Three essays on finance and economic development

There is a large body of literature stressing the importance of developing financial markets, including stock markets, to enhance countries' growth rates. In the first essay, I argue that the relationship between stock markets and growth is exaggerated and that the simple act of opening a formal stock market is not a good predictor of whether a country will experience economic growth. This is evaluated using two Bayesian econometric methods, Extreme Bounds Analysis (EBA) and Bayesian Model Averaging (BMA) by regressing growth between 2002 and 2007 on stock market openings between 1960 and 1999. The findings indicate that the opening of a stock market does not influence economic growth. In the second essay the Schumpeterian innovation life-cycle is used to argue that firms will be more likely to raise funds through the stock market rather than the bond market if they are engaged in radically new technologies. This argument is placed within the context of the dominant theories of capital structure. Empirically, I test this relationship of innovative activity to equity issuance by using patents as a proxy for innovation from a dataset covering 1970 to 1992 regressed on whether a firm raised funds through the bond or stock market. I find statistically significant evidence using a dichotomous probit model that the industries with higher innovative/patenting activities are significantly more likely to raise funds through stock market issuance than firms without innovative activity. The third essay evaluates the relationship between access to credit and the private credit to GDP ratio. I argue that two measures of inclusiveness, total access and the equality of access, are positively related with private credit and financial development. The newly released Global Financial Index database from the World Bank allows for the first time the ability to effectively test the impact of access and inequality of access. I find significant evidence that the total percentage of people in the financial sector is associated with, and unequal access to finance leads to, a lower private credit/GDP ratio.

THREE ESSAYS ON FINANCE AND ECONOMIC DEVELOPMENT by Chase Parker DeHan A dissertation submitted to the faculty of The University of Utah in partial fulfillment of the requirements for the degree of Doctor of Philosophy Department of Economics The University of Utah December 2013 Copyright © Chase Parker DeHan 2013 All Rights Reserved The University of Utah Graduate School STATEMENT OF DISSERTATION APPROVAL The dissertation of Chase Parker DeHan has been approved by the following supervisory committee members: Cihan Bilginsoy , Chair 08/13/2013 Date Approved Richard Fowles , Member 08/13/2013 Date Approved Thomas Maloney , Member 08/13/2013 Date Approved Korkut Erturk , Member 08/13/2013 Date Approved Philip Arestis , Member 08/13/2013 Date Approved and by Thomas Maloney , Chair of the Department of Economics and by David B. Kieda, Dean of The Graduate School. ABSTRACT There is a large body of literature stressing the importance of developing financial markets, including stock markets, to enhance countries' growth rates. In the first essay, I argue that the relationship between stock markets and growth is exaggerated and that the simple act of opening a formal stock market is not a good predictor of whether a country will experience economic growth. This is evaluated using two Bayesian econometric methods, Extreme Bounds Analysis (EBA) and Bayesian Model Averaging (BMA) by regressing growth between 2002 and 2007 on stock market openings between 1960 and 1999. The findings indicate that the opening of a stock market does not influence economic growth. In the second essay the Schumpeterian innovation life-cycle is used to argue that firms will be more likely to raise funds through the stock market rather than the bond market if they are engaged in radically new technologies. This argument is placed within the context of the dominant theories of capital structure. Empirically, I test this relationship of innovative activity to equity issuance by using patents as a proxy for innovation from a dataset covering 1970 to 1992 regressed on whether a firm raised funds through the bond or stock market. I find statistically significant evidence using a dichotomous probit model that the industries with higher innovative/patenting activities are significantly more likely to raise funds through stock market issuance than firms without innovative activity. iv The third essay evaluates the relationship between access to credit and the private credit to GDP ratio. I argue that two measures of inclusiveness, total access and the equality of access, are positively related with private credit and financial development. The newly released Global Financial Index database from the World Bank allows for the first time the ability to effectively test the impact of access and inequality of access. I find significant evidence that the total percentage of people in the financial sector is associated with, and unequal access to finance leads to, a lower private credit/GDP ratio. TABLE OF CONTENTS ABSTRACT………………………………………………………………………. iii LIST OF TABLES………………………………………………………………... vii LIST OF FIGURES………………………………………………………………. viii ACKNOWLEDGEMENTS………………………………………………………. ix Chapters 1. INTRODUCTION……………………………………………………………. 1 1.1 Stock Markets and Growth: A Re-Evaluation.……………………….. 4 1.2 Schumpeterian Innovation and Equity Issuance……………………… 5 1.3 Private Credit and Unequal Access…………………………………… 6 1.4 References……………………..……………………………………… 8 2. STOCK MARKETS AND GROWTH: A RE-EVALUATION……………… 10 2.1 Abstract……………………………………………………………….. 10 2.2 Introduction…………………………………………………………… 11 2.3 Finance-Growth Nexus……………………………………………….. 14 2.4 Stock Markets and Growth…………………………………………… 19 2.5 Empirical Methodology………………………………………………. 24 2.5.1 Extreme Bounds Analysis…………………………………….. 27 2.2.2 BMA Overview……………………………………………….. 30 2.6 Data…………………………………………………………………... 34 2.7 Results………………………………………………………………... 37 2.7.1 EBA Results…………………………....................................... 37 2.7.2 BMA Results………………………………….......................... 39 2.8 Conclusion………………………………………………………........ 46 2.9 References………………………………………………..................... 47 3. SCHUMPETERIAN INNOVATION AND EQUITY ISSUANCE.………… 58 3.1 Abstract……………………………………………………................. 58 3.2 Introduction…………………………….............................................. 59 vi 3.3 Theories of Capital Structure……………………………..................... 62 3.3.1 Trade-Off Theory………………………………….................... 62 3.3.2 Market Timing……………………............................................. 66 3.3.3 Pecking Order……………………............................................. 67 3.4 Innovation Life Cycles in the Vein of Schumpeter……....................... 69 3.5 Financing………………....................................................................... 73 3.6 The Life Cycle and Capital Structure Theories……………................. 78 3.6.1 Market Timing…………………................................................ 78 3.6.2 Trade-Off…………………........................................................ 79 3.6.3 Pecking Order…………………................................................. 82 3.7 Empirical Methodology……………………………............................ 83 3.7.1 Data…………………………………........................................ 86 3.7.2 Industry Level Variables……………………............................. 88 3.7.3 Market Control Variables……………………........................... 92 3.8 Results……………………………………………….......................... 98 3.8.1 Base Model…………………………………............................ 99 3.8.2 Industry Returns Cohort……………………............................ 103 3.8.3 Fama-French Cohort……………………................................. 104 3.8.4 Time Trend Cohort…………………….................................... 106 3.8.5 Reduced Sample……………………........................................ 107 3.9 Conclusion………………………………………………................... 108 3.10 References……………………………………………….................. 110 4. PRIVATE CREDIT AND UNEQUAL ACCESS………………….……...... 125 4.1 Abstract……………………………………………………................ 125 4.2 Introduction…………………………….............................................. 126 4.3 Theory and Review of Literature……………………………............. 129 4.3.1 Information Sharing………………………………….............. 133 4.3.2 Creditor Rights…………………….......................................... 135 4.3.3 Access to Credit……………………........................................ 136 4.3.4 Total Access…………………….............................................. 140 4.3.5 Unequal Access……………………......................................... 143 4.4 Data……............................................................................................. 147 4.4.1 Dependent Variable…………………...................................... 148 4.4.2 Access Variables…………………........................................... 148 4.4.2.1 Total Access…………………....................................... 150 4.4.2.2 Unequal Access………………….................................. 151 4.4.3 Control Variables………………….......................................... 152 4.4.4 Market Conditions…………………........................................ 156 4.5 Model……………………………...................................................... 157 4.6 Results………………………………………………........................ 159 4.7 Conclusion……………………………………………….................. 168 4.8 References……………………………………………….................. 171 5. CONCLUSION……………………………………………….……………. 181 LIST OF TABLES 2.1. Countries in Sample………………………………………………………… 52 2.2. Variable Descriptions…………………………………………………......... 53 2.3. EBA Results…………………………………………………....................... 54 2.4. BMA Summary Output…………………………………………………...... 55 3.1. Data Summary………………………………………………….................... 116 3.2. Industry Fundraising and Patenting…………………………....................... 119 3.3. Data Description.…………………………………………………............... 121 3.4. Base Model…………………………………………………........................ 122 3.5. Base Plus Industry Returns……………………………………………........ 122 3.6. Fama-French Estimations…………………………………………….......... 123 3.7. Results Including Time Trend…………………………………………….... 123 3.8. Reduced Sample…………………………………………………................. 124 4.1. Data Descriptions…………………………………………………................ 175 4.2. Private Credit/GDP Regressions……………………………………............. 177 4.3. Private Credit/GDP Regressions with Information Sharing………………… 177 4.4. Private Credit/GDP Regressions…………………………............................. 178 4.5. Private Credit/GDP Regressions with Information Sharing….……............... 179 4.6. Regressions with GINI Coefficient….…….................................................... 180 LIST OF FIGURES 2.1. Stock Market Openings Over Time…………………………………............ 52 2.2. Selected Posterior Distributions…………………………………………...... 56 2.3. BMA Image Plot…………………………………………………................. 57 3.1. Trade-Off Theory…………………………………………………................ 114 3.2. Output Over the Life Cycle…………………………………........................ 114 3.3. Stock Issuance Over Time………………………………………………….. 115 3.4. Trade-Off Including Radical Industries…………………………………….. 115 3.5. Bond and IPO Activity Over Time: Total……………................................... 117 3.6. Bond and IPO Activity Over Time: Total $ Proceeds…………………........ 117 3.7. Fundraising…………………………………………………......................... 118 4.1. Private Credit ~ Access…………………………………………………...... 176 4.2. Private Credit ~ Ineq.Access……………………………………………….. 176 ACKNOWLEDGEMENTS There are numerous people that have lent assistance throughout my graduate career. The first thanks goes to my dissertation chair, Cihan Bilginsoy, for helping me throughout the process of writing this dissertation and preparing me for my eventual career as an academic; for this I am eternally grateful. I would also like to thank Richard Fowles for introducing me to Bayesian statistics and his help with the creation of my paper "Stock Markets and Growth: A Re-Evaluation." Thomas Maloney also provided numerous helpful comments and guidance throughout my career as a graduate student for which I am grateful. Korkut Erturk and Philip Arestis provided numerous comments in completing my dissertation, many of which have helped turn these essays into substantially better research. I would also like to thank Yavuz Yasar for recommending I read Mumtaz Keklik's research on the Schumpeterian innovation life cycle that ended up transforming my paper "Schumpeterian Innovation and Equity Issuance." I would also like to thank Steve Bannister, Codrina Rada, Rudi Von Arnim, and Lance Girton for their valued conversations. Most of all, I would like to thank my wife, Karie Warfield, for standing by my side and being there forever and always. CHAPTER 1 INTRODUCTION A developed and efficient financial system has a number of functions through which financial intermediaries are able to influence growth. According to the Levine (1997) classifications, these functions include allocating resources, mobilizing savings, reducing risks, facilitating transactions, and decreasing costs to monitor firms. Each of these functions plays its part in how financial development can influence growth. Gurley and Shaw (1955), Goldsmith (1969), McKinnon (1973), and Hicks (1969) argue that an efficient transfer of funds from surplus units to deficit units is necessary for stimulating economic growth. Without an efficient transfer of funds, entrepreneurs would be unable to obtain the necessary funds to expand their businesses, therefore lowering the level of growth. It is from this theoretical basis that this dissertation has emerged. The body of literature on the relationship between finance and growth would seem to suggest that financial development is the lynchpin to prosperity. Unfortunately, developing financial markets has not released the floodgates of economic growth. The essays contained herein evaluate the impact stock markets have as well as some specific conditions to better inform the development of effective financial systems. Empirical work on the relationship between finance and growth did not emerge in earnest until the early 1990s. Following the work of King and Levine (1993)-where 2 they found that the initial level of financial development can predict later rates of economic growth-a number of studies emerge confirming the results that more financial development is associated with (or causes) economic growth. Within the finance-growth empirical literature, the research can be divided into promoting market or bank-based financial systems, whereby increasing either element should result in higher growth. Additional cross country results supporting the notion that increasing overall financial development include Levine (2002)-whose findings were that there is a strong connection between financial development and growth regardless of whether the country has a bank or market-based financial system-and Demirguc-Kunt and Maksimovic (2002), who showed that overall financial development helps to explain the growth of firms. These studies evaluated the financial system as a whole, while a number of other studies took a narrower approach by examining the role of stock markets. These results show that stock markets are positively associated with economic growth (Atje & Jovanovic, 1993; Demirguc-Kunt & Maksimovic, 1998; Levine & Zervos, 1998). It is this last category of research that has prompted many developing countries to open a stock market and expect economic prosperity to follow. Alternatively, there exists a growing body of research questioning whether financial development is always and everywhere beneficial for growth. The idea suggested by Robinson (1952) is not that finance leads growth, but rather that financial systems act in response to economic conditions. As the economy is expanding, firms and households will have more demand for financial services. Responding to the increase in demand, more financial intermediaries and financial services will emerge. Outside of reverse causality, Lucas (1988) argued that finance is irrelevant for growth. He posits 3 that economists tend to overstate the impact of finance, and this idea was the reasoning for the exclusion of all financial considerations from a theoretical growth model. While the dominant theme behind the relationship between finance and growth is of a positive causal relationship there exists a smaller, yet powerful literature contending that this is overstressed. For example, Ram (1999) used a sample of 95 countries, finding that the correlation is weakly negative or negligible. A number of other studies have been able to show that causality does not always run from finance to growth, with results being found that the direction of causality can also run in the reverse direction (Ang & McKibbin, 2007; Arestis & Demetriades, 1997; Demetriades & Hussein, 1996). The results provided by Arestis, Demetriades, and Luintel (2001) of banks being more effective in promoting economic growth-and arguing that the effects of stock markets on growth have been exaggerated by cross country studies-sets the stage for the first essay evaluating the impact of opening a stock market. The three papers of my dissertation evaluate, theoretically and empirically, how certain elements of financial development achieve their aims. The first essay, "Stock Markets and Growth: A Re-Evaluation," fits into the literature evaluating whether stock markets are effective in promoting growth, finding insufficient evidence in support of the claims that the simple act of opening a stock market increases growth. The second essay, "Schumpeterian Innovation and Equity Issuance," furthers the first essay by evaluating particular circumstances where firms chose to raise funds through the stock market rather than bank channels. This is completed by presenting theory and evidence for the hypothesis that stock markets are more often used by firms in innovative industries rather than mature industries. The closing essay of the dissertation, "Private 4 Credit and Unequal Access," further evaluates the types of financial institutions most effective in facilitating growth. This is evaluated by using two new variables on the level of access to financial services regressed on the most widely used metric of financial development, private credit to GDP ratio. The robust results lend support for my hypothesis that higher levels of unequal access to finance leads to lower levels of private credit. 1.1 Stock Markets and Growth: A Re-Evaluation Building on the large body of literature stressing the importance of developing financial markets, including stock markets, to enhance countries' growth, the first essay of this dissertation evaluates the link between the simple act of opening a stock market and economic growth. I argue that the relationship between stock markets and growth is exaggerated and that the simple act of opening a formal stock market is not a good predictor of whether a country will experience economic growth. While it is possible that in some instances opening a stock market can influence growth, I do not find any evidence that opening a stock market will have a broad impact. This research uses two Bayesian econometric methods, Extreme Bounds Analysis (EBA) and Bayesian Model Averaging (BMA), to discover if there are meaningful links between opening a stock market and growth in developing economies. Superior to traditional cross-sectional regressions, these methodologies allow for determining the true impact of certain variables. The impact of opening a stock market is tested by regressing growth between 2002 and 2007 on stock market openings between 1960 and 1999. Using similar explanatory variables as many other studies, I find a zero, or weakly 5 negative, relationship between the opening of a stock market and growth in developing countries. This essay finds that, on average, opening a stock market does not have any influence on growth, but does not show that stock markets will never influence growth. Stock markets may be more effective in countries with certain underlying characteristics. Intuitively, I believe that countries with higher levels of innovation will have conditions more amenable to an effective stock market. As such, determining when firms use stock markets to finance their activities is necessary. The second essay of the dissertation further evaluates this issue by positing that firms raise capital differently over the Schumpeterian innovation life-cycle. 1.2. Schumpeterian Innovation and Equity Issuance I hypothesize that highly innovative firms-those with high risk, yet higher potential return-will be more likely to raise funds through stock markets than bond issuance. Using the Schumpeterian innovation life-cycle as a theoretical framework, I argue that in the beginning, a company with a radically new innovation is more likely to raise funds through equity issuance until it becomes an acceptable loan for bankers with a limited return (interest rate). This is all placed within the context of, and does not conflict with, the dominant theories of firms' capital structure: the Trade-Off, Market timing, and Pecking Order Models. Empirically, I test this relationship of innovative activity to equity issuance by using patents as a proxy for innovation from a dataset covering the 1970 to 1992, encompassing 26,784 instances where firms raised funds through either the bond or stock 6 market. This independent variable is then regressed on the ratio of funds raised through equity to total funds raised. I find statistically significant evidence using a dichotomous, probit model that the industries with higher innovative/patenting activities are significantly more likely to raise funds through stock market issuance than firms without innovative activity. The opening of a stock market is a distinct change in development; other changes in financial development are much more subtle. The measurement of financial development is difficult; as such, we as economists have created a number of metrics. The widest used variable in recent research is the private credit to GDP ratio. This variable has grown to being used as synonymous with financial development when evaluating the impacts of financial development on economic growth and other human development indicators such as inequality and poverty. Causality issues aside, it is important to understand the underlying elements of this widely used variable. As such, the third essay of this dissertation revisits Djankov, McLiesh, and Shleifer (2007) on the determinants of private credit by evaluating the equality of access to financial services. 1.3. Private Credit and Unequal Access Building on the numerous studies evaluating how financial development affects economic growth, it is then important to understand important characteristics of financial development. One such characteristic of financial development is the level of access to financial institutions; increasing the level of access to financial institutions should increase the level of credit in the economy. In addition to access' impact on credit, the distribution of access across income groups is also important. Using the newly released 7 Global Financial Index database from the World Bank, this is able to be empirically tested for the first time. An additional contribution of this piece is through the construction of a new variable measuring the distribution of access among income groups. This differs from the traditional measurement of access as being the percentage of the population with accounts at financial institutions; my constructed metric is able to measure the percentage difference between having accounts at formal financial institutions between income groups. The empirical section of this essay follows the model specifications as set forth by Djankov et al. (2007), using the same control and explanatory variables with private credit set as the dependent variable. Each individual model specification is identical, only adding the two access to finance variables. Essentially, I am able to replicate the results of Djankov et al., although the access variables begin to dominate the results of the other control variables, indicating that access is an important element behind the growth of the financial sector. While determining the exact institutional structure to influence the accessibility is beyond the scope of this essay, it is evident that bringing the poor into the formal financial sector is an important policy choice. The idea of promoting an inclusive financial sector would imply that opening a stock market would be unnecessary as the poor do not typically purchase common stock. Within the topic of financial development, the three essays of this dissertation evaluate a separate issue in order to help determine the ideal policies for developing the financial sector of the economy. As will be seen, each of these essays substantially contributes toward our understanding of macro level financial reforms in developing countries. An interesting relationship between these essays is that the equality of access 8 to financial services is irrelevant for an effective functioning stock market. If opening a stock market was the gateway to economic growth, we would have found that formulating policies targeting the elite in the economy would be an ideal policy. However, we find the opposite result in that the equality of access matters and that simply opening a stock market is not enough to guarantee economic growth. Each of these essays opens the door for future research into effective ways to form financial sector policies. These avenues of future research will be discussed in the concluding remarks. 1.4. References Ang, J., & McKibben, W. (2007). Financial liberalization, financial sector development and growth: Evidence from Malaysia. Journal of Development Economics, 84, 215-233. Arestis, P., Demetriades, P. O., & Luintel, K. B. (2001). Financial development and economic growth: The role of stock markets. Journal of Money, Credit and Banking, 33, 16-41. Arestis, P., & Demetriades, P.O. (1997). Financial development and economic growth: Assessing the evidence. Economic Journal, 107, 783-799. Atje, R., & Jovanovic, B. (1993). Stock markets and development. European Economic Review, 37, 632-640 Demetriades, P.O., & Hussein, K.A. (1996). Does financial development cause economic growth? Time series evidence from sixteen countries. Journal of Development Economics, 51, 387-411. Demirguc-Kunt, A., & Maksimovic, V. (1998). Law, finance, and firm growth. Journal of Finance, 53, 2107-2137. Demirguc-Kunt, A., & Maksimovic, V. (2002). Funding growth in bank-based and market-based financial systems: Evidence from firm-level data. Journal of Financial Economics, 65, 337-363 Djankov, S., McLiesh, C., & Schleifer, A. (2007). Private credit in 129 countries. Journal of Financial Economics, 84, 299-329. 9 Goldsmith, R. (1969). Financial structure and development. New Haven, CT: Yale University Press. Gurley, J., & Shaw, E. (1955). Financial aspects of economic development. American Economic Review, 45, 515-538. Hicks, J. (1969). A theory of economic history. Oxford: Oxford University Press. King, R., & Levine, R. (1993). Finance and growth: Schumpeter might be right. Quarterly Journal of Economics, 108, 717-737. Levine, R. (1997). Financial development and economic growth: Views and agenda. Journal of Economic Literature, 35, 688-726. Levine, R. (2002). Bank-based or market-based financial systems: Which is better? Journal of Financial Intermediation, 11, 398-428. Levine, R., & Zervos, S. (1998). Stock markets, banks, and economic growth. American Economic Review, 88, 537-558. Lucas, R. (1988). On the mechanics of economic development. Journal of Monetary Economics, 22, 3-42. McKinnon, R. (1973). Money and capital in economic development. Washington, DC: Brookings Institution. Ram, R. (1999). Financial development and economic growth: Additional evidence. Journal of Development Studies, 35, 164-174. Robinson, J. (1952). The rate of interest and other essays. London: Macmillan. CHAPTER 2 STOCK MARKETS AND GROWTH: A RE-EVALUATION 2.1. Abstract There is a large body of literature stressing the importance of developing financial markets, including stock markets, to enhance countries' growth rates. A number of empirical studies have come up with conflicting evidence as to how effective stock markets are in facilitating economic growth. Dominant economic theory posits that a stock market more efficiently allocates capital to productive projects. I argue that this relationship between stock markets and growth is exaggerated and that the simple act of opening a formal stock market is not a good predictor of whether a country will experience economic growth. This is evaluated using two Bayesian econometric methods, Extreme Bounds Analysis (EBA) and Bayesian Model Averaging (BMA), to discover if there are meaningful links between opening a stock market and growth. In light of the conflicting results using classical econometric techniques, these Bayesian methodologies are able to reduce the uncertainty inherent in estimating any relationship. The impact of opening a stock market is tested by regressing growth between 2002 and 2007 on stock market openings between 1960 and 1999 using similar explanatory variables as many other 11 studies. While it is possible that in some instances opening a stock market can influence growth, I do not find any evidence that opening a stock market has any impact. 2.2. Introduction There exists a large body of research on the relationship between finance and growth; much of this points to a positive relationship, where more developed financial markets are found in countries with higher levels of growth. The McKinnon-Shaw (Gurley & Shaw, 1955; McKinnon, 1973) hypothesis posits that efficient financial markets are able to mobilize savings and allocate capital to the productive sectors of the economy, therefore facilitating growth. That finance causes growth has been empirically displayed a number of times using different datasets and econometric methodologies (Demirguc-Kunt & Levine, 1996; Levine, 1991; Minier, 2009). The basic policy advice coming from this theoretical and empirical literature is to develop financial markets and watch the economy grow. However, not all researchers are convinced that the relationship is as clear cut or that the direction of causality is accurate (Lucas, 1988; Ram, 1999; Rousseau & Wachtel, 2005). Joan Robinson posited that financial services act as a response to economic conditions, where markets will develop in order to accommodate a growing economy (Robinson, 1952). While it is acknowledged that efficient financial markets help to facilitate transfers of capital to productive sectors, the narrower impact of stock markets is not so certain. Stock markets facilitate funds transfers to a different set of companies and investors than banking channels do. The operation of a stock market may not provide a net benefit if banks are able to provide the services necessary (Arestis, Demetriades & Luintel, 2001). I argue that the relationship 12 between stock markets and growth is exaggerated and that opening a stock market is not a good predictor of whether an economy will grow. On the heels of the financial development literature and the dominant ideology of the liberalization of capital and trade accounts, policy prescriptions pushed the opening of stock markets. These recommendations precipitated the large number of openings during the 1990s, as shown in Figure 2.1. That so many stock market openings occurred in the 1990s out of distinct policy changes is indicative that these openings were not driven by economic growth. The dominant ideology of the period was that the absence of a stock market hindered growth opportunities that could be partially solved by the opening of a formal exchange. The number of openings during this period provides a natural experiment in the effect that if stock markets do cause economic growth, the periods subsequent to the openings should be accompanied by per capita growth. For this reason, I am empirically testing whether the act of opening a stock market has had a subsequent impact on the economies of various countries. In empirically testing whether the simple act of opening a stock market influences growth, I use two complimentary Bayesian methodologies-Extreme Bounds Analysis (EBA) and Bayesian Model Averaging (BMA). EBA is a global sensitivity analysis able to determine the precise bounds a variable can take based on a given dataset. This is a robustness check in that many variables can have coefficient values that are both positive and negative. Knowing the bounds allows for the conclusion that, given a dataset, some variables will always carry a positive or negative coefficient. If a variable has bounds that do not cover zero, there is no question as to its relationship. BMA compliments this methodology because of the uncertainty inherent in model selection. Debate often arises 13 over use of certain variables, or combination of variables, in model specifications. To alleviate some of the uncertainty, BMA computes the probability that a particular model will be the best model given the dataset. Then, based on the models selected, BMA takes a weighted average to find the posterior distributions of the variables. Rather than relying on T-statistics, BMA assigns the probabilities that variables are statistically different from zero. Ranging from zero for a variable that does not appear in any model specification to one for a variable in every selected model, the variables considered to be more important in the determination of the dependent variable will have higher posterior probabilities. The empirical results of this paper are straightforward. I find little evidence that the act of opening a stock market has any influence on growth. This result holds with the use of each of the estimation methodologies. The values of opening a stock market are fragile with extreme bounds falling on both sides of zero. BMA finds that the two measures of stock markets have posterior probabilities of 0.0 and are not statistically different from zero. In fact, BMA does not select either variable to be included in any of the top models selected. Neither methodology is able to find any evidence that the opening of a stock market has any influence on growth. The essay is organized first by discussing the theoretical and empirical literature on broadly measured financial development, followed by a targeted discussion on the impacts of stock markets on growth. The empirical section follows with a detailed explanation of the Bayesian techniques used, presenting evidence that the impact of opening a stock market has little impact on growth. 14 2.3. Finance-Growth Nexus Early theorists, such as Gurley and Shaw (1955), Goldsmith (1969), and McKinnon (1973) (formulating what is known as the McKinnon-Shaw hypothesis), argued that a developed financial sector is a crucial element in economic growth. The process of channeling funds from surplus units with excess capital to deficit units in need of capital should lead to increased productivity and growth. The McKinnon-Shaw hypothesis argues that an underdeveloped financial sector will constrain growth since entrepreneurs with profitable opportunities would be unable to access the capital necessary to grow their company, leaving numerous growth opportunities unexplored. Hicks (1969) argument is similar in that within development, the financing of innovative technologies is crucial, but requires an illiquid investment with higher risks to investors. The introduction of efficient financial markets has the effect of lower costs of financing these enterprises through increasing liquidity, resulting in higher levels of capital available to entrepreneurs. Lessening financial constraints for technologically advanced firms will lead to increased productivity and growth. Extending the McKinnon-Shaw hypothesis, further benefits of financial development have been identified and synthesized into three main functions. These are to allocate resources, mobilize savings, and reduce risks. Tobin and Brainard (1963) argued that well-functioning financial systems will lead to a more effective allocation of capital. In efficient capital markets, investors are better able to evaluate investments, thereby allowing entrepreneurs the ability to access capital at more favorable terms. This allocative efficiency is attained through the ability of financial intermediaries to obtain information at lower costs and being able to move faster to fund profitable projects. 15 Wicksell (1935) theorized that financial markets play the important role of matchmaker between savers and borrowers. The savings of individual households are typically not enough to fund large projects, so financial intermediaries have the ability to pool capital, making the entire aggregated amount available for lending. Many profitable investments require long-term commitment, but many investors are wary of tying their capital up for long periods of time. Whereas an underdeveloped financial system would require investors to invest over the long run, developed and liquid financial markets allow investors the ability to liquidate their investment in the event they need cash quickly. Liquid financial markets allow for the long-term financing of projects with short-term capital. Building on the early theoretical literature on the finance-growth nexus, empirical studies emerged in earnest in the 1990s. Prominent work by King and Levine (1993) supports the view that financial development positively influences growth, controlling for other factors that affect long-run growth. This study focused on banking variables, including credit to the private sector divided by GDP, to proxy for the level of financial development, and setting the dependent variable as economic growth. Demirguc-Kunt and Maksimovic (1998) extend this by showing that countries with more efficient legal systems will have more firms using funds from financial institutions. In an environment with an effective legal system and well-functioning financial markets, the countries experienced higher levels of productivity of capital, facilitating firm growth. The empirical work on countries' legal origins positively influencing financial development and economic growth has been confirmed a number of times using cross-sectional 16 (Levine 1998, 1999), time series (Djankov et al. 2007), and dynamic panel (Beck and Levine 2002, 2004; Levine, Loayza & Beck, 2000) methodologies. While the relationship between finance and growth was becoming well known, the early studies did a poor job of correcting for endogeneity; the assessment of causality was setting up cross-sectional regressions and concluding that the explanatory variables caused changes in the dependent variable. Thus the need to account for endogeneity issues was accomplished first through the use of instrumental variables. Demirguc-Kunt and Maksimovic (2002) and Levine (2002) find, using legal structure as an instrumental variable, that the development of financial markets influences the level of economic growth. Both studies' results did not provide different results for firms' access from either a bank or market-based financial system; it did not matter what the primary source of financing is, but rather how developed the financial systems are. Strengthening the results, a number of studies assessing the causality have arisen using time series models. For example, Choe and Moosa (1999), using VARs and Granger causality, found that causality runs from financial development to Growth during the period 1970-1992 in Korea. This result primarily supported the role of financial intermediaries rather than capital markets. While a large amount of literature finds a positive relationship between financial development and economic growth, not all researchers share this opinion. A number of studies propose that there are specific conditions where financial development will positively influence growth. Rousseau and Wachtel (2002), using a fixed effects panel model, found that in countries with high inflation, financial development does not seem to increase growth. The inflation threshold is that in countries with annual inflation rates 17 under 8%, the effects of financial development are significantly positive; conversely, when inflation rates are above 13%, there does not appear to be a relationship, while inflation rates in between are ambiguous. Rioja and Valev (2004) found that financial development's influence depends on how developed the countries are when implementing the reforms. Using a Generalized Method of Moments (GMM) model, Rioja and Valev divide their sample of 74 countries into three groups based on their starting level of financial development. Their findings indicate that finance has a robust positive impact on countries that are already well developed, while an ambiguous relationship exists in less developed countries. These threshold results are also found by Deidda and Fattouh (2002) regarding the level of per capita income. With the full 80 country sample, a positive relationship between financial development and economic growth is found. However, financial development appears to only influence growth in wealthier countries, but not in low income countries. These results all present evidence that financial development is not a blanket solution that helps every country in every situation; they highlight that there are circumstances where financial development may be beneficial and others when it is not. In light of the studies finding that finance is not a blanket solution to development problems, a number of researchers contend that financial development is irrelevant, having no influence on growth. A famous example is the Robert Lucas comment that finance is "very badly overstressed" and "is not inclined to be apologetic" for its exclusion from his growth model (Lucas, 1988, p. 6). This opinion is empirically supported by Ram (1999) who, in using a sample of 95 countries, finds that the correlation between financial development and growth is weakly negative or negligible. 18 There were also similar findings when separating the countries by income cohorts as well as grouping the countries according to growth rates. Extending the research contesting that financial development may not be the gateway to growth, Rousseau and Wachtel (2005) examine the relationship between financial depth and economic growth using cross sectional and panel data for 84 countries between 1960 and 2003. Three different measures of financial depth were used, with the updated time period not presenting as robust of findings as earlier studies. Their principle finding is that while the relationship may have existed through the early 1990s, it appears to have diminished in the later periods. The authors compared the relationship between finance and growth to the Phillips Curve, where an observed relationship was believed to be an empirical regularity before the relationship disintegrated. These conclusions were to act as a reminder that the correlations between finance and growth may well represent cross-country differences rather than a causal relationship. Although a statistical relationship between finance and growth may exist, some theorists question the direction of causality. Joan Robinson contended that finance does not lead growth, but rather that financial systems act in response to economic conditions (Robinson, 1952). As the economy is expanding, firms and households will have more demand for financial services, which will be provided by profit-seeking financial institutions. A number of other studies have been able to show that causality does not always run from finance to growth, with results being found that the direction of causality can also run in the reverse direction (Ang & McKibbin, 2007; Demetriades & Hussein, 1996; Demetriades & Luintel, 1996). Arestis and Demetriades (1997) find that causality varies substantially across countries. Their results showed that in Germany, causality 19 runs from financial development to growth, while the opposite result was found for the United States for the period 1979 to 1991. 2.4. Stock Markets and Growth Where many of the above studies evaluate the link between financial development and growth, our concern is specifically related to how-and whether-stock markets can facilitate economic growth. The theoretical basis behind the introduction of stock markets largely follows the McKinnon-Shaw argument behind the development of financial markets. Specifically, the opening of a stock market will increasingly allocate capital for long-run investments with short-run capital due to the ease with which investors are able to remove their investment. This should effectively reduce transaction costs, increase capital accumulation, and lead to higher levels of economic growth. Building on the early theoretical work on the relationship between finance and growth, Levine (1991) expands the McKinnon-Shaw hypothesis to specifically model stock markets' impact on per capita growth. This model is able to demonstrate that stock markets help to facilitate technological innovation and economic growth through an easy transfer of ownership that does not disrupt the productive capabilities or cash flows of firms. Levine emphasizes the positive impact of stock market liquidity on long-run growth, empirically showing that taxes on stock market transactions are associated with lower economic growth. Atje and Jovanovic (1993) have similar findings in that stock markets are associated with higher income levels as well as positive economic growth effects. This finding supports the view that of the elements of financial development that are most 20 effective are stock markets as similar growth effects were not observed for bank lending. These empirical results were performed in a similar manner to Levine (1991), evaluating liquidity for a sample of 94 countries with annual observations during the period 1960- 1985. Levine and Zervos (1998) and Rousseau and Wachtel (2000) expand the sample size of the earlier studies and, using different econometric techniques, come to similar conclusions as the earlier studies. Both papers show that stock market liquidity and banking development can predict growth levels. On the other hand, neither study was able to find evidence that stock market capitalization as a percentage of GDP has a relationship with growth. Levine and Zervos (1998) were also unable to find evidence that any of the financial indicators used had any relation with private savings rates, concluding that the data showed that stock markets play a different role than banks. Acknowledging that banks and stock markets play different roles, Arestis et al. (2001) finds that both stock markets and banks are able to promote economic growth. However, using a time series analysis over the period 1972-1998, the authors find that the stock market provides relatively little impact on growth as compared to that received from banks. They argue that the impact of stock markets is exaggerated by the use of cross-country growth regressions. Directly assessing the possible endogeneity between stock market development and growth, Caporale, Howells, and Soliman (2005) found that causality runs from stock markets to growth. These results were found using Vector Auto-Regression models and WALD tests to the effect that growth is influenced through effects on investment efficiency. Bringing capital to the firms in technologically advanced industries with high profit opportunities is a fundamental reason behind why stock markets should exist. 21 Beck and Levine (2002) support the notion of stock markets' principle purpose being to channel funds to high productivity sectors. Empirical results showed that a more developed financial system is correlated with higher levels of economic growth regardless of whether the country's financial system is bank-based or market-based as each type of financial service is designed to accommodate different sectors. These results are also unable to find evidence that stock market capitalization causes higher levels of economic growth, concluding that it is not the number of companies listed on the stock exchange, but rather it is that the stock market exists. The size of the stock market is irrelevant to growth so long as it is able to facilitate the transfer of capital to the high productivity sectors. The question of whether it is the size of the stock market or whether its existence is the important factor has begun to be addressed in recent years. Baier, Dwyer, and Tamura (2004) evaluated the effect of opening a stock market on productivity growth as measured by total factor productivity (TFP) and its subsequent impact on economic growth. They argued that the mechanism through which a stock market influences economic growth is not through capital accumulation, but is from changes in the growth rate of productivity. Baier et al. did not find any statistical difference in economic growth in the periods before and after the stock market opens. The support they found for opening a stock market is that productivity growth (TFP) increases in the period after the opening. However, they did acknowledge that point estimates suggest that countries have slower economic growth after opening an exchange. Even in the presence of conflicting results, they concluded that opening an exchange generates faster productivity and economic growth. Minier (2009), on the other hand, looked directly at the effect a 22 stock market had on growth, showing a statistically significant positive result of growth in the first 5 years after opening a stock market. This study was conducted by comparing growth rates for the 5 years before and after the opening of the stock market. Enders (2004) has shown that such tests are poorly designed because successive values of GDP are serially correlated; some of the effects of a pre-stock-market economy could carry over to the period after the stock market is opened. Baier et al. (2004) and Minier (2009) provided evidence that the opening of a stock market has a positive impact on productivity and growth. In contrast to these studies promoting the benefits of opening a stock market, a number of researchers have theorized and provided empirical results questioning the efficacy of opening a stock market. The most prominent is that of the destabilizing effects of opening a stock market. Keynes (1936) has argued that stock markets present too many speculative opportunities that are not complimentary to a stable, growing economy. Keynes' view was that stock markets were like a casino in which investors placed bets without full knowledge of the underlying components, pushing asset prices away from their fundamental value. After it is realized that these prices are above their fundamental value, the readjustment can have serious consequences on the real economy. This sentiment is echoed by Kindleberger (1978) where excessive speculation and high levels of leverage can cause a mania to occur. The immediate result is pushing asset prices up before a sudden loss of confidence causes a panic in the market where, if not controlled, can spiral into a crash. The instability of financial markets, contend Keynes and Kindleberger, can bleed over to the real economy as investors will remove their funds from being able to be put to productive use because of fear of losing their 23 investments. As such, firms become constrained by not being able to access capital, resulting in depressing the economy. Both felt as though the benefits of stock markets may not outweigh the costs to the economy. Where many of the studies pointed towards stock markets benefits being increased liquidity and the ease with which investors are able to remove their capital, Bhide (1993) found increased liquidity in stock markets holds hidden costs. His findings indicated that liquidity discourages internal monitoring because of information asymmetry problems. With a liquid market for a company's stock, there is little incentive for stockholders to monitor the firm's managers since dissatisfied investors have the ability to quickly rid themselves of their holdings at little cost. That a liquid market discourages monitoring by large investors causes a societal drag in that agency costs are amplified. These results were surprising to many, yet were subsequently supported. Harris (1997) found that the relationship between stock market activity and growth is weak at best for developing countries. He did find that there is a positive and statistically significant result for stock market activity on growth in the developed countries, but he "… finds no hard evidence that the level of stock market activity helps to explain growth in per capita output" (Harris, 1997, p. 139) for anyone other than the most developed countries. Harris' sentiments were echoed by Singh (1997), who argued that the expansion of stock markets through the 1980s and 1990s in the developing world hindered rather than assisted growth and development. This result insists that the opening of a stock market undermined the benefits accrued from removing the financial repression policies within developing countries. Singh also provided numerous historical 24 examples of economies that experienced high levels of growth in the absence of a functioning stock market. In light of the conflicting evidence on the impacts of opening a stock market, the question of whether the simple act of opening a stock market even influences growth is not solved. I hypothesize that the relationship between the simple act of opening a stock market and growth is overstressed. That there was a large number of openings in the 1990s indicates that these were not "organic" in the sense that this were distinct policy choices rather than emerging as Joan Robinson hypothesized that financial systems arise out of necessity to accommodate a growing economy. These were distinct policy choices essentially saying "open a stock market, the economy will grow" rather than emerging out of a growing economy. This acts as a natural experiment as we are able to test whether these openings have resulted in increased growth for these countries. I am empirically testing whether the countries that opened a stock market had higher rates of growth relative to the countries without a stock market. This will be able to effectively test whether it is the presence of a stock market that increases growth. 2.5. Empirical Methodology While a number of empirical studies point to a positive relationship between financial development and growth, there is a sizable literature contending that the relationship is irrelevant at best and negative at worst. This uncertainty is not unsurprising due to the numerous empirical methodologies and datasets that are used in the literature. Bayesian econometric methodologies have arisen to address these issues surrounding model uncertainty. Depending on the methodology employed, parameter 25 estimates can change, especially considering a total of 2 different specifications are possible (where k is the number of explanatory variables). Since we are uncertain of the specific impact of certain variables on our dependent variable, Bayesian statistics assigns probabilities. Changing model specifications can lead to changes in maximum likelihood estimates, which is addressed through the generation of models and variables with higher posterior probabilities. In order to address this uncertainty surrounding the impact of opening a stock market on growth in poorer countries, I use two complementary Bayesian methodologies: Extreme Bounds Analysis (EBA) and Bayesian Model Averaging (BMA). EBA is able to compute the range of possible values of maximum likelihood estimation over every possible model specification and all specified combinations of explanatory variables. Knowing that these extreme bounds (minimum to maximum) cover every possible value a coefficient may take, given a set of variables, is a stringent test. In order to "pass," the variable's bounds must not cover zero, being either strictly negative or positive. The result is that those variables that do pass are statistically different from zero, being concluded that this relationship will be different from zero given the data and specifications. BMA is also able to address model uncertainty by averaging over a set of Bayesian estimates and assigning posterior probabilities of a specific model being the best fit. BMA then averages the coefficients of variables for the models they appear in. These estimates are a posterior distribution, with the posterior mean being the expected value (EV) of the variable. BMA also relies on assessing the posterior probability that the variables' impact is not zero; variables not included in any models are assumed to have a minimal impact as they would have a posterior probability of zero. 26 I estimate an empirical growth model paying special attention to the impact of stock market openings. Levine and Renelt (1992) examined the empirical work on growth using a new (at the time) procedure known as Extreme Bounds Analysis to test for robustness on the empirics of the growth literature. Results from these estimations were that most variables were fragile, only finding two robust correlations with growth: share of investment in GDP and ratio of international trade to GDP. Sala-i-Martin (1997) acknowledged the benefits of the approach, but provided strong criticisms, saying "…the test is too strong for any variable to pass it….Thus, giving the label of non-robust to all variables is all but guaranteed." (Sala-i-Martin, 1997, p.179) He then averaged over two million regressions a likelihood-weighted sum of normal cumulative distribution functions. This rudimentary approach was the predecessor of Sala-i-Martin, Doppelhofer, and Miller (2004), which used a model averaging approach known as Bayesian Averaging of Classical Estimates (BACE). This process takes the Bayesian concept of averaging across models and combines it with classical OLS estimations. Adding to the literature on Bayesian estimations of growth theory is Fernandez, Ley, and Steel (2001), who used Bayesian Model Averaging to confirm the conclusions of Sala-i- Martin (1997) in that some variables are robust, having some explanatory power, rather than the limited conclusions of Levine and Renelt (1992). Applying BMA to the Sala-i- Martin dataset reduced much of the uncertainty of the initial estimates. Fernandez went substantially further as clear interpretations of the data were able to be inferred. This paper takes a purely Bayesian approach, combining the well-known techniques used by Levine and Renelt (1992) and Sala-i-Martin et al. (2004). Where my approach differs from Sala-i-Martin et al. (2004) is that instead of averaging over 27 classical estimations with BACE, I use Bayesian Model Averaging (BMA) which averages over Bayesian estimates. BACE is derived from this purely Bayesian technique of BMA and is typically used when presenting to those not familiar with Bayesian techniques. The reason for the divergence is that BMA is better equipped to handle model uncertainty. Additionally, the strength of statistical packages (R) in recent years has allowed for the creation of detailed image plots for a more intuitive presentation of the results. Given the Sala-i-Martin (1997) criticisms discussed earlier of the Extreme Bounds approach, it is still a useful tool and, combined with BMA, presents a compelling way to view the data. If a variable is robust with bounds that do not cover zero, it is then absolutely certain of the direction of correlation. This combination of techniques uses EBA to test for robustness and BMA for assessing "importance." There are numerous studies evaluating the determinants of growth, but I make no attempt at reconciling them. While some results are presented on the relationships between certain indicators and growth, the impact of this paper is the determination of whether the simple act of opening a stock market can positively influence growth. 2.5.1. Extreme Bounds Analysis The variations in empirical results summarized in the literature above are not surprising. Econometric analyses often differ due to varying data, model selection, or statistical techniques. In this paper, I begin with a global sensitivity analysis as introduced by Leamer (1978, 1982, 1983, 1985, 1997). Extreme Bounds Analysis (EBA) is a Bayesian methodology of global sensitivity analysis able to compute the range of values of a coefficient. This is done in the context of linear regression models by 28 computing the extreme values using maximum likelihood estimation procedures under all possible combinations of variables in the dataset. EBA computes coefficients for all possible linear combinations for a set of variables; the extreme bounds are the maximum and minimum posterior mean estimates. This methodology is a rigorous test of robustness as few variables are able to survive and show a definitive impact. EBA computes the possible bounds for the posterior mean for a normal linear regression model, given by the standard equation: Y = X + (2.1) In this equation, the X matrix contains the variables to be included in the model specification. In EBA parlance, these variables are referred as being "free" variables. Free variables are those that are not properly specified and not associated with a prior specification. These variables are those that would typically be included in a model specification. These would be the list of variables that would be reported in the final results, while showing combinations within this set. This X matrix would also necessarily include the variables of interest in addition to the control variables. The selection of these variables would be to include those variables traditionally used in a classical model specification within the literature. As with any model specification, the selection of variables is subject to debate as any combination of variables (sometimes selected in a seemingly arbitrary way) can be used. However, there always exist additional variables that could be included in a model specification. When we exclude a variable from a model, the expected impact of this 29 variable on the dependent variable will be precisely zero. In this way, all regressions use priors; excluding a variable from a model essentially says that the econometrician believes the variable is not important and will not be considered to have an impact. The problem is that some of these variables can alter the impact of variables in the X matrix on the dependent variables. EBA is able to account for the range of values the explanatory variables will have on the dependent variable. The variables that are typically dropped in a model specification are referred to as "doubtful" variables and are added to the standard linear regression as the Z matrix. Y = X + Z + (2.2) Interest is in the set of coefficients, but the selection of variables in the Z matrix are able to influence the range with which B values are able to take. The choice of a set of Z variables, called "doubtful" variables, allows for the calculation of the extreme posterior values for the coefficients associated with the X variables. That sets of variables (Z) might be dropped from a regression induces a coherent prior, rather than an improper prior for free variables, on the coefficient vector, . Whereas in traditional cross-sectional specifications we would just drop these variables, they are still included as doubtful variables to see their influence on the free variables. Setting a variable to doubtful is a twist on proper prior specification by setting the prior mean equal to zero, representing the belief that these variables could be dropped from the model specification. Because the mean of these doubtful variables is set at zero, the doubtful 30 variables' extreme posterior means would necessarily include zero. For this reason, the variables of interest must always be free; otherwise, they would never pass. With a large number of explanatory variables there are an exponentially large number of doubtful/free combinations. As such, we group variables into various categories according to our prior beliefs about their impact. The combinations would start with including only the variables of interest as the free variables, with all others set as doubtful. This will give the widest bounds as the combination of explanatory variables is the greatest. The other combinations would be to include a set of variables that are typically included in a classical estimation as free. There are circumstances where debate about the types of variables are most important arise; a solution with EBA would be to use one set of variables as free, with the others doubtful in one specification and switching these around for the next. Using every possible combination of variables is cumbersome and unwieldy, so the selection of variables must be reduced to a reasonable number. 2.5.2. BMA Overview The second estimation procedure used is Bayesian Model Averaging (BMA) as discussed by Raftery, Painter, and Volinsky (2005) and Hoeting, Madigan, Raftery, and Volinsky (1999). Since EBA provides the bounds a coefficient can take, BMA quantifies the value of many different models to compute the posterior distributions to help select the best models. These Bayesian methodologies are complementary in that one checks and verifies robustness, while the other ranks models according to their explanatory power. In econometrics, one of the most difficult and contentious issues is in model 31 selection. This uncertainty can lead to misspecification errors and erroneous conclusions. A major issue in the use of linear regression models with ordinary least squares is that the inclusion of additional variables is not discounted in the search for a high . With the rise of high computer power and statistical software it becomes possible to run thousands of models while adding or subtracting variables until the desired result is discovered (as shown by EBA). In BMA, however, the economist reduces this uncertainty by including all variables that could have an impact and allowing an established algorithm to select the most appropriate models. Utilizing the leaps and bounds algorithm developed by Raftery et al. (2005), the exponential number of models is reduced to a workable model space. Leaps and bounds returns a set of the best models that are then ranked according to their Bayesian Information Criteria (BIC). These models that make it in the set of best models are assigned posterior probabilities of being the best model out of the set. Those ranked higher, with lower BIC values, receive higher posterior probabilities. The variables included in these models are assigned posterior probabilities of being included in the best models, with coefficients being the posterior distributions. This differs greatly from the traditional quest for statistical significance in that rather than arbitrarily adding or subtracting variables, the established algorithm chooses the most representative models. Underlying BMA is the desire to average over all possible models. However, the computations required to perform the exponential number of computations is unwieldy and was long limited by computing power. Two difficulties for the implementation of BMA arise; the first involves computing the integrated likelihood function, which is obtained by integrating the unknown parameters. The second is averaging across all 32 models, which is exponentially large. The first computational problem is solved by using the BIC approximation that does not rely on difficulty to compute high dimensional integral, while the second is corrected by the leaps and bounds algorithm. It is unnecessary to compute the models that stand no chance of being the best model. Thus arose the need to reduce the number of models into a workable set, rather than the exponentially large number of models to average across. Reducing the models is accomplished by the fast leaps and bounds algorithm as introduced by Furnival and Wilson (1974), and made applicable to BMA by Raftery (1995). Once the models have been narrowed down, selection of the best models comes by way of using the Bayesian Information Criteria (BIC). This process is similar to evaluation by the Akaike Information Criterion (AIC) but differs in that the penalty for adding variables under BMA is much less than under AIC. The following equation gives the calculation for BIC where is the value of the computed and is the number of independent variables regressed on model k. = log 1 − + log (2.3) Adding variables is discounted in BMA as the models are penalized for adding explanatory variables and rewards models with better explanatory power. The lower the value of BIC, the better the data fit, with the best model having the lowest BIC. Ranking the models according to BIC solves the primary problem of computing integrals in multiple dimensions. After the reduction of the model space and ranking of models according to BIC, it must be acknowledged that the top ranked model may not represent 33 the data the best. For those things we are uncertain about, we assign probabilities. As such, we are able to compute posterior probabilities that a particular model will be the best. In a BMA analysis, there are two posterior probabilities that are computed. The first is the probability that a particular model specification will be the best, while the second is the probability that a specific variable will be in the model that tells the best story. Posterior probabilities of being the best model are calculated as Δ = Δ , (2.4) where is the unknown quantity of our dependent variable and is a given matrix of available data. Δ , is the posterior distribution of given the model and is the posterior probability that provides the best fit. BMA determines this posterior distribution of as a weighted average of the posterior distributions of the models. The models with the lowest BIC will have the highest posterior probabilities down to the lowest probabilities being assigned to the lowest ranked model in the reduced leaps and bounds model space. Once the posterior probabilities of the models being the best fit have been calculated, BMA computes the posterior probabilities of the individual variables. The computations are the probability that the coefficient attached to the variable is not zero. The higher this value is, the higher the likelihood that the variable differs from zero. Variables with posterior probabilities of 100% are included in every model. The posterior probabilities of differing from zero are the sum of the model posterior 34 probabilities the variables are included in. Variables with posterior probabilities equal to zero are not included in any models and, with the employed data set, cannot be concluded to have much of an effect on the dependent variable. The expected values of the coefficients are referred to as the posterior mean, which is a weighted average of the posterior means from each model. These values are weighted by their posterior probabilities, where those with a higher likelihood of being in the best model carry a higher weight on the posterior mean. For variables with little impact, these posterior distributions will be centered on zero. The lower the posterior probability, the higher the likelihood is that the variables impact will have no effect on the dependent variable. For the variables that are statistically different from zero, the density of the coefficients should fall on either side of zero. The standard deviation provides for a level of confidence that the distribution is significantly different from zero. If a one standard deviation change from the posterior mean does not cover zero and the variable has a high posterior probability, it can be considered an important indicator. 2.6. Data If stock markets are able to promote, or stall, growth, the opening of a stock market should see subsequent changes in growth rates. The conditions through the 1980s and 1990s that saw a large number of stock markets open was a distinct change in sentiments rather than emerging out of a growing country's need for additional financial services. As such, this natural experiment allows for the empirical testing of whether a stock market impacts growth. In formally testing my hypothesis that opening a stock market does not create a positive shift in economic growth, I use both EBA and BMA 35 methodologies. I use a dataset with 82 countries and 32 independent variables that are typically found in the finance-growth literature. This dataset will serve both EBA and BMA methodologies. Stock markets have been hypothesized to be able to increase growth through a number of mechanisms and should be accompanied by a permanent increase in growth. The absence of differences in annual growth rates would be indicative that just opening a stock market is not enough to cause growth. Conversely, a negative relationship would provide evidence that the speculative costs of stock markets outweigh the benefits. The dependent variable in this study is the average annual per capita GDP growth rate between 2002 and 2007 as measured in 2000 US Dollars. This time period was chosen to smooth any fluctuations and to give time between the last openings of the sample in 1999 and the start of the growth period in 2002. The time spacing was necessary as there may be a lag between the time a market opens and increases in financial activity will extend over to economic growth. Using two variables-Stock.Dummy and Years.Open-I am able to estimate the impact of opening a stock market on growth. Stock.Dummy is a dummy variable equal to one if a stock market is present and zero if not. This simplistic measure is able to test the simple question of whether the existence of a stock market accelerates growth. It is necessary to include countries that do not currently have a stock market as a control group. There are 59 countries identified that have opened a stock market between 1960 and 2000. 36 countries that do not currently have a stock market have been included as a control group, for a total of 95 countries as shown in Table 2.1. Because of limitations on data, the total sample has been reduced to 82 countries. 36 Although the simplistic dummy variable should be enough to assess whether the presence of a stock market causes a permanent increase in growth, there is likely a lag between the time a stock market is opened and growth is impacted. Years.Open is a measure counting the number of years the stock market has been open in 2010. Countries without a stock market have a value of zero. In this way, we place a higher weight on stock markets that have been open longer. Of the countries with a stock market, the lowest number of years open is 11 since the cutoff for inclusion of opening a stock market was the year 2000. The implicit assumption is that the longer a stock market has been open, the more of an impact it is able to have as funds get channeled through the real economy. The longer the market has been open, the more time it has had to cause a permanent shift. If the existence of a stock market influences growth, positively or negatively, one of these two variables should play a significant role in the regressions. The existing growth literature has discovered a large number of variables with significant effects on growth. These were then narrowed by selecting the variables with the widest availability across countries and highest reliability, resulting in 30 control variables. Variables used in this model are closely related to those used in the growth literature, but may be specified slightly differently. Variable descriptions are shown in Table 2.2 and take into account such issues as property rights, human capital, infrastructure, and monetary assets and flows. As with any econometric exercise, there is always room for debate about whether the correct variables were used; it is possible there are considerations that influence growth that could have been excluded. Some specific variables that have been suggested are other aspects of stock or bond markets. There are a number of measures of financial 37 development that would have been able to be used, but many of these could have masked the effects of stock markets. Specifically, aggregated measures of financial development, such as private credit/GDP or Liquid Liabilities/GDP are influenced by stock markets and would be inappropriate for use because of the possibility of masking the effects of stock markets. It was also decided not to use any of a number of metrics on bond markets because of the purpose of this study being to evaluate the impact of opening a stock market. While tempting to use variables related to stock market development (liquidity or stock market capitalization), the choice was made to exclude these as they would detract from the purpose of defining the contribution of opening a stock market. If the existence of a stock market is what is important, as argued by Beck and Levine (2002), the opening of a stock market should be accompanied by subsequently higher levels of growth. 2.7. Results 2.7.1. EBA Results The combinations of free and doubtful variables have been broken into: a social/political set where these social and political effects can be captured as free variables; a financial set that sets the financial variables as free with the others as doubtful; and one with both the stock market dummy and years open set as free with the others set as doubtful. The reason for running multiple EBA estimates is to check for robustness and to address the possibility of not including adequate variables. In our search for the value of a particular parameter, we first wish to know the direction of correlation. If it is possible to generate both positive and negative coefficients, doubt 38 arises as to the value generated. The free variables whose bounds do not cover zero are robust in that the coefficient will always have the same sign regardless of how the model is specified. The EBA model uses all 32 of the independent variables as discussed in the above section and summarized in Table 2.3. Where there is not an entry in the table the variable was set as doubtful, while each value reported is set as a free variable. Those variables that are robust and do not cover zero have bolded results in the table. These variables include: GDP in 1992, life expectancy, rural population, government effectiveness, export index, expected levels of schooling, and the Human Development Index. To development scholars, the fact that these variables are robust should not be a surprise. What may be surprising to some is the direction of these coefficients. Life.Expect and Exp.School are both negatively robust; indicating that as schooling and life span increase, growth is expected to decrease. Additionally, percentage of population living in rural areas (Rural.Pop) carries a positive value, indicating that highly urbanized countries have lower growth rates. One interpretation behind these results is that controlling for the level of HDI, the negative coefficients attached to education and life expectancy could indicate an unbalanced HDI. For example, if HDI remains constant and education levels were to rise, life expectancy would necessarily drop. In economies with unbalanced HDIs one indicator would be significantly larger than the others. When a scenario where education is outstanding, but with lackluster per capita income and life expectancy, the economy would be unable to leverage the higher levels of education and face bottlenecks from the elements with low levels. These human development factors are a rapidly emerging field 39 of study and with more data availability in the future, research could further address and quantify the discrepancies between human development and growth. Outside the explanation given by the individual factors of HDI being included in the same models, there is the possibility of poorer countries "catching up." This would provide support for convergence as theorized by Solow (1956) and supported by Barro and Sala-i-Martin (1992). These theories posit that poorer countries will have faster growth rates because they have the ability to use production techniques pioneered in the developed world and that diminishing returns to capital and labor are not as strong as they are in rich countries. The coefficients attached to education, life expectancy, and rural population are characteristics of poor countries and can be attributed to this catching up factor. Each of the models shows that while the bounds may vary, they will always cover zero. Although fragile, the maximum likelihood estimate for both is negative, indicating a weak result that opening a stock market is associated with lower levels of economic growth. This is especially evident in Model 2 as the positive values are only generated far into the tail. Knowing that these variables can generate a coefficient in any direction warrants a fuller investigation into their effect on growth; Bayesian Model Averaging is able to quantify the impact. 2.7.2. BMA Results The 32 explanatory variables included in this model can yield 2 (4,294,967,296) total models. Even having the "right" dataset may not yield an accurate model due to the vast possibilities for model selection. It is for this reason that many 40 economists are unsure of other economists' results and why procedures such as BMA have arisen. The model presented in this paper attempts to objectively quantify the elements important in accelerating growth. The results of BMA return the top 73 models narrowed by leaps and bounds; Table 2.4 displays the top five models. The p!=0 column is the posterior probability of that variable being included in the model, EV is the BMA posterior mean, and SD is the posterior standard deviation of each variable. The values for , BIC, and the posterior probability for each model being the best model are shown on the bottom of Table 2.4. The higher the posterior probability, the higher the likelihood the impact of the variables is not zero. The coefficients on each model can vary widely, and therein lies the averaging portion of BMA; the expected value of the mean (EV) is computed by a weighted average of each coefficient and the posterior probability when the variable enters into a model. When a variable carries a posterior probability of zero, this indicates that it was not included in any of the models and is not statistically different from zero. Therefore, their expected value is equal to zero, having no impact on growth. Figure 2.2 shows selected variables' posterior distributions. The vertical black line at zero is the probability of the variable not being in a model with the curve being the coefficient's model averaged posterior density given that the variable is included in a model. This curve has been scaled such that the height is the probability of being included in a model; the heights of both the model averaged posterior density and posterior probability will equal one. The plots without a vertical black line are those considered to be included in a model 100% of the time according to posterior probabilities. Charting this output of the BMA posterior distribution provides slightly more information than the summary and offers another way of viewing the results. What 41 should be immediately noticed is that both stock market variables have large spikes at 0.0, indicating that these variables are not included in any of the models and are not statistically different from zero. On the other hand, there are a number of variables that show posterior distributions that are significantly taller than the spike indicating that their likelihood of being included is greater than being excluded. Another feature of the BMA statistical package in R is the ability to produce an image plot as shown in Figure 2.3. This output is easier to understand for those not familiar with the procedure. Variables are listed on the vertical axis with the rankings of the models along the horizontal axis. The variables selected are shown in their rows by being either red or blue. Red indicates a positive influence while blue indicates a negative correlation. Variables included in a particular model (1-73) are highlighted in either red or blue while those not included are not highlighted. The width of the column is proportionate to its posterior probability for each model; higher ranking models have higher posterior probabilities and wider columns. The total width of all columns is equal to the cumulative posterior probability as reported in Table 2.4. The image plot (Figure 2.3) is quite influential in showing which variables are important indicators due to the abundance of color associated with these variables. The variables with high posterior probabilities-GDP.1992, Avg.Infl, Life.Exp, RQ, nrbloan, Exp.Index, Exp.School and HDI-appear prolifically in the image plot with signs consistent with theory. The exceptions are Exp.School and Life.Exp for the reasons discussed above. Comparing the EBA results to BMA is a two-stage process in verifying the importance of certain variables. If a variable has bounds that do not cover zero and high posterior probabilities, it can be concluded that it is an important indicator without 42 ambiguity as to the direction of correlation. This dataset has four variables that are robust with high posterior probabilities; these are Exp.Index, HDI, Exp.School, and Life.Exp. Each of these variables is robust as measured by EBA and important as indicated in BMA. Exp.Index is an index of exports with the base year set at 100. Countries that have imported more than they exported have a value less than 100, while countries with a trade surplus have values greater than 100. Exp.Index has a posterior probability of 100, is included in every model, and has a positive correlation with an expected value of .043. As confirmed by EBA, this variable is robust in that it does not cover zero and is an extremely important indicator of growth in my sample of countries. This provides the expected result that countries with higher levels of exports will tend to grow faster. The Human Development Index (HDI) is a composite statistic of life expectancy, education, and income computed by the United Nations Development Programme. This variable has a posterior probability of 100.0 and carried the largest coefficient with an expected value (EV) of 25.68 and Maximum Likelihood (from EBA) of 32.443. The size of the coefficient is indicative of the weight this variable holds, the other part being due to the small size of the HDI variables (between 0 and 1). Two other variables deemed important are Exp.School and Life.Expect, which are used in the computation of HDI. Exp.School is a measure of the expected years of schooling. This variable proxies for human capital as the assumption is that higher levels of education lead to a more productive workforce and higher rates of growth. Importantly, the posterior probability is 64.5 with an expected value of -0.347, and this variable was found to be robust by EBA. This negative coefficient lies in contrast to our expected result. The most logical 43 explanation is in regards to the collinear aspects as described above. The same negative relationship is observed between Life.Expect and growth. Life.Expect is the life expectancy at birth in number of years. It would be expected that this variable would have a positive effect on growth due to this variable being a component of human capital. A healthier society should be more productive over the long run as there is less time spent away from work because of personal illness or an illness of a family member. The posterior probability of being included is 83.3 with an EV of -0.16 and a robust Maximum Likelihood -0.1995. Because Life.Expect, Exp.School, and HDI are highly collinear, it is challenging to derive individual conclusions on these variables. Possible explanations discussed above regarding the directions of correlation are that an unbalanced HDI results in slower growth or that the poorer countries with lower values on the individual components are "catching up." One question that arises is how BMA handles multicollinearity issues. The collinearity is handled in the model because they are not perfectly collinear and by the fact that multicollinearity does not reduce the predictive power of the model as a whole. Multicollinearity only affects calculations of individual variables as it may not give valid individual results. However, when the collinear variables are bundled together, the aggregate effects estimate is reliable and adequate for my purposes. Since I am not attempting to quantify the individual effects, the best way to deal with this multicollinearity issue is to leave the model as is. GDP.1992 captures the absolute level of GDP in 1992 and exhibits a posterior probability of inclusion at 18.6. This variable captures the size of the economy, and since it does not measure the number of people, does not incorporate relative richness. With an 44 expected value of 1.166e-11, it has an extremely small coefficient due to scaling as GDP is measured in billions. With a standard deviation of 1.018e-11, the majority of this variable's distribution is robust; it is only in one tail that it becomes negative. Because of the small size of the coefficient, EBA was unable to give bounds on this variable, but when combining the methodologies, absolute size of GDP is an important determinant of per capita GDP growth. GE and Rural.Pop are special cases: they have robust values as shown by EBA but are not important as shown by BMA. Just knowing that a particular variable is not fragile does not necessarily mean that it has much of an impact, as shown for these two variables. These variables, for the subsets of free/doubtful combinations, were not fragile. However, that these variables did not show high posterior probabilities is indicative that these variables are not an element of the top models. The most plausible explanation for this phenomenon is that these variables are reacting with other variables, where the results are skewed. Variables need to be able to pass both methodologies in order to be certain of the direction of correlation and that the impact differs from zero. Avg.Infl, nrbloan and offdep had fragile extreme bounds, but were shown to be important by the BMA methodology. Inflation's (Avg.Infl) impact on growth has been well documented in the literature. My results show that it is important (posterior probability equal to 71.8) and most likely has a positive correlation (EV of .02 and a SD of .015), meaning that higher levels of inflation indicate higher levels of growth.1 Non-resident bank loans (nrbloan) was also fragile, carrying a negative coefficient (-0.229), 1 This is an interesting result as this relationship is surely not linear, nor always positive. One explanation behind this result could be that it is an artifact of the time period examined where there was a high level of relative stability with only three outliers-Angola, Democratic Republic of the Congo and Liberia- experiencing extremely high inflation (above 200% annually). Angola was actually one of the faster growing in the sample with an average growth rate of 14.1% and average inflation of 222.3%. 45 yet was included in 31.5% of the narrowed models. This result implies that higher levels of offshore bank loans to GDP correlate with lower levels of per capita growth. The offshore deposits to domestic deposits ratio (offdep) follows the same traits with a posterior probability of 40.3, yet has a standard deviation that covers zero (EV of -0.033 and a SD of 0.044). These three variables appear important but offer low levels of certainty as to their sign because the extreme bounds cover zero. RQ, regulatory quality index computed by the World Bank Governance Indicators, has fragile extreme bounds but entered into the BMA output as reasonably important with a posterior probability of 29.5. The expected value coefficient for RQ is -0.519, yet it has a standard deviation of 0.88. This places a significant portion of the coefficient distribution on the other side of zero, effectively questioning its robustness. The less stringent checks for robustness as provided by BMA are able to question, and probably discard, this variable as an important determinant of growth. Rule of Law, RL, was another variable that did not pass the weaker robustness check in BMA but is of less concern because its posterior probability of 10.0 barely registers. The most interesting thing to note, and the focus of this paper, is that the variables Years.Open and Stock.Dummy show a posterior probability of being included at exactly zero-meaning that neither of these variables is important in explaining growth. Other financial and social considerations appear to be more important in understanding growth. This means that both measures of stock markets do not display any explanatory power and, therefore, cannot be concluded to be an important indicator of growth. The reliability of this methodology in predicting growth is rather high, reporting in excess of 0.5 on 46 the individual models. This fit is in keeping with much of the results in the existing growth literature. 2.8. Conclusion In light of the large body of literature that has arisen as to the benefits of financial development on growth, this paper empirically evaluates the relationship between opening a stock market and growth. Basic economic theory posits that effective, functioning financial markets will lead to more projects to be funded and fuel economic growth. However, a small number of theorists believe that the rapid expansion of financial services may not be the gateway it was hoped. When evaluating whether a stock market is beneficial, the benefits must be weighed against the costs. The first argument questioning the effectiveness of stock markets is that active stock markets can lead to instability, causing a drag on long run economic growth. The second argument that carries merit is that developing countries are unable to fully leverage the benefits to be gained from an active stock market, so they will have little to gain from one opening. It is this second argument that I have argued is the limit to the effectiveness of opening a stock market. In formally testing my hypothesis that stock markets are irrelevant for growth, I use two Bayesian econometric methods: Extreme Bounds Analysis and Bayesian Model Averaging. EBA is a global sensitivity analysis able to compute the extreme bounds with which a variable is able to take. EBA results show that the opening of a stock market, and the number of years that market has operated have fragile and nonrobust effects on economic growth. Bayesian Model Averaging has the ability to objectively take a dataset 47 and determine the probabilities with which variables are different from zero by averaging over a reasonable subset of possible models. Both the stock market variables exhibited posterior probabilities of being included in the models with the best fit of 0.0, showing that the simple act of opening a stock market is not indicative of growth (in either direction) in this sample of countries. The results from this study provide significant evidence that opening a stock market has little, if any, influence on growth. While there are certainly cases where stock markets can be beneficial to an economy, there is insufficient evidence that stock markets on the whole are able to promote growth. That some countries may be able to better leverage the use of a stock market deserves more attention in the literature. Individual characteristics of the instances where stock markets deliver societal benefits are likely the most important determinant of whether a stock market will influence growth. Further research is needed to evaluate the specific conditions where opening a stock market and other elements of financial development are beneficial for human development. 2.9. References Ang, J.B., & McKibben, W.J. (2007). Financial liberalization, financial sector development and growth: Evidence from Malaysia. Journal of Development Economics, 84, 215-233. Arestis, P., & Demetriades, P.O. (1997). Financial development and economic growth: Assessing the evidence. Economic Journal, 107, 783-799. Arestis, P., Demetriades, P.O., & Luintel, K. (2001). Financial development and economic growth: The role of stock markets. Journal of Money, Credit and Banking, 33, 16-41. Atje, R., & Jovanovic, B. (1993). Stock markets and development. European Economic Review, 37, 632-640. 48 Baier, S., Dwyer Jr., G., & Tamura, R. (2004). Does opening a stock exchange increase economic growth? Journal of International Money and Finance, 23, 311-331. Barro, R. J., & Sala-i-Martin, X. (1992). Convergence. Journal of Political Economy, 100, 223-251. Beck, T., & Levine, R. (2002). Industry growth and capital allocation: Does having a market or bank-based system matter? Journal of Financial Economics, 64, 147- 180. Beck, T., & Levine, R. (2004). Stock markets, banks, and growth: Panel evidence. Journal of Banking and Finance, 28, 423-442. Bhide, A. (1993). The hidden costs of stock market liquidity. Journal of Financial Economics, 34, 31-51. Caporale, G.M., Howells, P., & Soliman, A.M. (2005) Endogenous growth models and stock market development: Evidence from four countries. Review of Development Economics, 9, 166-176. Choe, C., & Moosa, I.A. (1999). Financial system and economic growth: The Korean experience. World Development, 23, 433-448. Deidda, L., & Fattouh, B. (2002). Non-linearity between finance and growth. Economic Letters, 74, 339-345. Demetriades, P.O., & Hussein, K.A. (1996). Does financial development cause economic growth? Time series evidence from sixteen countries. Journal of Development Economics, 51, 387-411. Demetriades, P.O., & Luintel, K. (1996). Financial restraints in the South Korean miracle. Journal of Development Economics, 64, 459-479. Demirguc-Kunt, A., & Levine, R. (1996). Stock markets, corporate finance, and economic growth: An overview. The World Bank Economic Review, 10, 223-239. Demiguc-Kunt, A., & Maksimovic, V. (1998). Law, finance, and firm growth. Journal of Finance, 53, 2107-2137. Demiguc-Kunt, A., & Maksimovic, V. (2002). Funding growth in bank-based and market-based financial systems: Evidence from firm-level data. Journal of Financial Economics, 65, 337-363. Djankov, S., McLiesh, C., & Schleifer, A. (2007). Private credit in 129 countries. Journal of Financial Economics, 84, 299-329. 49 Enders, W. (2004). Applied econometric time series (2nd ed.). Hoboken, NJ: John Wiley & Sons, Inc. Fernandez, C., Ley, E., & Steel, M. (2001). Model uncertainty in cross-country growth regressions. Journal of Applied Econometrics, 16, 563-576. Furnival, G.M., & Wilson, R.W. (1974) Regressions by leaps and bounds. Technometrics, 16, 499-511. Goldsmith, R.W. (1969). Financial structure and development. New Haven, CT: Yale University Press. Gurley, J., & Shaw, E.S. (1955). Financial aspects of economic development. American Economic Review, 45, 515-538. Harris, R. (1997). Stock markets and development: A re-assessment. European Economic Review, 41, 139-146. Hicks, J.R. (1969). A theory of economic history. Oxford: Oxford University Press. Hoeting, J.A., Madigan, D., Raftery, A., & Volinsky, C.T. (1999). Bayesian model averaging: A tutorial with discussion. Statistical Science, 14, 382-417. Keynes, J.M. (1936). The general theory of employment, interest and money. London: Macmillian. Kindleberger, C.P. (1978). Manias, panics and crashes. New York: Basic Books. King, R. & Levine, R. (1993). Finance and growth: Schumpeter might be right. The Quarterly Journal of Economics, 108, 717-737. Leamer, E.E. (1978). Specification searches: Ad hoc inference with non-experimental data. New York: Wiley and Sons. Leamer, E.E. (1982). Sets of posterior means with bounded variance priors. Econometrica, 50, 725-736. Leamer, E.E. (1983). Let's take the con out of econometrics. American Economic Review, 73, 31-43. Leamer, E.E. (1985). Sensitivity analysis would help. American Economic Review, 75, 308-313. Leamer, E.E. (1997). Revisiting Tobin's 1950 study of food expenditure. Journal of Applied Econometrics, 12, 533-553. 50 Levine, R. (1991). Stock markets, growth, and tax policy. Journal of Finance, 46, 1445- 1465. Levine, R. (1998). The legal environment, banks, and long-run economic growth. Journal of Money, Credit and Banking, 30, 596-613. Levine, R. (1999). Law, finance, and economic growth. Journal of Financial Intermediation, 8, 8-35. Levine, R. (2002). Bank-based or market-based financial systems: Which is better? Journal of Financial Intermediation, 11, 398-428. Levine, R., Loayza, N., & Beck, T. (2000). Financial intermediation and growth: Causality and causes. Journal of Monetary Economics, 46, 31-77. Levine, R. & Renelt, D. (1992). A sensitivity analysis of cross-country growth regressions. American Economic Review, 82, 942-963. Levine, R. & Zervos, S. (1998). Stock markets, banks, and economic growth. American Economic Review, 88, 537-558. Lucas, R. (1988). On the mechanics of economic development. Journal of Monetary Economics, 22, 3-42. McKinnon, R. (1973). Money and capital in economic development. Washington, DC: Brookings Institution. Minier, J. (2009). Opening a stock exchange. Journal of Development Economics, 90, 135-143. Raftery, A.E. (1995). Bayesian model selection in social research. Sociological Methodology, 25, 111-163. Raftery, A.E., Painter, I., & Volinsky, C.T. (2005). BMA: An R package for Bayesian model averaging. R News: The Newsletter of the R Project, 5, 2-8. Ram, R. (1999). Financial development and economic growth: Additional evidence. Journal of Development Studies, 35, 164-174. Rioja, F., & Valev, N. (2004). Does one size fit all? A reexamination of the finance and growth relationship. Journal of Development Economics, 74, 439-447. Robinson, J. (1952). The rate of interest and other essays. London: Macmillan. 51 Rousseau, P., & Wachtel, P. (2000). Equity markets and growth: Cross-country evidence on timing and outcomes, 1980-1995. Journal of Banking and Finance, 24, 1933- 1957. Rousseau, P. & Wachtel, P. (2002). Inflation thresholds and the finance-growth nexus. Journal of International Money and Finance, 21, 777-793. Rousseau, P., & Wachtel, P. (2005). Economic growth and financial depth: Is the relationship extinct already? SSRN Working Paper Sala-i-Martin, X. (1997). I just ran two million regressions. American Economic Review, 87, 178-183. Sala-i-Martin, X., Doppelhofer, G., & Miller, R. (2004). Determinants of long-term growth: A Bayesian Averaging of Classical Estimates (BACE) Approach. American Economic Review, 94, 813-835. Singh, A. (1997). Financial liberalisation, stockmarkets and economic development. The Economic Journal, 107, 771-782. Solow, R.M. (1956). A contribution to the theory of economic growth. The Quarterly Journal of Economics, 70, 65-94. Tobin, J., & Brainard, W. C. (1963). Financial intermediaries and the effectiveness of monetary controls. American Economic Review, 53, 383-400. Wicksell, K. (1935). Lectures on political economy, (Vol. 2). London: Routledge and Kegan Paul. 52 Table 2.1: Countries in Sample Country Year Open Country Year Open Country Year Open Albania Algeria Angola Armenia Bahrain Belarus Belize Benin Bhutan* Bolivia Botswana Brunei Darussalam Burkina Faso Burundi Central African Republic Chad China Comoros Congo, Dem. Rep. Congo, Rep. Costa Rica Cote d'Ivoire Croatia Cyprus Djibouti Dominican Republic Ecuador Egypt, Arab Rep. El Salvador Equatorial Guinea Eritrea* Ethiopia 1996 1999 N/A N/A 1989 1998 N/A 1998 1993 1990 1989 N/A 1998 N/A N/A N/A 1990 N/A N/A N/A 1976 1998 1991 1996 N/A N/A 1970 1997 1965 N/A N/A N/A Fiji Gabon Gambia, The Georgia Ghana Guinea Guinea-Bissau Haiti Honduras Iceland Iran, Islamic Rep. Jamaica Jordan Kazakhstan Kiribati* Kyrgyz Republic Lao PDR Lesotho Liberia Lithuania Macedonia, FYR Madagascar Malawi Mali Malta Marshall Islands* Mauritania Mauritius Micronesia, Fed. Sts.* Moldova Mongolia* Mozambique N/A N/A N/A 1999 1990 N/A 1998 N/A 1990 1986 1968 1969 1999 1993 N/A 1995 N/A N/A N/A 1993 1995 N/A 1996 1998 1992 N/A N/A 1989 N/A 1994 1991 1999 Namibia Nepal Nicaragua Niger Nigeria Oman* Papua New Guinea Romania Russian Federation Samoa* Saudi Arabia* Senegal Seychelles* Slovak Republic Solomon Islands Sudan Suriname Swaziland Tajikistan Tanzania Thailand Togo Tonga Trinidad and Tobago Tunisia Turkmenistan* Uganda Uzbekistan* Vanuatu* Yemen, Rep. Zambia 1992 1993 1990 1998 1961 1989 1999 1995 1992 N/A 1985 1998 N/A 1991 N/A 1995 1994 1990 N/A 1996 1975 1998 N/A 1981 1969 N/A 1997 1991 N/A N/A 1993 *Countries not included in model due to insufficient data: total countries include is 82. 0 10 20 30 40 50 1879 1889 1899 1909 1919 1929 1939 1949 1959 1969 1979 1989 1999 2009 More Frequency Year period ending at year listed Figure 2.1 Stock Market Openings over Time 53 Table 2.2: Variable Descriptions Variable Name Description Mean Std Dev Avg Growth Dependent variable - average growth of per capita GDP from 2002- 2007, expressed as percentage. 2000 constant dollars (World Bank) 4.305 4.104 Years.Open Number of years stock market has been open. Equal to 0 if no stock market 12.93 12.3 Stock.Dummy Dummy equal to 1 if country has stock market, 0 otherwise n/a n/a Low.Mid Dummy variable equal to 1 if lower middle quartile country, 0 otherwise (as defined by the World Bank) n/a n/a Upper.Mid Dummy variable equal to 1 if upper middle quartile country, 0 otherwise (as defined by the World Bank) n/a n/a Low.Income Dummy variable equal to 1 if lowest income quartile country, 0 otherwise (as defined by the World Bank) n/a n/a Per.Cap.GDP Per capita GDP in 1992, 2000 constant dollars (World Bank) 1946.3 3785.79 GDP.1992 Absolute level of GDP in 1992, 2000 constant dollars (World Bank) 19.2 bil 70.54 bil Avg.Infl Average Inflation for period 1997-2002, expressed as percentage (World Bank) 24.54 79.25 Tax.Rate Total amount of taxes payable by businesses after accounting for deductions and exemptions as a percentage of profits, 2007 - chosen for completeness of data. (World Bank) 57.36 54.76 FDI.Flow Absolute value of foreign direct investment flows (UNCTAD) 1082.62 5415.99 Paved.Road Percent of roads in country that are paved, average from 1995 - 2005 (World Bank) 37.99 29.93 Adj.Savings Adjusted gross savings - difference between Gross National Income and public and private consumption, 2002 (World Bank) 16.55 14.34 Ag.Land Percent of land area dedicated to agricultural produce, 2002 (World Bank) 41.53 22.45 Ag.Value.Added Total amount of agricultural value added per worker, constant 2000 dollar (World Bank) 3082.25 9314.6 Life.Expect Life Expectancy at birth in years, 2002 (UNESCO) 62 10.46 Pop.Growth Average population growth 1997-2002 expressed as a percentage (UNESCO) 1.64 1.21 Cell.Phone Number of cell phones per 100 people, 2002 (World Bank) 12.44 18.02 Curr.Acct Current account balance, BOP, 2002 (World Bank) 0.62 5.23 Rural.Pop Rural Population as percent of total population, 2002 (World Bank) 54.87 20.6 CC Control and Corruption Index, 2002 (World Governance Indicators, World Bank) -0.48 0.67 RL Rule of Law Index, 2002 (World Governance Indicators, World Bank) -0.5 0.73 RQ Regulatory Quality, 2002 (World Governance Indicators, World Bank) -0.39 0.74 GE Government Effectiveness, 2002 (World Governance Indicators, World Bank) -0.45 0.72 VA Voice and Accountability, 2002 (World Governance Indicators, World Bank) -0.41 0.79 Nrbloan Offshore bank loans relative to GDP, 2002 (BIS Statistical Index via Demirguc-Kunt and Levine, 2009) 0.49 2.99 Exp.Index Export index with 2000 as base year set at 100, 2000 (UNCTAD) 107.95 27.92 Exp.School Expected years of schooling at birth (UNDP Human Development Index Report) 10.66 3.05 Fertility Total fertility rate (births per woman), 2002 (UNESCO) 3.8 1.78 HDI Human Development Index, 2000 (UNDP Human Development Index Report) 0.55 0.17 Dbacba Deposit money bank assets / (deposit money + central bank) assets, 2002 (IMF International financial statistics via Demirguc-Kunt and Levine, 2009) 0.75 0.24 Bcbd Private credit by deposit money banks as a share of demand, time and saving deposits in deposit money banks, 2002 (IMF International financial statistics via Demirguc-Kunt and Levine, 2009) 0.81 0.49 offdep Offshore bank deposits relative to domestic deposits, 2002 (BIS Statistical Index via Demirguc-Kunt and Levine, 2009) 3.64 28.75 54 Table 2.3: EBA Results Variable Name Maximum Likelihood Point Estimate Model 1 Low High Model 2 Low High Model 3 Low High Constant (Int) 2.1857 -29.308 28.4340 -7.6931 9.3720 -38.4143 44.3200 Years.Open -0.0125 -0.099 0.0780 -0.0692 0.0728 -0.1827 0.1961 Stock.Dummy -0.6104 -2.467 1.8001 -1.8219 0.5597 -3.8707 3.6396 Low.Mid -3.0917 Upper.Mid -2.6088 Low.Income -2.5896 Per.Cap.GDP 0.0000 GDP.19922 0.0000 Avg.Infl 0.0283 Tax.Rate -0.0038 FDI.Flow 0.0000 -0.00051 0.00038 Paved.Road 0.0086 -0.0074 0.0408 Adj.Savings -0.0450 -0.08878 0.08054 Ag.Land 0.0135 Ag.Value.Added -0.0001 Life.Expect -0.1995 -0.0729 -0.3587 Pop.Growth 0.6186 -0.6322 1.1535 Cell.Phone -0.0296 -0.1075 0.0695 Curr.Acct -0.1970 -0.94909 0.13024 -0.7776 0.5806 Rural.Pop 0.0341 0.0011 0.0654 CC -0.4090 -2.1138 1.7308 RL -0.8849 -5.9948 4.412 -2.4408 0.2125 RQ -0.8106 -2.9101 0.0887 GE* 1.9580 0.1727 3.9121 VA -0.5770 -1.4836 0.6655 Nrbloan -0.6539 -3.8377 3.8902 -3.4009 2.7470 Exp.Index 0.0520 0.01331 0.0844 0.0267 0.0680 Exp.School -0.5328 -0.9953 -0.1850 Fertility -0.7244 -1.1120 0.4108 HDI 32.4433 11.7857 54.1040 Dbacba -2.1391 -5.2162 3.3638 Bcbd 1.1509 -0.7079 2.6319 offdep -0.0106 -0.4708 0.0364 Bolded values are robust, not covering zero. 2 Rounding to only four decimal places generates a 0.0000 Maximum Likelihood estimate. The current EBA code is unable to give the exact bounds, but appears as though they cover zero 55 Table 2.4: BMA Summary Output 73 models were selected. Best 5 models (cumulative posterior probability = 0.1767 ): p!=0 EV SD model 1 model 2 model 3 model 4 model 5 Intercept 100.0 -1.463e+00 4.225e+00 5.833e-01 2.477e+00 -1.157e+00 3.753e-01 -2.579e-01 Years.Open 0.0 0.000e+00 0.000e+00 . . . . . Stock.Dummy 0.0 0.000e+00 0.000e+00 . . . . . Low.Mid 0.0 0.000e+00 0.000e+00 . . . . . Upper.Mid 0.0 0.000e+00 0.000e+00 . . . . . Low.Income 0.3 4.632e-03 1.080e-01 . . . . . per.cap.GDP 21.1 -5.490e-05 1.180e-04 . . -2.808e-04 . -2.702e-04 GDP.1992 18.6 2.275e-12 5.234e-12 . . . 1.197e-11 . Avg.Infl 71.8 2.017e-02 1.507e-02 . . 3.028e-02 . 2.976e-02 Tax.Rate 0.0 0.000e+00 0.000e+00 . . . . . FDI.Flow 78.4 1.365e-04 9.202e-05 1.657e-04 1.693e-04 1.749e-04 . 1.663e-04 Paved.road 22.1 6.308e-03 1.373e-02 . 2.956e-02 . . . Adj.Savings 19.7 -1.062e-02 2.504e-02 . . . . . Ag.Land 0.0 0.000e+00 0.000e+00 . . . . . Ag.Value.Added 16.7 -1.464e-05 3.674e-05 . . . . . Life.Expect 83.3 -1.626e-01 1.027e-01 -2.276e-01 -2.473e-01 -1.569e-01 -2.172e-01 -1.629e-01 Pop.Growth 0.0 0.000e+00 0.000e+00 . . . . . Cell.Phone [phone]e-03 2.334e-02 . . . . . Curr.Acct 0.0 0.000e+00 0.000e+00 . . . . . Rural.Pop 4.4 1.759e-03 9.491e-03 . . . . . CC 4.3 -6.466e-02 3.323e-01 . . . . . RL 10.0 -1.604e-01 5.256e-01 . . . . . RQ 29.5 -5.186e-01 8.766e-01 -1.938e+00 -1.835e+00 . -1.860e+00 . GE 0.0 0.000e+00 0.000e+00 . . . . . VA 10.0 -1.172e-01 3.920e-01 . . . . . nrbloan 31.5 -2.286e-01 3.661e-01 . . . . . Exp.Index 100.0 4.332e-02 1.306e-02 4.189e-02 3.905e-02 3.935e-02 4.348e-02 4.051e-02 Exp.School 64.5 -3.477e-01 3.210e-01 -5.769e-01 -6.250e-01 . -6.098e-01 -4.448e-01 Fertility 0.0 0.000e+00 0.000e+00 . . . . . HDI 100.0 2.568e+01 1.006e+01 3.364e+01 3.193e+01 1.974e+01 3.313e+01 2.711e+01 dbacba 0.0 0.000e+00 0.000e+00 . . . . . bcbd 1.4 1.397e-02 1.455e-01 . . . . . offdep 40.3 -3.286e-02 4.371e-02 . . -8.733e-02 . -7.810e-02 nVar 6 7 7 6 8 r2 0.465 0.492 0.489 0.459 0.513 BIC -2.485e+01 -2.471e+01 -2.419e+01 -2.392e+01 -2.376e+01 post prob 0.046 0.043 0.033 0.029 0.027 56 Figure 2.2: Selected Posterior Distributions 57 Figure 2.3: BMA Image Plot CHAPTER 3 SCHUMPETERIAN INNOVATION AND EQUITY ISSUANCE 3.1. Abstract I hypothesize that highly innovative firms-those with high risk, yet higher potential return-will be more likely to raise funds through stock markets than bond issuance. Using the Schumpeterian innovation life-cycle as a theoretical framework, I argue that that in the beginning, firms with radically new innovations are more likely to raise funds through equity issuance until it becomes an acceptable loan for bankers with a limited return (interest rate). This is all placed within the context of, and does not conflict with, the dominant theories of firms' capital structure: the Trade-Off, market timing, and Pecking Order Models. Empirically, I test this relationship of innovative activity to equity issuance by using patents as a proxy for innovation from a dataset covering 1970 to 1992, encompassing 25,064 instances where firms raised funds through either the bond or stock market. This independent variable is then regressed on the ratio of funds raised through equity to total funds raised. I find statistically significant evidence using a dichotomous, probit model that the industries with higher innovative/patenting activities are 59 significantly more likely to raise funds through stock market issuance than firms without innovative activity. 3.2. Introduction Do firms care about their capital structure? Do investors? With the introduction of the Modigliani and Miller (1958) Irrelevance Proposition, it was argued that in perfect capital markets, firms will be indifferent to their capital structure. Competing theories of capital structure emerged in light of the restrictive assumptions imposed by Modigliani- Miller and the difficulties in empirically testing this idea. One aspect of capital structure literature that does receive attention from any of the dominant theories is that firms have different capital requirements over their life cycle. These financing requirements and constraints will influence whether firms will raise capital through debt or equity. I hypothesize that firms' external capital decisions change depending upon where they fall within the innovation life cycle as conceptualized by Joseph Schumpeter in Capitalism, Socialism, and Democracy (Schumpeter, 1942). I argue that in the beginning of the life cycle, firms that are highly innovative will be more likely to raise funds through the stock market than mature firms that are further into the life cycle. This is based around the argument that banks ration credit, limiting the amount of funds to newer industries with higher levels or risk, even though there is a high potential return. Even in the presence of a high risk premium, banks have asymmetric returns as they are exposed to losing their entire investment in the event of default while returns are limited to a fixed interest rate. 60 Along the innovation life cycle, the industries emerging are classified as being radically innovative because they products or processes they are promoting are radical departures from those currently seen. This period is marked by large amounts of innovation, which steadily decreases as the industry reaches stagnation. Innovations tend to be clustered in the beginning of the innovation life cycle (Keklik, 2003). Competition intensifies as the product makes it to market, causing less innovative firms to drop out. After the weaker firms fall, successful firms become more attractive for debt financing. During the time when there is a large amount of competition to "produce or perish," radically innovative firms have large capital requirements and do not have access to the same financial instruments that mature firms traditionally use. The supply constraints imposed by lenders does not reduce these firms' requirements for capital. They will continue to seek capital and will have a larger portion of equity to debt in their capital structure. I posit that stock market issuance is used more by firms during the early stages of the life cycle because of constraints on their ability to use debt. My theory as to choices made by radically innovative firms does come into conflict with any of the dominant theories of capital structure-Trade-Off, Pecking Order, and market timing. The Trade-Off theory (Kraus and Litzenberger, 1973) posits that firms will balance the tax advantages of debt with the increased probabilities of bankruptcy as they become more leveraged; it is postulated that there is an optimal level of leverage that maximizes the firm's value, and taking on more debt than the optimal will result in a lower valuation. My theory states that radically innovative industries are more likely to default and will result in a lower optimal level of leverage. If the firms still have additional capital requirements above the optimal leverage ratio, they will raise 61 all additional funds through equity/stock markets. Myers and Majluf (1984) introduce the Pecking Order theory as the main competitor to the Trade-Off theory. Rather than concerns over balancing costs and benefits, Myers and Majluf introduce asymmetric information, positing that investors believe firms' managers hold more information and would not be issuing stock unless they believe it to be overvalued. Because of the information problem (perceived or otherwise), Myers (1984) hypothesizes that adverse selection will occur, resulting in investors discounting the stock offering. Because of the discount placed on equity, stock market financing is only used as a last resort by firms unable to raise capital from any other source. Being constrained by debt markets, radically innovative firms are then more likely to need to use stock markets for capital because of the lack of options. I am able to empirically show how firms in innovative industries have a higher likelihood of raising funds through the stock market, controlling for portfolio returns and other market conditions. While the data do not allow for a precise positioning of firms within the life cycle, I provide empirical results in support of my theory that firms at the beginning of the innovation life cycle will be more likely to raise funds through equity than firms at the end of the life cycle. This is performed using patents as a proxy for innovation, operating on the assumption that more innovation occurs in the early stages of the life cycle. Probit estimation methods with clustered standard errors are used to estimate this relationship between innovation and firm choice as to whether debt or equity financing is used. The results are straightforward; I find significant evidence that the more innovative a firm is, the more likely it is to use the stock market to raise funds 62 when it is seeking external capital. These results are consistent with most of the literature on capital structure. 3.3. Theories of Capital Structure As the first widely accepted theory of capital structure, Modigliani and Miller (1958) showed that the value of the firm is not affected by how the firm is financed. The theory is that firms will raise external capital through whichever avenue is the least expensive, bringing the most capital at the lowest cost. The implication is that it does not matter what the underlying capital structure is; whether firms raise funds through equity or debt and how they pay dividends is irrelevant to firm value. Modigliani and Miller reached this Irrelevance Proposition under some crucial assumptions: perfect capital markets in the absence of taxes, bankruptcy costs, asymmetric information, adverse selection, and agency costs. While a ground breaking piece of research, the introduction of various market imperfections has given rise to a number of competing theories as to why firms finance themselves the way they do. 3.3.1. Trade-Off Theory The basic arguments behind which theory of capital structure is ideal are derived from disagreements over which imperfections are most important. These imperfections include agency costs, asymmetric information, bankruptcy costs, and taxes. Following is a discussion of the dominant theories of capital structure: Trade-Off, market timing, and Pecking Order. Modigliani and Miller (1963) acknowledged that the benefits gained from taxes are not insignificant, but that "under our analysis the tax advantages of debt 63 are the only permanent advantage" (Modigliani and Miller, 1963, p. 434). This is concluded by a reaffirmation of the 1958 hypothesis with the conclusion that these tax benefits are still small and, once investors' personal taxes are factored in, there are circumstances where other forms of finance may be cheaper for the firm than pure debt issuance. Empirically, Modigliani-Miller is difficult to test with many researchers unable to find reasonable evidence that it holds up using modern statistical methods (Frank & Goyal, 2007). Although it has long since been realized that perfect capital markets with perfect information do not exist, understanding Modigliani-Miller can help an understanding of how imperfections can distort markets. Arguing that taxes are an important factor in how firms finance themselves, Kraus and Litzenberger (1973) introduced the Trade-Off Theory of capital structure. The mix of financing depends on the tax savings and the states in which a firm would become insolvent. There is significant tax advantages for firms that are gained by issuing debt that far outweigh any of the costs incurred by investors' personal taxes. Paying interest on outstanding debt is tax deductible and lowers the cost to service the debt. On the dark side of leverage, Kraus and Litzenberger introduce bankruptcy costs into consideration. Modigliani and Miller (1958) assumed that firm value does not depend on how certain they are to repay their debt obligations; the value of a firm is not affected by its leverage since bankruptcy penalties do not exist in perfect capital markets. However, under the Trade-Off theory, as the leverage ratio increases, the value of a firm begins to fall because of the increased probability of becoming insolvent. The basic Trade-Off models predicted much higher debt levels than were actually observed (Miller, 1977). This is addressed by the discussion of whether firms are able to 64 costlessly restructure their debt at any given time. Kane, Marcus, and McDonald (1984) and Brennan and Schwartz (1984) separately developed continuous time models incorporating the imperfections of uncertainty, taxes, and bankruptcy costs, but ignoring transactions costs. Modeling uncertainty reduced the optimal leverage ratio, but still predicted values that were much larger than were actually observed. These models were still unable to explain the stickiness of restructuring their leverage ratio. Introducing transactions costs, Fischer, Heinkel, and Zechner (1989) found that even in the presence of small transactions costs, firms allow their leverage ratio to "float" within certain bounds, only adjusting when the ratio moves outside those bounds. That small transaction costs will lead to stickiness in adjustments of the leverage ratio makes it more difficult for smaller, less liquid firms to raise capital through the market. Investors purchasing a firms' bonds or stocks in the market from an illiquid company will face greater transactions costs, thus reducing the desire to purchase these securities. Discussions surrounding whether the Trade-Off theory can predict leverage ratios, and whether firms attempt to reach them, are relevant as a test of predictive power. Henessey and Whited (2005) and Strebulaev (2007) dispute that firms are underlevered relative to the predictions of the Trade-Off theory. These results present new estimates of the bankruptcy costs and tax benefits, providing leverage ratios that are consistent with current corporate debt levels. Other considerations have arisen out of the behavioral finance literature around confidence levels of company managers. Hackbarth (2008) presented a theoretical model, concluding that overconfident managers will raise m