period did not lead to a concentrated market as in other jurisdictions. However, it is essential to ensure that these gains are harnessed and preserved. Reforms such as Credit Referencing should be improved so as to reduce the adverse impact of credit default on the economy. 5.3 Limitation of the Study The first limitation of the study has to do with the sample period. The study intended to cover the period, 2000 to This would have helped to gain a much more comprehensive insight on the determinants of bank profitability in Ghana. However, data unavailability limited the sample period to only 2003 to Subsequently, this period was further reduced to 6 years, that is, 2007 to 2012 because of the econometric method employed; system GMM. The System GMM requires a large number of cross-sectional observations as against a shorter time period (Roodman, 2009). The number of commercial banks in the sector is only 23, and therefore the sample period was reduced to ensure that the number of instrumental variables does not proliferate. The second limitation has to do with the number of instruments used in the estimation of the parameters. As stated in Roodman (2009), it is ideal that when using system GMM, the number of instrument must be equal to the number of cross sectional observations. Roodman 49

50 (2009) argued where this requirement is not satisfied, the Hansen-test for over-identification can be weakened. The present study violated this requirement. The number of instruments was more than the cross-sectional observation. Specifically, the study used 29 instruments as against 23 cross-sectional observations (refer to Appendix E) implying the instruments exceeded the number of observation by 6. However, the Hansen-test was not weakened by this violation but remained robust. The present study adopted a model used by Garza-Garcia (2012) however; it excludes two of his explanatory variables, that is, Liquidity Risk and Interest rate volatility due to unavailability of data on them for banks in Ghana. Data on these variables would have improved on the comprehensiveness of the study. Another limitation of the study has to do with the coefficient of the F-Test. From Table 4, the coefficients of the F-Test were very low even though they were significant. Lastly, the Ghanaian financial sector is made up of Commercial, Rural Banks, Microfinance Institutions and other non-financial institution and the stock market as well. However, the present study only dwelled on the commercial banks and the findings are thus limited to only commercial banks and not a generalization for the entire financial sector. 5.4 Further Study The present study uses persistence of profit (POP) to measure banking competition in Ghana as well using Data Envelopment Analysis (DEA) to measure banking efficiency. As an extension, the present study I propose future studies to consider measuring banking competition using the Panzar-Ross Model and parametric methods such as Stochastic Frontier Analysis (SFA) to measure banking efficiency. This should help to augment the present study and provide basis for critically assessing the results. 50

51 References Ametefe, F., Aboagye, A. Q. Q. and Sarpong-Kumankoma, E, (2011) Housing and Construction Finance, Deposit Mobilization and Bank Performance in Ghana, Journal of Property Research, 28 (2) Arellano, M., and Bond, S.R, (1991) Some Test of Specification for Panel Data: Monte Carlo Evidence and an Application to Employment Equation, Review of Economic Studies, Arellano, M., and Bover, O, (1995) Another Look at the Instrumental Estimation of Error- Components, Journal of Economics, 68, Arpa, M., Giulini, I., Ittner, A., and Pauer, (2001) The Influence of Macroeconomic Development on Austrian Banks: Implications for Banking Supervision, BIS Papers, No. 1. Athanasouglou, P., Brissimis, P. and Delis, M, (2005) Bank-Specific, Industry-Specific and Macroeconomic Determinants of Bank Profitability, Bank of Greece Working Paper No 25. Baltagi, B, H. (2001) Econometric Analysis of Panel Data. 2nd edn. John Wiley & Sons: Chichester. Banker, R., Charnes, A., and Cooper, A.W, (1984) Some Models for Estimating Technical and Scale Efficiency Inefficiencies in Data Envelopment Analysis, Management Science Baum, C. F., Schaffer, M. E., and Stillman, S, (2003) Instrumental Variables and GMM: Estimation and Testing, The Stata Journal 3, Bawumia, M., Belnye, F., and Ofori, M. E, (2005) The Determinants of Banks Interest Spread in Ghana: An Empirical Analysis of Panel Data, Bank of Ghana Working Paper No 2005/09. Berger, A.N. (1995) The Profitability-Structure Relationship in Banking: Test of Market Power and Efficiency-Structure Hypothesis, Journal of Money, Credit and Banking, 27,

52 Berger, A. N., Bonime, S. D., and Covitz, D. M, (2000) Why are Bank Profits so Persistent? The Role of Product Market Competition, Informational Opacity, and Regional/Macroeconomic Shocks, Journal of Banking and Finance 24, Berger, A. N., and Hannan, T. H, (1997) Using Efficiency Measures to Distinguish Among Alternative Explanations of the Structure- Performance Relationship in Banking, Managerial Finance, 23, Bernanke, B.S. (1983), Nonmonetary Effect of the Financial Crisis in the Propagation of the Great Depression. American Economic Review, 73(3), Bikker, J. A., and Hu, H, (2000) Cyclical Patterns in Profits, Provisional and Lending of Banks, DNB Staff Reports, No 86/2002. Bikker, J. A., Shaffer, S., and Spierdijk, L, (2012) Assessing Competition Using the Panzar- Rosse Model: The Role of Scale, Costs and Equilibrium, The Review of Economics and Statistics, 94 (4), Blundell, R., and Bond, S, (1998) Initial Conditions and Moment Restrictions in Dynamic Panels Models, Journal of Econometrics 87, Bond, R.S., and Greenberg, W, (1976) Industry Structure, Market Rivalry and Public Policy: A Comment, Journal of Law and Economics, 19 (1), Bond, S. (2002) Dynamic Panel Data Models: A Guide to Micro Data Methods and Practices, Cemmap Working Paper, CWP, 09/02. Boyd, J. H., Levine, R., and Smith, B. D, (2001) The Impact of Inflation on Financial Sector Performance, Journal of Monetary Economics 47(2), Bourke, P. (1989) Concentration and Other Determinants of Bank Profitability in Europe, North America and Australia Journal of Banking and Finance, 13, Brooks, C. (2008) Introductory Econometrics for Finance. 2 nd edn. Cambridge University Press, Cambridge. 52

53 Brownbridge, M., and Gockel, A. F, (1996) The Impact of Financial Sector Policies on Banking in Ghana, Institute of Development Studies Buchs, T., and Mathisen, J, (2005) Competition and Efficiency in Banking: Behavioural Evidence from Ghana, IMF Working Paper, No 05/17. Casu, B., and Molyneux, P, (2003) A Comparative Study of Efficiency in European Banking, Journal of Applied Economics, 35, Charnes, A., Cooper, W. W, and Rhodes, E, (1978) Measuring Efficiency of Decision Making Units. European Journal of Operational Research, 2, Cooper, W. W., Seiford, L. M, and Zhu, J, (2004) Handbook on Data Envelopment Analysis, Kluwer International Series Demirguc-Kunt, A., and Huizinga, H, (1999) Determinants of Commercial Bank Interest Margins and Profitability: Some International Evidence, World Bank Economic Review, 13, Garcia-Herrero, A., Gavila, S., and Santabarbara, D, (2009) What Explains the Low Profitability of Chinese Banks? Banco de Espana Working Paper, No 0910 Garza-Garcia, J. G. (2011) Determinants of Bank Performance in Mexico: Efficiency or Market Power Centre For Global Finance Working Paper Series No: 03/11. Garza-Garcia, J. G. (2012) Does Market Power Influence Bank Profit in Mexico: A Study on Market Power and Efficiency, Applied Financial Economics 22, Ghana Banking Survey (2012) Enhancing Customer Value to Sustain Profitable Growth. PricewaterhouseCoopers Survey. Ghana Banking Survey (2008), Raising the Bar: Increase in the minimum Capital Requirements, and Implications for the Industry. PricewaterhouseCoopers Survey, Goddard, J., Molyneux, P., and Wilson, J.O.S (2001) European Banking Efficiency, Technology and Growth, Wiley and Sons: England. 53

54 Goddard, J. P., Molyneux, P, and Wilson, J. O. S. (2004) The Profitability of European Banks: A Cross-Sectional and Dynamic Panel Analysis, Manchester School, 72, Goldberg, L. G, and Rai, A, (1996) The Structure Performance Relationship of European Banks, Journal of Banking and Finance, 20 (4), Maudos, J. (1998) Market Structure and Performance in Spanish Banking Using a Direct Measure of Efficiency, Applied Financial Economics, 8, Molyneux, P., and Thornton, J., (1992) Determinant of European Bank Profitability: A Note Journal of banking Finance, 16, Molyneux, P. and Forbes, W, (1995) Market Structure and Performance in European Banking Applied Economics, 27, Mc-Williams, A., and Smart, D. L, (1993) Efficiency v. Structure Conduct Performance: Implication for Strategy and Research Journal of Management, 19 (1), Miller, S. M and Noulas, A. G, (1997) Portfolio Mix and Large-Bank Profitability in the USA, Applied Economics, 29, Mirzaei, A., Liu G., and Moore, T, (2011) Does Market Structure Matters on Bank Profitability and Stability? Emerging verses Advanced Economies, Economics and Finance Working Paper Series No: Nissan, E. (2003) Relative Market Power verses Concentration as Measure of Market Dominance: Property and Liability Insurance, Journal of Insurance Issues, 26 (2), Perry, P. (1992) Do Banks Gain or Lose from Inflation, Journal of Retail Banking, 14, Revell, J. (1979) Inflation and Financial Institutions, Financial Times: London. Rhoades, S. A. (1985) Market Share as a Source of Market Power: Some Implications and Evidence, Journal of Economics and Business, 37, Roodman, D. (2009) How to do xtabond2: An Introduction to Difference and System GMM in Stata, Stata Journal, 9,

55 Shepherd, W.G (1982) Economies of Scale and Monopoly profits in Industrial Organization, Antitrust and Public Policy (Ed) J.V Cravien, Klumer Niihoff, Boston, pp Sufian, F., and Noor, A. K. N. M. (2012) Determinants of Bank Performance in a Developing Economy: Does Bank Origin Matters? Global Business Review, 13 (1), International Monetary Fund, (2002) Financial Soundness Indicators: Analytical Aspects and Country Practices, Occasional Paper 212, Washington. Wooldridge, J. M. (2001), Applications of Generalized Method of Moments Estimation. Journal of Economic Perspective, 15, Zouari, A., Mensi, S. (2010) Efficient Structure versus Market Power: Theories and Empirical Evidence, International Journal of Economics and Finance, 2(4). 55

56 APPENDICES APPENDIX A Name of Banks and Year and length used in the Panel Sources: Ghana Banking Survey,

57 APPENDIX B Mathematical Notations of Variables No Variables Formula 1 Return on Average Asset (ROA) Earnings (Net Profit before Tax) / Total Asset (%) 2 Return on Average Equity (ROE) Earnings (Net profit before Tax) / Total Asset (%) 3 Market Share (MS) Total Asset of Bank i / total asset of all the banks (%). where i represent an individual bank. 4 Hirfindahl Hirschman Index (HHI) Summed square of market Share ( 2 ). 5 Ratio of Non-Performing Loans Non-Performing Loans/ Total Asset (%) (NPL) 6 Capital Adequacy Total Capital /Total Asset (%) 7 X- Efficiency (ESX) Expressed as Total cost over Total Revenue using Data Envelopment Analysis (DEA). The EMS software was used. 8 Scale Efficiency (ESS) ESX under variable return to Scale (VRS) / ESX under Constant Return to scale (CRS) 9 GDP Annual Growth of Aggregate GDP obtained from World Bank database. 10 Consumer Price Index (CPI) Inflation rates obtained from Ghana Statistical Database. 57

58 APPENDIX C TABLES Table 1.Descriptive Statistic (all figures are expressed as a % except HHI) Table 2. Correlation matrix of ROA and the Independent Variable Variable ROA MS HHI NPL CAP EXS ESS GDP CPI ROA 1 MS HHI NPL CAP EXS ESS GDP CPI Sources: Ghana Banking Survey,

59 Table 3. Correlation matrix of ROE and the Independent Variables Variable ROE MS HHI NPL CAP EXS ESS GDP CPI ROE 1 MS HHI NPL CAP EXS ESS GDP CPI Sources: Ghana Banking Survey,

60 where ROA and ROE stands for Return on Average Asset and Return on Average Equity respectively. MS represents Market Share and HHI for Herfindahl-Hirschman Index, NPL for Ratio of Non- Performing Loans, CAP for Capital to Asset ratio. EXS stands for X-efficiency, ESS for scale efficiency, GDP for Annual growth of Aggregate Gross Domestic product and CPI for Inflation. 60