Master Thesis MSC Finance, corporate finance track Kyra Berendsen 10362142 June 2017
Thesis supervisor: Prof dr. A.W.A. Boot
RE-EXAMINING THE
FINANCE-GROWTH
NEXUS
2
STATEMENT OF ORIGINALITY
This document is written by Student Kyra Berendsen who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
3
ABSTRACT
This thesis researched the relationship between finance and growth. The model that was used for this research is based on previous literature, as this makes it possible to compare results. This model includes some important control variables that are known to affect growth rates. Using data from the World Bank for all countries over the period 1981-2015 different instrumental variables (IV) / two stage least squares (2SLS) regressions were performed. The results for the linear model were overwhelmingly positive. The non-linear model indicated that there exists an inverse U-shaped relationship between finance and growth.
4
TABLE OF CONTENTS
I INTRODUCTION 5
II LITERATURE REVIEW 8
III METHODOLOGY 16
IV DATA AND DESCRIPTIVE STATISTICS 20
V RESULTS 24
VI ROBUSTNESS CHECKS 32
VII CONCLUSION AND DISCUSSION 36
VIII REFERENCES 38
5
I. INTRODUCTION
In the last couple of years the value added by financial sector has grown extensively all over the world mostly due to rising profits and compensation (Cournède et all, 2015). Then the financial crisis of 2007 demonstrated that a dysfunctional could have disastrous consequences. The crisis was a shock to people all over the world and led to increased financial regulation. Furthermore, the financial crisis showed that the financial sector is not a stand-alone entity, but rather is an integrated part of the economy, connected to many other industries.
The financial sector has always been vital for the functioning of the economy. This is because the different functions of the financial system were designed to solve market frictions (Beck, 2011). The system plays a role in the transmission of money. Finance or credit allows entrepreneurs to invest in projects that would otherwise be out-of-reach. Banks allow depositors to make payments. The system covers different kinds of financial risks. In addition financial development may promote economic growth through capital accumulation and technological progress.
While in theory the financial system was created to solve certain market frictions, it was not necessarily certain what the relationship between finance and growth was. The discussion about the finance and growth relationship began with the direction of causality. Schumpeter (1959) argued that production requires investments implying that finance leads to growth, while Robinson (1952) argued that finance follows where growth goes. Because both variables are endogenously determined it is hard to establish a causal relationship and theory could support either direction.
Goldsmith (1969) was one of the first to empirically assess the relationship. The troubles he encountered during his research were the lack of control variables and number of available observations and the endogenous aspect of both finance and economic growth. A couple of years later both Rajan and Zingales (1996) and King and Levine (1993) developed models to deal with these problems. Most of the papers thereafter used some sort of variation of those models and the majority of these papers found a positive relationship from finance to growth, thereby confirming that the financial sector function as was intended when it was originally designed.
6 Because most researchers found the same results there was consensus about the direction and the discussion began to shift to the different types of financial structure. There are two distinctive financial structures: a market-based structure and a bank-based structure. As the name already implies, a market-based structure is one where financial markets are more dominant, as is the case in the US, and a bank-based structure is one where banks are more dominant, as is the case in Germany. Both systems have comparative advantages over the other and the goal of research was to determine which system was better suited to promote growth. More recently Demirgüç-Kunt et all (2013) argued that neither system is particularly better but that as a country becomes more financially developed the system transitions with it.
Then the financial crisis showed that it not necessarily always is the case that finance is good for economic growth. New research emerged that demonstrated that the financial system might be dysfunctional as it might advocate economic instability. Economic instability causes financial crises and can lead to pricing bubbles. After the crisis people began to raise questions about the dominance of the financial sector in total economic activity. Could there be such a thing as too much finance?
As is evident from the real world consequences and the magnitude of empirical work about this topic establishing the relationship between finance and economic growth is important. I will try to contribute to understanding that relationship with this thesis. The model used to test this relationship is based on earlier research so that results could easily be compared, but instead of pure cross-sectional data time-series data will be used together with an instrumental variables approach to minimize any endogeneity problems. In addition to that there is now enough data after the crisis available to thoroughly test whether or not there indeed can be too much finance by employing a U-test to test for an (inverse) U-shaped relation between finance and growth. Using data from the World Bank Group that ranges from 1981 until 2005 for all countries all over the world I will try to answer the following research question: what is the relationship between finance and growth and is this relationship non-linear?
This thesis is organized as follows. The next chapter gives an overview of the most important literature regarding the finance-growth relationship. I have tried to present the literature in a chronological order, beginning at the original objective of the financial system and concluding with the most recent empirical work and a hypothesis development. The chapter thereafter explains the methodology used in great detail. The data used will be described in the following chapter. Then the results will be presented and discussed. After that I will perform some
7 robustness checks to confirm the results. Finally, the last chapter concludes this thesis and discusses the main shortcomings and recommendations for future research.
8
II. LITERATURE REVIEW
This chapter describes the literature related to the finance growth relationship. The first section focuses on the positive aspects of the financial system. This system was designed to solve market frictions and thereby fosters economic growth. Early empirical work confirms this positive relationship. The section thereafter summarizes the discussion about financial structure. Then follows a section that describes the main arguments why, after reaching a certain level of financial development, finance might impact economic growth negatively. This chapter concludes with a hypothesis development
A. FINANCIAL SYSTEM: SOLVING MARKET FRICTIONS AND
THEREBY FOSTERING ECONOMIC GROWTH
The International Monetary Fund (2008) defines a financial system as a set of institutional units and markets that interact, typically in a complex manner, for the purpose of mobilizing funds for investment and providing facilities, including payment systems, for the financing of commercial activity. A financial system is said to be more developed when it is better able to fulfill its functions. According to Levine (2005) these five functions are to produce information about possible investments and allocate capital, to monitor these investments and exert corporate governance, to facilitate the trading, diversification and management of risk, to mobilize and pool savings, and to ease the exchange of goods and services. These functions were designed to solve market frictions and thereby foster economic growth.
One of these market frictions is information asymmetry. Information asymmetry arises when one party in a transaction has more or better information than the other party. Because a buyer does not know the real value of an asset, the seller cannot capture all potential value and this presents an opportunity for a financial intermediary to capture the remaining value (Allen, 1990).
Another issue is the free rider problem which arises because information is a public good. The free rider problem is the problem that individuals will not take part in certain profitable activities when they are not incentivized (Stigler, 1974). Both banks and financial markets have ways to deal with the free rider problem. Banks avoid duplication of information, are able to specialize so that they are able to produce information and allocate capital more efficient and quicker, and banks can exploit cross-sectional similarities and re-use information (Levine,
9 2005). Markets on the other hand, when efficient and large in size, consists of uninformed and informed traders. When the proportion of uninformed traders is large enough, the informed traders could hide the fact that they know more and therefore they have enough incentives to trade (Kyle, 1974).
The monitoring function of the financial sector makes sure that the interests of investors are protected and that firms maximize shareholder value. If this is not ensured investors are less willing to fund new projects and this will decrease efficiency and capital allocation (Jensen & Meckling, 1976). Other issues that needs close monitoring is the issue of small versus big shareholders (blockholders). According to research blockholders might reduce the threat of intervention, lower liquidity, or they may extract private benefits, all of which increase economic costs (Edmans, 2014).
The financial system has three solutions to deal with risk compared to individuals. It diversifies risk, smooth risk and it reduces liquidity risk. Acemoglu and Zilibotti (1997) developed a model for the process of diversifying risk. They argue that in early stages of financial development the spreading of risk is limited. There is a desire to avoid highly risky investments which slows down capital accumulation, and the inability to diversify introduces a high amount of uncertainty. Such a system is subject to a higher variability of output than other, more developed systems. A more developed system eventually has more steady growth. In addition, King and Levine (1993) show that cross-sectional risk diversification can stimulate innovative activity.
Bencivenga and Smith (1991) developed a model whereby financial intermediation reduces the fraction of savings held in unproductive liquid assets and to prevent misallocations on invested capital due to liquidity needs. By providing liquidity, risk averse savers can hold bank deposits rather than liquid but unproductive assets. Banks provide a mixture of liquid and illiquid longer term investments, and therefore provide an individual insurance against risk.
The mobilizing and pooling of savings reduces both transaction costs and information asymmetries. The more developed the financial system is, the more it is able to mobilize and pool savings of individuals. Additional benefit of the pooling of savings is that production processes are not constrained to be small scaled (Sirri & Tufano, 1995). Many investments require a lot of capital and this is usually not feasible for a single investor.
As argued by Adam Smith (1776) specialization leads to productivity improvements. Greenwood and Smith (1996) have modeled the connections between exchange, specialization,
10 and innovation. They argue that more specialization requires more transactions, which may be costly. More developed financial systems usually have lower transaction costs, and therefore facilitate greater specialization, which in turns increases growth. Another benefit of lower transaction costs is that they favor long term investments (Levine, 2005). If these long term investments are also more effective than short term investments this will lead to higher growth rates.
One of the first to empirically assess the relationship between financial development and growth was Goldsmith (1969) hinting on a parallel between financial development and economic growth. However, limitations of his work are the lack of control variables and observations in his sample and that correlation does not necessarily imply causation (Levine, 1997). Both finance as well as growth are endogenously determined, making it hard to empirically establish causality (Butler & Cornaggia, 2010).
An important model to deal with these problems was developed by Rajan and Zingales (1996). Their rationale is that in more financially developed countries the cost of external financing to a firm is cheaper. By using a difference in difference approach they test whether industries who are more dependent on external finance grow faster in a system that is more financially developed. This rationale was confirmed by their results.
Another important paper in the finance-growth nexus is by King and Levine (1993). Using data on 80 countries over a 29 year period and employing various measures of financial development they find that financial development is positively related to economic growth, physical capital accumulation, and economic efficiency improvements. Levine and Zervos (1998) expanded this model by including stock markets measures and also find a positive relationship.
While most papers argue that finance is important for growth, other papers doubt that finance is the main determinant and argue that property rights are the main variable. Johnson et all (2002) look at entrepreneurs in Eastern Europe. They test whether secure property rights are sufficient for investment. They find that while the availability of banks loans matters for growth, but only after property rights are perceived to be secure. A few years later Acemogly and Johnson (2005) use an instrumental variables approach and find that when financial development is compared to private property rights, that the latter matters more for economic growth.
11 After establishing that most empirical work found a positive relationship between finance and growth, the discussion shifted to which financial structure is better suited to enhance growth. There are basically two main financial structures. One is the market-based system, where the financial market is dominant and this structure is most prevalent in the US. The other is the bank-based system, where the banks are more dominant and this structure is most prevalent in countries like Germany. The discussion about the financial structure is different from the discussion about financial development. While indicators of financial development may be roughly the same, separate indicators of financial structure may be quite different (Stulz, 1999).
Banks and markets have different ways of turning savings into investments. Banks usually provide loans and markets issue securities. Different industries rely on different finance structures. Industries with a lot of tangible assets to pledge as collateral rely more on banks while industries which rely on human capital and who do not have a lot of collateral to pledge will rely more on markets (Gambacorta et all, 2014). In addition, firm size is also of importance for funding. Because of the large fixed costs associated with markets small firms depend more on banks (Gambacorta et all, 2014).
The functions of the financial system might not work as well when only looking at either banks or markets. Markets might still suffer from the free rider problem. Key roles of banks include screening and monitoring, which are simply too costly for individual investors (Mayer, 1990). Grossman and Hart (1980) make another argument in regard to free riding with takeovers. There is no sufficient takeover threat for dysfunctional management because most shareholders hold on to their shares to benefit from improvements made by the new acquirer.
A problem with banks is that they have a lot of private information about firms and they can use this information to their advantage (Stulz, 1999). Banks further dislike risk and might therefore make decisions that do not maximize firm value. On top of that banks require collateral. This is detrimental for a lot of R&D intensive firms, who do not have a lot of tangible assets.
Empirical results are mixed. Arestis et all (2001) examine the relationship between the stock market development and economic growth, controlling for the banking system. They find that the stock market effects economic growth more than the banking system. Chakraborty and Ray (2006) however find that while neither system is better for growth, they do find some advantages for a bank based system. These advantages make per capita GDP higher in countries with a bank based system.
12 More recently evidence emerged that suggested that a country is not either a bank or a market based system, but that the system transitions as the country becomes more developed. Demirgüç-Kunt et all (2013) examine previous theoretical work and in addition to that also do some empirical tests and both lead to the same conclusion. Banks are more suited for standard finance agreements with low risk and financial markets are better suited for non-standard finance agreements with higher risk. When a country develops economically the demand for these agreements develops with it: the financial agreements become more complex in nature and therefore financial markets become more important than banks. Demirgüç-Kunt et all (2013) conclude that it depends on the circumstances which system is better and that countries that do not experience a lot of economic growth may simply have the wrong type of finance.
C. TOO MUCH FINANCE?
The financial crisis showed that the financial system may not work as intended and that credit does not necessarily foster economic growth in all cases. The empirical research after the crisis focused on the notion of too much finance. Researchers began examining the possibility that under certain circumstances finance may actually be harmful for economic growth. In this section I will summarize the most important papers that explain the channels through which finance may be harmful for economic growth and concluding with the most important work testing the finance-growth relationship taking into account that finance might have a negative impact.
The first channel through which finance may reduce economic growth is the magnification of economic costs in a number of ways. One of these ways the financial sector does so is by changing their role from merely intermediating to being active participants themselves. They are actively creating new products. One example of such a product is the credit default swap (CDS). A CDS is a type of financial derivative, similar to default insurance contracts. According to Stulz (2009) there are three main arguments why such products magnify economic costs: a) they support pricing bubbles, b) financial intermediaries have many positions in these products themselves, resulting in high systematic risk, and c) the complexity of these products makes it easier to manipulate the market.
Another channel through which economic costs are magnified is the corporate governance system. Tarraf (2011) argued that the main cause of the financial crisis was aggressive risk-taking. The corporate governance system was originally designed to prevent this, but failed to do so during the crisis. The main problem with the current system is the incentive system. Rajan
13 (2005) already warned about these altered incentives before the crisis. Before the crisis, technological change lowered transaction costs, fastened the processing of information and introduced new financial techniques. In addition, the financial sector experienced trends of deregulation and the creation of new institutions such as hedge funds. This altered managers’ incentives. Managers’ compensation is related to risk and to the performance of peers. Due to limited downside risk and a high upside potential managers will take more risk. In addition, peer based compensation might lead to manipulating behavior.
Another aspect that increases risk-taking are the debt guarantees that regulators issue to to-fail banks or other institutions. This means that institutions that are considered too-big-to-fail are so large that their default would have disastrous consequences for the economy of the country where that institution is based and that they therefore should be supported when facing difficulties. While the intention of these debt guarantees was to prevent any negative effects, Denk et all (2015) found that banks and other institutions because of these guarantees would lend excessively, thereby reducing economic efficiency.
Third, the highly profitable financial sector attracts too many people. Tobin (1984) argued that the financial sector induces a misallocation of talent and that the social costs of the industry are greater than the benefits. This argument was enriched by Beck et all (2014). They argue that due to comparative advantages some countries specialize in the export of finance. They find that while pure intermediation activities have a positive effect on growth, all other financing activities only increase volatility.
Related is the work by Cecchetti and Kharroubi (2015). They model the interaction between financial and real growth while simultaneously introducing skilled workers into this model. These workers can be hired either by the financial sector to increase financing activity or by entrepreneurs to improve returns and productivity. When most workers are hired by the financial sector the entrepreneur has not enough workers left to work for him so that he cannot in more risky and high return projects and is therefore left with low return projects. This results in a lower total factor productivity growth of the overall economy.
Bolton et all (2011) also model entrepreneurs and finance workers but then in relation with informed dealers and entrepreneurs. They argue that the opaqueness of the over-the-counter- market leads to a negative externality where only less valuable projects will be traded. This will lead to rent extraction by the dealer. Because of this too much talent will work in the financial industry in equilibrium.
14 Fourth, the financial sector increased the financial instability which ultimately slows or lowers economic growth. The financial instability hypothesis developed by Minksy (1992) predicts that during economically expansive times the investments in risky assets will go up. This is supported by Jorda et all (2013) and Jiménez et all (2014), who examine credit granted by financial intermediaries and banks in particular. They argue that when there are economically good times the interest rates are low. This implies that the return for risk-free investments is low as well. Financial intermediaries are therefore more likely to invest in riskier projects, increasing financial instability. Furthermore, Jiménez et all (2014) found that banks that have a lot of agency problems especially are prone to supply more credit to risky firms. This in turn increases the changes of default for these banks.
Another problem that leads to instability are pricing bubbles. These bubbles usually start when people are overoptimistic with regard to future prospects, they expect that prices will continue to rise in the same way as they do when the economy is in an uplift. Another cause is explained by Epstein and Crotty (2012). They argue that one of the functions of the financial system is to create liquidity. Through demand and supply this liquidity helps to determine the value of an asset. But they argue that when the financial sector becomes too developed, or too powerful, this function of liquidity provision does not lead to price discovery, but rather to price creation. The activities of the financial intermediaries determine what the price will be. This in turn leads to price run ups and eventually pricing bubbles. These price bubbles can lead to overinvestment and thereby a decreasing economic efficiency (Dupor, 2005).
Because of these negative effects of finance on economic growth, researchers began re-examining the finance growth relationship, now focusing on the possibility of a non-linear relationship. Law and Singh (2014), Cecchetti and Kharroubi (2012) and Arcand et all (2015) use different empirical approaches and find a non-linear relationship between finance and growth when finance reaches 90-100 percent of GDP. So in the beginning stages of financial development the positive effects outweigh the negative, but after the level of financial development reaches around 90 percent of GDP the negative effects outweigh the positive and the relationship between finance and growth becomes negative.
D. HYPOTHESIS DEVELOPMENT
In summary, most empirical work before the crisis found a positive relationship between finance and growth. After the financial crisis researchers began examining the possibility of a non-linear relationship, specifically an inverse U-shape. The reason of this difference in results
15 may be that early researchers never considered a non-linear relationship because there were never any big troubles with the financial system. This changed however after the financial crisis. The crisis exposed a lot of existing problems with the current system. Researchers then slowly started reasoning why there might be a non-linear relationship and started empirically testing if this indeed was the case. Because of this I will formulate two hypotheses:
Hypothesis 1: when testing a linear model, there is a positive relationship between finance and growth
Hypothesis 2: when testing a non-linear model, there is an inverse U-shape relationship between finance and growth
16
III. METHODOLOGY
This chapter describes the model used to test the finance-growth relationship and the variables needed.
A. MODEL AND STATISTICAL METHOD
The model used in this thesis is based on the model by King and Levine (1993). This model has been used in different versions throughout the finance-growth literature so it was deemed the most appropriate for comparison reasons. The basic model is:
𝐺𝑟𝑜𝑤𝑡ℎ𝑖,𝑡 = 𝛼0+ 𝛽1𝐹𝐼𝑁𝑖,𝑡+ 𝛾𝑋𝑖,𝑡+ 𝑢𝑖,𝑡
- Growth is the measure of economic growth for country i, in period t - FIN is the measure of financial development for country i, in period t
- X represents the set of explanatory control variables for country i, in period t - U is the error term
As discussed before previous empirical work had some econometric problems. King and Levine (1993) for instance used a basic OLS cross-country regression. Due to causality issues this might not be the most appropriate technique. They tried to solve this by using the average of the dependent variable and initial values of the independent variables. Arestic and Demetriades (1997) further argue that cross-country regressions only measure the average effect. Long-run causality can not only vary across countries, but also the long-run relationships themselves may vary. They therefore conclude that a time-series analysis may be better to establish a causal relationship.
For this thesis the model by King and Levine (1993) will be transformed at two points: panel data will be used instead of cross sectional data and an instrumental variables (IV) or two stage least squares (2SLS) regression will be conducted instead of a standard OLS regression. With a panel data analysis one looks at two dimensions: cross-sectional and longitudinal. Multiple cases or in this instance countries are observed over two or more time periods. An additional advantage of panel data is that it is possible to control for some degree of omitted variable bias (OVB). Panel data can control for OVB by seeing the changes in the dependent variable over time. It can control OVB that differ between cases but are constant over time (time-fixed effects) or for OVB that differ over time but are constant between cases (random effects).
17 An IV/2SLS approach is used because it addresses the main issues with a simple OLS regression: omitted variables bias and reverse causality. Because there is a chance of either of these problems it is better to use an IV approach. An IV regression is estimated in two stages. The instrument predicts the main independent variable in the first stage and that prediction of the independent variable is then used in the second stage to predict the dependent variable. For this to be valid, one needs an instrument that is highly related to the independent variable, but is uncorrelated to the dependent variable. Because there are endogenous variables in the model the generalized method of moments (GMM) estimation method will be used.
The model will be estimated in multiple versions over different time frames. The first few versions of my model will be following early empirical work and only estimate the model in a linear version. Then the results are compared to these earlier papers. To address the potential of an inverse U-relationship a squared term will be added to the model. Just checking the significance of this squared term is not enough so therefore a U-test by Lind and Mehlum (2010) will be done. The model will then have the following form:
𝐺𝑟𝑜𝑤𝑡ℎ𝑖,𝑡 = 𝛼0+ 𝛽1𝐹𝐼𝑁𝑖,𝑡+ 𝛽2𝐹𝐼𝑁𝑖,𝑡2 + 𝛾𝑋𝑖,𝑡+ 𝑢𝑖,𝑡
Lind and Mehlum (2010) stated the following null and alternative hypotheses:
𝐻0: ( 𝛽1+ 2𝛽2𝐹𝐼𝑁 𝑀𝐼𝑁 ≤ 0) ∪ (( 𝛽1+ 2𝛽2𝐹𝐼𝑁 𝑀𝐴𝑋 ≥ 0)
𝐻1: ( 𝛽1+ 2𝛽2𝐹𝐼𝑁 𝑀𝐼𝑁 > 0) ∪ (( 𝛽1+ 2𝛽2𝐹𝐼𝑁 𝑀𝐴𝑋 < 0)
B. MAIN VARIABLES
The main variables used in this thesis are the dependent variable and the proxies for financial development. These are:
- GDP per capita growth (log of GDP per capita constant US$ 2010): the GDP growth rate measures how fast the economy is growing. It compares the country’s gross domestic product to the previous period’s outcome. The level of GDP consists of four components: personal consumption, investment, government spending and net trade. The growth rate is an indicator of economic health and is therefore an important measure.
- Liquid liabilities (M3, % of GDP): will be used as one of the proxies for financial development. M3 is known as broad money and include all readily available forms
18 of money. It is money that can be used to buy goods and services without incurring any costs. Included are bank deposits, savings deposits and shares of mutual funds or market funds held by residents.
- Domestic credit to private sector by banks (% of GDP): will be used as a proxy for banking activity. It refers to the financial resources for the private sector provided by banks. This can be through loans, purchases of securities or accounts receivables. - Market capitalization (% of GDP): will be used as a proxy for financial market
activity. Market capitalization is the total value of a company’s outstanding shares, or in this case the total value of shares of all domestic companies combined. To know the market capitalization the total shares outstanding are multiplied by the current market prices.
C. CONTROL VARIABLES
Control variable in a regression are those variables that are kept constant and unchanged throughout the sample period. These variables are included into regression model because they strongly influence results. To only test the relationship of the dependent and independent variables control variables need to be included in the model. The control variables included in my model are based off of previous literature. These variables are:
- Government spending (% of GDP): includes all government consumption, investment, and transfer payments. As said before GDP consists of four components: consumption, investment, trade and government spending. Financial development only has effect on consumption and investment so it is important to add trade and government spending as control variables to net measure the effects of said variables on GDP growth.
- Inflation (GDP deflator, annual %, log transformed): a sustained increase in the general price level of goods and services in an economy over a period of time. When prices rise the demand in an economy will drop and the GDP growth rate will slow down. Because inflation is related to GDP growth it is an important variable to include in the regression.
- Trade openness (% of GDP): the sum of exports and imports of goods and services measures as a share of GDP. Trade openness raises economic growth. Because financial developments affects economic growth through consumption and investment and not trough trade, it is included as a control variable.
19 - Secondary school enrollment (gross %, log transformed): a proxy for human capital and it measures the total enrollment in secondary education. Human capital is strongly related with economic growth and should therefore be included in the model. Human capital is the knowledge, skill sets and motivation that people have, which provide economic value. Economic growth is an increase in an economy’s ability to produce output. A higher level of human capital raises this ability.
20
IV. DATA AND DESCRIPTIVE STATISTICS
This chapter describes the database used and gives a first look at the data characteristics.
A. DATABASE
Data is gathered from the World Bank database. The World Bank is an organization consisting of 189 member countries that provide loans to countries all over the world, totaling $60 billion in 2015. Stated goal of the organization is to reduce poverty. They further strive to promote foreign investment, international trade and capital investment. The organization was founded in 1944 to help countries recover from the world war. The focus today is on developing countries.
The World Bank has extensive databases on financial development and economic growth. The main database used for this thesis is the database called World Development Indicators. This database is chosen because most previous literature used this database as well. All countries included in this database are also used here (for a list see Appendix). The most recent year in the database is 2015. The full time frame chosen is 1981-2015. 1981 was chosen as a starting point because much data was missing in the years before. Missing in this database is the measure of liquid liabilities, or M3. This variable is gathered from the financial structure database from Beck et all (2000). The financial structure database was last updated in 2015 so the last year the M3 variable was available was 2014.
B. DESCRIPTIVE STATISTICS
Table 1 describes the descriptive statistics for the complete sample. As can be seen from the table quite a few observations are missing, especially for the variables market capitalization and inflation. The observations available are respectively 2002 and 3875. GDP growth has the most observations with 6072. Further it can be seen that the measures of financial development are on around 50 percent of GDP on average, while trade openness is almost 85 percent and government spending is only 16 percent of GDP. Another aspect that stands out is the variation for most variables as can be seen from the standard deviation and the difference between the minimum and maximum value. For example, market capitalization can be almost as low as 0 percent of GDP, but also as high as 1254 percent of GDP with a standard deviation of 92. The high standard deviations justifies splitting the sample in low and high income countries.
21 Table 2 and 3 in the appendix are descriptive statistics for low and high income countries only. As expected the means of all variables except inflation are lower in low income countries than in high income countries. Further can be noted that the standard deviations for the measures of financial development are much higher in high income countries than in low income countries, thus implying higher variation for high income countries. Lastly, the maximum value for government spending is higher for low income countries than it is for high income countries.
C. CORRELATION MATRIX
Table 2 shows the correlation matrix of the variables used. The measures of financial development do not have a high correlation with GDP growth. Their correlation averages at around 26 percent. The variable representing human capital is correlated the most with GDP growth with a correlation of 62 percent. As expected the inflation measure has a negative correlation with all other variables. The variables of liquid liabilities (M3) and credit to the private sector by banks (Credit) have the highest correlation (77 percent). The correlation of M3 and market capitalization is remarkably lower at 36 percent, but the correlation between credit and market capitalization is high again at 70 percent.
Measurement N Mean Std. Dev. Min Max
GDP per capita, constant US$ 2010, log transformed 6072 8.29 1.53 4.75 11.89 % of GDP 5270 51.42 41.53 .00 399.11 % of GDP 5571 40.14 36.20 .00 312.12 % of GDP 2002 58.24 92.06 .01 1254.47 % of GDP 5798 84.87 53.34 .02 531.74 % of GDP 5643 16.49 7.94 .00 156.53 Gross %, log transformed 4535 4.06 .71 .97 5.10 GDP deflator, annual %, log transformed 3975 2.08 1.43 13.49- 10.19 Inflation
Table 1: Descriptive statistics
Notes: This table represents the descriptive statistics for the entire sample. Time period is 1981-2015. For a full list of countries included see appendix.
Market capitalization Trade openness Government spending Secondary school enrollment Variables
GDP growth M3
22
D. INTRODUCTION TO DATA
Figure 1 shows the graphs of the main variables over time. The three lines represent the averages for the entire sample, for high income countries only and for low income countries only. The trends for all variables are quite similar for every (sub)sample. The only variable that is showing some variation is market capitalization. The spikes and downturns in the market capitalization graph represents economic uplifts or recessions. Usually not all countries are affected at the same time, so it makes sense that the graphs do not always follow the same trends exactly.
Table 4 Correlation matrix
GDP growth M3 Credit
Market
capitalization Trade openness
Government spending Sec. school enrollment Inflation 1.00 0.21 1.00 0.32 0.77 1.00 0.26 0.36 0.70 1.00 0.05 0.38 0.40 0.24 1.00 0.20 0.03 0.12 0.05 0.24 1.00 0.62 0.15 0.21 0.14 0.03 0.31 1.00 0.09- 0.39- 0.39- 0.24- 0.25- 0.14- 0.14- 1.00 Government spending
Sec. school enrollment Inflation Variables GDP growth M3 Credit Market capitalization Trade openness
23
Figure 1 Graphs of main variables over time
7 8 9 10 1981 2015 year
All countries High income countries Low income countries
Average GDP per capita (constant US$ 2010, log transformed)
20 40 60 80 100 1981 2015 year
All countries High income countries Low income countries
Average M3 (% of GDP) 20 40 60 80 1981 2015 year
All countries High income countries Low income countries
Average credit (% of GDP) 20 40 60 80 100 120 1981 2015 year
All countries High income countries Low income countries
24
V. RESULTS
This chapter summarizes the results of the various regressions used to test the two hypothesis. First, the results from the linear IV regressions for all three measures of financial development are discussed. Second, the linear IV regression results that compare low versus high income countries. A discussion about the non-linear IV regressions follows that and this chapter is closed with a comparison with non-linear IV models of low versus high income countries.
A. LINEAR IV RESULTS
Table X shows the regression results for the entire sample of countries and all three measures of financial development. The sample was split by different time periods. The results were obtained using a generalized method of moments IV model with fixed effects. A lagged value of financial development was used as instrument. Time dummies were also included but not reported.
The regressions run on the entire period show significantly positive results at the 1 percent level for all three measures of financial development. The measure of liquid liabilities (M3) has a coefficient of .0024, credit to the private sector from banks (credit) has a coefficient of .0077 and market capitalization has a coefficient of .0018. From the coefficients it can be seen that credit has the highest influence. When the level of credit is changed by 1 percent, the dependent variable is expected to change by .77 percent.
These results confirm the first hypothesis based on earlier empirical work of a positive relationship when testing a linear model. Because the model used is based on King and Levine (1993) it is not surprising that the results are similar. Both results are significant at the 1 percent level, the only difference being the size of the coefficients as the coefficients of King and Levine are bigger (0.024 – 0.034). This is probably because different empirical techniques were used. Were King and Levine used pure cross-country regression, this thesis used panel data and an IV approach.
What stands out is that in the separate periods the coefficient of liquid liabilities becomes insignificant. This probably is because of a difference in statistical calculations due to a smaller sample size. Furthermore, the coefficient of credit increases from .0047 to .0076 between the two periods, while the coefficient of market capitalization decreases from .0023 to .0020. Similarly, the R squared of credit is higher than the R squared of market capitalization .
25 Measurement % of GDP .0024 (.0006) *** .0001 (.0001) .0001- (.0001) % of GDP .0077 (.0008) *** .0047 (.0011) *** .0076 (.0010) *** % of GDP .0018 (.0006) *** .0023 (.0011) ** .0020 (.0008) *** % of GDP .0026 (.0005) *** .0009 (.0006) .0009 (.0007) .0002 (.0001) *** .0034 (.0007) *** .0021- (.0014) .0002 (.0000) *** .0012- (.0008) .0016 (.0009) * % of GDP .0032 -(.0023) .0016 (.0024) .0154 (.0059) *** .0016- (.0003) *** .0063 (.0032) ** .0009 (.0084) .0006- (.0003) ** .0023 -(.0027) .0246 (.0086) *** % gross, log 1.2320 (.0257)*** 1.1827 (.0257) *** 1.8769 (.0736) *** .0140 (.0032) *** .9718 (.0277) *** 1.1785 (.1562) *** .0117 (.0039) *** 1.5544 (.0448) *** 1.9722 (.0850) *** % yearly, log .0545- (.0134) *** .0372- (.0129) ** .0060 (.0240) .0097- (.0016) *** .0076- (.0161) .0212 (.0396) .0005 (.0015) .0023- (.0027) .0202- (.0308) .5770 .5942 .4154 .9979 .5823 .2613 .9989 .6182 .4635 2153 2240 620 1035 1061 172 1118 1179 448
Table 3: IV regression results full sample
Notes: coefficients in this table are the result of an instrumental variables regression using the GMM option because of endogenous regressors. Instrument was a lagged value of financial development. Every regression included time dummies for 5 non-overlapping years (not reported here). Standard errors are in parantheses. The symbols are * = 90 percent significance, ** = 95 percent significance and *** = 95 percent significance.
Time period Variables M3 Credit Market capitalization R2 N 1981-2015 1981-2000 2001-2015 Trade openness Government spending
Sec. school enrollment
26 The included control variables give mixing results. As expected the coefficient for inflation is either significantly negative or insignificant. According to Sarel (1996) there is a structural break in the inflation-growth relationship. Below an 8 percent inflation rate there is (almost) no effect on growth and above an 8 percent rate there is a significant effect. Similarly, trade openness is either significantly positive or insignificant. Although a positive relationship between trade openness and growth was always prevalent in the (theoretic) literature, this might also be the case of poor measures and wrong methods (Rodriquez and Rodrik, 2000). Furthermore the coefficient for human capital (secondary school enrollment) has a positive significant effect for all regressions. Lastly, the coefficient for government spending changes from insignificant to significantly positive to significantly negative, depending on the time period and measure of financial development. This difference can be explained because the type of government expenditures changes per year and country resulting in different relationships with economic growth (Barro, 1990).
B. LINEAR IV RESULTS LOW VERSUS HIGH INCOME COUNTRIES
Table 4 shows the results of the linear IV regression for low and high income countries separately. These regressions were estimated in the same way as the IV regressions for the entire sample and the only difference is that the sample is now split in low and high income countries (see also tables 3 and 4 in appendix). There are more low income countries for the measures of liquid liabilities and credit, but high income countries are in the majority for the market capitalization measure.At first sight one can see that the significance levels and signs differ for the measures of financial development when comparing the two sample groups. Liquid liabilities are insignificant for low income countries, but are significantly negative for high income countries. Besides statistical differences due to a smaller sample size a reason for this might be that the level of liquid liabilities is much higher for high income countries (an average of 68 percent of GDP, see appendix) than for low income countries (an average of 36 percent of GDP). This might be a case of too much finance as described in the literature review. At low levels of liquid liabilities the money is invested effectively but when this level rises this investment becomes more and more ineffective and ultimately lowers economic growth. This is similar for the coefficients of credit. The coefficients for market capitalization are both significantly positive at the 1 percent level. Finally, while the R squared drops for low income countries when comparing credit to market capitalization, the R squared increases for high income countries.
27 Measurement
Low High Low High Low High
% of GDP .0000 (.0001) .0002- (.0001) ** % of GDP .0091 (.0007) *** .0008- (.0007) % of GDP .0064 (.0014) *** .0008 (.0002) *** % of GDP .0002 (.0001) *** .0003 (.0001) *** .0048 (.0006) *** .0005- (.0005) .0071 (.0009)*** .0017- (.0004) *** % of GDP .0009- (.0003) *** .0015- (.0003) *** .0073- (.0015) *** .0117- (.0160) *** .0714- (.0096)*** .0351- (.0031) *** % gross, log .0143 (.0028) *** .0117 (.0101) .7124 (.0234) *** .6430 (.0795) *** .8002 (.0729)*** .8292 (.0825) *** % yearly, log .0080- (.0015) *** .0034- (.0016) ** .0700- (.0129) *** .0428- (.0160) *** .2115- (.0262) *** .0680- (.0155) *** .9971 .9932 .5795 .2181 .4718 .3977 1337 816 1404 836 280 340
Notes: coefficients in this table are the result of an instrumental variables regression using the GMM option because of
endogenous regressors. Instrument was a lagged value of financial development. Every regression included time dummies for 5 non-overlapping years (not reported here). Standard errors are in parantheses. The symbols are * = 90 percent significance, ** = 95 percent significance and *** = 95 percent significance.
Time period
Table 4: IV regression results low versus high income countries
1981-2015 Inflation R2 N M3 Credit Market capitalization Trade openness Government spending Sec. school enrollment
28
C. NON-LINEAR IV RESULTS
The regression results for the non-linear models are shows in table 5. All three measures of financial development were included in the various models. The sample was split in different time periods. A generalized method of moments IV model with fixed effects was used to mitigate the problem of endogenous variables. The instrument used was a lagged value of the included measure of financial development. Time dummies were also included but not reported and were most of the time insignificant. The outcomes of the U-test by Lind and Mehlum (2010) are also reported.
What stands out is that the coefficients for the M3 variable are mostly insignificant except for the older period where the coefficient for M3 is significant at the 5 percent level. The result is that the U-test for the entire period and the recent period experiences a trivial failure to reject the null hypothesis which states a monotone or linear relationship. The U-test for the old period does find an inverse U-shaped relationship although it is only significant at the 10 percent level.
On the other hand the measures of credit, credit squared, market capitalization and market capitalization squared are all significant at the 1 percent level and have the expected signs (negative for the squared version of the measure). Subsequently all p-values from the U-test are significant at the 1 percent level, except for the U-test for credit in the recent period. Taken together these results confirm the second hypothesis of an inverse U-shaped relationship between financial development and economic growth.
When comparing the two measures of financial structures (credit and market capitalization) a few things stand out. The R squared for market capitalization for the entire period is much higher than the R squared for credit (.8819 and .6004 respectively). This is because the R squared for market capitalization changes quite a bit from the old period to the recent (.4340 in the old period and .9826 in the recent period). These results are supportive of the theory by Demirgüç-Kunt et all (2013) who argue that the financial system transitions as a country becomes more developed. Furthermore the slopes for market capitalization are steeper than the slopes for credit (-1.9469 versus -.0215 during the older period respectively).
29 Measurement % of GDP .0024 (.0016) .0099 (.0045) ** .0005 (.0020) .0000 (.0000) .0001- (.0000) .0000 (.0000) % of GDP .0186 (.0021) *** .0170 (.0025) *** .0149 (.0036) *** .0001- (.0000) *** .0001- (.0000) *** .0001- (.0000) ** % of GDP .1612 (.0127) *** .3041 (.1111) *** .0661 (.0085) *** .0007- (.0001) *** .0013- (.0005) *** .0004- (.0000) *** % of GDP .0022 (.0005) *** .0009 (.0006) .0008 (.0017) .0038 (.0007) *** .0034 (.0007) *** .0057 (.0106) .0000- (.0008) .0013- (.0008) .0031- (.0011) *** % of GDP .0020- (.0022) .0000- (.0022) .0968 (.0146) *** .0014 (.0030) .0043 (.0032) .0540 (.0467_ .0029- (.0027) .0021- (.0025) .0995 (.0145) *** % of GDP 1.2169 (.0277) *** 1.1031 (.0279) *** .4958 (.1534) *** .9380 (.0330) *** .9079 (.0290) *** .6513- (1.0971) 1.6399 (.0463) *** 1.4692 (.0524) *** 1.2526 (.1285) *** % of GDP .0575- (.0134) *** .0265- (.0130) ** .6392 (.0663) *** .0043 (.0162) .01179 (.0163) 1.4527 (.6158) ** .1207- (.0226) *** .0678- (.0209) *** .2294 (.0488) *** P-value Trivial failure .0000 *** .0000 *** .0789 * .0001 *** .0037 *** Trivial failure .0635 * .0000 *** Slope .0024 .0186 .1612 .0099 .0170 .3092 .0006 .0149 .0661 Slope .0045 .0378- 1.6508- .0294- .0215- 1.9469- .0087 .0215- .7095-.5766 .6004 .8819 .5709 .5867 .4340 .5978 .6215 .9826 2153 2240 620 1035 1061 172 1118 1179 448
Table 5: IV regression and U-test results full sample
Notes: coefficients in this table are the result of an instrumental variables regression using the GMM option because of endogenous regressors. Instrument was a lagged value of financial development. Every regression included time dummies for 5 non-overlapping years (not reported here). The U-test test the presence of an inverse U-shaped relationship (default) or an U-shaped relationship (indicated by a U). Standard errors are in parantheses. The symbols are * = 90 percent significance, ** = 95 percent significance and *** = 95 percent significance.
R2 N U-test M3 squared
Credit squared
Market capitalization squared
Lower bound Upper bound Market capitalization
Trade openness
Government spending
Sec. school enrollment
Inflation
1981-2015 1981-2000 2001-2015
M3
Credit
30
D. NON-LINEAR IV RESULTS LOW VERSUS HIGH INCOME COUNTRIES
Table 6 shows the results of the non-linear IV regression for low and high income countries separately. These regressions were estimated in the same way as the non-linear IV regressions for the entire sample and the only difference is that the sample is now split in low and high income countries (for a full list of these countries and for regression results for the separate time periods see Appendix tables 1, 6 and 7).Compared to table 5 were the entire sample is regressed the liquid liability becomes significant when the sample is split in high and low income countries. The measures of credit (squared) and market capitalization (squared) stay significant compared to the previous table. However, there is one big difference. The U-test for credit (squared) for high income countries indicates a U-shape relationship instead of an inverse U-shaped relationship. This means that for high income countries, the level of credit decreases economic growth at low levels of financial development but then increases economic growth at higher levels. The U-test for low income countries does indicate an inverse U-shaped relationship, so this combination might account for the rejection of the null hypothesis in the previous table.
Comparing low to high income countries one could notice some differences between the results. The measure for liquid liabilities has a higher economic impact for low income countries (.0094) than for high income countries (.0046), even though both are significant. However, the U-test does not indicate an inverse U-shape for low income countries, but it does for high income countries. Another difference is that the coefficient for credit is significantly positive for low income countries and significantly negative for high income countries. As stated earlier, this negative coefficient for high income countries may be the result of too much finance as described in the literature review because high income countries are usually more financially development than low income countries. The coefficients for market capitalization do have the same signs and significance levels for low and high income countries. In this case however, the slopes for low income countries are steeper than the slopes for high income countries, especially the slope for the upper bound (-2.4105 and -.6941 respectively). This indicates that an increase in market capitalization has more effect on low income countries than on high income countries.
The results of the included control variables differ both in sign and significance when the sample is separated in high and low income countries. Trade openness for example is significantly positive for low income countries and significantly negative or insignificant for
31 high income countries. This suggests that not only the level of trade matters but also the structure of trade.
Measurement
Low High Low High Low High
% of GDP .0094 (.0022) *** .0046 (.0015) *** .0000- (.0000) * .0000- (.0000) ** % of GDP .0189 (.0019) *** .0070- (.0023) *** .0001- (.0000) *** .0001 (.0000) *** % of GDP .1506 (.0139) *** .0724 (.0104) *** .0014- (.0002) *** .0003- (.0000) *** % of GDP .0056 (.0001) *** .0018- (.0005) *** .0043 (.0006) *** .0007- (.0006) .0105 (.0012) *** .0058- (.0014) *** % of GDP .0091- (.0015) *** .0092- (.0029) *** .0071- (.0015) *** .0108- (.0035) *** .0289 (.0147) ** .0372 (.0124) *** % of GDP .7033 (.0262) *** .5491 (.0844) *** .6694 (.0251) *** .6929 (.0789) *** .5618 (.0913) *** 1.1551- (.3711) *** % of GDP .0895- (.0138) *** .051- (.0152) * .0620- (.0129) *** .0475- (.0161) *** .0424 (.0370) .2786 (.0594) *** P-value .3850 .0127 ** .0013 *** .0014 *** U .0000 *** .0000 *** Slope .0094 .0044 .0189 .0070- .1504 .0724 Slope .0011- .0088- .0097- .0288 2.4105- .6941-.5516 .2342 .5795 .2063 .5933 .9758 1337 816 1404 836 280 340
Notes: coefficients in this table are the result of an instrumental variables regression using the GMM option because of endogenous regressors. Instrument was a lagged value of financial development. Every regression included time dummies for 5 non-overlapping years (not reported here). The test test the presence of an inverse shaped relationship (default) or an U-shaped relationship (indicated by a U). Standard errors are in parantheses. The symbols are * = 90 percent significance, ** = 95 percent significance and *** = 95 percent significance.
Lower bound Upper bound R2 N Inflation M3 U-test Credit squared Market capitalization Market capitalization squared Trade openness
Government spending Sec. school enrollment
Table 6: IV regression and U-test results low versus high income countries
Variables Time period
1981-2015
M3 squared Credit
32
VI. ROBUSTNESS CHECKS
This chapter checks the robustness of the previous results using different measurements for both the variable credit as well as market capitalization. The other specifications used to check the results are first defined before comparing the results to the main variables.
A. NEW VARIABLES
Two new variables are introduced that are comparable to credit and market capitalization:
- Domestic credit to private sector (% of GDP): instead of the financial resources to the private sector provided by banks this variable measures the resources by all financial corporations. As said before, the financial sector expanded increasingly in the last couple of years. This includes new financial corporations and this variable could therefore potentially differ from the credit provided by banks alone. Difference in results would show the influence of these other corporations.
- Stocks traded (% of GDP): is the value of shares traded times the total number of shares traded for both domestic and foreign shares. It is essentially the same measure as market capitalization except that foreign shares traded in a particular country are now included. Difference in results would now show the influence of foreign stocks traded in a country.
B. ROBUSTNESS CHECK RESULTS
Table 7 and 8 show the results for the robustness check for the linear and non-linear regressions for the entire sample and for different time frames. For the linear regressions the credit measures are all significantly positive. Quite surprisingly though, the economic significane of the total credit measure is lower than the economic significance of the credit by banks measure (.0057 and .0076 respectively). Similarly, while the measures of market capitalization are significantly positive during all periods, the measures of stocks traded are insignificant for the entire period and the recent period and only significantly positive for the old period.
33 For the non-linear regressions all results are significant, although some are only at the 10 percent level. Again, the results for both the credit measures are quite similar, but the results for the market measures differs in some cases quite substantially. For example, the slope for the upper bound for the entire time frame for market capitalization is -1.6508 while the slope for stocks traded is -4.049 implying a greater impact on economic growth, which is not surprising as it is well documented that foreign direct investment has a significant impact on growth (see for example Alfaro et all, 2004). This might also explain why the significance for stocks traded is lower in the linear model.
34 Measurement % of GDP .0077 (.0008) *** .0047 (.0011) *** .0076 (.0010) *** % of GDP .0066 (.0006) *** .0049 (.0010) *** .0057 (.0008) *** % of GDP .0018 (.0006) *** .0023 (.0011) ** .0020 (.0008) *** % of GDP .0002 (.0010) .0088 (.0035) ** .0008- (.0010) .5942 .5903 .4154 .3385 .5823 .5570 .2613 .1627 .6182 .6166 .4635 .4251 2240 2233 620 770 1061 1055 172 258 1179 1178 448 512 2001-2015 R2 N Credit total Stocks traded
Notes: coefficients in this table are the result of an instrumental variables regression using the GMM option because of endogenous regressors. Instrument was a lagged value of financial development. Every regression included time dummies for 5 non-overlapping years and standard control variables (not reported here). Standard errors are in parantheses. The symbols are * = 90 percent significance, ** = 95 percent significance and *** = 95 percent significance.
Credit by banks
Market capitalization
Variables
Table 7: IV regression results different measurements
Time period
35 Measurement % of GDP .0186 (.0021) *** .0170 (.0025) *** .0149 (.0036) *** .0001- (.0000) *** .0001- (.0000) *** .0001- (.0000) ** % of GDP .0184 (.0020) *** .0192 (.0025) *** .0140 (.0031) *** .0001- (.0000) *** .0001- (.0000) *** .0001- (.0000) *** % of GDP .1612 (.0127) *** .3041 (.1111) *** .0661 (.0085) *** .0007- (.0001) *** .0013- (.0005) *** .0004- (.0000) *** % of GDP .2613 (.0201) *** .0259 (.0077) *** .1277 (.0143) *** .0023- (.0002) *** .0001- (.0001) * .0010- (.0001) *** P-value .0000 *** .0000 *** .0000 *** .0000 *** .0001 *** .0000 *** .0037 *** .0771 * .0635 * .0216 ** .0000 *** .0000 *** Slope .0186 .0184 .1612 .2613 .0170 .0192 .3092 .0256 .0149 .0140 .0661 .1277 Slope .0378- .0351- 1.6508- 4.049- .0215- .0303- 1.9469- .0570- .0215- .0209- .7095- 1.8296-.6004 .5967 .8819 .1724 .5867 .5833 .4340 .1159 .6215 .6202 .9826 .7635 2240 2233 620 770 1061 1055 172 258 1179 1178 448 512
Credit by banks squared
Credit total squared
Market capitalization squared
Stocks traded squared
Notes: coefficients in this table are the result of an instrumental variables regression using the GMM option because of endogenous regressors. Instrument was a lagged value of financial development. Every regression included time dummies for 5 non-overlapping years and standard control variables (not reported here). Standard errors are in parantheses. The symbols are * = 90 percent significance, ** = 95 percent significance and *** = 95 percent significance.
U-test Lower bound Upper bound Credit by banks Credit total Market capitalization Stocks traded R2 N
Table 8: non-linear IV regression results different measurements
Variables Time period
36
VII. CONCLUSION AND DISCUSSION
The financial sector plays a crucial role in the world that we live in today. What happens in that sector result in changes throughout the whole economy. Because of its real-world consequences understanding the finance-growth relationship is important. This thesis tried to increase the knowledge about this relationship. The first thing to do was to research the current literature- both theoretical and practical. This gave a summary of current knowledge about the topic. By using a two stage least squares approach and an extensive database for a long period of time, the relationship between finance and growth was tested.
The first version of the model was linear and tested the three measures of finance over different time periods. The majority turned out to be significantly positive, which is in line with earlier work testing linear models. In a second version of the model the entire sample was split between low income and high income countries. This was done because the level of financial development differs between these two groups, so they might also have different effects. These results showed that some measures of finance indeed have different effects on economic growth depending on whether the sample consisted of low or high income countries.
To confirm more recent theories that imply that there could also be too much finance, a non-linear version of the same model was tested. To test for significance of the non-non-linearity a U-test was employed. In this case, the U-U-test U-tests for an inverse U-shaped relationship. This means that at low levels of financial development finance fosters growth, but at higher levels of development finance decreases growth. Results vary depending on the measure of financial development. While the measure of liquid liabilities showed an insignificant relationship, the measures of credit and market capitalization do show a significant relationship. In addition, the U-tests for these measures confirm an inverse U-shaped relationship. The same set of regressions were then also conducted for the split sample of high and low income countries, showing no big differences except for the credit measure. Instead of an inverse U-shaped relationship, the U-test indicates an U-shaped relationship for credit for high income countries. In summary, the majority of these results confirm research indicating an inverse U-shaped relationship.
As with any research, this thesis has some important limitations. First of all, the variables used to measure financial development do not cover the level of financial development completely. They are mostly related to financial depth. This has also been pointed out by earlier papers, but
37 as of today no new variables have been initiated. On the other hand, the consistent use of the same variables between papers makes it easier to compare results. Furthermore, the variable for market capitalization has significantly less observations than the other variables. A smaller sample size might undermine the reliability of the results. However, the low number of observations is not something that could be easily changed. All available data is already included in the World Bank database and missing data is simply the result of no data available for that country in that particular year. Financial markets only recently developed in many countries. Lastly, while this thesis tried to minimize endogeneity problems with advanced econometric techniques it did not solve these problems completely. For instance, lagged values of the financial development measures were used as instrument. Even though it was rejected that this was a weak instrument, there might still exist better suited instruments with greater predictive power.
With regard to policy implications one needs to be extremely careful. The results of this thesis cannot be directly translated to economic policy. For example, the results of the non-linear regression do not imply that a country that has a high level of financial development that is in the upper bound of the relationship could decrease the level of finance and immediately increase growth rates.
That being said, there are a variety actions that can be taken to address failures in the financial system that can benefit economic growth. When dealing with an excessive amount of credit that is being lend out, one could increase regulation. One example is to increase the amount of capital buffers required. Furthermore, regulation could be developed that addresses the too-big-to-fail theory by splitting banks up so that failure of one bank does not start a series of disastrous events for the economy. Another issue that needs more regulation is the compensation and distortive incentives in the financial system, as they increase risk. In addition, special focus needs to be given to decreasing the instability of the financial sector and its interconnectedness with the rest of the economy to ensure that trouble in the financial sector does not translate to the rest of the economy.
