University of Ghana http://ugspace.ug.edu.gh STOCK MARKETS, BANKS AND GROWTH IN SUB- SAHARAN AFRICA BY PETER SAITOTI NGOTHO (10509736) THIS THESIS IS SUBMITTED TO THE UNIVERSITY OF GHANA, LEGON, IN PARTIAL FULFILLMENT OF THE REQUIREMENT FOR THE AWARD OF MPHIL ECONOMICS DEGREE. JULY 2016 University of Ghana http://ugspace.ug.edu.gh DECLARATION I Peter Saitoti Ngotho, hereby declare that “Stock markets, Banks and Growth in Sub-Saharan Africa” is the outcome of my research under the guidance of my supervisors. The whole or part of this thesis has not been presented in this or any other university for any academic award. I have duly acknowledged any works consulted in the course of writing this thesis by way of in- text citation and referencing. I assume full responsibility for any other shortcomings …………………………………………. PETER SAITOTI NGOTHO (10509736) ……………………………………….. DATE i University of Ghana http://ugspace.ug.edu.gh CERTIFICATION We hereby certify that “Stock Markets, Banks and Growth in Sub-Saharan Africa” was supervised as per the approved University of Ghana procedures. ……………………………………….. ………………………………….. PROF. PETER QUARTEY DR. EMMANUEL A. CODJOE (SUPERVISOR) (SUPERVISOR) ………………………………………. ………………………………….. DATE DATE ii University of Ghana http://ugspace.ug.edu.gh ABSTRACT Economic literature proposes stock market development and bank development as growth- inducing avenues. However, the long run and causal linkage focussing on the interaction between stock market development, bank development, and economic growth in Sub-Saharan Africa (SSA) has not received much attention among scholars. This study provides some insights on the long run and causal relationship between stock market development, bank development and economic growth for a group of 12 selected SSA countries for the period 1990-2014. Our sample had 13 selected SSA countries but we took out Zimbabwe, which had an outrageous mean inflation of 46.5 million percent Using the Pedroni cointegration test, this study established that stock markets and bank development do not have long run trends with economic growth for the selected SSA countries irrespective of the bank development measure used. To analyse the causal linkages, this study conducted Wald tests after running panel system GMM regressions. We used the system GMM estimator since it realizes more efficiency than the difference GMM estimator does. For the case of stock markets and bank development with economic growth and using bank credit ratio as the measure of bank development, this study found evidence supporting the supply-leading hypothesis. Further, the study established no causality between bank development and stock market development. Lastly, we concluded that the causality between stock market development, bank development, and growth for the chosen SSA economies is sensitive to the choice of bank development measure. The two alternative measures used were the broad money ratio and the domestic credit ratio. Based on the findings of this study, we recommend formulation of policies that promote further stock markets and bank development to support economic growth for the chosen SSA countries. iii University of Ghana http://ugspace.ug.edu.gh DEDICATION I dedicate this thesis, primarily, to God, Father Almighty for giving me this opportunity and sustaining me through the entire process. Secondly, I dedicate it to my mother Eunice Tito Mzee for her prayers, advice and going against all odds to finance my education. Thirdly, I dedicate this thesis to my fiancée Ms. Rebecca Kagendo and my handsome son Jayson Lenaola Saitoti for enduring two tough years of my absence. iv University of Ghana http://ugspace.ug.edu.gh ACKNOWLEDGEMENTS I will be forever grateful to the Almighty God, who has consistently placed the right opportunities along my path. His Grace has indeed been sufficient and His mercy on me, I cannot wish away. I would also like to appreciate the insightful and thought provoking comments from my supervisors Prof. Peter Quartey and Dr. Emmanuel A. Codjoe. I specifically thank them for finding time to read my work and making comments where necessary. Their constructive criticism invaluably contributed to making this study a success. In addition, I would like to express my sincere gratitude to Prof. Peter Quartey for giving me one of his publications that proved very useful to this thesis. I am also grateful to Dr. Ebo Turkson, Dr. Nketiah Amposah and the other members of staff at the Department of Economics, the University of Ghana for their comments and support. In addition, I want to sincerely express gratitude to Dr. Joseph Mucahi Muniu of the School of Economics, Kenyatta University for his comments and encouragement. I am particularly grateful to the African Economic Research Consortium in collaboration with the Government of Kenya (AERC-GoK) joint capacity building project on policy analysis and financial management for the National Treasury, for awarding me a full scholarship and fully financing this program. To my colleagues at the Department of Economics, thank you for making my stay in Ghana a memorable one. Special gratitude goes to my lovely mum, Eunice Tito Mzee for her unwavering support, prayers and encouragement throughout my life. To my lovely fiancée, Rebecca Kagendo, thank you for your encouragement and prayers. I am deeply humbled by the part each one of you played towards a successful completion of this study, God bless you all. v University of Ghana http://ugspace.ug.edu.gh TABLE OF CONTENTS Content Page DECLARATION ............................................................................................................................. i CERTIFICATION .......................................................................................................................... ii ABSTRACT ................................................................................................................................... iii DEDICATION ............................................................................................................................... iv ACKNOWLEDGEMENTS ........................................................................................................... v TABLE OF CONTENTS .............................................................................................................. vi LIST OF FIGURES ........................................................................................................................ x LIST OF TABLES ..........................................................................................................................xi ACCRONYMS .............................................................................................................................. xii CHAPTER ONE ............................................................................................................................. 1 INTRODUCTION ................................................................................................................ 1 1.1 Background ............................................................................................................. 1 1.2 Problem Statement .................................................................................................. 5 1.3 Research Objectives ................................................................................................ 8 1.4 Significance of the Study ........................................................................................ 8 1.5 Scope of the Study ................................................................................................ 10 1.6 Organization of the study ...................................................................................... 10 CHAPTER TWO .......................................................................................................................... 11 OVERVIEW OF GROWTH PER CAPITA, STOCK MARKETS AND BANK DEVELOPMENT FOR THE SELECTED SSA COUNTRIES ........................................... 11 2.1 INTRODUCTION ................................................................................................ 11 vi University of Ghana http://ugspace.ug.edu.gh 2.2 The Economic Growth Trends in the Selected Sub-Saharan Africa since the 1990s11 2.3 Brief Overview of Stock Market Development in the Selected Sub-Saharan Africa Economies since the 1990s .................................................................................................. 15 2.4 The Structure of the Banking Sector in the Study Countries .................................. 18 2.5 Trends in Bank Development in the Study Countries since the 1990s .................... 19 2.6 Conclusion ............................................................................................................ 21 CHAPTER THREE ...................................................................................................................... 22 LITERATURE REVIEW .................................................................................................... 22 3.1 Introduction .......................................................................................................... 22 3.2 Theoretical Review ............................................................................................... 22 3.2.1 Patrick’s Stages of Development Hypothesis.................................................. 23 3.2.2 McKinnon-Shaw Hypothesis.......................................................................... 24 3.2.3 Theory of Financial Intermediation with Delegated Monitoring ..................... 26 3.3 Empirical Literature .............................................................................................. 27 3.3.1 Research on Stock Market Development and Growth ..................................... 27 3.3.2 Literature on Bank Development and Growth ................................................ 31 3.3.3 Literature on both Stock Markets, Banks, and Growth.................................... 41 3.4 Conclusion ............................................................................................................ 45 Chapter Four ................................................................................................................................. 47 The Methodology ................................................................................................................ 47 4.1 Introduction .......................................................................................................... 47 4.2 Theoretical Framework ......................................................................................... 47 4.2.1 Introduction ................................................................................................... 47 4.2.2 The AK Endogenous Growth Model .............................................................. 48 vii University of Ghana http://ugspace.ug.edu.gh 4.3 The Empirical Model ............................................................................................ 51 4.3.1 Measurement of Variables.............................................................................. 52 4.4 Panel Unit Roots Tests .......................................................................................... 57 4.4.1 The Choi (2001) Test ..................................................................................... 59 4.4.2 The Im et al. (2003) Stationarity test for Panel Data ....................................... 60 4.5 Panel Cointegration ............................................................................................... 60 4.6 Causality tests ....................................................................................................... 62 4.7 Diagnostic tests ..................................................................................................... 65 4.7.1 The Sargan Test ............................................................................................. 65 4.7.2 Autoregressive Test ....................................................................................... 66 4.7.3 Endogeneity Test ........................................................................................... 66 4.7.4 The Heteroskedasticity Test ........................................................................... 66 4.7.5 Time Effects, Breusch & Pagan Test and Hausman Specification Test ........... 67 4.8 The Sources of Data .............................................................................................. 67 4.9 Conclusion ............................................................................................................ 68 CHAPTER FIVE .......................................................................................................................... 70 ANALYSIS AND DISCUSSION........................................................................................ 70 5.1 Introduction .......................................................................................................... 70 5.2 Summary Statistics ................................................................................................ 70 5.3 Panel Unit Root Results ........................................................................................ 73 5.4 Panel Cointegration Results .................................................................................. 75 5.5 Pedroni Cointegration Tests Using Alternative Measures of Bank Development ... 81 5.5.1 Broad Money Ratio ........................................................................................ 81 5.5.2 Domestic Credit Ratio .................................................................................... 83 viii University of Ghana http://ugspace.ug.edu.gh 5.6 Correlations .......................................................................................................... 84 5.7 Panel Causality Results ......................................................................................... 85 5.8 Causality Analysis Using Alternative Measure of Bank Development ................... 91 5.8.1 Broad Money Ratio ........................................................................................ 91 5.8.2 Domestic Credit Ratio .................................................................................... 95 5.9 Other Diagnostic Tests .......................................................................................... 98 5.9.1 Endogeneity Test ........................................................................................... 98 5.9.2 The Heteroskedasticity Tests .......................................................................... 99 5.9.3 Time Effects, Breusch & Pagan Test and Hausman Specification Tests. ....... 100 5.10 Conclusions. ....................................................................................................... 101 CHAPTER SIX ........................................................................................................................... 103 SUMMARY, CONLUSION AND POLICY RECOMEDATIONS ................................... 103 6.1 Introduction ........................................................................................................ 103 6.2 Summary and Conclusions .................................................................................. 103 6.3 Policy Implications ............................................................................................. 106 6.4 Limitations of the Study and Suggestions on Further Research Areas .................. 107 References ................................................................................................................................... 109 APPENDICES ............................................................................................................................ 118 ix University of Ghana http://ugspace.ug.edu.gh LIST OF FIGURES Figure Page Figure 2.1: Trends in Growth per Capita in the Selected Countries in SSA (1990-2014) ...... 14 Figure 2.2: Trends in Turnover Ratio in the Selected Countries in SSA (1990-2014) ........... 17 Figure 2.3: Trends in Bank Credit Ratio in the Selected countries in SSA (1990-2014) ....... 20 x University of Ghana http://ugspace.ug.edu.gh LIST OF TABLES Table Page Table 2.1: Stock Markets, Banks and Growth Performance for the Selected SSA Countries (1990-2014) ........................................................................................................................ 13 Table 4.1: Summary of Variables, Measurement and Data Sources………………………. ..68 Table 5.1: Summary statistics for Real GDP per capita and other predictor variables........... 71 Table 5.2: Panel Unit Roots Results .................................................................................... 74 Table 5.3. Pedroni Heterogeneous Panel Cointegration Tests(Bank Credit Ratio) ................ 77 Table 5.4. Pedroni Heterogeneous Panel Cointegration Tests (Broad Money Ratio) ............ 82 Table 5.5. Pedroni Heterogeneous Panel Cointegration Tests (Domestic Credit Ratio) ........ 83 Table 5.6: Correlations ........................................................................................................ 85 Table 5.7: Panel VAR Models. One Step System GMM results (Bank Credit Ratio) ........... 88 Table 5.8: Panel VAR Models. One Step System GMM results (Broad Money Ratio) ......... 92 Table 5.9: Panel VAR Models. One Step System GMM results (Domestic Credit Ration) ... 95 xi University of Ghana http://ugspace.ug.edu.gh ACCRONYMS ADF Augmented Dickey-Fuller ARDL Autoregressive Distributed Lag ASEA African Securities Exchanges Association BD Bank Development BMR Broad Money Ratio EAP East Asia and Pacific DMR Domestic Credit Ratio ECA Europe and Central Asia FEM Fixed Effects Model GDP Gross Domestic Product GFC Gross Fixed Capital Formation GMM Generalized Method of Moments GNP Gross National Product GOVT Government Spending IMF International Monetary Fund INF Inflation IPS Im, Pesaran & Shin IRF Impulse Response Functions LAC Latin America and Caribbean LGDP Natural Logarithm of Real GDP Per Capita xii University of Ghana http://ugspace.ug.edu.gh MENA Middle East and North Africa MEM Macro-Economy Meter OECD Organization for Economic Cooperation and Development OLS Ordinary Least Squares OPN Openness to Trade REM Random Effects Model SEM Stock Exchange of Mauritius SIC Schwarz Information Criteria SMD Stock markets Development SSA Sub-Saharan Africa SVAR Structural Vector Autoregressive Model TOP Openness to Trade TR Turnover Ratio US United States USA United States of America VAR Vector autoregressive Model VDC Variance Decomposition VECM Vector Error Correction Model WAMU West African Monetary Union WDI World Development Indicators WEO World Economic Outlook xiii University of Ghana http://ugspace.ug.edu.gh CHAPTER ONE INTRODUCTION 1.1 Background Financial sector development has been a significant player in the economic growth process in SSA. Nonetheless, the financial sector in SSA is still not sufficiently developed. Numerous foreign banks are operating in SSA signifying the limited or no barriers to entry to the sector. However, their presence has not had a major effect on competition, as it remains low in this region. Little competition and inefficiency of the financial sector in SSA limits the effect the financial sector can have on growth. The macroeconomic and financial environment have been stable, and there are expectations for higher economic growth. Innovations in the banking sector like mobile and agency banking promise to reduce inefficiencies and promote more competition in the industry. Financial innovations might enable the banks to play an enhanced role in supporting economic growth (Mlachila et al., 2013). Economists widely agree that economies with advanced financial sectors tend to experience faster economic growth. Theoretically, Bencivenga & Smith (1991) demonstrated that economies with well-developed financial sectors grew faster that those with undeveloped financial sectors. Further, developed financial systems result in lower information and transaction costs. Reduced transactions and information costs leads to better apportionment of resources and improvements in productivity (Greenwood & Jovanovic, 1990; Boyd & Prescott, 1986). 1 University of Ghana http://ugspace.ug.edu.gh In a historical perspective, the debate pitting supply-leading finance1 on one hand and the demand-following finance2 on the other has raged on since the 19th century. Odhiambo (2008, p.706) cited earlier works by Bagehot (1873) and Schumpeter (1912) who predicted that the financial sector leads the real sector development. However, evidence that the real sector leads the financial sector emerged on the onset 1950s pioneerd by Robinson (1952). The supply- leading finance received more support in the early 1970s from McKinnon (1973) and Shaw (1973). This debate continues to raise mixed reactions among scholars focussing on specific countries and regions in the world. The finance-growth literature has postulated two main conduits through which financial development can promote economic growth. The two avenues, widely covered in the literature, are stock markets and banking sector development (see inter alia, Beck & Levine, 2004; Adjasi & Biekpe, 2006; Ndako, 2010; Tachiwou, 2010; Adebola & Dahalan, 2011; Adusei, 2013). Theoretically, the stock markets and bank development ameliorate transactions and information costs, thereby promoting economic activities. A reduction in transactions and information costs improve the reallocation of savings from unproductive assets to capital accumulation and diversification of risks thus stimulating investment and growth. The stock markets provides alternative channels of savings and investment for surplus spending units of the economy. As opposed to short-term lending by banks, stock markets lend long- term. Stock markets also improve the efficiency of allocation of capital to profitable projects. Stock market development has led to the theoretical thinking that stock markets can exert an 1 Supply-leading finance refers to the hypothesis that developments in the financial sector take place before development in the real sector. 2 Demand-following finance refers to the hypothesis that economic development leads to the development in the financial sector. 2 University of Ghana http://ugspace.ug.edu.gh independent influence on growth. Growth and development of stock markets reflect an increase in investor confidence in the performance of the local economy (Tachiwou, 2010). For the companies and corporations, stock markets provide an avenue to raise more capital in the primary market. It also provides owners of stocks an opportunity to sell their stake in companies on the secondary market. Investors who miss the opportunity of buying stocks, bonds and treasury bills in the primary market, have a chance to do so at the secondary market. Companies can thus raise additional capital on the stock exchange and invest in profitable projects, which would otherwise lack funds and thus enhance growth. On the other hand, banks perform the functions of mobilization of savings, reducing transactions cost, the transformation of maturities and sharing of risks. In this regard, bank development ameliorates information asymmetry and lower the cost of banking transactions, and thereby improve upon the allocation of credit. Banks have the expertise and economies of scale to gather information, develop credit scores and risk profiles of borrowers over time, thereby, reducing adverse selection and moral hazard. By so doing, they ensure credit allocation to creditworthy borrowers. Improvements in the efficiency of allocation of credit to productive and profitable projects increases physical capital accumulation and ultimately economic growth. The last two decades have witnessed the development of stock markets and banks in Sub- Saharan Africa (SSA) following financial liberalization and reforms in the 1990s. Several countries in SSA privatized inefficient and loss-making government-owned banks. Also, the SSA Region continues to face numerous challenges on the one hand and a wealth of opportunities on the other. Its investment environment is weak coupled with challenges in governance and institutional ability. Further, economic growth has averaged between 5 and 7 percent since the 1990s, (Mlachila et al., 2013) 3 University of Ghana http://ugspace.ug.edu.gh Stock markets role in the economic development has gained increased attention from researchers in the last two decades. The New York, London, and Frankfurt stock markets are just a few examples of well-developed stock markets in the world. Over the last five years, there have been numerous Euro Bond issues by SSA countries. Ghana, Nigeria, Senegal, Zambia and Kenya have issued Euro bonds on some of these markets to the tune of $950 million, $525 million, $200 million, $750 million and $2.815 billion respectively (IMF, 2011 & 2013). In SSA, the number of stock markets has dramatically increased from only eight in 1990 to twenty in 2012. There is also one regional stock exchange for the West African Monetary Union. These stock markets are still underdeveloped, compared to well-developed stock markets in Asia, Europe and North America. Empirically, studies focussing exclusively on the causal linkage between stock market development and economic growth have yielded mixed results in SSA. Prominent in the findings is the support for the supply-leading hypothesis and a bi-directional causal flow (see inter alia, Enisan & Olufisayo, 2009; N'zue, 2006). Literature on the causality between bank development and economic growth also revealed mixed findings. Globally, Jung (1986) argued that the supply-leading finance dominates the demand-following finance in developed and developing countries. In developing countries, Christopoulos, & Tsionas (2004) established that the banking industry development leads to economic growth. Conversely, Demetriades & Hussein (1996) reported the existence of a two-way causation between bank development and the growth of real GDP per capita (hereafter, growth per capita). In SSA, there is support for demand-following hypothesis, supply-leading hypothesis, a bi-directional causal flow and no causality at all (see among others, Ghirmay, 2004; Odhiambo, 2008; Agbetsiafa, 2004; Odhiambo, 2007; Quartey & Prah, 2008). 4 University of Ghana http://ugspace.ug.edu.gh Studies including both stock markets and bank development in growth regressions emerged in the late 1990s pioneered by Levine & Zervos (1998). Causality tests have also been implemented in models that include both stock markets and bank development. Given the mixed causlity results established by scholars focussing exclusively on either stock market development or bank development, it is not surprising that the findings here are mixed too. Specifically, there is support for both the demand-following and supply-leading hypothesis (see Ndako, 2010; Adebola & Dahalan 2011). This therefore calls for an examination of the long run and causal relationship between both stock market development, bank development and economic growth and forms the focus of this study. The SSA region has experienced a considerable growth of stock markets and banks over the last two decades. The financial sector has also witnessed significant reforms spearheaded by the IMF and the World Bank over that period. Despite the above views on the issue of causality between stock market development, bank development and economic growth, this area has not received the attention it deserves in SSA. This study attempts to fill this gap by proposing to analyze the long run and causal relationship between stock market development, bank development and growth in the selected SSA countries. The study will include other variables that are critical in explaining economic growth such as physical capital, government expenditure, inflation and openness to trade as control variables. This study will be on 13 selected countries in SSA because of data availability. See Appendix I for a list of the countries included in this study. 1.2 Problem Statement Most SSA countries carried out financial and economic reforms spearheaded by the World Bank and the IMF in the 1990s. The changes have influenced financial sector development and 5 University of Ghana http://ugspace.ug.edu.gh efficiency, particularly the banking sector. The capital base and risk management strategies for banks have developed. The banking sector in SSA had enough strength to withstand the recent global economic crisis that destabilized the banking sectors of developed economies like the US (Mlachila et al., 2013). In an attempt to diversify saving avenues and sources of capital, some countries went ahead and established stock markets. However, most of these stock markets remain shallow and highly illiquid. The stock turnover ratio for instance, remains relatively low for most countries in our sample. Aside the Johannesburg stock exchange in South Africa, that has had turnover ratios above the 50% mark, others remain low. The highest turnover ratio for the Zimbabwe stock exchange, the Nairobi securities exchange, the Nigeria stock exchange and the Lusaka stock exchange is 29.40%, 38.54%, 29.3% and 21.66% respectively. All other stock markets in our sample have a turnover ratio consistently lower than 20% for the sample period. Recent issues of Euro bonds by Ghana, Nigeria, Zambia and Kenya among others show evidence of the shallow financial markets in SSA, which are incapable of lending such massive amounts to government. Although there exists numerous works on the issue of causality involving growth and financial development, most of the works involve only bank development measures (see among others, Quartey & Prah, 2008; Agbetsiafa, 2004; Odhiambo, 2008; Adusei, 2013). Other works have exclusively focussed on the causal linkage between stock market development and growth (for more details consult, Enisan & Olufisayo, 2009; N'zue, 2006). It is therefore intellectually appealing to see how the causal relationship plays out when we interact stock markets and bank development proxies in the economic growth regressions for the selected SSA countries. There are notable studies that have established a causal link between stock market development, bank development and economic growth in select economies in SSA. Ndako 6 University of Ghana http://ugspace.ug.edu.gh (2010) found evidence for a bi-directional causal flow between bank development and economic growth using the bank credit ratio to measure bank development. Using the shares traded ratio and the turnover ratio as the stock market development proxies, the study found evidence for the demand-following hypothesis between stock market development and economic growth in South Africa. On the other hand, Adebola & Dahalan (2011) established support for the supply-leading finance for both stock market development and bank development with economic growth in Nigeria. Although Ndako (2010) and Adebola & Dahalan (2011) have established that stock market development, bank development, and economic growth have a causal linkage, they are national-level studies. The issue of causality involving both stock market development, bank development and economic growth is still not clear in SSA region. It is critical to investigate the direction of causality between stock market development, bank development and economic growth in the selected SSA countries. In other words, does accounting for country-specific effects change the results of the national-level studies earlier done in this area? The limited works available on the causality between stock market development, bank development and economic growth inspired this study. This study aims to contribute to the finance-growth literature in SSA in the following three ways. First, this study analyses the presence or lack thereof of cointegration between stock market development, bank development and growth in the selected SSA countries. Also, we include stock markets and bank development measures in our growth per capita regressions before testing for Granger- causality. Based on our reading of the available literature, no study has simultaneously investigated the causal relationship between stock markets development, bank development, and economic growth in a panel of selected SSA countries. At best, what we have come across are the two, national-level time series studies on South Africa and Nigeria. 7 University of Ghana http://ugspace.ug.edu.gh Secondly, we use a panel data set that takes advantage of the time series and cross-sectional aspects of the data. Panel data sets bring in additional variation and increases the efficiency of the estimates. The findings of this study will thus be more robust than time series and cross section results. Finally, our study uses a more recent and updated data set. The study will control for other determinants of economic growth like physical capital, government spending, inflation and openness to trade. 1.3 Research Objectives The primary aim of this study is to investigate the causal relationship between stock market development, bank development and growth per capita in the selected Sub-Saharan Africa countries. More fundamentally, our study seeks to achieve the following specific objectives. (i) To investigate whether a long run relationship between stock market development, bank development and growth exist in the selected SSA countries. (ii) To determine the direction of causality between stock market development, bank development and growth per capita in the selected SSA countries. (iii) To examine whether the causal relationship between stock market development, bank development, and growth per capita in the selected SSA countries is sensitive to the choice of the bank development proxy. 1.4 Significance of the Study The view that SSA is the next growth frontier is not in doubt. Sub-Saharan Africa is home to emerging economies like Kenya, South Africa, and Nigeria among others. The prospects of economic growth in SSA are bright. The question on many people’s lips is the source of the much-anticipated growth. Will stock markets and banks contribute to this growth or support it? This study will contribute to the literature by revealing whether there exist a long run and stable 8 University of Ghana http://ugspace.ug.edu.gh relationship between stock market development, bank development and growth in the selected SSA countries. In addition, this study will determine direction of causality between stock market development, bank development and growth per capita in the selected SSA countries in a panel framework. Lastly, this study will also reveal the sensitivity of the causal relationship between stock market development, bank development and growth per capita to a change of the measure of bank development. We have witnessed the growth of banks and stock markets in SSA. Banks have grown especially after the financial liberalization reforms spearheaded by the World Bank and the IMF. Stock markets have also increased from an initial 8 in 1990 to 13 in the sample in 2014. However, as far as literature this study has reviewd is concerned, there has been no study on the long-term and causal relationship between stock markets, banks, and growth in the selected countries in SSA. This study attempts to fill that gap in the literature. Understanding the causal linkage between stock market development, bank development and economic growth will be of great importance to policy makers. We expect that the findings of this study will influence the premium policy makers place on instituting financial sector and economic growth reforms. They get to know what further reforms to implement in the financial sector to enable it play an enhanced role in development. The findings of this study will inform policy required to promote further stock market development, bank development and economic growth. A paramount goal of this study is to reveal the evidence of causality between stock market development, bank development and growth per capita in selected countries in SSA. The results of this study will also add to the finance-growth literature and serve as a reference for future studies in this area. 9 University of Ghana http://ugspace.ug.edu.gh 1.5 Scope of the Study This study is limited to 13 selected SSA countries with operational stock markets, and whose data on key variables is readily available. The establishment of six stock markets in the sample took place before 1990. The creation of the rest occurred in the year 1990 or after. We provide more details on the particular year of establishment of the stock markets in our sample on Appendix II. The year of establishment of the stock markets in our sample notwithstanding, consistent data on the measure of stock markets development we intend to use is only available beginning 1990. For this reason, the study covers the period from 1990 to 2014. 1.6 Organization of the study With this introduction and motivation in mind, the rest of the study proceeds as follows. Chapter 2 gives a brief overview of recent developments in growth per capita, the stock markets, and bank development. Chapter 3 provides a review of relevant literature in this area. Chapter 4 discusses the data and methodology adopted by this study to fulfill its objectives. The findings of this study and subsequent discussion will be presented in chapter 5. Chapter 6 concludes the study, provides recommendations and areas for further research. 10 University of Ghana http://ugspace.ug.edu.gh CHAPTER TWO OVERVIEW OF GROWTH PER CAPITA, STOCK MARKETS AND BANK DEVELOPMENT FOR THE SELECTED SSA COUNTRIES 2.1 INTRODUCTION The current chapter gives a brief overview of growth per capita, stock markets and bank development for the selected countries in SSA and compares them to other regions in the World. Specifically, this study presents averaged figures for growth per capita, stock markets and bank development computed using data from the World development indicators online data base of the World Bank. The chapter contains six main sections. The present section introduces the chapter. Section 2.2 covers recent developments in economic growth. The subsequent section presents the trends in stock market development. We introduce the structure of the banking sector in section 2.4 and trends in bank credit ratio in section 2.5 while Section 2.6 concludes the chapter. 2.2 The Economic Growth Trends in the Selected SSA Countries since the 1990s SSA has experienced impressive growth per capita since the 1990s supported by a refined macroeconomic environment, regulatory and trade reforms, lower levels of armed conflicts, conducive commodity price movements, the discovery of new resource deposits and debt forgiveness (Mlachila et al., 2013). This study observes that growth for our sample has outperformed the one for SSA. Specifically, the level of real GDP per capita averaged $1,919.5 for our sample and $873.5 for SSA for the period 1990-2014. The average growth per capita of the same for our sample is also higher than that of SSA. Specifically, growth per capita averaged 2% for our sample and 0.8% annually for SSA over the sample period 1990-2014. This study will compare the average level of real GDP per capita and growth per capita for 11 University of Ghana http://ugspace.ug.edu.gh specific countries in the sample to the sample and SSA average below. Table 2.1 highlights this information, with all figures adjusted to one decimal place. The World Bank classifies economies as either high income, upper middle-income, low- income and lower middle-income countries based on estimates of their gross national incomes. The generation of the gross national income estimates is based on the World Bank Atlas method. The World Bank revises these classifications on July 1st every year. Based on the July 2015 classification, economies with a gross national income of above $4,125 but lower than $12,736 make up the upper middle-income group. In our sample, Botswana, Mauritius, Namibia, and South Africa fall into the upper middle- income group. The lower middle-income group comprises of economies with a gross national income above $1,045 but below $4,125. Ghana, Kenya, Nigeria, Swaziland, and Zambia fall into the lower middle-income group in our sample. Lastly, countries whose gross national income is less than $1,045 fall in the low-income group. Our sample also contains low-income countries namely, Malawi, Tanzania, Uganda and Zimbabwe. Turning our attention to the growth performance of the particular economies in our sample, we note significant differences. For the upper middle-income countries in the sample, Botswana and Mauritius attained an average level of real GDP per capita of $5,119.6 and $4,853.5 and average growth per capita of 2.8% and 3.7% over the sample period that is higher than that of the sample and SSA. Similarly, Namibia has also realized good economic performance for the period 1990-2014. Her average level of real GDP per capita of $3,447 exceed the one for the sample but has the same average growth per capita 2% as the sample. Namibia’s average level of real GDP per capita and average growth per capita is higher than that of the entire SSA. 12 University of Ghana http://ugspace.ug.edu.gh Lastly, in the upper middle-income group of our sample, South Africa’s growth performance for the sample period beginning 1990 to 2014 has also been impressive. South Africa recorded an average level of real GDP per capita of $5,285.3 higher than the one for the sample and SSA average. Her average growth per capita of 0.7% is, however, lower than that of the sample and SSA average respectively. Table 2.1: Stock Markets, Banks and Growth Performance for the Selected SSA Countries (1990-2014) Variable Sample SSA EAP ECA LAC MENA Real GDP per capita 1919.5 873.5 4560.2 17649.3 5008.5 4152.6 (Constant 2005 US$) Real GDP per capita growth 2 0.8 2.9 1.4 1.6 2.2 (annual %) Stocks traded, turnover ratio 7.8 20 87.2 77.1 36.3 42.1 of domestic shares (%) Domestic credit to private 24.7 32.9 126.2 108.4 32.8 38.6 sector by banks (% of GDP) Source: Author’s computation using STATA 13 and data from WDI, 2016. In the lower middle-income group of our sample, Ghana and Nigeria recorded an average level of real GDP per capita of $504 and $729.5 that is way below the sample average. However, their mean growth per capita of 2.9% and 3.1% are slightly above the sample and SSA average. In the same vein, the average level of real GDP per capita and growth per capita for Kenya of $546.05 and 0.7% is lower than the sample and SSA average respectively. This shows that Kenya’s economy is performing poorly compared to the sample and the entire SSA. In addition, Swaziland’s average level of real GDP per Capita of $2,255.4 exceeds the average for the sample and SSA. Even though Swaziland’s average growth per capita of 1.6% is lower than the one for the whole sample, it is higher than the SSA’s average, for the sample period 1990- 2014. 13 University of Ghana http://ugspace.ug.edu.gh Zambia concludes the lower middle-income countries in the sample. Zambia’s average level of real GDP per capita of $707.3, is lower than the sample and SSA average. Her average growth per capita for the period under review has been 1.86% lower the sample average but higher than the SSA average. Concerning the low-income countries in the sample, the average level of real GDP per capita and growth per capita for Malawi of $224.9 and 1.7% is lower than the sample average. However, her growth per capita is greater than the SSA average for the period 1990-2014. Finally, Tanzania and Uganda have lower mean level of real GDP per capita at $423.1 and $308.2 than the sample and SSA average. Their average growth per capita of 2.2% and 3.3%, however, exceeds the sample and SSA average during the period under review. Figure 2.1: Trends in Growth per capita in the Selected Countries in SSA (1990-2014) Source: Author’s computation using STATA 13 and data from WDI, 2016 14 University of Ghana http://ugspace.ug.edu.gh Lastly, Zimbabwe had an average level of real GDP per capita of $550.1 and mean growth per capita of -1.1, both below the sample and the SSA mean. Figure 2.1 gives the trends in growth per capita for the low-income countries, lower middle-income countries, and upper middle- income countries in our sample, the sample average and the SSA average. Since the focus of our study is stock markets, banks, and growth, the next section reviews recent developments and trends in stock market development and bank development. 2.3 Overview of Stock Market Development in the Selected SSA Economies (1990-2014) The number of stock markets in our sample has increased from an initial 8 in 1990 to 13 in 2014. In comparison, the SSA region has seen a rise in the number of stock markets from 8 in 1990 to 20 in 2014. The total number of listings in our sample as of 2015 was 1,001 compared to 1,098 for the SSA region. It is also important to note in our sample, the Johannesburg stock exchange is the oldest and the largest in SSA, established in 1887. According to Jefferis & Mbekeani (2000), the Johannesburg stock exchange ranked 17th in the world and 3rd in emerging markets in 1998. Its market capitalization was equivalent to 75% of African stock markets capitalization. The Johannesburg stock exchange leads with 402 listings followed by the Nigerian stock exchange with 189. The Swaziland Stock exchange has the lowest listings our sample, i.e., 7. For more details on the list of stock markets in our sample, the date they were established and their number of listings, see Appendix II. The stock turnover ratio (hereafter, turnover ratio) is a commonly used measure of stock market development. It is one of the measures of stock markets liquidity. The value of shares traded as a proportion of market capitalization gives the turnover ratio. The value of shares traded equals the price per share multiplied by all shares traded. Market capitalization, on the other hand, relates to the price per stock multiplied by all the stocks listed on the stock markets. 15 University of Ghana http://ugspace.ug.edu.gh Jefferis & Mbekeani (2000) observed that, despite the fast increase in the size and number of the stock markets in SSA, these markets remain highly illiquid. They continued to argue that these stock markets are fractured and experience low technology penetration. These challenges reduce their efficiency and ability to influence growth in their own economies. Farid (2013) also noted that a small clique of blue chip companies dominates the stock markets in SSA. The sample average turnover ratio that is lower than the SSA average confirms that stock markets in our sample are indeed illiquid. Specifically, Table 2.1 shows that the average turnover ratio for our sample is 7.8%, which is significantly lower than the SSA average of 20% for the sample period 1990-2014. Both the sample average and the SSA average of the turnover ratio are way below other regions in the world. For example, the Middle East and North Africa (MENA), Europe and Central Asia (ECA), East Asia and Pacific (EAP) and the Latin America and Caribbean (LAC) regions outperform our sample and SSA region by reporting average turnover ratios of 42.1%, 77.1%, 87.2% and 36.3% in that sequence for the period under review. Concerning the liquidity of specific countries in the sample, the stock exchanges of Johannesburg, Zimbabwe, Nigeria and Nairobi with average turnover ratios of 37.8%, 10.3%, 8.5% and 8.2% respectively, have average turnover ratios above the sample average, over the period under review. Of these, it is only the Johannesburg stock exchange, which has an average turnover ratio above the SSA average. The stock exchanges of Mauritius, Botswana, Lusaka, Malawi, Namibia, Ghana, Dar es Salaam, Swaziland and Uganda have average turnover ratios below the sample average. Their average turnover ratios for the period are 5.8%, 5.3%, 3.8%, 3.8%, 3.6%, 3.2%, 2.6%, 0.9% and 0.6% respectively. The Johannesburg stock exchange is the most liquid in our sample while the Uganda securities exchange is the least liquid in the sample based on their average turnover ratios. 16 University of Ghana http://ugspace.ug.edu.gh Figure 2.2: Trends in Turnover Ratio in the Selected Countries in SSA (1990-2014) Source: Author’s computation using SATA 13 and data from WDI, 2016 This study uses data on the average US dollar-adjusted total returns for the last three years from the investing in Africa website (Hoover, 2016), to classify stock markets in our sample as either high, poor or very poor performers. In this regard, we rank stock markets, which had a positive average three-year return as high performers. Also, this study ranks stock markets with an average three-year return of between -4.5% and -18.7% as poor performers. Very poor performing stock markets are the ones, which had an average three-year return of less than - 18.7%. In this regard, the Dar es Salaam stock exchange and the Lusaka stock exchange are high performers with an average three-year return of 18.1% and 14.7%. The cut-offs indicated above are arbitrary. 17 University of Ghana http://ugspace.ug.edu.gh The poor performers group comprises of the stock exchanges of Nairobi, Malawi, Botswana, Johannesburg and Uganda. Their average US dollar three-year average returns were -4.5%, -8.6%, -9.2%, -14.7% and -18.7% respectively. Lastly, the very poor performing stock markets are the stock exchanges of Mauritius, Nigeria, Lusaka, Zimbabwe and Ghana. Their average US dollar adjusted average three-year returns were -20.9%, -41.9%, -42.4%, -50.7% and -52% respectively. Figure 2.2 presents the turnover ratio trends for stock markets in our sample based on the classification highlighted above. 2.4 The Structure of the Banking Sector in the Study Countries Appendix III presents the structure of the financial system supervised by their respective central banks in our sample. We obtained the information contained in Appendix III from the respective central bank’s websites as referenced in the reference list, on March 12, 2016. The structure provided thereon is limited to the information available on their website on that date. Kenya had the highest number of commercial banks, 42, closely followed by Tanzania with 40 and Ghana with 31 banks. Botswana, Namibia, and Swaziland had the lowest number of banks in our sample with 10, 9 and 4 banks respectively. In addition, South Africa hosted 15 branches of foreign banks. Ghana had 140 registered rural community banks. It should, however, be noted that having a high number of banks is not an indication of higher bank development or stability. South Africa, Kenya, and Ghana had 38, 8 and 4 representative offices in that order. The other countries in the sample did not have representative offices as of March 12, 2016. Concerning non-bank financial institutions, Nigeria, Ghana, Mauritius and Kenya lead their peers in the sample with 117 and 65 institutions while both Ghana and Kenya had 20 non-bank financial institutions each respectively. 18 University of Ghana http://ugspace.ug.edu.gh South Africa had 5 non-bank institutions while Malawi and Swaziland had 4 each. Botswana and Namibia did not have non-bank financial institution regulated by their respective central banks. Regarding microfinance institutions, Nigeria and Ghana were far ahead of the rest with 941 and 576 microfinance institutions respectively. Following from a distance were Zambia, Kenya, and Zimbabwe with 37, 12 and 8 microfinance institutions respectively. South Africa, Tanzania, Mauritius, Namibia, Botswana, Uganda, Malawi, and Swaziland did not have microfinance institutions regulated by their respective central banks. Lastly, Nigeria had the highest number of registered forex bureaus, i.e., 2,991. Ghana came a distant second with 413 forex bureaus while Uganda and Tanzania followed closely with 246 and 222 registered forex bureaus. Kenya, Zambia, and Botswana had 79, 73 and 52 registered forex bureaus respectively. At the lower end, South Africa, Namibia, Mauritius, and Swaziland had 17, 10, 5 and 1 registered forex bureaus in that sequence. Malawi and Zimbabwe did not have registered forex bureau as per the information available on their website. It is necessary to note that, most commercial banks are authorized by their respective central banks to undertake forex transitions. 2.5 Trends in Bank Development in the Study Countries since the 1990s Credit allocated to the private sector by banks as a share of nominal GDP is a widely used measure of bank development. The average bank credit ratio from Table 2.1 clearly shows that bank development in the sample lags behind that of SSA and other regions. Precisely, the sample average of the bank credit ratio is 24.7% for the sample period 1990-2014. In comparison, the sample average is slightly lower than the SSA average of 32.9%, the LAC average of 32.8% and the MENA average of 38.6%. The sample average is, however, way below the EAP average of 126.2% and the ECA average of 108.4%. 19 University of Ghana http://ugspace.ug.edu.gh Figure 2.3: Trends in Bank Credit Ratio in the Selected Countries in SSA (1990-2014) Source: Author’s computation using STATA 13 and data from WDI, 2016 Regarding the specific countries in our sample, Mauritius, South Africa and Namibia have bank credit ratio averages of 65.7%, 65.1% and 42.1% that are higher than the mean for the sample and the SSA average respectively. Zimbabwe and Kenya have an average bank credit ratio of 25.7% and 25% greater than the sample average but below the SSA average. It follows that, Botswana, Swaziland, Nigeria, Ghana, Malawi, Tanzania, Uganda and Zambia at 19.2%, 18.8%, 14.9, 10.8%, 9.5%, 9%, 8.2% and 8% respectively have average bank credit ratios less than both the sample and the SSA average. Figure 2.3 gives the trends in the average bank credit ratio for our sample and SSA over the period 1990-2014. 20 University of Ghana http://ugspace.ug.edu.gh 2.6 Conclusion This chapter presented a brief overview of economic growth, the stock markets and bank development for the selected SSA countries. We realize that, over the period 1990-2014, the sample average level of real GDP per capita has been more than that of the SSA region. However, the sample average has been lower than the average for LAC, MENA, EAP and ECA. However, over the period under review, only the EAP and the MENA regions have experienced a faster growth per capita than the sample average. Concerning stock markets and bank development, we note that our sample average falls below all other regions mentioned above. We proceed to review the literature on stock markets, banks, and growth in the subsequent chapter. 21 University of Ghana http://ugspace.ug.edu.gh CHAPTER THREE LITERATURE REVIEW 3.1 Introduction This chapter presents a review of the relevant literature stock market development, bank development and economic growth. This chapter contains four main sections. The next section explores the theoretical ways in which stock markets and bank development can contribute to economic growth. Section 3.3 looks at empirical studies on stock markets, banks and economic growth. The empirical literature comprises three subsections. First, we review the literary works on stock market development and economic growth. The next subsection deals with the literature on bank development and economic growth. Lastly, we discuss works that model both stock markets and bank development with economic growth. Section 3.4 concludes our literature review. 3.2 Theoretical Review We can trace the discourse on the causality between stock market development, bank development and economic growth to the 19th century. Three competing perspectives exist in past works concerning this causal relationship. Odhiambo (2008, p.706) cited earlier works by Bagehot (1873) and Schumpeter (1912) who predicted that the financial sector development leads the real sector development. Literature refers to this opinion as the supply-leading finance. It relates to the setting up of financial institutions, liabilities and assets well before demand for them arise. Robinson (1952) confronted the supply-leading hypothesis by suggesting that the real sector development leads the financial sector development. Literature refers to this point of view as 22 University of Ghana http://ugspace.ug.edu.gh the demand-following hypothesis. Demand-following hypothesis implies that the demand for financial services by entrepreneurs and savers in the real sector leads to the development of financial institutions, assets, and liabilities. The third view is that both financial development and economic growth Granger-cause each other. It is important to note that even though the contribution of stock markets and bank development to economic growth is well documented in the literature, there exist no unified theoretical linkage between the stock markets and bank development with economic growth. In this regard, we discuss notable theoretical works that have attempted to make theoretical predictions on the conceptual linkage between finance and growth. 3.2.1 Patrick’s Stages of Development Hypothesis Patrick (1966) attempted to reconcile the demand-following and supply-leading finance by advancing the stages of development hypothesis. In his opinion, demand-following and supply- leading finance, take place at different stages of the economic growth process. The study points out that supply-leading finance plays a key role before industry-propelled growth begins. Precisely, they argued that supply-leading finance is prevalent at the traditional stages of the economic growth process. At the traditional stages of the economic growth process, the agricultural sector’s contribution to GDP is usually high. In this regard, financial development can encourage technological development, improvement of production processes and investment before modern economic growth gets underway. Patrick (1966) pointed out that as advanced growth gets underway, the importance of the supply-leading finance diminishes and demand-following finance dominates. They cited Japan in the period 1870 to World War one as a classic case of the ordering of supply-leading and demand-following finance. The modern banking system was set up in Japan in 1870 with 23 University of Ghana http://ugspace.ug.edu.gh government support. The banks focus was on funding foreign trade, local commerce, and agriculture. The setting up of a modern banking system before modern industrialization points to supply-leading finance. The focus shifted to demand-following finance in the mid-1890s through funding of textile and consumer goods industries. Patrick (1966) observed that the contribution of financial development on capital stock is arguably the most important link between finance and growth. Financial intermediaries enhance the allocation of capital, new additions to capital and the rate of capital accumulation through financial intermediation. Through diversification of financial assets, financial intermediaries encourage people to work, save and invest. The substitution of unproductive assets such as land to more productive financial assets through financial development accelerates the growth of real income. Not all savers are entrepreneurs and entrepreneurs do not have the capacity to save enough to self-finance their investment. Financial intermediaries therefore, ensure allocation of savings to most talented entrepreneurs through financial intermediation. The study concluded that it is the successful intermediation of savings to investment, which promotes growth. 3.2.2 McKinnon-Shaw Hypothesis The supply-leading hypothesis received strong support in finance-growth nexus in the early 1970s from McKinnon (1973) and Shaw (1973). McKinnon (1973) put forward a complementarity hypothesis between physical capital and the demand for real money balances. He examined supply-leading finance, Keynesian policies and the objectives they were set to achieve. Their study discovered that the low-interest rates regime, directed credit schemes, and interest rate ceilings among other policies led to financial repression. The resulting financial 24 University of Ghana http://ugspace.ug.edu.gh repression hindered stock markets and bank development that are imperative for sustainable economic growth. The complementarity hypothesis offers the channel through which money can influence savings, investment, and economic growth. Money acts as a conveyor belt through which physical capital accumulation takes place. Low-deposit rates meant low returns on savings and other interest-earning assets. Investors had to accumulate money before they can have enough for investment in projects considered lumpy and indivisible in developing countries. McKinnon (1973) recommended financial liberalization, which involves abolishment of interest rates ceilings, credit controls, directed credit regimes and a reduction in the-the role of government in the financial sector. Elimination of financial repression would lead to an increase in deposit rates and attract outside money into the banks. McKinnon (1973) proposed an increase of the nominal deposit rates or reducing inflation to achieve high and positive real deposit rates. Positive real deposit rates would encourage savings as people will accumulate their savings in the banks before investing in lumpy projects. The higher savings would lead to increased investment and economic growth. The complementarity hypothesis, therefore, envisages a situation where the stock markets and banks develop through increased savings and investments and promote economic growth. Since the financial sector development preceded economic growth, we view the complementarity hypothesis as support for a supply-leading hypothesis. Shaw (1973) put forward a ‘debt intermediation’ view whose attention mainly focused on the role of financial intermediaries and the depth of the financial sector to enhance the efficiency of allocation of credit. They argued that improvements in credit appropriation would result from hooking up disintegrated capital markets in developing countries and reducing 25 University of Ghana http://ugspace.ug.edu.gh uncertainties and unpredictability of future returns on assets. They concentrated on inside money where a positive relationship exists between financial intermediation and motivations to save. A positive relationship between financial intermediation and quality of investment should exist too. Moreover, financial intermediaries play the roles of mobilization of savings, transformation of maturities, diversification of risk, reduction of transaction costs and economies of scale in offering credit that promotes economic growth. 3.2.3 Theory of Financial Intermediation with Delegated Monitoring Apart from the debate on the direction of causality, the theory has also focused on how stock markets and banks can affect economic growth through lower transaction and information costs. To promote economic growth, credit has to be made available to investors at affordable lending rates. The cost of financial services has to be reasonable too. One of the justifications of high loan rates in developing countries is the high rates of default. Diamond (1984) developed the financial intermediation theory with delegated monitoring. Under financial autarky, they argued that investors would have to identify savers in the economy and borrow from them. Every saver would have to monitor the borrower who he has lent money to and receive payments directly from them. The cost of monitoring borrowers in such a scenario could not only be significant but prohibitive. Diamond (1984) also focussed on the solutions to moral hazard within the financial sector. Moral hazard arises since lenders cannot observe the activities of borrowers. Borrowers can misuse borrowed funds while others do not intend to repay the loans. Solving moral hazard lowers the default risk and translates to lower cost of loans and economic growth. Shareholders of firms and savers in the case of banks transfer the very expensive monitoring of borrowers to financial intermediaries including banks and stock markets. Shareholders are the owners of 26 University of Ghana http://ugspace.ug.edu.gh stock whose interest is to maximize returns on their investments. Savers in the case of banks provide savings, which the banks lend to their borrowers. Diamond (1984) demonstrated that the cost associated with this delegation declined as the number of approved loans increase up to a point where financial intermediation becomes feasible. Diversification within financial intermediaries results in further cost reductions. The reduction of moral hazard in lending by the financial sector leads to proper utilization of borrowed funds and improvements in productivity of investment. Increases in productivity of investment promote economic growth. Earlier, Patrick (1966) also argued that the productivity of investment is an important avenue through which economic growth benefits from financial development. Though theory lumps up stock market development and bank development as financial development, empirical literature has shown that they can have independent effects on growth (see Levine & Zervos, 1998; Beck & Levine, 2004; Tachiwou, 2010) among others. We proceed to review the empirical literature of the finance-growth nexus below. 3.3 Empirical Literature 3.3.1 Research on Stock Market Development and Growth The financial sector comprises of banks, stock markets, money market funds among others. Although the stock markets are an essential part of the financial sector, little attention has been devoted to studying their long-term and causal linkage with economic growth. The increasing significance of the stock markets to growth has seen some studies in this area. Economic researchers have applied different econometric techniques and posted interesting and mixed results. Using the Vector error correction model (hereafter, VECM), different sample periods and countries, Tachiwou (2010), Nowbutsing & Odit (2010) and N’zue (2006) found support for a supply-leading hypothesis. 27 University of Ghana http://ugspace.ug.edu.gh Tachiwou (2010) for instance, found that stock market development exerts positive causal effect on growth per capita for the West African Monetary Union (WAMU). The study also reported a very quick adjustment to long-term growth from short-term deviations. In addition, they argued that the stock markets reveal the perceptions of the investors about any given economy. Further, deeper stock markets that are liquid imply a high confidence in the economy and makes the stock markets a critical driver of growth. Similarly, Nowbutsing & Odit (2010) established that stock market development Granger- causes growth per capita in both the short-term and the long-term in Mauritius. They also reported a quick return to long run equilibrium from short-term deviations on the basis magnitude of the error correction term. A closer look at the works of Tachiwou (2010) and Nowbutsing & Odit (2010) does not clearly reveal whether they tested for the presence of cointegration before adopting the VECM. N’zue (2006) follows the expected procedure by checking for the presence of cointegration before analysing the causal linkage between growth and stock market development in Cote D’Ivoire. Using the Johansen & Juselius (1990) cointegration procedure, they established no evidence of cointegration between stock market development and economic growth in a bivariate framework for the study. However, their results revealed that stock market development and economic growth, including government spending, public investment, development aid, foreign direct investment, and inflation as control variables exhibit a long run relationship. In addition, they found evidence of a one-way causal flow from stock market development to economic progress based on the VECM. Using the ARDL bounds test model, Enisan & Olufisayo (2009) found evidence of a stable long run relationship between stock market development and economic growth in South Africa 28 University of Ghana http://ugspace.ug.edu.gh and Egypt. They also concluded that economic growth and stock market development do not have a long run relationship in Zimbabwe, Kenya, Morocco, and Cote D’Ivoire. Further, they found evidence to support a supply-leading hypothesis in the republic of South Africa and Egypt using a VECM. Using a Vector autoregressive (VAR) model, Enisan & Olufisayo (2009) also found support for a two-way causal flow between growth and stock market development in Zimbabwe, Kenya, Cote D’Ivoire, and Morocco. Their study, however, uses market capitalization ratio and shares traded ratio to proxy stock market development. The market capitalization ratio and shares traded ratio relates the size of the stock market and trading volumes to nominal GDP. However, the size of the stock market is not a good predictor of economic growth as shown in the literature notably by Levine and Zervos (1998). On the other hand, the forward-looking nature of investors affects shares traded ratio as a proxy of the stock market development. Apart from investigating the direction of causality, researchers have worked on the general effect of stock market development to economic growth. In this regard, Adjasi & Biekpe (2006) improve upon earlier literature by introducing market capitalization ratio3, shares traded ratio4, and turnover ratio5 sequentially into growth regressions. They use GMM for dynamic panel to analyse the contribution of stock market development to economic growth in 14 selected African countries. Market capitalization ratio and the turnover ratio has no significant effect on 3 Market capitalization ratio= The value of all stocks listed on the stock market as a proportion of nominal GDP 4 Shares traded ratio= The value of all stocks sold at the stock exchange as a share of nominal GDP 5 Turnover ratio= The value of all shares traded at the stock exchange as a proportion of the value of all shares listed at the stock exchange 29 University of Ghana http://ugspace.ug.edu.gh growth for their whole sample. The shares traded ratio had a positive and significant effect on economic growth for their entire sample. Adjasi & Biekpe (2006) also established that stock market development has no contribution to economic growth for low and lower middle-income countries in their sample. However, the contribution to economic growth for the upper middle-income countries is significant and positive in their sample. Based on capitalization of the market classification, it is only market capitalization ratio and shares traded ratio, which contribute to economic growth significantly in moderately capitalized markets in their sample. One limitation of their study, which they duly acknowledge, is that their sample size is small and that a larger sample size is desirable for GMM. It is also not clear whether their results are robust to initial income and human capital development, as they did not control for them. Cooray (2010) estimated a stock markets augmented Mankiw et al. (1990) production function and found that stock market development’s contribution to growth per capita is positive. Their study included a sample of 35 developing economies for 12 years. Their paper adopted OLS, GMM and an alternative estimation technique that gives minimum weight to outliers in carrying out the analysis. Further, they established that the choice of the stock markets variable does not affect the contribution of stock markets to growth per capita. Consistent with Mankiw et al. (1990), they concluded that human capital’s contribution to growth per capita is positive and significant. Their study also reported evidence in support of conditional convergence. In conclusion, the literature we have reviewed appears skewed towards a supply-leading hypothesis (Tachiwou, 2010; Nowbutsing & Odit, 2010; N’zue, 2006). There is also evidence of a bi-directional causal flow between stock market development and economic growth as reported by Enisan & Olufisayo (2009). Concerning the general contribution of stock market 30 University of Ghana http://ugspace.ug.edu.gh development to growth, Adjasi & Biekpe (2006) and Cooray (2010) established a positive relationship. This literature supports the view that stock market development is essential for economic growth. We now turn our attention to bank development and growth, an area that has witnessed much of the debate. 3.3.2 Literature on Bank Development and Growth As stated earlier, much of the focus on the finance-growth nexus has concentrated on bank development and economic growth. There exists some country-level and cross-country studies with varying methodologies and sample periods. These studies have reported mixed results. While some appear to support a supply-leading hypothesis, others support a demand-following hypothesis. There are some studies too who found evidence of a bi-directional causal flow while some found no causal relationship at all. We present a detailed review of this literature hereafter. 3.3.2.1 Studies Supporting the Supply-Leading Hypothesis Even though a good number of studies have found evidence supporting a supply-leading finance, their methodologies, sample periods and countries included varies among authors. Using a sample of 19 advanced and 37 developing countries and simple Granger causality, Jung (1986) found that supply-leading finance dominates demand-following finance in developing countries. They established moderate support for the stages of economic development hypothesis proposed by Patrick (1966). On the other hand, Agbetsiafa (2004) found evidence of a long-term relationship between bank development and economic growth for seven out of eight SSA countries in his sample using Johansen & Juselius (1990) cointegration tests. The study covers the period beginning 1963 to 31 University of Ghana http://ugspace.ug.edu.gh 2001. Based on the finding of cointegration, they found support for the supply-leading hypothesis in Ghana, Zambia, Senegal, Nigeria and Togo in a VECM framework. Similarly, Ghirmay (2004) using a sample of thirteen Sub-Saharan Africa countries and adopting both Johansen (1988) and Johansen & Juselius (1990) cointegration procedure, established evidence of cointegration between bank development and economic progress in 12 of the 13 countries in their sample. On causality, based on a VECM, they established that financial intermediary development causes growth in eight out of the 13 countries. They concluded that African nations, in general, could, realize faster growth by further developing their financial systems. Even though Agbetsiafa (2004) and Ghirmay (2004) have eight and thirteen countries respectively in their sample, they conduct cointegration and causality tests on the countries separately. On the other hand, Odhiambo (2007) using Johansen & Juselius (1990) cointegration procedure established cointegration between bank development and growth per capita in Tanzania. The study also reported that the supply-leading finance is dominant in Tanzania. Lastly, Chukwu & Agu (2009) found evidence of cointegration between bank development and economic growth in Nigeria for the period 1971 to 2008. They also conduct causality analysis between economic growth and banking industry development using a multivariate VECM for Nigeria. Using bank deposit liabilities and loan deposit ratio, their study established support for the supply-leading hypothesis in Nigeria. Cointegration tests are not limited to country-level research. Pedroni (1999 & 2004) extended tests for cointegration to panel data. Christopoulos, & Tsionas, (2004) combined both the Johansen (1988) cointegration tests and the Pedroni heterogeneous panel cointegration test in their work. In both cases, they concluded that a long-term relationship between banking 32 University of Ghana http://ugspace.ug.edu.gh industry development and economic growth exist for their sample. Concerning causality, their study reported support for the supply-leading hypothesis. Their results are robust to endogeneity bias, physical capital, and inflation. They also made use of the Pedroni (2000) fully modified OLS estimator whose advantage is to account for the endogeneity of predictor variables. The long-term nature of causality they established implies that financial development effects on economic expansion does not happen immediately. Similarly, Acaravci et al. (2009) established support for the supply-leading hypothesis using domestic credit ratio to measure bank development. Their study used the Pedroni (1999 & 2004) cointegration test, which revealed no stable long run relationship between bank development and growth per capita. For this reason, they estimated the causal relationship using a time-stationary VAR model. Toda & Yamamoto (1995) developed a Granger non-causality procedure robust irrespective of the cointegrating characteristics of the series. Hassan et al. (2011) apply this method to 47 developed and 121 developing countries. They found support for the supply-leading finance in SSA and the EAP regions. A significant point of departure from earlier studies is that Hassan et al. (2011), group the countries into six regions namely, the MENA, ECA, LAC, EAP, SSA and South Asia. They also group the high-income countries into OECD high-income countries and high-income non-OECD economies. The groupings bring together economies with almost similar cultures, stages of financial development and legal structures as opposed to grouping them entirely as either developed or developing economies. Another alternative causality procedure we have encountered in the literature is the Geweke (1982) linear dependence and feedback framework. Calderon & Liu (2003) apply this approach to a sample of 109 countries drawn from both developed and developing countries. Their study 33 University of Ghana http://ugspace.ug.edu.gh covers 35 years from 1960-1994. They also tested stages of development hypothesis put forward by Patrick (1966). To get rid of business cycle effects, they average the data set over seven non-overlapping five-year periods and three non-overlapping ten-year periods. They found evidence of supply-leading finance for the whole sample of 109 countries. The supply-leading finance dominates the demand-following finance for a sub-sample of low and middle-income countries. Although the supply-leading finance contributes strongly to growth per capita for this sub-sample at earlier stages of economic development, i.e., 1960- 1979, there is little evidence that the demand-following finance dominates at latter stages i.e. 1980-1994. Instead, a simultaneous feedback causality emerges at last stages of economic development for the low and middle-income sub-sample. In conclusion, therefore, the low and middle-income sub-sample data set does not fully conform to the stages of economic development hypothesis. This could be due to unfinished economic transformation from least developed to developed economies, (Calderon & Liu, 2003). 3.3.2.2 Studies Suggesting a Demand-Following Hypothesis Despite the fact that theory predicts a supply-leading finance, particularly for the developing countries, empirical evidence supporting the demand-following finance is available in the literature. It is also necessary to note that some cross-country studies have established supply- leading finance and demand-following finance in different countries of their samples. Similar works, therefore, may appear under the supply-leading finance sub-section above and here under the demand-following finance. Notable studies that have found support for demand-following hypothesis include (Agbetsiafa 2004; Odhiambo, 2007; Quartey & Prah, 2008; Chukwu & Agu, 2009). Even though these 34 University of Ghana http://ugspace.ug.edu.gh studies conclude that real sector development leads the banking industry development, there are notable similarities and differences among them. All these studies use the Johansen and Juselius (1990) cointegration procedure and the VECM for causality analysis. Agbetsiafa (2004) found support for demand-following finance in 2 out of 8 countries i.e. Kenya and Cote D’Ivoire. Similarly, Odhiambo (2007) established support for the demand-following hypothesis in South Africa using broad money ratio as the bank development proxy. Odhiambo (2007) also concluded that real sector development leads the banking industry development in Kenya using the currency to M1 ratio and bank credit ratio. Odhiambo (2007) therefore, confirmed the findings of Agbetsiafa (2004) that the real sector leads the banking industry in Kenya. Similarly, Chukwu & Agu (2009) found that the real sector development leads bank development in Nigeria using bank credit ratio and broad money ratio to measure of bank development. On the other hand, Quartey & Prah (2008) obtained similar findings for Ghana using broad money ratio to measure bank development. Furthermore, Odhiambo (2008) reinforces the demand-following hypothesis for Kenya using a trivariate VECM. Their study included savings in addition to bank development and growth in the causality analysis. Studies adopting panel methodologies reached similar conclusions as highlighted here below. Rachdi (2011) estimated a panel VECM using the system GMM estimator and found support for the demand-following finance for the MENA region. Similarly, Acaravci et al. (2009) using the system GMM estimator on 24 select SSA countries, concluded that growth per capita Granger-causes financial depth. Lack of cointegration between financial depth and growth per capita guides Acaravci et al. (2009) to adopt a VAR model. Their findings are the same, nonetheless. 35 University of Ghana http://ugspace.ug.edu.gh 3.3.2.3 Studies Suggesting a Bi-Directional Causality The two competing arguments on the causation between bank development and growth are the supply-leading and the demand-following hypothesis. On the middle are studies supporting the view that both economic growth and bank development Granger-cause each other. A number of studies adopted the Johansen & Juselius (1990) cointegration test and a VECM for causality analysis. Agbetsiafa (2004) for instance, using this procedure found support for a bi-direction causal flow between bank development and economic growth in Zambia, South Africa, Kenya, Togo, Ghana, and Nigeria. Furthermore, Ghirmay (2004) also reported a bi-directional causality for six out of thirteen countries in their sample. Similarly, Abu-Bader & Abu-Qarn (2008) established a bi-directional causal link between bank development and growth per capita in Egypt for the sample period 1960-2001. They also reported that finance Granger-causes growth per capita directly and indirectly through investment. Further, their study provided support for a long-term relationship between bank development and growth in Egypt. Their findings are in line empirical literature that suggests that bank development expedite growth by increasing resources available for investment and the efficiency of investment. Similarly, Odhiambo (2007) established support for a bi-directional causal flow between bank development and growth in Tanzania using the currency to M1 ratio to measure bank development. They also reported similar findings for Kenya using the broad money ratio to measure bank development. The main shortcoming of their study is its choice of the broad money ratio as the bank development proxy. Ghirmay (2004) notes that a rise in broad money ratio reflects the degree of monetization of the economy. Theoretically, however, financial development contributes to growth through increased resource mobilization and improved 36 University of Ghana http://ugspace.ug.edu.gh productivity of investment. Ghirmay (2004) therefore, holds the view that the broad money ratio does not adequately meet the theoretical link between bank development and growth. In addition, Demetriades & Hussein (1996) carry out causality analysis for 16 developing countries using time series data spanning at least 27 years. Broadly, their results show evidence of a two-way causality between banking industry growth and growth per capita. They also performed the Johansen cointegration tests that revealed support for cointegration between bank development and growth per capita, in 13 out of the 16 nations in their sample. As stated earlier, Calderon & Liu (2003) applied the Geweke (1982) linear dependence and feedback framework to a sample of 109 advanced and developing countries. They found support for a two-way causal flow between bank growth and economic progress for the developing and developed sub-sample of their data. Hassan et al. (2011) also found support for a bi-directional causality between bank development and growth per capita for the MENA, LAC, ECA and South Asia, OECD high-income countries and high-income non-OECD countries. Separately, Rachdi (2011) applying a system GMM estimator found that a two-way causal flow between bank development and economic growth per capita exists for OECD countries. The difference GMM estimator developed by Holtz-Eakin et al. (1988), to address potential endogeneity issues where researchers lack good instruments, has also been used in literature to estimate causal relationships. The major contribution of the GMM estimator to econometrics is the use of lags of endogenous variables as internal instruments. Adusei (2013) applied it to a panel of 24 African countries, with at least two nations from each of the geographical regions in Africa for the period 1981-2010. To test for Granger-causality, their study used the final prediction ECM and the pairwise Granger causality test. Using the domestic credit ratio and 37 University of Ghana http://ugspace.ug.edu.gh the broad money ratio, they concluded that bank development contributes positively to growth per capita. In addition, they also reported evidence of a bi-directional causal flow between banking sector development and growth in Africa. Acaravci et al. (2009) used a panel cointegration test but obtain similar results. Specifically, their study adopted the Pedroni (1999 & 2004) panel co-integration and Holtz-Eakin et al. (1988) GMM causality test to analyse cointegraion and the causal relationship between banking industry development and growth per capita for 24 selected SSA countries. They found no support for cointegration between bank development and growth per capita, however, their study obtained support for a bi-directional causal link between bank credit ratio and expansion of real GDP per capita. 3.3.2.4 Evidence on no Causal Relationship Perhaps more interestingly, Quartey & Prah (2008) did not find support for causality in any direction between growth per capita and bank development in Ghana, using bank credit ratio, domestic credit ratio, and a ratio bank credit to domestic credit ratio as the proxies for banking industry development. Their study also reported that there exist no long run relationship between bank development and growth per capita using the above bank development proxies. In conclusion, empirical results on causation between bank development and economic growth fall into four broad groups. First, there is a group of results supporting the supply-leading finance. Secondly, another group of findings is in favour of the demand-following hypothesis. Thirdly, another group of researchers are of the view that causality between bank development and economic growth is bi-directional. Lastly, a study by Quartey & Prah (2008) found no causal relationship between bank development and growth in Ghana. It is also important to 38 University of Ghana http://ugspace.ug.edu.gh note that the choice of a bank development proxy affects causality between bank development and economic growth growth (see inter alia Abu-Bader & Abu-Qarn, 2008; Quartey & Prah, 2008; and Acaravci et al., 2009). 3.3.2.5 Studies on the General Contribution of Bank Development to Growth The OLS estimator assumes variables in the regression are exogenous. The bi-directional causation between bank development and growth makes it difficult to satisfy this assumption. Despite this shortcoming, some cross-country studies have adopted the OLS estimator. King & Levine, (1993) applied the OLS estimator to data from 80 countries including both developed and developing countries. They reported that the broad money ratio at the beginning of the period is a good indicator of the level of efficiency of capital apportionment, physical capital accumulation and economic growth use for the next decade and thirty years respectively. King & Levine (1993) results are robust to controls such as initial income, government expenditure, openness to trade, human capital development, and the inflation rate. Their study, however, does not simultaneously examine stock market development. Beck & Levine, (2004) criticized the cross-section approach used by King & Levine, (1993) to the extent that it did not account expeditiously for the simultaneity and omitted variable bias. Cross-sectional approaches also, do not exploit the time series elements of the data. Averaging data over decades also leads to some minor information loss (Beck & Levine, 2004). Similarly, Ndebbio (2004) performs cross-country analysis for 34 selected SSA nations using OLS in multiple regression. They found out that financial deepening contributes positively to overall growth per capita in SSA. Further, they also reported that shallow financial markets and lack of well- functioning stock markets in SSA hinder the financial sector from fully supporting growth per capita. 39 University of Ghana http://ugspace.ug.edu.gh Beck et al. (2000) make significant improvements to the work by King & Levine (1993) by using system GMM for dynamic panel. This methodology controls for the endogeneity of all predictor variables, omitted variables, and simultaneity bias. It also explores the time series dimensions of the data. They had a sample of 77 countries including both developed and developing countries. Their study established that bank development stimulates growth through improvements in productivity. The effect of financial development, physical capital and savings is not clear from their findings. Research on the effect of bank development on economic growth has involved numerous measures of bank development. This raises the question of whether there exist a perfect measure of financial development, given the conflicting results using different proxies. Adu et al. (2013) attempted to fill that gap in the literature by investigating the presence of cointegration between economic growth, bank development, and real deposit rates in Ghana. They covered the period beginning 1961 through to 2010 and employed eight measures of financial development namely; the bank credit ratio, the bank credit to domestic credit ratio and the currency ratio. Other Bank development indicators used included the ratio of currency to broad money, the domestic credit ratio, and currency as a share of nominal GDP, the bank deposit liabilities ratio and the broad money ratio. Using the principal component analysis, they constructed four indexes of financial development, which included a combination of the proxies mentioned above, and included a financial liberalization dummy in the analysis. Adu et al. (2013) adopted the Autoregressive Distributed Lags (ARDL) bounds tests, to run the cointegration tests. The cointegration test revealed that bank credit ratio and bank credit to domestic credit ratio promote economic growth in Ghana. On the other hand, broad money ratio does not contribute to growth. In this regard, their study concluded that the effect of 40 University of Ghana http://ugspace.ug.edu.gh financial development on growth in Ghana depends on the choice of the measure of financial development. They also established that the effect of bank development on growth is responsive to the empirical specification. 3.3.3 Literature on both Stock Markets, Banks, and Growth 3.3.3.1 The Causal link between Stock Markets and Bank Development with Growth In our reading of literature, we have only encountered a few works that include both stock market development, bank development, and economic growth in their causal analysis. Nonetheless, the few we have come across reported mixed results. Yang & Yi (2008) and Adebola & Dahalan (2011) are in favour supply-leading finance hypothesis. On the other hand, Ndako (2010) found evidence for a bi-directional causal flow as well as a demand-following hypothesis based on the stock market development proxy used. We present a brief review of the evidence in the subsequent paragraphs. Yang & Yi (2008) adopted a superexogeneity approach to determine the direction of causation between stock market development, bank development and economic growth in Korea for the period 1971 to 2002. They found support for a unidirectional causal flow from stock market development and bank development to economic growth. Their conditional equation for economic growth and their marginal equation for financial development do not suffer from autocorrelation as proven by the Breusch-Godfrey LM correlation test. Their study also used the cumulative sum of squares, recursive residuals, and recursive coefficients and established that there are no structural breaks in their conditional equation. In sum, their findings support the finance-led growth argument. The modeling of the joint causal relationship between stock markets, banks and economic progress appears limited in Africa. However, the role of stock markets in the process of 41 University of Ghana http://ugspace.ug.edu.gh economic development is gaining traction in Africa and so is the interest in including both the stock markets and bank development variables in growth regressions. For instance, Adebola & Dahalan (2011) found support for the supply-leading finance in Nigeria. They applied the ARDL approach developed by Pesaran et al. (2001) to time series data for Nigeria. Their cointegration results reveal that stock markets and bank development contribution to economic growth is positive but not significant in the short run but significantly positive in the long-term. Similarly, Ndako (2010) modelled the causal relationship between both the stock markets and bank development with economic growth in South Africa using data spanning 1983 quarter one to 2007 quarter four. Their study found evidence of a long-term relationship between stock markets and bank development with economic growth in South Africa based on the Johansen (1988) and Johansen and Juselius (1990) cointegration tests. Specifically, they found evidence of cointegration using the turnover ratio and shares traded ratio to measure stock market development. Using the market capitalization ratio, however, their study found no evidence of cointegration. After establishing the presence of cointegration between stock market and bank development with economic growth, Ndako (2010) proceeded to carry out causality tests using a VECM. Their results revealed a bi-directional causal flow between banking industry growth and economic growth using bank credit ratio as a bank development proxy. Further, using turnover ratio and the shares traded ratio as the stock market development proxies, the paper found that economic growth leads to stock market development. Further, employing Variance Decompositions (VDCs) and Impulse Response Functions (IRFs), their study revealed that stock market and bank development has short run effects on economic growth. Stock market and bank development also explains future economic growth. However, based on Structural 42 University of Ghana http://ugspace.ug.edu.gh Vector Autoregressions (SVAR), they reported little support for the existence of a long run relationship between financial development and economic growth in South Africa. 3.3.3.2 Studies on the Contribution of Stock Markets and Bank Development to Growth This sub-section deals with literature that models the contribution of the stock markets and bank development together to economic growth. As stated earlier, the inclusion of both stock markets and bank development in economic growth regressions started in the late 1990s. Levine & Zervos (1998) pioneered the simultaneous study of the stock markets, bank development, and economic growth covering the period 1976-1993 and included 47 countries. They reported a positive and significant contribution of stock markets and bank development to economic growth, investment and productivity using cross-sectional analysis. Levine and Zervos (1998) also found that the relationship between stock markets and bank development with savings is not significant. They noted that it is not the mere introduction of stocks in the market that affect growth but rather the ease of trading those stocks as confirmed by their findings. However, their work receive a fair share of criticism from Beck & Levine (2004) for adopting OLS in their analysis, which fails to account for unobserved country heterogeneity and simultaneity and omitted variable bias. Levine and Zervos (1998) also use initial values of financial development indicators. Beck & Levine (2004) argue that use of initial values causes loss of information and a probable loss of consistency. Beck & Levine (2004) used the system GMM estimator and reported that stock market development and bank development promote growth per capita for a sample of 40 developed and developing countries. They prefer the system GMM estimator to the difference GMM estimator. They point out that the difference GMM estimator has the following weaknesses; A) It eliminates the country heterogeneity that happens to be a focus of many cross-country 43 University of Ghana http://ugspace.ug.edu.gh studies. B) In small samples, it gives biased estimates with low precision. C) It may also increase biases by reducing the signal to noise ratio. Beck & Levine (2004) make a significant contribution to literature by using a system GMM estimator that improves upon consistency and efficiency of the coefficients. They reported better findings using the system estimator than both the difference and the level estimator. Using an alternative estimator developed by Calderon et al. (2002), they obtained consistent results to those of the system GMM estimator. The Calderon et al. (2002) alternative system estimator minimizes the over-fitting shortfalls of the two-step estimator and obtains heteroskedasticity consistent standard errors. Furthermore, Seetanah et al. (2008), covering the period 1991-2007, found a positive and significant influence of stock markets and banking industry growth on economic growth in 27 developing countries. They adopted a panel VAR approach that accounts for potential endogeneity, causation issues, and country-specific effects. Their study estimated a panel VAR using the GMM after eliminating fixed effects. Their study found that bank development has a higher effect on growth than the effect of stock market development as shown by its higher coefficient. Further, they also report that stock markets and bank development affect growth indirectly through investment channels. They also found evidence that stock market development and bank development play a complementary role in enhancing growth. However, Naceur & Ghazouani (2007) unlike Beck & Levine (2004) and Seetanah et al. (2008) found no evidence that stock markets and bank development promote growth. Based on a study of 10 MENA countries, they reported that stock markets and bank development is not useful in explaining growth. In addition, they also estimated a dynamic panel model with system 44 University of Ghana http://ugspace.ug.edu.gh GMM estimators. Based on their findings, they suggested that stock markets and banks had to experience more growth, for them to have a significant influence economic on growth. 3.4 Conclusion Our review of the literature reveals several notable points. First, on stock market development and economic growth, evidence exists supporting both the supply-leading finance hypothesis and a two-way causation. The literature also suggests that stock markets are necessary for economic growth. Secondly, on bank development and growth, evidence in favour of the supply-leading finance, demand-following finance, a bi-directional causation and indeed no causality exists as we have discussed above. Although most researchers conclude that bank development promotes economic growth, others found that banks need to experience further development before they can have a long-term effect on economic growth. Thirdly, though the literature on the causal linkage between stock market development, bank development and economic growth is scanty, the available literature suggest a supply-leading finance, demand-following finance, and a two-way causation. Some studies have concluded that both stock markets and banks are necessary for long-term growth. Other studies too have found out that stock markets and banks are not useful in explaining long-term growth. The limited literature on the causal linkage between stock market development, bank development and economic growth in SSA necessitate further research. To this end, this study intends to bring more clarity to the long run relationship and the direction of causality between stock market development, bank development, and economic growth by extending the analysis from the two national-level studies mentioned earlier to 13 selected SSA countries. 45 University of Ghana http://ugspace.ug.edu.gh Our study seeks to contribute to knowledge by using a recent panel data set and adopting the system GMM estimator that accounts for endogeneity. We also include a measure of stock market development to avoid the omitted variable bias. We will also examine the sensitivity of the causal linkage between stock market development, bank development and economic growth to a change in the measure of bank development. The next chapter presents the methodology we intend to use to analyse the aforementioned long run and causal relationship. 46 University of Ghana http://ugspace.ug.edu.gh Chapter Four The Methodology 4.1 Introduction Borrowing from the literature review preceding this chapter, this chapter presents the methodology and data that our study intends to use to achieve our objectives. This chapter contains nine sections. The current section introduces the chapter. Section 4.2 presents the theoretical framework that shows the conceptual link between bank and stock market development and growth. Section 4.3 presents the empirical model and measurement of variables. In section 4.4, we describe the stationarity tests we intend to rely on to determine the order of integration of the variables involved. We describe the panel cointegration technique we plan to use to examine the long run relationship between the stock markets and bank development and growth per in Section 4.5. We describe the causality test we intend to undertake in section 4.6. Section 4.7 presents the other diagnostic tests we plan to carry out. We presents the sources of data in section 4.9 and the subsequent section concluded the chapter. 4.2 Theoretical Framework 4.2.1 Introduction Financial development has no growth effects in the traditional growth theory. In this regard. Financial intermediation does not affect the rate of growth of capital per capita and output per capita. It only has level effects as seen in the Solow growth model. The Solow model identified technical progress as the primary determinant of economic growth but did not explain the drivers of technological advances. The Solow model, therefore, treats technological advancement, the sole driver of growth, as exogenous. The development of new endogenous 47 University of Ghana http://ugspace.ug.edu.gh growth models has shown that we can have a sustained growth in the absence of exogenous technical advancement. This reignited interest in the debate on finance and growth in the 1990s (Pagano, 1993). Levine (2002) shows that efficiency and quality of institutions, preferences, income distribution and technology promote sustainable economic progress. 4.2.2 The AK Endogenous Growth Model We reproduce the AK model used by Pagano (1993 p. 614), to demonstrate theoretically how financial development can contribute to growth. We adopt the AK model because it presents a model of an economy that can realize the growth of output per capita permanently in the absence of exogenous technical progress. The share of capital in output in the AK model is one. The AK model is better than the traditional growth theory in which the capital share of production is less than one hence suffering from diminishing marginal returns to capital. The diminishing marginal returns to capital in the traditional growth theory hurts growth. Diminishing marginal returns to capital refers to the concept of decreasing marginal output as more capital is employed. However, for the AK model, production takes place under constant returns to scale where marginal returns to capital remain constant. Consider the AK model where aggregate capital stock affects economic growth linearly as expressed below. 𝑌𝑡 = 𝐴𝐾𝑡 … … … … … … … … … … … … … … … … … … … … … … … … . . (1) Where 𝑌𝑡 is output in year t, A is marginal productivity of capital and 𝐾𝑡 is capital stock in year t. Pagano (1993) assumes that one good is produced which can be consumed or invested and that population grows at a constant rate. The good depreciates at a rate δ each period if invested with investment function expressed as follows. 48 University of Ghana http://ugspace.ug.edu.gh 𝐼𝑡 = 𝐾𝑡+1 − (1 − 𝛿)𝐾𝑡 … … … … … … … … … … … … … … … … … … . . (2) The capital markets clearing condition requires that in equilibrium savings be equal to investment, in an economy with no government. Pagano (1993) further assumes that a given share of savings is lost in the process of intermediation. Given that the share of savings lost is say α we can express the proportion of savings left for investment say Ө as 1-α=Ө. Pagano (1993) argues that the loss occurs through the spreads between the deposit and lending rates, brokerage and commission fees by stockbrokers and inefficiency of the financial institutions among other expenses. Therefore, the proportion of all savings invested Ө multiplied by the total savings in the economy is equated to investment as shown below. In this regard, the capital market clearing condition becomes, Ө𝑆𝑡 = 𝐼𝑡 … … … … … … … … … … … … … … … … … … … … … … … … . . (3) In the AK model, the capital per capita growth rate equals the rate of increase of output per capita since there is no exogenous technical progress. The expression of the growth rate in the following period is as follows. 𝑌𝑡+1 𝐾𝑔𝑡+1= -1 = 𝑡+1 − 1 … … … … … … . … … … … … … … … … … … … (4) 𝑌𝑡 𝐾𝑡 We can express the growth rate of output per capita at the steady state as follows, 𝐼 𝑔 = 𝐴 − 𝛿 … … … … … … … … … … … … … … … … … … … … … … … . (5) 𝑌 Using the capital market clearing condition in equation (3) and denoting the savings rate S/Y as s, we have, 𝑔 = Ө𝐴𝑠 − 𝛿 … … … … … … … … … … … … … … … … … … … … … … … (6) 49 University of Ghana http://ugspace.ug.edu.gh Equation (6) reveals that financial development can affect economic growth through three channels. First, financial development can accelerate economic growth by increasing the proportion of savings invested Ө through improved financial intermediation, efficiency, and reduced transaction and information costs. Secondly, financial development can lead to higher economic growth through increases in the marginal productivity of capital A. Earlier, Patrick (1966) and Diamond (1984) theoretically suggested that financial development accelerates economic growth through improvements in the productivity of capital. Thirdly, financial development can promote economic growth by increasing the savings rates s as highlighted below. The theoretical underpinning of our study is the contribution of stock market development and bank development to growth through enhancing the proportion of savings invested Ө and increases on the marginal productivity of capital A. 4.3.2.1 Efficiency of Capital Allocation Financial intermediaries ensure that capital goes to profitable projects with the highest expected returns. They achieve this by collecting information, which they use to appraise alternative projects competing for funds. Financial institutions also encourage investments in high risk, high returns projects through sharing of risks. These actions improve the capital allocation, efficiency and increases in the productivity of capital A. Productivity gains hastens the rate of economic growth (Pagano, 1993). Through monitoring of borrowers to ensure prudent utilization of borrowed funds, banks ameliorate the problem of moral hazard. Theoretically, a reduction of moral hazard in the credit market improves productivity and accelerates economic growth (Greenwood & Jovanovic, 1990). Empirically, Calderon & Liu (2003) report that financial development promotes economic growth through improvements in productivity. Their findings confirm the theoretical prediction by Pagano (1993) above. 50 University of Ghana http://ugspace.ug.edu.gh 4.3.2.2 The Savings Rate Effect Pagano (1993) found an ambiguous effect of financial development on savings. Empirically, Beck et al. (2000) also reported similar findings. Financial development offers more channels of investment to households and firms. It also reduces the gap between the deposit rates and the loan rates. An increase in savings channels and higher returns on savings may increase the savings rate and by extension economic growth. On the other hand, the presence of insurance markets may lead to lower precautionary savings. The resultant low savings rate may slow down growth. Investors faced with uncertainties about rates of returns from the stock exchange and can solve that through investing in a diversified portfolio. The effect of the uncertainties in returns on savings is not clear. However, according to Pagano (1993), an economy will experience higher economic growth rates if the productivity gains outweigh any reductions in the savings rate. 4.3 The Empirical Model Panel data arises from observing the same cross-sectional units over multiple periods. The units could be countries, households or firms. This study analyses panel data for the selected SSA countries. The availability of stock markets and bank development data solely determines inclusion of countries into this study. The data spans 25 years from 1990 to 2014. According to Beck & Levine (2004), panel data enjoys several advantages over pooled cross-section data. First, it accounts for individual heterogeneity, an aspect not properly accounted for by pooled cross-section data. Secondly, panel data comes with more data points, reduced collinearity among variables, improved efficiency, higher degrees of freedom and more variability. Thirdly, by using panel data models, we account for simultaneity bias and the routine use of lagged outcome variables in regressions involving growth as the outcome variable. 51 University of Ghana http://ugspace.ug.edu.gh The empirical model for this study derives from the theoretical framework above and its prediction that stock market and bank development can increase the proportion of savings invested and marginal productivity of capital. Additionally, we include other determinants of economic growth i.e. physical capital, government expenditure, inflation and openness to trade in our empirical model below. We construct our model based on the following functional form. 𝑌𝑖𝑡 = 𝑓(𝐵𝐷𝑖𝑡, 𝑆𝑀𝐷𝑖𝑡, 𝐺𝐹𝐶𝑖𝑡, 𝐺𝑂𝑉𝑇𝑖𝑡,𝐼𝑁𝐹𝑖𝑡, 𝑂𝑃𝑁𝑖𝑡) … … … … … … . . (7)  𝑌𝑖𝑡 Represents the natural logarithm of real GDP per capita (Constant 2005 international US dollars for country i in year t.  𝐵𝐷 Measures bank development, SMD measures stock market development, GFC measures gross fixed capital formation, GOVT measures government expenditure, INF measures inflation and OPN measures openness to trade. We can rewrite a reduced version of the above augmented production function as follows: 𝑌𝑖𝑡 = 𝛼𝑖 + 𝛽𝑖𝑋𝑖𝑡 + 𝜀𝑖𝑡 … … … … … … … … … … … … … … … … … … . (8) Where:  𝑋𝑖𝑡 Represents predictor variables including measures of stock markets and bank development and control variables.  𝜀𝑖𝑡 Represents the residuals. We have defined 𝑌𝑖𝑡 above. 4.3.1 Measurement of Variables Economic Growth: This study will analyze the long run relationship and causal link between stock market development, bank development and economic growth for a sample of 13 selected 52 University of Ghana http://ugspace.ug.edu.gh SSA countries. We take inspiration from Agbetsiafa (2004) and Demetriades & Hussein (1996) and use real GDP per capita to measure economic growth6. We will obtain data for real GDP per capita from the World development indicators online database of the World Bank. Stock market Development: Economic theory emphasizes that stock markets contribute to economic growth by ameliorating information asymmetry and lowering cost of transactions. However, Beck & Levine (2004) observed that there is no direct measure to the extent by which stock markets achieve the above objective. In cognizance of the lack of a forthright connection between theory and measurement, we follow Beck & Levine (2004) by using the turnover ratio as a proxy for market liquidity to measure stock market development7. Stock markets liquidity reflects the ease of selling and buying stocks in the stock market. Liquid markets accelerate economic growth thus making turnover ratio a good indicator of the effect of stock markets on growth. The turnover ratio equals the value of shares traded divided by market capitalization. This proxy for stock market development has price in the numerator and the denominator, and is therefore not affected by the forward-looking nature of investors in the stock exchange. It measures the volume of stocks traded relative to the size of the stock market. There are alternative proxies of stock market development as the shares traded ratio and the market capitalization ratio. The shares traded ratio measures the activity in the stock markets relative 6 Adusei (2013), Odhiambo (2007), Ndebbio (2004) and Cooray (2010) also employ real GDP per capita to measure economic growth. 7 N'zue (2006), Cooray (2010) and Ndako (2010) also use the turnover ratio to measure stock market development. 53 University of Ghana http://ugspace.ug.edu.gh to nominal GDP. The market capitalization ratio measures the size of the stock markets relative to the size of the economy. The shares traded ratio and the market capitalization ratio; therefore, do not gauge the liquidity of the stock markets. The forward-looking nature of investors also affects the market capitalization ratio as a proxy for stock market development. Levine and Zervos (1998) demonstrated that the ease of trading stocks (liquidity) is a better predictor of growth than a mere introduction of stocks in the stock markets. We obtained the turnover ratio data from the World development indicators online database of the World Bank, the African Securities Exchanges Association (hereafter, ASEA) Financial Data Base and the Stock Exchange of Mauritius (hereafter, SEM) historical data portal. Bank Development: Economic theory predicts that bank development contributes to economic progress through a reduction of information costs and transaction costs. Once again, according to Beck & Levine (2004), there is no direct measure of the extent to which bank development ameliorates information and transaction costs. We recognize the lack of a straightforward link between economic theory and measurement. In this regard, our study follows Levine and Zervos (1998) and Beck & Levine (2004) by using bank credit ratio to proxy for bank development8. Bank credit ratio is measured as the domestic credit to the private sector by banks as a share of nominal GDP. Higher levels of bank credit ratio imply more flow of savings to the private sector and thus, a high degree of financial intermediation and development of the financial 8 Agbetsiafa, (2004), Acaravci et al. (2009), Demetriades & Hussein (1996) and Hassan et al. (2011) also use bank credit ratio to measure bank development. 54 University of Ghana http://ugspace.ug.edu.gh intermediaries (Seetanah et al., 2008). This measure does not include credit allocated to government and government agencies. It also excludes credit given by development banks. The main advantage of using bank credit ratio is that it reflects more accurately allocation of loans to the private firms by the banking industry. Some studies including Christopoulos & Tsionas (2004) and Odhiambo (2008) have used other measures of bank development such as the broad money ratio. The main shortcoming of the broad money ratio given by broad money to nominal GDP ratio is that it measures the degree of monetization of the economy. Theory, however, predicts that financial development affects growth by improving resource mobilization and investment via improvements in productivity. In this regard, we agree credit advanced to the private firms by banking industry, resonates well with theoretical predictions. Although credit by banks to the private firms does not capture developments outside the banking sector, a high degree of financial development in SSA has taken place in the banking industry. Credit to the private sector therefore, is a very representative proxy for bank development in SSA (Ghirmay, 2004). This study will obtain data on bank credit ratio from the World development indicators online database of the World Bank. We will also use two popular measures of bank development to determine whether the causal relationship between stock market development, bank development, and economic growth is sensitive to the choice of bank development proxy. In this regard, we adopt the broad money ratio and the domestic credit ratio as the additional bank development measures. Other studies that have also used the broad money ratio to measure developments in the banking sector include (Odhiambo, 2007; Chukwu & Agu, 2009; Quartey & Prah, 2008). Some works that have used the domestic credit ratio to measure bank development include (Beck et al., 2000; Acaravci et al., 2009; Hassan et al., 2011). We obtain data for the broad money ratio and the 55 University of Ghana http://ugspace.ug.edu.gh domestic credit ratio from the World development indicators online database of the World Bank. Physical Capital: The Solow model predicts that an increase in the level of physical capital increases the steady-state lever of capital per effective worker and output per effective worker. An empirical study by Mankiw et al. (1990) proves that accumulation of physical capital contributes positively to economic growth. We follow Christopoulos & Tsionas, (2004) and control for physical capital defined as gross fixed capital formation as a proportion of nominal GDP. We obtain this data from the World development indicators online database of the World Bank for ten countries in the sample. We obtain data for Botswana, Malawi, and Zambia from the World economic outlook online database of the IMF. Government Expenditure: There are a number of fiscal policy options available to government, key among them being government expenditure and taxation. For simplicity, we use government expenditure to control for fiscal policy effects on growth. Government expenditure is defined as the general government final consumption expenditure as a share of nominal GDP by the World Bank. Other studies that have used government spending in growth regressions are (Levine, 2002; N'zue, 2006; Adusei, 2013). We obtain the government spending data for all countries except Zambia, from the World development indicators online database of the World Bank. We obtained data for Zambia from the Macro Economy-Meter Database (hereafter, MEM). Inflation: Price instability measured by inflation can have detrimental effects on economic growth, stock markets and bank development. Central banks use various monetary policy tools to achieve a given inflation target. To control for the effect of monetary policy on economic growth, stock markets and bank development therefore, we take inspiration from Beck & 56 University of Ghana http://ugspace.ug.edu.gh Levine (2004) and use inflation for this purpose. We obtain inflation figures from World development indicators online database of the World Bank for ten countries. Inflation data for Namibia and Uganda were obtained from the World economic outlook online database of the IMF. Lastly, we collect inflation figures for Zimbabwe from the Macro Economy-Meter Data Base. Openness to Trade: We also control for openness to trade, proxied by the summation of imports and exports of goods and services as a share of nominal GDP. Beck & Levine (2004), also control for openness to trade in their study. We obtain the exports and imports of goods and services as a share of nominal GDP data from World development indicators online database of the World Bank. 4.4 Panel Unit Roots Tests This study will apply panel stationarity tests to all the variables in our specifications. The aim of the panel stationarity tests is to determine the order of integration in the variables and avoid any biases that arise from using non-stationary data. The use of stationary series helps us avoid reporting spurious results. We obtain the general panel unit root equations from Stock & Watson (2007): 𝑌𝑖𝑡= б𝑌𝑖𝑡−1+ ϒ𝑋𝑖𝑡 +𝑈𝑖𝑡 … … … … … … … … … … … … … … … … … … . (9) The variable 𝒀𝒊𝒕 has a unit root if |б|=1. The variable is weakly stationary if |б|<1. Taking away Yt-1 from both sides gives: Δ𝑌𝑖𝑡= (б-1) 𝑌𝑖𝑡−1 + ϒ𝑋𝑖𝑡 +𝑈𝑖𝑡 … … … … … … … … … … … … … … … . (10) Let б-1=η so that the ADF-type specification now becomes: 57 University of Ghana http://ugspace.ug.edu.gh Δ𝑌 𝜆𝑖𝑡= η𝑌𝑖𝑡−1+ ϒ𝑋𝑖𝑡+∑𝑗=1 𝜃𝑖𝑗∆𝑌𝑖𝑡−1 + 𝑈𝑖𝑡 … … … … … … … … … . . . (11) Where:  𝑌𝑖𝑡 Represents each variable under consideration.  λ represents the lag length  𝑋𝑖𝑡 stands for the panel-specific fixed effects or panel-specific fixed time effects  ϒ represents the corresponding vector of coefficients  i stands for the cross-sectional units while t represents time. The unit root test involves exploring the following null and alternative hypothesis: 𝐻0: η=0 (The series is non-stationary) 𝐻1: η < 0 (The series is stationary) Taking inspiration from Acaravci et al. (2009), we will also adopt two stationarity tests for heterogeneous panels to determine the order of integration of our variables. Heterogeneous panel unit roots tests are appropriate for our data set since the different countries included are at varying stages of economic, stock markets and bank development. The two-panel unit root tests adopted are, the Choi (2001) and the Im, Pesaran & Shin (IPS, 2003)9 stationarity test. Both tests provide better results than the Levine et al. (2002) unit root test. The Levine et al. (2002) unit root test makes an unrealistic assumption that the coefficients of the autoregressive term are homogenous across all individual units. The advantage of the Choi (2001) and the IPS 9 Because of the limitations inherent in panel unit root tests, most studies use more than one unit root test e.g. (Agbetsiafa, 2003; Odhiambo, 2008; Abu-Bader & Abu-Qarn, 2008; Demetriades & Hussein, 1996) 58 University of Ghana http://ugspace.ug.edu.gh (2003) over the Levine et al. (2002) is that they allow the autoregressive term to vary across the individual units. Christopoulos & Tsionas (2004) also uses the IPS (2003) stationarity test for panel data. 4.4.1 The Choi (2001) Test Choi (2001) uses the following model which we reproduce from Acaravci et al. (2009, p.20). 𝑦𝑖𝑡 = 𝑑𝑖𝑡 + 𝑥𝑖𝑡 … … … … … … … … … … … … … … … … … … … … … . . (12) In equation (12) above:  (𝑖 = 1, … , 𝑁; 𝑡 = 1, … , 𝑇𝑖)  𝑑 = 𝛽 + 𝛽 + ⋯ + 𝛽 𝑡𝑚𝑖𝑖𝑡 𝑖0 𝑖1 𝑖𝑚𝑖  𝑥𝑖𝑡 = 𝛼𝑖𝑥𝑖(𝑡−1) + 𝜇𝑖𝑡 Choi (2001) permits the separate time series variables to take varying sample sizes and varying specifications of stochastic and non-stochastic components based on i. Choi (2001) tests that following null and alternative hypothesis. 𝐻0: 𝛼𝑖 = 1 (The series is non-stationary) 𝐻1: |𝛼𝑖| < 1 (The series is stationary) Choi (2001) develops the following Fisher Augmented Dickey-Fuller test statistic: 1 ∑𝜂Z= −1𝑖=1 ɸ (𝜌𝑖)~𝑁(0,1) … … … … … … … … … … … … … … … … (13) √𝑛 Where ɸ-1 represents the inverse of the standard normal cumulative distribution function. 59 University of Ghana http://ugspace.ug.edu.gh 4.4.2 The IPS (2003) Stationarity test for Panel Data The IPS (2003) test gives room for the coefficients of the autoregressive term to vary across all individual units in the data. In other words, it allows for heterogeneity among individual units. The test makes use of a standardized t-bar statistic. The statistic involves the estimation of separate unit root tests for all individual units then it gets an average of their individual ADF t statistics. The computation of the 𝑡̅ statistic, done through Monte Carlo simulations, uses the following specification: √𝑁(𝑡 −1𝑖𝑇 − 𝑁 ∑ 𝑁 ̅ 𝑖=1 𝐸(𝑡𝑖𝑇)) 𝑡 = … … … … … … … … … … … … … … (14) √𝑁−1 ∑𝑁𝑖=1 𝑉𝑎𝑟(𝑡𝑖𝑇) Where:  𝑡𝑖𝑇 Represents the separate ADF t-statistics for the N cross sectional units.  𝐸(𝑡𝑖𝑇) Represents the mean of 𝑡𝑖𝑇 while Var (𝑡𝑖𝑇) stands for the variance of 𝑡𝑖𝑇. IPS (2003) assumes independence of the entire cross-sectional unit. By implication, the assumption implies no short-run or long run cross-equation correlation. 4.5 Panel Cointegration This study will adopt the Pedroni (1999 & 2004) heterogeneous panel cointegration tests, hereafter (Pedroni co-integration test), consistent with Christopoulos & Tsionas (2004) and Acaravci et al. (2009). This test is a residual based test. The countries included in this study are at different levels of stock markets and bank development and economic growth. These differences may create country heterogeneity. In this regard, we choose the Pedroni co- integration test since it permits for heterogeneity arising from linear trends, slope coefficients and individual effects across countries. 60 University of Ghana http://ugspace.ug.edu.gh Pedroni (2004) modeled an equation of the following form: 𝑦𝑖𝑡 = 𝛼𝑖 + 𝜋𝑖𝑡 + 𝜏𝑖𝑥𝑖𝑡 + 𝜀𝑖𝑡 … … … … … … … … … … … … … … … . (15) Where: 𝑦𝑖𝑡 and 𝑥𝑖𝑡 represents real GDP per capita, proxies of stock markets and bank development and control variables for country i in year t respectively and should not contain a unit root in their first difference. The parameters 𝛼𝑖 and 𝜋𝑖 give room for country specific effects and individual linear effects respectively. Pedroni (1999) points out that,𝛼𝑖 the country specific or fixed effect parameter can vary across the countries in the sample. The slope coefficients 𝜏𝑖 also can vary across countries (Acaravci et al., 2009). Pedroni (1999) developed asymptotic distributions and analyzed small sample performance of seven tests statistics that form the basis for his panel co-integration. We can categorize the test statistics he developed into two broad groups. The first group, which focuses on the ‘within’ aspect of panel data, referred hereafter as a panel, comprises of four test statistics, namely: panel parametric-ADF, panel-V, panel-non-parametric and panel-rho. Pedroni (1999) mentions that the panel-ADF, panel-PP and panel-rho correspond to the Phillips-Peron ADF, t and rho statistics respectively. The second group that emphasizes on the ‘between’ aspect of panel data, hereafter (group), comprises of three test statistics namely: group-pp, group-ADF, and group- rho. Pedroni (1999) notes that the group-rho, group-PP and group-ADF correspond to the Phillips-Peron rho, PP and ADF statistics respectively. The group-rho, panel-rho, and panel- ADF are the most powerful for a sample with at least 20 observations. According to Pedroni (1999), the within statistics which he refers to as the panel cointegration statistics are outcomes of estimators which result from effectively pooling the autoregressive 61 University of Ghana http://ugspace.ug.edu.gh coefficients across different cross-sectional units for the stationarity tests on the estimated residuals. The within statistics assumes that 𝛼?̂? is the same for all cross-sectional units. In this regard, the Pedroni panel co-integration test involves examination of the following null and alternative hypothesis for the within dimension: 𝐻0: 𝛼?̂? = α = 1 for all i (No co-integration) 𝐻1 = 𝛼?̂?= α<1 for all i (Presence of co-integration) The between-statistics, which Pedroni (1999) refers to as the group mean panel cointegration statistics, on the other hand, result from estimators that only average the separately estimated coefficients for each cross-sectional unit i. In this case, the assumption that 𝛼?̂? is the same for all cross-sectional units is relaxed. For the case of the between dimension, the following null and alternative hypothesis apply: 𝐻0: 𝛼?̂? = 1 for all i (No co-integration) 𝐻1 = 𝛼?̂?<1 for all i (Presence of co-integration) We fail to reject the null hypothesis of no cointegration if the computed statistics are less than the tabulated ones. 4.6 Causality tests Pedroni (1999 & 2004) panel cointegration test is a test of the presence or absence of a stable long run relationship. The tests does not indicate direction of causality. According to Acaravci et al. (2009) the findings of a panel cointegration tests influences the choice of a VAR framework. In this regard, if we fail to reject the null hypothesis of no cointegration, then it 62 University of Ghana http://ugspace.ug.edu.gh will be more appropriate to estimate causality using a time-stationary VAR framework. The presence of cointegration will call for the adoption of a time-stationary error correction VAR framework. According to Patrick’s (1966) stages of development hypothesis, causality between financial development and economic growth can be either way depending on the stages of development of the economy. As such, we cannot rule out potential endogeneity in our model. Several estimation techniques address the problem of endogeneity. The Instrumental variable and two stages least squares estimators account for endogeneity. They, however, requires external instruments that should not correlate with the residuals but be correlated with the endogenous variable. Getting instruments that satisfy the above requirement is a difficult task, especially when faced with more than one endogenous variable. In this regard, we will opt for the GMM estimator since it uses lags of the endogenous regressors as internal instruments. GMM is applicable for panels with large cross-sectional observations (N) and shorter periods (T). Unavailability of good external instruments also makes GMM, which uses internal instruments, convenient. It also produces robust results in the presence of fixed effects, heteroskedasticity and endogeneity (Roodman, 2009). Also, Baum et al. (2003) observed that the GMM estimator has higher efficiency gains than the instrumental variable and the two-stage least squares estimator, in the presence of heteroskedasticity. Baum et al. (2003) argued that the GMM estimator obtains efficient estimates by using orthogonality conditions. Empirical literature presents two forms of the GMM estimator. The difference estimator put forward by Holtz-Eakin et al. (1988) and the system estimator put forward by Arellano & Bover 63 University of Ghana http://ugspace.ug.edu.gh (1995). Holtz-Eakin et al. (1988, p.1373) specified a time-stationary VAR model of the following form: 𝑚 𝑚 𝑌𝑖𝑡 = 𝛼0 + ∑ 𝛼𝑗𝑌𝑖𝑡−𝑗 + ∑ 𝛽𝑗𝑋𝑖𝑡−𝑗 + 𝜇𝑖 + 𝜀𝑖𝑡 … … … … … … … . (16) 𝑗=1 𝑗=1 𝑚 𝑚 𝑋𝑖𝑡 = 𝛿0 + ∑ 𝛿𝑗𝑌𝑖𝑡−𝑗 + ∑ 𝛾𝑗𝑋𝑖𝑡−𝑗 + 𝜂𝑖 + 𝑣𝑖𝑡 … … … … … … … . (17) 𝑗=1 𝑗=1 Where 𝑌𝑖𝑡 has been defined earlier and, 𝑋𝑖𝑡 represents a proxy of either stock markets and bank development and the control variables. 𝜀𝑖𝑡 and 𝑣𝑖𝑡 represent the respective error terms. 𝜇𝑖 and 𝜂𝑖 stand for country specific effects. The analysis of whether stock markets and bank development granger causes growth, involves examination of the joint hypothesis that 𝛽1 = 𝛽2 = ⋯ = 𝛽𝑚 = 0 . However, the lagged outcome variable maybe correlated with the lagged residuals resulting in endogeneity. In addition, correlation between the fixed effects and the residuals results in biased estimates (Nickell, 1981). To get rid of the fixed effects, Holtz- Eakin et al. (1988) advice the use of the first difference operator that leaves us with the following equations: 𝑚 𝑚 ∆𝑌𝑖𝑡 = ∑ 𝛼𝑗∆𝑌𝑖𝑡−𝑗 + ∑ 𝛽𝑗∆𝑋𝑖𝑡−𝑗 + ∆𝜀𝑖𝑡 … … … … … … … … … . . (18) 𝑗=1 𝑗=1 𝑚 𝑚 ∆𝑋𝑖𝑡 = ∑ 𝛿𝑗∆𝑌𝑖𝑡−𝑗 + ∑ 𝛾𝑗∆𝑋𝑖𝑡−𝑗 + ∆𝑣𝑖𝑡 … … … … … … … … … . (19) 𝑗=1 𝑗=1 64 University of Ghana http://ugspace.ug.edu.gh The difference estimator gives consistent estimates by addressing the issue of endogeneity of some predictor variables. This is after taking away some of the inconsistency through the first difference operator. The estimation of the difference estimator uses equations 18 and 19. The system estimator developed by Arellano & Bover (1995) includes additional assumptions that the differenced instruments are not correlated with the fixed effects. This additional assumption makes use of more instruments possible, and the system estimator realizes more efficiency gains (Roodman, 2006). The system GMM estimator uses the level equations 16 and 17 and the difference equations 18 and 19 in a system. For the system estimator, the instruments utilized in the difference regressions are the lagged levels and the instruments for the level regressions most recent differences. We will adopt the system estimator developed by Arellano & Bover (1995) over the difference estimator for three reasons. First, the system estimator is more efficient compared to the difference estimator based on its ability to use more instruments. Secondly, the difference estimator gets rid of the country heterogeneity, a shortcoming addressed by the system estimator. Thirdly, small samples biases the difference estimator, a problem dealt with by the system estimator (Blundell & Bond, 1998; as cited in Beck & Levine, 2004). 4.7 Diagnostic tests 4.7.1 The Sargan Test The Sargan test is a j test statistic used to verify the validity of overidentifying restrictions in an econometric model. The problem of over-identification may arise from the utilization of the instruments mentioned above. Overidentification, results, when the instruments used, exceed the number of endogenous variables. Assuming the overidentifying restrictions are valid GMM yields consistent estimates. To verify the validity of the overidentifying restrictions, we will 65 University of Ghana http://ugspace.ug.edu.gh run the Sargan test. The null hypothesis of the Sargan test is that overidentifying restrictions are valid. It follows, therefore that the alternative hypothesis is that the overidentifying restrictions are not valid. 4.7.2 Autoregressive Test Absence of the second-order serial correlation in the differenced error terms is a critical assumption of the GMM estimators. In this regard, we will perform the Arellano and Bond autocorrelation test, to check for the presence or absence of second-order autocorrelation. The null hypothesis for this test is no second order autocorrelation. 4.7.3 Endogeneity Test We will estimate equations 16, 17, 18 and 19 as a system using the xtabond2 command written for STATA by Roodman (2006). This estimation requires specification of endogenous and exogenous variables separately. To meet this condition, we will perform an endogeneity test to classify our regressors as either endogenous or exogenous for all the nine specifications in our causality analysis. We will adopt the Hausman endogeneity test for this purpose. The null hypothesis of the Hausman endogeneity test is that the variable of interest is exogenous. It follows that the alternative hypothesis is that the variables of interest are endogenous. 4.7.4 The Heteroskedasticity Test This study will adopt the Breusch-Pagan/ Cook-Weisberg test heteroskedasticity. The null hypothesis of this heteroskedasticity test is homoscedasticity while the alternative hypothesis is heteroskedasticity. In the event that we detect heteroskedasticity in any of our nine causality specifications, we will include the robust command to correct for it. 66 University of Ghana http://ugspace.ug.edu.gh 4.7.5 Time Effects, Breusch & Pagan Test and Hausman Specification Test This study will test for time effects in all our nine causality specifications. The study will control for time in the specifications where time effects is detected. In addition, this study will adopt the Breusch & Pagan Lagrangian Multiplier test test for random effects. The null hypothesis for this test is that the pooled model is appropriate. The pooled model assumes that individual heterogeneity is not significant. The random effects model (hereafter, REM) assumes that the country specific differences are significant but not correlated with the other regressors. A correlation between the country specific effects and other regressors means that the REM is inappropriate. In the event that the Breusch & Pagan test indicates that the REM is appropriate, this study will proceed to run the hausman specification test to determine the appropriate model. The null hypothesis of the hausman specification test is that the REM is appropriate. Its alternative hypothesis is that the fixed effects model (hereafter, FEM) is appropriate. The FEM model is appropriate when the individual heterogeneity is correlated with the other regressors. 4.8 The Sources of Data According to the 2015 United Nations Development Programme, Human Development report, there are 46 countries in SSA. The SSA region had a combined nominal GDP of $1.712 trillion in 2015. The estimated land area covered by SSA is 23,638,000 square kilometres with an estimated population of 953.4 million people of which 37% live in urban areas (WDI, 2015). Of the 46 countries in SSA, only 20 of them had established stock markets as of the year 2012. This study focussed on 13 of these countries whose data on stock markets and bank development variables is readily available. Table 4.1 provides a summary of the variables used in this study, their measurement and sources of data. 67 University of Ghana http://ugspace.ug.edu.gh Table 4.1: Summary of Variables, Measurement and Data sources Variable Measurement and Supporting Literature Data Source(s) Economic GDP per capita (constant 2005 US$) WDI, 2015 Growth (Adusei, 2013; Odhiambo, 2007; Ndebbio, 2004; Cooray, 2010) Stock Market Stocks traded, turnover ratio (%) WDI, 2015 Development (N'zue, 2006; Cooray, 2010; Ndako, 2010) ASEA,2015 SEM, 2015 Bank Domestic credit to private sector by banks (% of GDP) WDI, 2015 Development (Agbetsiafa, 2004; Acaravci et al., 2009; Demetriades & Hussein, 1996) Broad money (% of GDP) WDI, 2015 (Odhiambo, 2007; Chukwu & Agu, 2009; Quartey & Prah, 2008) Domestic credit to private sector (% of GDP) WDI, 2015 (Beck et al., 2000; Acaravci et al., 2009; Hassan et al., 2011) Physical Gross fixed capital formation (% of GDP) WDI,2015 Capital (Christopoulos, & Tsionas, 2004; Ndako, 2010) WEO, 2015 Government General government final consumption expenditure (% of WDI, 2015 Expenditure GDP) MEM, 2015 (Levine, 2002; N'zue, 2006; Adusei, 2013). Inflation Inflation, consumer prices (annual %) WDI,2015 (Beck & Levine, 2004: Adebola & Dahalan, 2011) WEO, 2015 MEM, 2015 Openness to Imports + Exports of goods and services (% of GDP) WDI, 2015 Trade (Beck & Levine, 2004; Quartey & Prah, 2008) Note: The sample period for the study is 1990-2014 though some variables have shorter data spans in some countries. In effect, the panel data for this study is unbalanced. Source: Author’s Compilation 4.9 Conclusion To summarize, this study will estimate a dynamic growth equation based on the AK growth model. We expect to experience estimation challenges such as endogeneity, autocorrelation, and heteroscedasticity. In this regard, we plan to use the Hausman test for endogeneity, 68 University of Ghana http://ugspace.ug.edu.gh Arellano & Bond autocorrelation test and Breusch-Pagan/Cook Weisberg test for heteroskedasticity and make corrections as necessary. This study will also test for time effects and control for time where necessary. In addition, this study will conduct the Breusch & Pagan test for random effects and the Hausman specification test. To evaluate the long run relationship between stock markets, banks, and growth, our study will use the Pedroni panel cointegration test as described above. We will also conduct causality test based on the system GMM estimator as we have highlighted above. As stated earlier, we favour the system estimator since its results are more efficient that the difference estimator. The GMM estimator allows the inclusion of lagged outcome variable as a predictor variable, addresses the issue of unobserved country heterogeneity and the problem of endogenous regressors. We proceed to discuss the results based on the Choi (2001) and IPS (2003) stationarity tests, Pedroni panel cointegration tests, system GMM estimation in the next chapter of this study. 69 University of Ghana http://ugspace.ug.edu.gh CHAPTER FIVE ANALYSIS AND DISCUSSION 5.1 Introduction In this chapter, we present the findings of this study and discuss the results. This chapter comprises of ten sections. The current section introduces the chapter. Section 5.2 gives the summary statistics of the variables we have included in our analysis. Panel unit roots tests are discussed in section 5.3. We present the results of the cointegration tests in fulfillment of the first objective of our study in section 5.4. In section 5.5, this study presents the cointegration results using alternative measures of bank development. We show the correlation results among the variables in section 5.6. Section 5.7 contains the results and discussion of the causality results that fulfill our second objective. To meet our third goal, we present causality results using alternative measures of bank development in section 5.8. Other diagnostic test results are presented in section 5.9, and section 5.10 summarizes this chapter. 5.2 Summary Statistics The summary statistics presented below are for 13 selected SSA countries for the sample period 1990-2014. The statistics of interest to us are the number of observations, maximum values, standard deviation, mean and minimum values of the variables. Table 5.1 displays these summary statistics. From our sample of 13 SSA countries, the real GDP per capita for the period beginning 1990 up to 2014 averaged $1,919.51 with a total number of 325 observations. We notice significant variations in the levels of real GDP per capita across the countries in our sample. Malawi, for instance, had the lowest level of real GDP per capita at $182.24 in 1992. On the other hand, Mauritius attained the highest level of real GDP per capita of $7,116.59 in 2014, the highest in our sample. 70 University of Ghana http://ugspace.ug.edu.gh Table 5.1: Summary statistics for Real GDP per capita and other predictor variables Variable Observations Mean Std. Dev. Min Max GDP per capita N = 325 1919.51 2007.76 182.24 7116.59 Bank Credit Ratio N = 314 24.67 21.76 3.09 108.06 Broad Money Ratio N = 314 34.8 22.18 7.6 151.55 Domestic Credit N = 310 30.8 34.4 3.09 160.12 Ratio Turnover Ratio N = 256 7.82 11.59 0.01 64.26 Physical Capital N = 321 19.75 6.85 2.00 37.02 Inflation N = 321 46500000.00 833000000.00 -0.30 14930000000.00 Government N = 320 16.15 5.15 2.05 31.82 Openness to Trade N = 324 77.47 32.4 26.61 202.85 Source: Author’s computation using STATA 13 and data from WDI, 2015 The number of observations for the turnover ratio (our proxy for stock market development) is 256. This figure is considerably lower than the 314 observations we obtained for the bank credit ratio, our most important proxy for bank development. Obtaining consistent and continuous data for stock markets variables is a big challenge in SSA. The fact that the establishment of stock markets did not take place in the same year also explains the shortfall in the data. Nonetheless, the turnover ratio had an average of 7.82% for the period 1990-2014. The average turnover ratio points to the low liquidity of stock markets in our sample. The Swaziland stock exchange had the lowest turnover ratio of 0.01% in the sample. As expected, the Johannesburg stock exchange had the highest turnover ratio of 64.26% in the sample. The minimum and maximum turnover ratio in the sample shows the gap that exists between countries in the sample concerning stock market development. It is also critical to note that South Africa has the largest and oldest stock exchange in our sample and SSA region. This study obtained 314 observations for our measure of bank development, i.e. bank credit ratio. The mean of the bank credit ratio for the sample was 24.67%. Going by the available data, 71 University of Ghana http://ugspace.ug.edu.gh Mauritius had the most developed banking sector with a bank credit ratio of 108.06% in 2013. Tanzania had the lowest bank credit ratio of 3.09% in the sample. The first alternative measure of bank development for this study, the broad money ratio, had an average of 34.8% with 314 observations. Uganda recorded the lowest level of the broad money ratio in 1990 at 7.6% in the sample. On the other hand, Zimbabwe had the highest level of broad money ratio at 151.55% attained in 2002. The broad money ratio figure in Zimbabwe in 2002 has to be interpreted with caution given the hyperinflation that Zimbabwe faced beginning 2001. The third measure of bank development i.e. the domestic credit ratio has 310 observations and an average of 30.8% for the sample period. South Africa had the highest level of domestic credit from the financial sector allocated to the private sector at 160.12% in 2007, in our sample. Tanzania, on the other hand, had the minimum allocation of credit to the private sector by the financial sector at 3.09% in 1996. Physical capital proxied by gross fixed capital formation as a proportion of nominal GDP had a mean of 19.75% with 321 observations. Physical capital for our sample ranges from a minimum of 2% and a maximum of 37.02% for Zimbabwe in 2005 and Zambia in 2004 respectively. Again, we notice a significant gap between the lowest value and the highest value of physical capital in the sample. We obtained 320 observations for government expenditure defined as the general government final consumption expenditure. Government spending averaged 16.15% for the sample. The minimum value of public expenditures in the sample was 2.05%, and the maximum was 31.82% for Zambia and Zimbabwe respectively. Just like physical capital, we obtained 321 observations for inflation. Inflation in our sample had an average of 46.5 million percent. This figure was profoundly influenced by inflation data for Zimbabwe, which experienced hyperinflation beginning 2001 peaking in 2008 at 14.93 72 University of Ghana http://ugspace.ug.edu.gh billion percent. We exclude Zimbabwe from the main estimations hereafter because of this hyperinflation that is so high compared to the rest of the countries in the sample. Therefore, we are left with 12 countries in the main estimation. Uganda experienced deflation in 2002 at - 0.3%, the lowest rate in the sample. As mentioned earlier, Zimbabwe has the maximum rate of inflation in our sample at 14.93 billion percent. Openness to trade measures the extent of international trade in the economy. We obtained 324 observations for openness to trade measured by summation of exports and imports of goods and services as a share of nominal GDP. Swaziland and Uganda had the maximum and minimum values of openness to trade in our sample at 202.85% and 26.61% respectively. The next section highlights the unit root test results for our outcome as well as predictor variables, excluding Zimbabwe. 5.3 Panel Unit Root Results The first goal of this study was to examine the existence of a long run relationship between growth, stock markets development, bank development and the control variables, which are physical capital, government expenditure, inflation and openness to trade. To meet this objective, we conducted unit root test, to identify the order of integration of the variables involved. In this regard, as discussed in the preceding chapter, we conducted the Choi (2001) and IPS (2003) stationarity tests. We report the results of these tests adjusted to three decimal places in Table 5.2. The findings of the unit roots tests reported in Table 5.2 are very similar for the two different stationarity tests. Lag length selection for the two unit root tests conducted was automatic based on the Schwarz information criteria. The null hypothesis for the two tests is that the individual series in the panel have a unit root. The alternative hypothesis is that at least one individual series in the group does not contain a unit root. From the results, this study cannot reject the null hypothesis of a unit root for the log of real GDP per capita, the broad money ratio and 73 University of Ghana http://ugspace.ug.edu.gh government expenditure in levels. This result means that these three variables are not stationary in levels. We also fail to reject the null hypothesis of a unit root for the domestic credit ratio using the Choi (2001) test. However, the results firmly rejected the null hypothesis of a unit root for these variables at the 1% level of significance, in their first difference. Table 5.2: Panel Unit Roots Results (Individual Intercept and Trend) Variables Fisher-ADF( Choi, 2001) IPS (2003) Levels Differences Levels Differences lgdp 0.001(4) -5.982(3)*** 0.103(4) -7.164(3)*** Bank Credit Ratio -2.468(4)*** -2.884(4)*** Broad Money Ratio -0.671(3) -7.932(3)*** -0.786(3) -9.307(3)*** Domestic Credit Ratio -1.146(4) -8.469(4)*** -1.653(4)** Turnover Ratio -6.277(4)*** -8.622(4)*** Physical Capital -1.975(1)** -1.891(1)** Government -0.987(4) -8.628(4)*** -1.206(4) -10.461(4)*** Inflation -8.466(3)*** -13.597(3)*** Openness to trade -2.406(2)*** -2.426(2)*** Notes: Maximum lag length automatically selected by SIC in parenthesis (). ***, **, * signifies the level of significance at 1%, 5% & 10% respectively. Source: Author’s computation using Eviews 9 and data from WDI, 2015 This study therefore arrived at the conclusion that the log of real GDP per capita, the broad money ratio and government expenditure are I(1) variables. We strongly rejected the null hypothesis of a unit roots for the series, bank credit ratio, turnover ratio, inflation and openness 74 University of Ghana http://ugspace.ug.edu.gh to trade at the 1% level of significance in their levels. This means that the series, turnover ratio, bank credit ratio, inflation and openness to trade were all stationary in levels. We, therefore, conclude that bank credit ratio, turnover ratio, inflation and openness to trade are I(0). On the other hand, this study rejected the null hypothesis of a unit root for the series domestic credit ratio and physical capital at the 5% level of significance in levels. This result implies that domestic credit ratio and physical capital were stationary in levels at 5% levels of significance. This study concluded that domestic credit ratio and physical capital are also I(0). These results are relevant for this study since, as mentioned earlier, running regressions on non- stationary series may lead to spurious results. In this regard, we converted the log of real GDP per capita, broad money ratio and government expenditure into their first difference to remove unit roots in main estimations. 5.4 Panel Cointegration Results This section presents the Pedroni cointegration results. This study reports the results of the Pedroni cointegration test adjusted to three decimal places, in Table 5.3. The Pedroni cointegration test is appropriate for our study since it yields consistent results irrespective of whether the series are I(0) or I(1). With this conclusion in mind, we now turned our attention to fulfilling the first objective of this study. The first objective of this study was to examine whether there exists a long run relationship between stock market development, bank development and growth controlling for physical capital, government expenditure, inflation and openness to trade. In Table 5.3, we report the weighted panel statistics in addition to the traditional seven statistics put forward by Pedroni (1999). Lag length selection for the Pedroni cointegration test was automatic based on the Schwarz information criteria. 75 University of Ghana http://ugspace.ug.edu.gh The null hypothesis of the Pedroni cointegration test is that there is no cointegration between stock market development, bank development, and growth per capita. The alternative hypothesis is that the aforementioned variables are cointegrated, i.e., they have a long run relationship. Table 5.3 gives eleven statistics and their respective significance levels. We conclude that there is cointegration among the variables if the majority of the eleven statistics were significant at the 5 percent level. We report three versions of the Pedroni cointegration test. On the second column of Table 5.3, we report the Pedroni cointegration test including the individual intercept only. The third column includes the Pedroni cointegration test with individual intercept and individual trend. Lastly, the fourth column of Table 5.3 reports the Pedroni cointegration test with no intercept or trend. From column two, the panel-pp and panel-ADF statistics are significant at 5% significance level. On the other hand, the weighted panel-pp, weighted panel-ADF and group-pp statistics are significant at 10% significance level. Assuming that the data has no trend, only two statistics are significant at the 5% percent significance level. Since a majority of the statistics, i.e. nine are not significant at 5%; this study cannot reject the null hypothesis of no cointegration for the case of individual intercept only. For the case of individual intercept and trend, in column three of Table 5.3, the panel-v and the weighted panel-v are significant at the 1% significance level. A closer look at that column reveals that only two out of eleven statistics are significant. This study therefore, could not reject the null hypothesis of no cointegration for the case of individual intercept and individual trend. Lastly, in column four, we assumed no trend or intercept. For this case, the panel-pp, panel- ADF, weighted panel-pp, panel-ADF, group-pp and group-ADF statistics are significant at 1% significance level. A majority of the statistics are significant, hence for the case of no intercept 76 University of Ghana http://ugspace.ug.edu.gh and trend, this study concludes that there is cointegration. However, this study concluded that there is no evidence of cointegration between growth per capita, stock markets and bank Table 5.3: Pedroni Heterogeneous Panel Cointegration Tests (Bank Credit Ratio) Individual Individual intercept No intercept or intercept and individual trend trend Panel Statistic Panel v-Statistic -2.043 3.566*** -4.739 Panel rho-Statistic 1.963 4.170 1.750 Panel PP-Statistic -1.924** 1.254 -4.893*** Panel ADF-Statistic -2.043** 1.205 -4.578*** Weighted Statistic Panel v-Statistic -1.287 3.580*** -4.757 Panel rho-Statistic 2.112 4.036 1.839 Panel PP-Statistic -1.481* 0.488 -3.902*** Panel ADF-Statistic -1.454* 0.514 -3.778*** Group Statistic Group rho-Statistic 3.781 5.293 2.951 Group PP-Statistic -1.138 -0.792 -5.364*** Group ADF-Statistic -0.969 -0.149 -4.295*** SIC lag length 1 to 4 0 to 3 1 to 4 Notes: Number of countries (N) is 12 and periods (T) is 25. ***, **, * signifies the level of significance at 1%, 5% & 10%. Automatic lag length selection based on Schwarz information criteria Source: Author’s computation using Eviews 9 and data from WDI, 2015 development since we found no evidence of cointegration in two out of the three alternative Pedroni cointegration test. Our cointegration results revealed that there exists no long run relationship between the variables above. More specifically, we found out that growth, the stock markets, and bank development do not have a common trend and long run equilibrium. 77 University of Ghana http://ugspace.ug.edu.gh The findings of this study are consistent with Ndako (2010) who also found no cointegration between economic growth, stock markets and bank development in South Africa, using market capitalization ratio as the stock market development proxy. Ndako (2010) however, established cointegration between the variables mentioned above using the turnover ratio and shares traded ratio as measures of stock market development. Ndako (2010) uses bank credit ratio to measures bank development in the three alternative cointegration tests. The findings of our study are also consistent with Acaravci et al. (2009), who established that there is no long run relationship between bank development and economic growth for a panel of twenty-four Sub-Saharan African countries. In this regard, our analysis implies that controlling stock market development does not alter the cointegration results. Enisan & Olufisayo (2009) also found no long run relationship between stock market development and economic growth in Cote D’Ivoire, Zimbabwe, Morocco, Kenya, and Nigeria. Our findings contradict Adebola & Dahalan (2011) and Andabai & Igbodika (2015), who found evidence of cointegration between the stock markets and bank development with economic growth in Nigeria using economic growth as the outcome variable. Ndako (2010), Adebola & Dahalan (2011) and Andabai & Igbodika (2015), are country-level studies for South Africa and Nigeria respectively. Ndako (2010) and Andabai & Igbodika (2015), use the Johansen (1988) cointegration test while Adebola & Dahalan (2011) uses the Pesaran et al. (2001) ARDL cointegration approaches. Both studies employ time series data sets while we use a panel of twelve countries. We use the Pedroni cointegration test appropriate for heterogeneous panels. The difference in methodological approaches and type of data sets could partially explain the differing findings. 78 University of Ghana http://ugspace.ug.edu.gh Theoretically, Patrick (1966) predicted that it is the successful intermediation of savings to investment, which promotes growth. Empirically, Ang & McKibbin (2007) established that financial development affects growth through increased intermediation of savings and efficient allocation of credit to profitable investments. Our findings of no cointegration imply that the perceived linkages between savings, investments and economic growth in the selected SSA sample is not stable. The cost of credit in the selected sample and by extension SSA is high thereby crowding out many small and medium enterprises from accessing credit. We also hold the view that the selected sample has a large informal sector who find it tough to obtain credit. This study therefore, hold the view that these market frictions lead to the breakdown or weakening of the link between savings, investments, and economic growth. This explains why stock market development, bank development, and growth per capita do not have a long run relationship for the selected SSA countries. Jefferis & Mbekeani (2000) argued that the SSA region has experienced fast growth in the number and size of stock markets. Nonetheless, these markets are highly fractured, not liquid and have little technological penetration. Their ability to influence growth is thus highly compromised. The challenges highlighted above limits the economic contribution of these markets to their economies. Farid (2013) observed that a small clique of blue chip companies dominates stock markets in the region. Stock markets in SSA are also small with respect to the size of their respective economies. The Nigerian stock markets makes up 8% of the country’s GNP while Kenya, Ghana, and Zimbabwe account for 25-35% of their respective GNPs. In contrast, Asian and Latin America stock markets account for up to 100% and some almost 200% of their GNPs (Jefferis & Mbekeani, 2000). We believe that the relative small sizes of the stock markets in our sample, 79 University of Ghana http://ugspace.ug.edu.gh relative to the sizes of their respective GNPs is also part of the reason for the inexistence of a long run relationship with growth. Mlachila et al. (2013) also holds the view that, the SSA region lags behind other developing regions regarding financial intermediation and access to banking services. Meager income levels and a weak infrastructure is the cause for the low access to financial services. The absolute size of most banking systems in the region is low. Government securities and liquid assets make up a huge share of all assets controlled by banks. Close to 60% of loans advanced by banks in SSA, mature within a year reflecting the short-term lending nature of these banks. The little competition in the banking sector results from the oligopolistic character of the banks in the region. The region has strikingly high-unbanked population by modern standards. Stock markets in our sample experience similar challenges and features to those highlighted by Jefferis & Mbekeani (2000). The banking sector for our sample too shares similar shortcomings as highlighted by Mlachila et al. (2013) above. We make this inferences based on the comparisons we presented in chapter two. With these challenges and gaps in mind, it is, therefore, not surprising to conclude that there is no long run relationship between stock markets and bank development and growth for our sample. Naceur & Ghazouani (2007) also found out that stock markets and bank development do not have a significant contribution to economic growth in a sample of 10 MENA region countries. They attribute these findings to inadequately developed stock markets and banks in the region. We also believe that stock markets and banks in our sample have not developed sufficiently to have a long run relationship with growth. 80 University of Ghana http://ugspace.ug.edu.gh 5.5 Pedroni Cointegration Tests Using Alternative Measures of Bank Development The third objective of this study was to examine whether using alternative measures of bank development alter the causal connection between the stock markets and bank development with growth. To fulfill this goal, this study conducted additional Pedroni cointegration tests with broad money ratio and domestic credit ratio as the alternative proxies for bank development. As stated earlier, this study have to run the Pedroni cointegration test as a guide to whether we run the causality test in a VAR or VECM model. 5.5.1 Broad Money Ratio We report the results of the Pedroni cointegration test using broad money ratio adjusted to three decimal places, in Table 5.4. As highlighted above, we report three alternatives of the Pedroni cointegration test. We report the Pedroni cointegration analysis including the individual intercept, the individual intercept and individual trend and no intercept or trend in columns two, three and four of Table 5.4. The null hypothesis and decision criteria are as stated earlier. In Table 5.4, this study used the broad money ratio as the proxy for bank development. From column two, the panel-pp, weighted panel-pp and the group-pp statistics are significant at 5%, 10% and 1% significance levels, respectively. Since only, two statistics out of eleven are significant at 5% significance level, this study arrived at the conclusion that there is no cointegration between growth, the stock markets, and bank development, in the case of individual intercept. In column three, the panel-v, weighted panel-v and group-pp statistics are significant at 1% significance level. In the case of individual intercept and individual trend, therefore, only three out of eleven statistics are significant. For that case too, there was no cointegration. The last column of Table 5.4, presents the Pedroni cointegration statistics for the case of no intercept or 81 University of Ghana http://ugspace.ug.edu.gh trend. In that case, the panel-pp, panel-ADF, weighted panel-pp, panel-ADF, group-pp and group-ADF statistics are significant at 1% significance level. A majority of the statistics are significant, hence for the case of no intercept and trend, this study concludes that there is cointegration. Table 5.4: Pedroni Heterogeneous Panel Cointegration Tests (Broad Money Ratio) Individual intercept Individual intercept No intercept and individual trend or trend Panel Statistic Panel v-Statistic -2.790 4.793*** -4.734 Panel rho-Statistic 2.158 4.133 1.712 Panel PP-Statistic -1.645** 1.325 -2.739*** Panel ADF-Statistic -0.863 1.235 -2.897*** Weighted Statistic Panel v-Statistic -2.115 2.959*** -4.764 Panel rho-Statistic 2.355 3.778 1.657 Panel PP-Statistic -1.335* -0.543 -3.131*** Panel ADF-Statistic -0.632 -0.369 -3.396*** Group Statistic Group rho-Statistic 4.145 5.399 2.881 Group PP-Statistic -2.372*** -2.627*** -4.448*** Group ADF-Statistic -0.258 -0.509 -4.207*** SIC lag length 1 to 4 0 to 3 1 to 4 Notes: Number of countries (N) is 12 and periods (T) is 25. ***, **, * signifies the level of significance at 1%, 5% & 10%. Automatic lag length selection based on Schwarz information criteria Source: Author’s computation using Eviews 9 and data from WDI, 2015 In a majority i.e., two out of three alternative Pedroni cointegration tests, this study finds no evidence of cointegration using the broad money ratio to proxy bank development. Based on these findings, this study concluded that stock market development, bank development and growth do not move together in the long run using the broad money ratio to proxy bank development. 82 University of Ghana http://ugspace.ug.edu.gh 5.5.2 Domestic Credit Ratio Table 5.5: Pedroni Heterogeneous Panel Cointegration Tests (Domestic Credit Ratio) Individual Individual intercept No intercept intercept and individual trend or trend Panel Statistic Panel v-Statistic -2.082 3.344*** -4.742 Panel rho-Statistic 2.026 4.254 1.845 Panel PP-Statistic -1.732** 1.426 -4.593*** Panel ADF-Statistic -1.930** 1.187 -4.316 Weighted Statistic Panel v-Statistic -1.503 3.145*** -4.759 Panel rho-Statistic 2.445 4.134 1.972 Panel PP-Statistic -0.659 0.704 -3.626*** Panel ADF-Statistic -0.789 0.292 -3.391*** Group Statistic Group rho-Statistic 4.157 5.418 3.209 Group PP-Statistic -0.089 -0.755 -4.683*** Group ADF-Statistic -0.301 -0.478 -3.727*** SIC lag length 1 to 4 0 to 3 1 to 4 Notes: Number of countries (N) is 12 and periods (T) is 25. ***, **, * signifies the level of significance at 1%, 5% & 10%. Automatic lag length selection based on Schwarz information criteria Source: Author’s computation using Eviews 9 and data from WDI, 2015 We present the results of the Pedroni cointegration tests using the domestic credit ratio as the measure of bank development in Table 5.5. In Table 5.5, in the case of individual intercept, two out of eleven statistics are significant, i.e., panel-pp and panel-ADF at 5% significance level. Given that, a majority of the statistics are not significant this study accepted the null hypothesis of no cointegration. The results remained the same when we included the individual intercept and individual trend i.e. the panel-v and weighted panel-v are significant at 1% significance levels. The results for the case of no intercept or no trend are the same for the three 83 University of Ghana http://ugspace.ug.edu.gh measures of bank development that are the bank credit ratio, the broad money ratio, and the domestic credit ratio, except that the panel-ADF statistics is not significant in the last case. Further, this study found no evidence of cointegration between growth, the stock markets and bank development irrespective of the proxy used for bank development. Based on the Pedroni cointegration test results discussed above, this study concluded that the measure of bank development did not alter the cointegration results. The proxy for bank development notwithstanding, the results show no existence of a long run relationship between stock markets and bank development with growth. In this regard, this study proceeds to estimate the causal relationship between stock markets and bank development and growth per capita in a time stationary VAR framework. 5.6 Correlations One of the objectives of our study was to examine the causation between stock markets and bank development with growth per capita. We, however, present the results of correlation analysis between growth per capita, the stock markets and bank development and the control variables adjusted to three decimal places, in Table 5.6 then move on and present the causality results thereafter. From Table 5.6, the correlation between growth per capita and bank development is low but positive for all the proxies of bank development except the broad money ratio, which is considerably higher. Bank credit ratio (bank), our main measure of bank development has the second lowest correlation coefficient (0.071) with growth per capita (dlgdp) among the three alternative measures of bank development. The change in broad money ratio (dbmr) has a correlation coefficient of (0.659) while the domestic credit ratio (dmr) has a correlation coefficient of (0.052), with growth per capita. 84 University of Ghana http://ugspace.ug.edu.gh Table 5.6: Correlations dlgdp bank dbmr dmr tr gfc dgov inf opn dlgdp 1.000 Bank 0.071 1.000 dbmr 0.659 0.145 1.000 dmr 0.052 0.900 0.120 1.000 tr -0.005 0.491 0.046 0.721 1.000 gfc 0.124 0.006 0.086 -0.079 -0.193 1.000 dgov 0.421 0.069 0.470 0.063 0.044 0.071 1.000 inf -0.006 -0.011 -0.028 -0.094 -0.203 -0.097 0.039 1.000 opn 0.082 0.231 0.098 0.012 -0.208 0.167 -0.011 0.064 1.000 Source: Author’s computation using STATA 13 and data from WDI, 2015 Stock market development, measured by the turnover ratio (tr) has a very low and almost negligible negative correlation with growth per capita, the correlation coefficient is (-0.005). We also observe a positive correlation between physical capital (gfc) and change government expenditure (dgov) with growth per capita. Economic theory predicts that inflation hinders economic growth. The negative correlation between growth per capita and inflation (inf) as shown in Table 5.6 appears to confirm this prediction. Openness to trade (opn) also has a positive correlation with growth per capita in our sample 5.7 Panel Causality Results The second objective of this study was to assess causality between growth per capita, the stock markets and bank development in the selected SSA countries. To achieve this aim, we estimated equation 16, 17, 18 and 19 using the system GMM. We report the estimates of these 85 University of Ghana http://ugspace.ug.edu.gh equations in Table 5.7. Both the stationarity and the panel cointegration tests conducted above directed us to analyze the causal link in a time stationary VAR approach. The conclusion of no cointegration meant that panel causality analysis using a VECM was not feasible. This study therefore used the log of real GDP per capita, the broad money ratio and government expenditure in the first difference to remove unit roots. The new series, therefore, are the growth per capita, the change in broad money ratio and change in government spending. The bank credit ratio and the turnover ratio together with the other control variables, i.e. physical capital, inflation and openness to trade were stationary in levels hence they entered the VAR regressions in levels. This study used the Schwarz information criteria to select the number of lags to be included in the VAR regressions. Based on the SIC criteria, this study used two lags of the outcome variable and the other predictor variables mentioned above for the level equations 16 & 17 and the difference equations 18 & 19, in chapter four, as a system. We specified economic growth, stock market development and bank development as the outcome variables since the direction of causality could be either way. To analyse the sensitivity of the causal relationship to a change of the bank development proxy, we used three alternative measures of bank development as mentioned earlier. In this regard, we had nine specifications, three for each measure of bank development. To ensure correct specification of our time stationary VAR models, we carried out two specification tests. These tests are the Sargan test of overidentifying restrictions and the Arellano-Bond test for first and second order autocorrelation. The xtabond2 command we used to obtain the system GMM estimators written for STATA by Roodman (2006) implements the Sargan and Arellano-Bond test and reports them automatically. 86 University of Ghana http://ugspace.ug.edu.gh Both the difference and system GMM use lags of the endogenous variables as internal instruments. However, the estimation of the difference and system GMM estimators using prominent software can result in questionably too many instruments (Roodman, 2009). Roodman (2009) argued that a researcher should mainly be concerned when the results contain an equal number of instruments as the number of observations. Too many instruments may end up overfitting the endogenous variables in the model. Roodman (2009) proposed collapsing some instruments by limiting the number of lags of the endogenous variables to be used as internal instruments. This study followed his advice and collapsed instruments where a proliferation of the same was suspected to have occurred. This study carried out four Wald tests. First, Wald test statistic one, Wald1 test statistic is a test of the overall significance of the model. The null hypothesis of this test is that all parameters in the model are jointly equal to zero. In that case, the model would not be appropriate for our analysis. The desirable findings is the alternative hypothesis that states that all parameters are jointly significantly different from zero. Secondly, Wald2 test statistic is a test of the null hypothesis that stock market development does not Granger-cause growth per capita or bank development respectively. Thirdly, Wald3 test statistic tests the null hypothesis that bank development does not have a causal effect on growth per capita and stock market development respectively. Lastly, Wald4 test statistic tests the null hypothesis that growth per capita does not Granger-cause stock markets and bank development respectively. All the above Wald test statistics follow a chi- squared distribution with (k-m) degrees of freedom. This study presents the results of the causality test in Table 5.7, 5.8 and 5.9. 87 University of Ghana http://ugspace.ug.edu.gh Table 5.7: Panel VAR Models. One Step System GMM results 𝒗𝒂𝒓𝒊𝒂𝒃𝒍𝒆𝒔 1 2 3 𝒅𝒍𝒈𝒅𝒑 𝒕𝒖𝒓𝒏𝒐𝒗𝒆𝒓 𝒃𝒂𝒏𝒌 𝒅𝒍𝒈𝒅𝒑 0.113 5.893 20.11 −𝟏 (0.092) (7.268) (15.40) 𝒅𝒍𝒈𝒅𝒑 -0.007 0.695 -11.95 −𝟐 (0.020) (5.828) (13.85) 𝒃𝒂𝒏𝒌 0.001** 0.031 1.609*** −𝟏 (0.000) (0.113) (0.326) 𝒃𝒂𝒏𝒌 -0.001** 0.036 -0.629* −𝟐 (0.000) (0.113) (0.329) 𝒕𝒖𝒓𝒏𝒐𝒗𝒆𝒓 -0.000** 0.624*** 0.153 −𝟏 (0.000) (0.087) (0.099) 𝒕𝒖𝒓𝒏𝒐𝒗𝒆𝒓 0.000** 0.275*** -0.116 −𝟐 (0.000) (0.092) (0.135) 𝒈𝒇𝒄 0.001* -0.040 0.856* −𝟏 (0.001) (0.084) (0.444) 𝒈𝒇𝒄 -0.001 -0.055 -0.919** −𝟐 (0.001) (0.055) (0.451) 𝒅𝒈𝒐𝒗𝒕 -0.002** -0.155 -0.483** −𝟏 (0.001) (0.213) (0.208) 𝒅𝒈𝒐𝒗𝒕 0.001 -0.039 -0.088 −𝟐 (0.001) (0.069) (0.109) 𝒊𝒏𝒇𝒍𝒂𝒕𝒊𝒐𝒏 -0.000 -0.085** -0.062 −𝟏 (0.000) (0.033) (0.132) 𝒊𝒏𝒇𝒍𝒂𝒕𝒊𝒐𝒏 0.000 0.064** 0.080 −𝟐 (0.000) (0.027) (0.163) 𝒐𝒑𝒆𝒏𝒏𝒆𝒔𝒔 -0.000 0.014 -0.094 −𝟏 (0.000) (0.030) (0.061) 𝒐𝒑𝒆𝒏𝒏𝒆𝒔𝒔 -0.000 -0.058** 0.151** −𝟐 (0.000) (0.027) (0.062) 𝑪𝒐𝒏𝒔𝒕𝒂𝒏𝒕 0.021 5.082*** -3.414 (0.014) (1.105) (2.456) 88 University of Ghana http://ugspace.ug.edu.gh Table 5.7 Cont’d 𝑾𝒂𝒍𝒅𝟏 563.55 *** 61524.18 *** 885.07 *** 𝑾𝒂𝒍𝒅𝟐 6.15** - 3.95 𝑾𝒂𝒍𝒅𝟑 6.36** 3.23 - 𝑾𝒂𝒍𝒅𝟒 - 0.67 2.80 𝑺𝒂𝒓𝒈𝒂𝒏 𝑻𝒆𝒔𝒕 97.02 131.93 7.19 𝑨𝑹(𝟐) 𝑻𝒆𝒔𝒕 -0.32 -1.44 1.04 𝑪𝒐𝒏𝒕𝒓𝒐𝒍𝒍𝒆𝒅 𝒇𝒐𝒓 𝑻𝒊𝒎𝒆 No No No 𝑶𝒃𝒔𝒆𝒓𝒗𝒂𝒕𝒊𝒐𝒏𝒔 210 183 202 𝑵𝒐 𝒐𝒇 𝑰𝒏𝒔𝒕𝒓𝒖𝒎𝒆𝒕𝒔 103 129 21 𝑵𝒖𝒎𝒃𝒆𝒓 𝒐𝒇 𝑪𝒐𝒖𝒏𝒕𝒓𝒚 12 12 12 Notes: Robust standard errors in parentheses. ***, **, * signifies the level of significance at 1%, 5% & 10%. Wald1, Wald2, Wald3, and Wald4 represent the joint significance of the whole model, lags of turnover ratio, bank credit ratio and growth per capita respectively. dlgdp and turnover measures economic growth and stock market development respectively. Source: Author’s computation using STATA 13 and data from WDI, 2015 Beginning with table 5.7, specification one, two and three gives the causality test with the growth per capita, the stock market development (turnover ratio) and the bank development (bank credit ratio) as the outcome variables. Wald1 test statistic is significant at the 1% level of significance in column two, three and four. In this regard, this study rejected the null hypothesis that all the parameters in the model are insignificantly different from zero for the three specifications. This means that our three VAR specification results in table 5.7 for causality analysis are good. The Sargan test statistics is not significant at any level of significance. Based on this, we fail to reject the null hypothesis that the overidentifying restrictions are valid. We, therefore, conclude the overidentifying restrictions are valid for the three specifications. As proposed by Roodman (2009), this study also reported the number of 89 University of Ghana http://ugspace.ug.edu.gh instruments for all our specifications. Even though having lower instruments is preferred, there is no general rule as to the specific number of appropriate instruments. Nonetheless, the number of instruments is lower than the number of observations for all our specifications. In all the nine specifications, this study collapsed some instruments since they were equal to or sometimes higher than the number of observations. None of the Arellano-Bond test statistics too is significantly different from zero in the three specifications. As a result, this study concluded that the three specifications do not suffer from second-order autocorrelation. Having concluded that our VAR models are well specified, this study proceed to discuss the causal inferences from specification one, two and three in table 5.7 with bank credit ratio as the measure of bank development. From specification one in table 5.7, Wald2 test statistic is significant at the 5% level of significance. This implies that stock market development Granger-causes growth per capita. This study therefore rejects the null hypothesis that there is no causality running from stock market development to economic growth in the selected SSA countries in favor of the alternative. From specification two, Wald4 test statistic is not significant at any conventional level of significance. Growth per capita therefore does not Granger-causes stock market development. Regarding the causal relationship between stock market development and growth per capita therefore, this study found evidence of a supply-leading finance in the selected SSA countries. This finding is consistent with Yang & Yi (2008) and Adebola & Dahalan (2011) who found support for the supply-leading finance. Wald3 test statistic is significant at 5% significance level. This study therefore, found evidence that bank development Granger-causes growth per capita for the selected SSA countries. From the specification three, Wald4 test statistic is not significant at any conventional level of significance. This means that there is no reverse causation from growth per capita to bank 90 University of Ghana http://ugspace.ug.edu.gh development for the selected SSA countries. In sum, concerning bank development and growth per capita, this study found evidence in support of the supply-leading finance. This finding is consistent with Yang & Yi (2008) and Adebola & Dahalan (2011) who found support for the supply-leading finance. The findings of this study reinforce the theoretical predictions by Patrick (1966) that at earlier stages of economic development, financial development leads economic growth. Given that the countries in our sample are at earlier stages of economic development mostly with huge contribution to their GDPs coming from Agriculture, the findings mirror the prediction of Patrick (1966) stages of development hypothesis. Financial development in our study comprises of stock markets and bank development. From specification two, Wald3 test statistic is not significantly different from zero. Based on this finding we conclude that bank development does not Granger-causes stock market development. Wald2 in specification three is not significant at any conventional level of significance. This shows that stock market development does not Granger-causes bank development. In sum, we find no evidence of causality in any direction between bank development and stock market development using the bank credit ratio to measure bank development. 5.8 Causality Analysis Using Alternative Measures of Bank Development 5.8.1 Broad Money Ratio To fulfill our third objective, this study proceeded to estimate the causal relationship between stock markets and bank development with growth per capita using two alternative measures for bank development. 91 University of Ghana http://ugspace.ug.edu.gh Table 5.8: Panel VAR Models. One Step System GMM results 𝒗𝒂𝒓𝒊𝒂𝒃𝒍𝒆𝒔 1 2 3 𝒅𝒍𝒈𝒅𝒑 𝒕𝒖𝒓𝒏𝒐𝒗𝒆𝒓 𝒅𝒃𝒎𝒓 𝒅𝒍𝒈𝒅𝒑 0.171* 13.52* 1.108 −𝟏 (0.089) (7.69) (18.14) 𝒅𝒍𝒈𝒅𝒑 -0.002 0.267 6.047 −𝟐 (0.014) (0.802) (13.44) 𝒅𝒃𝒎𝒓 0.001** 0.005 0.660* −𝟏 (0.000) (0.085) (0.394) 𝒅𝒃𝒎𝒓−𝟐 -0.000 -0.035 -0.397 (0.001) (0.034) (0.317) 𝒕𝒖𝒓𝒏𝒐𝒗𝒆𝒓 -0.000** 0.515*** 0.032 −𝟏 (0.000) (0.082) (0.094) 𝒕𝒖𝒓𝒏𝒐𝒗𝒆𝒓 0.000 0.361*** -0.103 −𝟐 (0.000) (0.072) (0.136) 𝒈𝒇𝒄 0.001*** 0.001 -0.289* −𝟏 (0.000) (0.058) (0.158) 𝒈𝒇𝒄 -0.001 -0.052 0.101 −𝟐 (0.001) (0.057) (0.166) 𝒅𝒈𝒐𝒗𝒕 -0.002** -0.250 0.522 −𝟏 (0.001) (0.169) (0.817) 𝒅𝒈𝒐𝒗𝒕 0.001 -0.051 -0.177 −𝟐 (0.001) (0.053) (0.865) 𝒊𝒏𝒇𝒍𝒂𝒕𝒊𝒐𝒏 -0.000*** -0.064*** -0.134 −𝟏 (0.000) (0.022) (0.133) 𝒊𝒏𝒇𝒍𝒂𝒕𝒊𝒐𝒏 0.000194 0.0276 0.148 −𝟐 (0.000) (0.023) (0.153) 𝒐𝒑𝒆𝒏𝒏𝒆𝒔𝒔 -0.000 0.030 -0.056 −𝟏 (0.000) (0.035) (0.046) 𝒐𝒑𝒆𝒏𝒏𝒆𝒔𝒔 0.000 -0.021 0.034 −𝟐 (0.000) (0.029) (0.049) 𝑪𝒐𝒏𝒔𝒕𝒂𝒏𝒕 0.028* 1.942 6.557** (0.015) (2.133) (3.028) 92 University of Ghana http://ugspace.ug.edu.gh Table 5.8 Cont’d 𝑾𝒂𝒍𝒅𝟏 268.86 *** 1221.05 *** 468.70 *** 𝑾𝒂𝒍𝒅𝟐 8.66** - 1.80 𝑾𝒂𝒍𝒅𝟑 4.35 1.14 - 𝑾𝒂𝒍𝒅𝟒 - 11.51*** 0.23 𝑺𝒂𝒓𝒈𝒂𝒏 𝑻𝒆𝒔𝒕 126.07 140.34 3.81 𝑨𝑹(𝟐) 𝑻𝒆𝒔𝒕 0.27 -1.92* 1.62 𝑪𝒐𝒏𝒕𝒓𝒐𝒍𝒍𝒆𝒅 𝒇𝒐𝒓 𝑻𝒊𝒎𝒆 No No No 𝑶𝒃𝒔𝒆𝒓𝒗𝒂𝒕𝒊𝒐𝒏𝒔 211 206 189 𝑵𝒐 𝒐𝒇 𝑰𝒏𝒔𝒕𝒓𝒖𝒎𝒆𝒕𝒔 127 150 19 𝑵𝒖𝒎𝒃𝒆𝒓 𝒐𝒇 𝑪𝒐𝒖𝒏𝒕𝒓𝒚 12 12 12 Notes: Robust standard errors in parentheses. ***, **, * signifies the level of significance at 1%, 5% & 10%. Wald1, Wald2, Wald3, and Wald4 represent the joint significance of the whole model and causality. dbmr change in broad money ratio respectively Source: Author’s computation using STATA 13 and data from WDI, 2015 The aim here was to see whether the choice of the bank development proxy affects the causal connection. First, we used the broad money ratio as the first alternative measure of bank development. We used two lags of each predictor variable selected using the Schwarz information criteria. Table 5.8 presents the results of the causality tests using the broad money ratio as the alternative measure of bank development. The Sargan and Arellano-Bond test statistics are not significantly different from zero based on the 5% significance level. Wald1 for joint significance of the overall models is also highly significant at 1% level of significance for all the three specifications. This means our three VAR models in specifications four, five and six are well specified. Turning our attention to the causality results based on Wald test statistics, we notice some similarities and differences from the earlier causality results. 93 University of Ghana http://ugspace.ug.edu.gh Concerning the causal linkage between stock market development and growth, Wald2 in specification four is significant at 5% significance level. This finding implies that stock market development Granger-causes growth per capita. Wald4 in specification five is significant at the 1% significance level. Growth per capita therefore, Granger-causes stock market development at 1% significance level. These results reveal that using the broad money ratio as the measure of bank development, we find support for a bi-directional causality in the case of stock market development and growth per capita. Regarding bank development and growth per capita, Wald3 test statistic in specification four is not significant at any significance level. This shows that banking industry development does not Granger-causes growth per capita. Wald4 in specification six, on the other hand, is not significant meaning growth per capita does not Granger-cause bank development. Using broad money ratio to measure bank development, this study finds no support for a causality in any direction between bank development and growth per capita. This finding is different from the earlier one where we used bank credit ratio as the proxy for bank development. The use of the broad money ratio to measure bank development alters the causality between bank development and growth per capita in the selected SSA countries. Using bank credit ratio, we found no evidence of causal linkage between bank development and stock market development. The findings using broad money ratio to measure bank development are similar. In this regard, we report no causality between bank development and stock market development. The insignificant Wald3 test statistic in specification five and Wald2 test statistic in specification six, informs this conclusion. In short, the use of broad money ratio as an alternative bank development proxy does not alter the causal relationship between stock market development and bank development. 94 University of Ghana http://ugspace.ug.edu.gh 5.8.2 Domestic Credit Ratio We now turned our focus to the causal relations using domestic credit ratio as the alternative proxy for bank development. Table 5.9 presents the causality results using domestic credit ratio as the measure of bank development. Once again, Wald1 test statistic is highly significant for specifications seven, eight and nine, meaning that the parameters in these respective specifications jointly are not equal to zero. No Sargan test statistics are significant in the three specifications. The overidentifying restrictions are therefore valid in all the three specifications. The Arellano-Bond test statistics are also not significant in any of the three specifications. This study therefore, cannot reject the null hypothesis of no second order autocorrelation for all three specifications. From specification nine, Wald2 test statistic is not significant at any conventional significance level. The insignificant Wald2 test statistic suggests no causality running from of stock market development to growth per capita. Wald4 test statistic in specification eight is not significant signifying that growth per capita does not Granger-cause stock market development in the sample. Regarding growth per capita and bank development, using the domestic credit ratio to measure bank development, this study finds no eveidence of causation in any direction.The results are different from the earlier ones where we found evidence supporting the supply- leading hypothesis for the case of bank credit ratio and bi-directional causality for the case of broad money ratio. 95 University of Ghana http://ugspace.ug.edu.gh Table 5.9: Panel VAR Models. One Step System GMM results 𝒗𝒂𝒓𝒊𝒂𝒃𝒍𝒆𝒔 1 2 3 𝒅𝒍𝒈𝒅𝒑 𝒕𝒖𝒓𝒏𝒐𝒗𝒆𝒓 𝒅𝒐𝒎𝒆𝒔𝒕𝒊𝒄 𝒅𝒍𝒈𝒅𝒑 0.123 6.087 10.22 −𝟏 (0.092) (6.473) (10.60) 𝒅𝒍𝒈𝒅𝒑 -0.004 1.637 -0.982 −𝟐 (0.014) (6.717) (0.617) 𝒅𝒐𝒎𝒆𝒔𝒕𝒊𝒄 0.001*** 0.0417 0.903*** −𝟏 (0.000) (0.046) (0.080) 𝒅𝒐𝒎𝒆𝒔𝒕𝒊𝒄 -0.001*** 0.026 0.046 −𝟐 (0.000) (0.050) (0.087) 𝒕𝒖𝒓𝒏𝒐𝒗𝒆𝒓 -0.000 0.659*** 0.154 −𝟏 (0.000) (0.100) (0.130) 𝒕𝒖𝒓𝒏𝒐𝒗𝒆𝒓 0.000* 0.146 -0.022 −𝟐 (0.000) (0.106) (0.124) 𝒈𝒇𝒄 0.001 -0.049 0.147 −𝟏 (0.001) (0.080) (0.228) 𝒈𝒇𝒄 -0.001 -0.075 -0.144 −𝟐 (0.001) (0.068) (0.289) 𝒅𝒈𝒐𝒗𝒕 -0.002** -0.294 -0.088 −𝟏 (0.001) (0.194) (0.080) 𝒅𝒈𝒐𝒗𝒕 0.001 -0.029 -0.091 −𝟐 (0.001) (0.062) (0.078) 𝒊𝒏𝒇𝒍𝒂𝒕𝒊𝒐𝒏 -0.000 -0.083*** -0.047 −𝟏 (0.000) (0.023) (0.035) 𝒊𝒏𝒇𝒍𝒂𝒕𝒊𝒐𝒏 0.000 0.064*** 0.029 −𝟐 (0.000) (0.023) (0.045) 𝒐𝒑𝒆𝒏𝒏𝒆𝒔𝒔 -0.000 -0.000 0.025 −𝟏 (0.000) (0.032) (0.046) 𝒐𝒑𝒆𝒏𝒏𝒆𝒔𝒔 0.000 -0.044* -0.006 −𝟐 (0.000) (0.025) (0.043) 𝑪𝒐𝒏𝒔𝒕𝒂𝒏𝒕 0.013 5.894*** 0.052 (0.011) (0.992) (3.275) 96 University of Ghana http://ugspace.ug.edu.gh Table 5.9 Cont’d 𝑾𝒂𝒍𝒅𝟏 576.40*** 32605.08 *** 13747.98 *** 𝑾𝒂𝒍𝒅𝟐 3.79 - 14.30*** 𝑾𝒂𝒍𝒅𝟑 17.60*** 19.50*** - 𝑾𝒂𝒍𝒅𝟒 - 0.88 4.02 𝑺𝒂𝒓𝒈𝒂𝒏 𝑻𝒆𝒔𝒕 100.55 134.36 71.99 𝑨𝑹(𝟐) 𝑻𝒆𝒔𝒕 -0.59 -0.24 -0.64 𝑪𝒐𝒏𝒕𝒓𝒐𝒍𝒍𝒆𝒅 𝒇𝒐𝒓 𝑻𝒊𝒎𝒆 No No No 𝑶𝒃𝒔𝒆𝒓𝒗𝒂𝒕𝒊𝒐𝒏𝒔 210 171 210 𝑵𝒐 𝒐𝒇 𝑰𝒏𝒔𝒕𝒓𝒖𝒎𝒆𝒕𝒔 103 130 79 𝑵𝒖𝒎𝒃𝒆𝒓 𝒐𝒇 𝑪𝒐𝒖𝒏𝒕𝒓𝒚 12 12 12 Notes: Robust standard errors in parentheses. ***, **, * signifies the level of significance at 1%, 5% & 10%. Wald1, Wald2, Wald3, and Wald4 represent the joint significance of the whole model and causality. Domestic measures the domestic credit ratio Source: Author’s computation using STATA 13 and data from WDI, 2015 Regarding bank development and growth per capita, Wald3 test statistic in specification seven is significant at 1% significance level. Banking industry development therefore Granger-causes growth per capita in the sample. Wald4 test statistic in specification nine is not significant. This study therefore finds support for the supply-leading hypothesis for the case of bank development measured by the domestic credit ratio and growth per capita. The use of alternative measures of bank development therefore, alters the causal linkage between bank development and growth in the selected SSA countries. On the causal relationship between bank development and stock market development, we found support for a bi-idirectional causal relationship. The significant Wald3 and Wald2 test statistic at 1% significance level in specification eight and nine attest to this. Bank development and 97 University of Ghana http://ugspace.ug.edu.gh stock market development Granger-cause each other using domestic credit ratio to measure bank development. In summary therefore, this study concluded that the causal relationship between stock markets and bank development with growth per capita is sensitive to the choice of bank development proxy. These findings are consistent with Acaravci et al. (2009), Agbetsiafa, (2004) and Quartey & Prah (2008) who found that the causal linkage between economic growth and bank development depends on the choice of bank development proxy. We intended to perform similar analysis using alternative measures of stock market development namely shares traded ratio and market capitalization ratio. However, lack of continuous data for the shares traded ratio and the market capitalization ratio made that attempt impossible. 5.9 Other Diagnostic Tests 5.9.1 Endogeneity Test This study performed the Hausman test for endogeneity for all the regressors in our nine causality specifications. The null hypothesis for this test is that the variables are exogenous. The alternative hypothesis is that the variables are endogenous. Appendix IV presents the p- values of the Hausman endogeneity test. From the results presented in Appendix IV, bank credit ratio, turnover ratio, government expenditure and openness to trade were endogenous in specification one. In the same specification, physical capital and inflation were exogenous. In specification one, growth per capita was the outcome variable. With regard to specification two, log of GDP per capita, bank credit ratio, physical capital, inflation and openness to trade were endogenous. On the other hand, government expenditure was exogenous. Specification 2 had the turnover ratio as its outcome variable. All regressors in specification three except the physical capital and inflation were endogenous. The bank 98 University of Ghana http://ugspace.ug.edu.gh credit ratio was the dependent variable in specification three. Concerning specification four, all predictor variables except inflation were endogenous. Specification 4 had growth per capita as the outcome variable. Regarding specification five, no regressor was exogenous. The turnover ratio was the dependent variable in specification five. The only variables that were exogenous in specification six are inflation and government expenditure. The rest of the regressors were endogenous. The dependent variable in specification six was change in broad money ratio. Concerning specification seven, the turnover ratio, domestic credit ratio, government expenditure and openness to trade were endogenous, whereas the physical capital and inflation were exogenous. The outcome variable in specification seven was growth per capita. Concerning specification eight, log of GDP per capita, the domestic credit ratio, physical capital, inflation and openness to trade were endogenous. The exogenous regressor in specification eight was government expenditure. The turnover ratio was the outcome variable in specification eight. The outcome variable in specification nine was the domestic credit ratio. In that specification, the log of GDP per capita, turnover ratio and government expenditure were endogenous variables. The exogenous variables in that regression were physical capital, inflation and openness to trade. Based on these endogeneity results, this study specified the endogenous and exogenous variables in the system GMM regressions appropriately. 5.9.2 The Heteroskedasticity Tests This study relied on the one-step system GMM results to carry out the causality analysis whose results have been discussed above. According to Beck & Levine (2004), the one-step system GMM assumes homoscedastic standard errors. The presence of heteroskedasticity therefore may bias the one-step results. In this regard, we conducted the Breusch-Pagan/Cook Weisberg 99 University of Ghana http://ugspace.ug.edu.gh test for heteroskedasticity. We present the results of this test in Appendix V. Based on the Breusch-Pagan/Cook Weisberg test for heteroskedasticity we conducted; we detected heteroskedasticity in all the nine specifications. We used the robust command to obtain heteroskedasticity consistent standard errors in those specifications. 5.9.3 Time Effects, Breusch & Pagan Test and Hausman Specification Tests. Based on the time effects tests we conducted on the nine causality specifications, time effects were not significant in any of the nine specifications. In this regards, this study did not control for time in any of the specifications. About the Breusch & Pagan Lagrangian Multiplier test for random effects, this study found that the pooled model was appropriate for specifications one, four, six and seven. The ols estimator could thus be used to estimate specifications one, four, six and seven. However, the presence of endogenous variables in those specifications as mentioned above makes the ols estimator inconsistent. The next available consistent estimator of these specifications is the IV estimator. However, the presence of heteroskedasticity in these specifications as mentioned above makes the IV estimator inefficient. Given these circumstances, this study adopted the GMM estimator that is efficient in the presence of heteroskedasticity. The Breusch & Pagan Lagrangian Multiplier test for random effects also revealed that the REM was appropriate for specifications two, three, five, eight and nine. This study had to run the hausman specification test to determine the appropriate model between the REM and the FEM. The results of the Hausman specification test supported the adoption of the FEM in all these five specifications. This means that the GMM estimator, also robust to fixed effects was appropriate for all our nine causality specifications. In this regards, we used the GMM 100 University of Ghana http://ugspace.ug.edu.gh estimator for all our nine causality specifications above. Appendix VI contains the time effects, Breusch & Pagan and Hausman specification tests results. 5.10 Conclusions. This chapter presented and discussed the analysis and findings of this study. This study began by presenting the summary statistics of the variables involved, followed by the panel unit root tests. We also presented causality tests and other diagnostic tests in this chapter. The results of the Choi (2001) and IPS (2003) unit root tests show that log of GDP per capita, broad money ratio and government spending are non-stationary in levels but stationary in their first difference. Bank credit ratio, domestic credit ratio, turnover ratio, physical capital, inflation and openness to trade are all stationary in levels. Thereafter, we presented the findings of the Pedroni cointegration tests. The Pedroni cointegration results revealed that there is no evidence of a long run relationship between stock market development, bank development and economic growth using either the bank credit ratio, the broad money ratio and the domestic credit ratio to measure bank development. Lastly, we presented the causality findings using three alternative measures of bank development. For our second objective, using bank credit ratio, we established evidence in support of the supply-leading finance in the case of stock market development, bank development and growth per capita. The results also revealed no causal relationship between bank development and stock market development. In the case of the broad money ratio as the proxy for bank development, we found evidence supporting a bi-directional causality in the case stock market development and growth per capita. For bank development and growth, however, there is no evidence of causality in any direction. This study also discovered that there exists no causality between bank development and stock market development. 101 University of Ghana http://ugspace.ug.edu.gh This study also used the domestic credit ratio as an alternative measure of bank development. In this case, this study found that, there exist no causality between stock market development and growth per capita in the selected SSA countries. Further, this study found support the supply-leading hypothesis in the case of bank development and growth per capita. This study also established that there exists a bi-directional causal flow between bank development and stock market development. Regarding the third objective of this study therefore, we roprted that the causal linkage between stock market development, bank development and growth per capita is sensitive to the choice of the proxy for bank development. The following chapter presents the summary, conclusions and policy recommendations of our study. 102 University of Ghana http://ugspace.ug.edu.gh CHAPTER SIX SUMMARY, CONLUSION AND POLICY RECOMEDATIONS 6.1 Introduction The present chapter presents the summary and conclusions of this study, policy recommendations, and limitations of the study and potential areas for future research. The current chapter comprises of four sections. The present section introduces the chapter. Section 6.2 presents the summary and conclusions of the study while the subsequent section gives policy recommendations. We present limitations to our study and suggest areas for further research in section 6.4. 6.2 Summary and Conclusions Policy makers, governments, multilateral and bilateral lenders, donors and civil society organizations have continuously been confronted with the challenges affecting growth in SSA. Research have proposed functioning stock markets and banks as one way of increasing the pace of economic growth. In this regard, many economist have encouraged the SSA region to adopt further development of stock markets and banks as a way of promoting growth (Ndebbio, 2004; Agbetsiafa 2004; among others). The comparative analysis we presented in chapter two, revealed that our sample of the selected SSA countries indeed needs further development of stock markets and banks since they lag behind the SSA region. This study contributes to literature, especially by considering the interaction of stock market development and bank development and their causal effects on growth per capita in the SSA context. This is important, since literature as we have highlighted in chapter three, has not paid much attention to this interaction. The primary objective of this study was to examine the causality between stock market development, bank development and growth per capita in the 103 University of Ghana http://ugspace.ug.edu.gh selected SSA countries. Our sample had 13 selected SSA countries but we took out Zimbabwe, which had an outrageous mean inflation of 46.5 million percent. An examination of the existence or lack thereof of a long run relationship between stock markets and bank development with growth per capita in selected SSA countries preceded the causality analysis. Lastly, we examined the sensitivity of the causal relationship to the change of the measure of bank development. Available empirical evidence presented in chapter three reveals mixed findings on the long run and causal relationship between stock markets and bank development with economic growth. As mentioned earlier, in the empirical review, much attention has gone to bank development and economic growth. There has been little attention devoted to economic growth and stock market development. The causal relationship between stock market development, bank development and economic growth has received the least attention. While some studies support the supply-leading hypothesis, others favour the demand-following hypothesis. In addition, some studies have reported a two-way causal flow while some have found no causal relationship at all. These mixed findings and the existence of limited literature on the causal relationship between stock market development, bank development and economic growth inspired our study. This study empirically estimated a dynamic panel growth model inspired by the AK endogenous growth model. The study adopted the Pedroni cointegration test to analyse the long run relationship between stock markets and bank development with growth. This study chose the Pedroni cointegration test since it accounts for endogeneity and allows for heterogeneity among the countries. On causality, this study used the system GMM estimator that also addresses shortcomings of the OLS, IV and traditional random and fixed effects estimators. In this regard, our methodology, anchored in literature involved the Pedroni cointegration tests 104 University of Ghana http://ugspace.ug.edu.gh and estimation of causality using the system GMM estimator in a time stationary VAR framework. The first objective of this study was to examine the existence of a long run relationship between stock markets and bank development with growth. The findings of this study do not support the existence of a long run relationship between stock markets and bank development with growth in the selected SSA countries, irrespective of the bank development proxy used. The second objective was on causality, and this study found evidence supporting the supply-leading hypothesis between stock market development, bank development and growth per capita. In addition, this study found no support for causality between bank development and stock market development. The bank credit ratio was used to measure bank development with respect to the second objective. The third objective of the study was to analyse the sensitivity of the causality findings to a change in the measure of bank development. To this end, the study used two alternative measures of bank development, i.e., the broad money ratio and the domestic credit ratio. Using the broad money ratio, the study found support for a bi-directional causal flow in the case of stock market development and growth per capita. In the case of bank development and growth per capita, the findings do not support the existence of causality in any direction. In addition, this study established no evidence of a causal flow between bank development and stock market development. Using the domestic credit ratio to measure bank development, the findings of this study revealed no causal linkage between stock market development and growth for the selected SSA countries. For the case of bank development and growth per capita, the study found evidence supporting the supply-leading finance. Lastly, there exist a bi-directional causal flow between bank development and stock market development. 105 University of Ghana http://ugspace.ug.edu.gh In sum, the long run relationship between stock market development, bank development and growth is not sensitive to the choice of the measure of bank development. However, the causal relationship between stock market development, bank development and growth per capita is sensitive to the choice of the measure of bank development. 6.3 Policy Implications The policy implications of this study are drawn using the bank credit ratio as the measure of bank development. We have highlighted the reasons we prefer this measure based on literature in chapter three above. The main findings of this study is that there is no long run relationship between stock market development, bank development and growth in the selected SSA countries. Based on the comparative analysis we presented in the overview chapter, we noted that the sample has experienced lower stock markets and bank development, compared to the SSA and other regions in the world. We believe the underdeveloped stock markets and banking sector in the sample inform these findings. On causality, this study found evidence that stock market development and bank development leads growth per capita. In this regard, the findings support the supply-leading hypothesis. Evidently, further development of the stock markets and banks in our sample and by extension SSA, will promote further economic growth. We therefore, suggest that policy makers and governments in the selected countries in SSA formulate and implement policy and regulatory reforms aimed at promoting further stock markets and bank development, since there is a huge potential for growth in these two sectors to support economic growth. Specifically, concerning the stock markets, we recommend simplification of listing procedures to increase listings on the stock markets, which as shown in chapter two, are low. To increase participation of small and medium enterprises in the stock markets, policy makers and 106 University of Ghana http://ugspace.ug.edu.gh governments should organize and encourage sensitization fora to these target groups. This study also recommends simplification of trading rules and procedures including promotion of mobile enabled trading of stocks to raise the trading volumes of stock markets in the sample. With respect to the banking sector, we recommend enhancing competition in the banking sector to reduce the cost of lending and expand access to credit. We further recommend a reduction of the government role in the economy to ameliorate the crowding out of the private sector in the credit market. 6.4 Limitations of the Study and Suggestions on Further Research Areas This study was limited to the domestic credit to the private sector by banks as a share of nominal GDP (bank credit ratio) as the measure of bank development. There are many players in diverse sectors in the private sector. The sectors where the private sector players participate have varied contributions to economic growth. However, disaggregated data on bank lending to the various economic sectors constituting the private sector is not readily available. The available data lumps up the entire private sector as a single unit without regard to the specific sectors in the economy. The study, therefore, did not consider the sectors in the economy that receive this credit. To clearly understand the long run relationship and the causality between bank development and growth, future studies should attempt to obtain disaggregated data on bank loans by economic sectors such as manufacturing, agriculture and service sectors, within the private sector. Further, the study did not consider alternative measures of stock market development due to lack of consistent data on these alternative measures such as the shares traded ratio and the market capitalization ratio. The data on the measure of stock market development (turnover ratio) used by this study, was also considerably lower in frequency compared to the other 107 University of Ghana http://ugspace.ug.edu.gh variables used. We have no doubt that future studies considering these areas will bring more insight into the existing body of knowledge. Future studies can also increase the number of countries and the sample period as data becomes available. 108 University of Ghana http://ugspace.ug.edu.gh References Abu‐Bader, S., & Abu‐Qarn, A. S. (2008). Financial Development and Economic Growth: Empirical Evidence from six MENA Countries. Review of Development Economics, 12(4), 803-817. Acaravci, S. K., Ozturk, I., & Acaravci, A. (2009). Financial Development and Economic Growth: Literature Survey and Empirical Evidence from Sub-Saharan African Countries. South African Journal of Economic and Management Sciences, 12(1), 11- 27. Adebola, S. S., & Dahalan, J. (2011). Financial Development and Economic Growth: The Role of Stock Markets and Banking Sector in Nigeria. Journal of Sustainable Development in Africa, 13(7). Adjasi, C. K., & Biekpe, N. B. (2006). Stock Market Development and Economic Growth: The case of Selected African Countries. African Development Review, 18(1), 144-161. Adu, G., Marbuah, G., & Mensah, J. T. (2013). Financial Development and Economic Growth in Ghana: Does the Measure of Financial Development Matter? Review of Development Finance, 3(4), 192-203. Adusei, M. (2013). Financial Development and Economic Growth: Evidence from Ghana. The International Journal of Business and Finance Research, 7(5), 61-76. African Securities Exchanges Association. (2015). Financial Data. Retrieved November 15, 2015, from http://www.african-exchanges.org/ Agbetsiafa, D. (2004). The finance growth nexus: Evidence from Sub-Saharan Africa. Savings and Development, 271-288. Andabai, P. W., & Igbodika, M. N. (2015). A Causality Analysis of Financial Deepening and Performance of Nigerian Economy (1990-2013). African Research Review, 9(4), 249- 263. 109 University of Ghana http://ugspace.ug.edu.gh Ang, J. B., & McKibbin, W. J. (2007). Financial Liberalization, Financial Sector Development and Growth: Evidence from Malaysia. Journal of development Economics, 84(1), 215- 233. Applegarth, P. V., Kansteiner, W. H., & Morrison, J. S. (2004). Capital Market and Financial Sector Development in Sub-saharan Africa. Report of the Africa Policy Advisory Panel, Center for Strategic and International Studies, Washington, DC. Arellano, M., & Bond, S. (1991). Some Tests of Specification for Panel Data: Monte Carlo Evidence and an Application to Employment Equations. The Review of Economic Studies, 58(2), 277-297. Arellano, M., & Bover, O. (1995). Another look at the Instrumental Variable Estimation of Error-Components Models. Journal of econometrics, 68(1), 29-51. Bank of Ghana. (2016). Register of Licensed Institutions. Retrieved March 12, 2016, from https://www.bog.gov.gh/index.php?option=com_content&view=article&id=110&Ite mid=123 Bank of Mauritius. (2016). Authorized Financial Institutions. Retrieved March 12, 2016, from https://www.bom.mu/?id=60007 Bank of Namibia. (2016). Bank Supervision. Retrieved March 12, 2016, from https://www.bon.com.na/ Bank of Tanzania. (2016). Banking Supervision. Retrieved March 12, 2016, from https://www.bot-tz.org/BankingSupervision/BankingSupervision.asp Bank of Uganda. (2016). Supervision. Retrieved March 12, 2016, from https://www.bou.or.ug/bou/supervision/overview.html Bank of Zambia. (2016). Financial Supervision. Retrieved March 12, 2016, from http://www.boz.zm/ 110 University of Ghana http://ugspace.ug.edu.gh Baum, C. F., Schaffer, M. E., & Stillman, S. (2003). Instrumental Variables and GMM: Estimation and Testing. Stata journal, 3(1), 1-31. Beck, T., & Levine, R. (2004). Stock Markets, Banks, and Growth: Panel Evidence. Journal of Banking & Finance, 28(3), 423-442. Beck, T., Levine, R., & Loayza, N. (2000). Finance and the Sources of Growth. Journal of financial economics, 58(1), 261-300. Bencivenga, V. R., & Smith, B. D. (1991). Financial Intermediation and Endogenous Growth. The Review of Economic Studies, 58(2), 195-209. Blundell, R., & Bond, S. (1998). Initial Conditions and Moment Restrictions in Dynamic Panel Data Models. Journal of econometrics, 87(1), 115-143. Boyd, J. H., & Prescott, E. C. (1986). Financial Intermediary-Coalitions. Journal of Economic Theory, 38(2), 211-232. Calderón, C., & Liu, L. (2003). The Direction of Causality between Financial Development and Economic Growth. Journal of development Economics, 72(1), 321-334. Calderon, C. A., Chong, A., & Loayza, N. V. (2002). Determinants of Current Account Deficits in Developing Countries. Contributions in Macroeconomics, 2(1). Central Bank of Botswana. (2016). Financial Institutions. Retrieved March 12, 2016, from http://www.bankofbotswana.bw/index.php/content/2009111815017-financial- institutions Central Bank of Kenya. (2016). Bank Supervision. Retrieved March 12, 2016, from https://www.centralbank.go.ke/index.php/bank-supervision Central Bank of Nigeria. (2016). Financial Institutions. Retrieved March 12, 2016, from https://www.cbn.gov.ng/Supervision/finstitutions.asp Central Bank of Swaziland. (2016). Bank Supervision. Retrieved March 12, 2016, from http://www.centralbank.org.sz/functions/banksupervision/index.php 111 University of Ghana http://ugspace.ug.edu.gh Choi, I. (2001). Unit Root Tests for Panel Data. Journal of international money and Finance, 20(2), 249-272. Christopoulos, D. K., & Tsionas, E. G. (2004). Financial Development and Economic Growth: Evidence from Panel Unit Root and Cointegration Tests. Journal of development Economics, 73(1), 55-74. Chukwu, J. O., & Agu, C. C. (2009). Multivariate Causality between Financial Depth and Economic Growth in Nigeria. African Review of Money Finance and Banking, 7-21. Cooray, A. (2010). Do Stock Markets lead to Economic Growth? Journal of Policy Modeling, 32(4), 448-460. Demetriades, P. O., & Hussein, K. A. (1996). Does Financial Development cause Economic Growth? Time-series evidence from 16 countries. Journal of development Economics, 51(2), 387-411. Diamond, D. W. (1984). Financial Intermediation and Delegated Monitoring. The Review of Economic Studies, 51(3), 393-414. Enisan, A. A., & Olufisayo, A. O. (2009). Stock Market Development and Economic Growth: Evidence from Seven Sub-Sahara African Countries. Journal of economics and business, 61(2), 162-171. Farid, S. (2013). Financial integration in African emerging markets. Paper presented at the African Economic Conference 2013. Johannesburg, South Africa Geweke, J. (1982). Measurement of Linear Dependence and Feedback between Multiple Time Series. Journal of the American statistical association, 77(378), 304-313. Ghirmay, T. (2004). Financial Development and Economic Growth in Sub‐Saharan African Countries: Evidence from Time Series Analysis. African Development Review, 16(3), 415-432. Greenwood, J., & Jovanovic, B. (1990). Financial Development and Economic Development. Economic Development and Cultural Change, 15, 257-268. 112 University of Ghana http://ugspace.ug.edu.gh Hassan, M. K., Sanchez, B., & Yu, J.-S. (2011). Financial Development and Economic Growth: New Evidence from Panel Data. The Quarterly Review of economics and finance, 51(1), 88-104. Historical Data. (2015). Stock Exchange of Mauritius Historical Data. Retrieved November 15, 2015, from http://www.stockexchangeofmauritius.com/historicaldata. Holtz-Eakin, D., Newey, W., & Rosen, H. S. (1988). Estimating Vector Autoregressions with Panel Data. Econometrica: Journal of the Econometric Society, 1371-1395. Hoover, R. (2016). Investing in Africa. Retrieved March 13, 2016, from http://investinginafrica.net/african-stock-markets/african-stock-market-performance/ Im, K. S., Pesaran, M. H., & Shin, Y. (2003). Testing for Unit Roots in Heterogeneous Panels. Journal of econometrics, 115(1), 53-74. International Monetary Fund (April, 2011). Regional Economic Outlook: Sub-Saharan Africa, Recovery and New Risks. World Economic and Financial Surveys, 0258-7440. Washington D.C International Monetary Fund (2013). Regional Economic Outlook: Sub-Saharan Africa, Building Momentum in a Multi-Speed World. World economic and financial surveys, 0258-7440. Washington D.C International Monetary Fund (2015). IMF World Economic Outlook Database. Retrieved November 15, 2015, from http://data.okfn.org/data/core/imf-weo Jahan, S. (2015). Human Development Report 2015. Work for Human Development. New York: United Nations Development Programme. Jefferis, K., & Mbekeani, K. (2000). New Forms Of Integration In Emerging Africa. Paper Presented at the A Policy Workshop Organized By The Oecd Development Centre At The Graduate Institute Of International Studies–Geneva Friday. 113 University of Ghana http://ugspace.ug.edu.gh Johansen, S. (1988). Statistical Analysis of Cointegration Vectors. Journal of economic dynamics and control, 12(2), 231-254. Johansen, S., & Juselius, K. (1990). Maximum Likelihood Estimation and Inference on Cointegration—with Applications to the Demand for Money. Oxford Bulletin of Economics and statistics, 52(2), 169-210. Jung, W. S. (1986). Financial Development and Economic Growth: International Evidence. Economic Development and Cultural Change, 34(2), 333-346. King, R. G., & Levine, R. (1993). Finance and Growth: Schumpeter Might be Right. The quarterly journal of economics, 717-737. Levine, R. (2002). Bank-Based or Market-Based Financial Systems: Which is better? Journal of financial intermediation, 11(4), 398-428. Levine, R., & Zervos, S. (1998). Stock Markets, Banks, and Economic Growth. American economic review, 537-558. Levin, A., Lin, C. F., & Chu, C. S. J. (2002). Unit root tests in panel data: asymptotic and finite- sample properties. Journal of econometrics, 108(1), 1-24. Macro Economy Meter. (2015). Zimbabwe Inflation. Retrieved November 25, 2015, from http://mecometer.com/whats/zimbabwe/inflation/ Macro Economy Meter. (2015). Zambia Government Expenditure as a Share of GDP. Retrieved November 25, 2015, from http://mecometer.com/whats/zambia/government- expenditure-percentage-of-gdp/ Mankiw, N. G., Romer, D., & Weil, D. N. (1990). A Contribution to the Empirics of Economic Growth: National Bureau of Economic Research. McKinnon, R. I. (1973). Money and Capital in Economic Development. Brookings Institution Press, Washington 114 University of Ghana http://ugspace.ug.edu.gh Mlachila, M., Park, S. G., & Yabara, M. (2013). Banking in Sub-Saharan Africa: The Macroeconomic Context. Banking in Sub-Saharan Africa, 7. N'Zue, F. F. (2006). Stock Market Development and Economic Growth: Evidence from Cote D'Ivoire. African Development Review, 18(1), 123. Naceur, S. B., & Ghazouani, S. (2007). Stock Markets, Banks, and Economic Growth: Empirical Evidence from the MENA region. Research in International Business and Finance, 21(2), 297-315. Ndako, U. B. (2010). Stock Markets, Banks and Economic Growth: Time Series Evidence from South Africa. The African Finance Journal, 12(2), 72-92. Ndebbio, J. (2004). Financial Deepening, Economic Growth and Development: Evidence from Selected Sub-Saharan African Countries, AERC Research Paper, 142. Nariobi (Kenya): African Economic Research Consortuim. Nickell, S. (1981). Biases in Dynamic Models with Fixed Effects. Econometrica: Journal of the Econometric Society, 1417-1426. Nowbutsing, B. M., & Odit, M. (2010). Stock Market Development and Economic Growth: The case of Mauritius. International Business & Economics Research Journal (IBER), 8(2). Odhiambo, N. M. (2007). Supply‐Leading versus Demand‐Following Hypothesis: Empirical Evidence from three SSA Countries. African Development Review, 19(2), 257-280. Odhiambo, N. M. (2008). Financial Depth, Savings and Economic Growth in Kenya: A Dynamic Causal Linkage. Economic Modelling, 25(4), 704-713. Pagano, M. (1993). Financial Markets and Growth: An Overview. European economic review, 37(2-3), 613-622. Patrick, H. T. (1966). Financial Development and Economic Growth in Underdeveloped Countries. Economic Development and Cultural Change, 14(2), 174-189. 115 University of Ghana http://ugspace.ug.edu.gh Pedroni, P. (1999). Critical Values for Cointegration Tests in Heterogeneous Panels with Multiple Regressors. Oxford Bulletin of Economics and statistics, 61(s 1), 653-670. Pedroni, P. (2000). Fully Modified Ols For Heterogeneous Cointegrated Panels. Panel Cointegration, and Dynamic Panels (pp. 93-130). Pedroni, P. (2004). Panel Cointegration: Asymptotic and Finite Sample Properties of Pooled Time Series Tests with an Application to the PPP hypothesis. Econometric theory, 20(03), 597-625. Pesaran, M. H., Shin, Y., & Smith, R. J. (2001). Bounds Testing Approaches to the Analysis of Level Relationships. Journal of applied econometrics, 16(3), 289-326. Quartey, P., & Prah, F. (2008). Financial Development and Economic Growth in Ghana: Is there a Causal Link? The African Finance Journal, 10(1), 28-54. Rachdi, H. (2011). The Causality between Financial Development and Economic Growth: A Panel Data Cointegration. International Journal of Economics and Finance, 3(1), 143. Reserve Bank of Malawi. (2016). Bank Supervision. Retrieved March 12, 2016, from https://www.rbm.mw/Supervision/BankSupervision/ Robinson, J. (1952). The Generalization of the General Theory”, In: the Rate of Interest and Other Essays, London: MacMillan. Roodman, D. (2006). How to do xtabond2: An introduction to difference and system GMM in Stata. Center for Global Development working paper(103). Roodman, D. (2009). A Note on the Theme of Too Many Instruments. Oxford Bulletin of Economics and statistics, 71(1), 135-158. Seetanah, B., Sawkut, R., Sannasee, V., & Seetanah, B. (2008). Stock Market Development and Economic Growth in Developing Countries: Evidence from Panel VAR Framework. Shaw, E. S. (1973). Financial Deepening in Economic Development. Oxford University Press, London 116 University of Ghana http://ugspace.ug.edu.gh South African Reserve Bank. (2016). Bank Supervision. Retrieved March 12, 2016, from http://www.resbank.co.za/RegulationAndSupervision/BankSupervision/Pages/SouthA fricanRegisteredBanksAndRepresentativeOffices.aspx Stock, J., & Watson, M. (2007). Introduction to Econometrics.: Pearson Education Inc. New York. Tachiwou, A. M. (2010). Stock Market Development and Economic Growth: The case of West African Monetary Union. International Journal of Economics and Finance, 2(3), 97. Toda, H. Y., & Yamamoto, T. (1995). Statistical Inference in Vector Autoregressions with Possibly Integrated Processes. Journal of econometrics, 66(1), 225-250. World Bank. (2014, June). World Development Indicators. Retrieved October 20, 2014, from http://data.worldbank.org/data-catalog/world-development-indicators World Bank. (2015, June). World Development Indicators. Retrieved October 20, 2015, from http://data.worldbank.org/data-catalog/world-development-indicators World Bank. (2016, June). World Development Indicators. Retrieved June 23, 2016, from http://data.worldbank.org/data-catalog/world-development-indicators Yang, Y. Y., & Yi, M. H. (2008). Does Financial Development Cause Economic Growth? Implication for Policy in Korea. Journal of Policy Modeling, 30(5), 827-840. 117 University of Ghana http://ugspace.ug.edu.gh APPENDICES Appendix I: List of the 12 Selected Sub-Saharan Africa Countries in Main Estimations 1. Botswana 8. South Africa 2. Ghana 9. Swaziland 3. Kenya 10. Tanzania 4. Malawi 11. Uganda 5. Mauritius 12. Zambia 6. Namibia 7. Nigeria Appendix II: Stock Markets Listings for the 12 Selected SSA countries in Main Estimations Country City Name Founded Listings 1. Botswana Gaborone Botswana Stock Exchange 1989 38 2. Ghana Accra Ghana Stock Exchange 1990 43 3. Kenya Nairobi Nairobi Securities Exchange 1954 46 4. Malawi Blantyre Malawi Stock Exchange 1995 13 5. Mauritius Port Louis Stock Exchange of Mauritius 1988 95 6. Namibia Windhoek Namibia Stock Exchange 1992 41 7. Nigeria Abuja/Lagos Nigerian Stock Exchange 1960 189 8. South Africa Johannesburg Johannesburg Stock Exchange 1887 402 9. Swaziland Mbabane Swaziland Stock Exchange 1990 7 10. Tanzania Dar es salaam Dar es salaam Stock Exchange 1998 23 11. Uganda Kampala Uganda Securities Exchange 1997 18 12. Zambia Lusaka Lusaka Stock Exchange 1994 22 Totals 937 Source: African Securities Exchanges Association (2015) 118 University of Ghana http://ugspace.ug.edu.gh Appendix III: Financial Structure for the 12 selected SSA Countries in the Main Estimations Countries Commercial Representative Non-Bank Microfinance Forex Banks Offices financial Institutions/ Bureaus institutions Banks Botswana 10 - - - 52 Ghana 31 4 65 576 413 Kenya 42 8 20 12 79 Malawi 11 - 4 - - Mauritius 22 - 20 - 5 Namibia 9 - - - 10 Nigeria 22 117 941 2991 South Africa 16 38 5 - 17 Swaziland 4 - 4 - 1 Tanzania 40 - 8 - 222 Uganda 25 - 8 - 246 Zambia 19 - 17 37 73 Totals 251 50 268 1566 4109 Source: Respective Central Bank’s Websites 119 University of Ghana http://ugspace.ug.edu.gh Appendix IV: Hausman Test for endogeneity Specification (1) (2) (3) (4) (5) (6) (7) (8) (9) Variables lgdp - 0.000 0.000 - 0.000 0.000 - 0.000 0.000 bank 0.000 0.000 - - - - - - - bmr - - - 0.000 0.000 - - - - domestic - - - - - - 0.000 0.000 - turnover 0.001 - 0.000 0.000 - 0.000 0.001 - 0.000 gfc 0.067 0.000 0.942 0.046 0.000 0.048 0.063 0.015 0.486 govt 0.000 0.062 0.013 0.000 0.047 0.068 0.000 0.167 0.033 inflation 0.914 0.000 0.901 0.907 0.000 0.238 0.913 0.010 0.407 openness 0.000 0.000 0.007 0.000 0.000 0.000 0.000 0.009 0.917 Notes: The figures presented in Appendix V are P-values for the Hausman endogeneity test Source: Author’s computation using STATA 13 and data from WDI, 2015 120 University of Ghana http://ugspace.ug.edu.gh Appendix V: Breusch-Pagan/Cook-Weisberg Heteroskedasticity Tests Results Specification Breusch-Pagan / Cook-Weisberg test for heteroskedasticity 1 25.65 (0.000) 2 50.91 (0.000) 3 81.78 (0.000) 4 29.63 (0.000) 5 59.91 (0.000 6 7.50 (0.006) 7 29.60 (0.000) 8 38.40 (0.000) 9 296.87(0.000) Notes: P-values are in parentheses. Source: Author’s computation using STATA 13 and data from WDI, 2015 121 University of Ghana http://ugspace.ug.edu.gh Appendix VI: Time Effects, Breusch & Pagan Test and Hausman Specification Tests Specification Test for Time Breusch/ Pagan Hausman Specification Test Effects Test 1 1.24 (0.209) 0.00(1.000) - 2 0.45(0.988) 187.11 (0.000) 20.13 (0.003) 3 0.43(0.992) 918.50 (0.000) 48.37 (0.000) 4 1.37(0.123) 0.00(1.000) - 5 0.50(0.977) 532.89 (0.000) 9.23 (0.1613) 6 1.25(0.203) 0.00(1.000) - 7 1.24(0.214) 0.00(1.000) - 8 0.79(0.747) 26.27 (0.000) 24.04 (0.001) 9 0.53(0.968) 651.28 (0.000) 123.53 (0.000) Notes: P-values are in parentheses. Source: Author’s computation using STATA 13 and data from WDI, 2015 122