Reassessing the “Too Much Finance” Hypothesis. Does the credit structure matter for economic development?

Reassessing the “Too Much Finance” Hypothesis.

Does the credit structure matter for economic

development?

MSC Thesis

University of Groningen

Faculty of Economic Development and Globalization

Supervisor: Prof. Dr. Dirk Bezemer

Co-assessor: Dr. Andreas Steiner

Student: Sipovalova Irina

Student number: 3803163

Abstract

The aim of this paper is to show that not only the level of credit is related to economic development, but also the structure of the credit. Based on the broader dataset of 124 countries over the period 1991- 2018, the non-monotone relationship between private-credit and economic growth is identified; more precisely finance starts having a negative effect on output growth when credit to the private sector reaches 85% of GDP. Based on data for 43 countries, it was shown that household credit stocks as well as the non- financial credit flows have a “diminishing” effect on economic growth.

Table of Contents

1. Introduction ... 4

2. Literature Review. ... 8

2. 1. Financial development and economic growth.... 8

2.2. Financial Development, economic growth and different use of credit in the economy ...11

2.3. Financial development, economic growth and Country Heterogeneity. ...13

3. Methodology ... 14

3.1. Variables ...14

3.2. Empirical model ...17

4. Data description ... 20

5. Empirical results ... 25

5.1. Private credit and economic growth ...25

5.2. Household credit, non-financial corporate credit and economic growth...31

5.3. Robustness check ...35

6. Conclusion ... 37

Reference List ... 39

Appendix A. List of variables. ... 43

Appendix B. List of countries included in the dataset ... 45

Appendix C. Histogram of variables included in the model... 47

Appendix D. GDP growth over the period 1991- 2018. ... 48

Appendix E. Test for heteroskedasticity and Hausman test ... 49

1. Introduction

A large empirical literature has established a positive link between financial development and economic growth by injecting liquidity into the economy through credit until the beginning of 2000s. Most recent research, on the other hand, suggested that the effect is ambiguous and uncertain.

The theoretical as well as empirical evidences proposed different explanations of the “vanishing effect”. More precisely the idea of too much finance has been put into place. The empirical studies have identified a threshold, above which the finance- growth nexus has a “diminishing effect” and financial development slows down output growth rather than stimulate economic development (Arcand et al. 2012, Beck et al. 2012, Cecchetti & Kharroubi 2013). The growth in non- intermediation activities by banking sector, growth of non- business credit exacerbated the “diminishing effect”. Moreover, the expansion of the financial sector leads to movements of skilled labor from R&D intensive sector, which has a positive effect on output growth to financial sector (Cecchetti & Kharroubi, 2013). Furthermore, the development of financial sector is not always coupled with sufficiently developed regulatory environment. The explanation provided by the previous literature underpins the diminishing effect of financial development of economic growth. The assumption regarding countries’ homogeneous response generates misleading results based on the research conducted by Ram (1999). A large empirical literature suggests that legal, regulatory and institutional frameworks are necessary for efficient resource allocation.

Furthermore, the recent empirical papers have established the importance of the distinction between credit directed to enterprises or to households. The evidences mostly support the idea that enterprise credit is productive credit, which tends to have a positive effect on economic growth. The effect of household credit is less conclusive due to the fact that it can stimulate the increase of debt- to- burden ratio and inflate assets market without having an impact on output growth. The stocks and flows of the credit is essential to consider since credit stocks are debt levels, which can harm economic growth. Credit flows, on the other hand, stimulate economic growth by injecting liquidity.

43 countries. The time period is dictated by data availability as well as the fact that after the breakdown of the Soviet Union fifteen new economies emerged. Since most of the countries are included into the analysis, the systematically missing data could stimulate biasedness of the coefficients. By looking at the data, the expansion of private credit and more prominently of household credit has been observed over the period. Therefore, a significant part of financial development was driven by increase in household credit. Based on the balanced panel data for 43 countries the average ratio of household credit to GDP has risen from approximately 35 percent in 2000 to 58 percent in 2018. Therefore, the rise in household credit can explain “the diminishing effect” of finance- economic growth nexus.

This paper possesses objectives similar to the ones analyzed by Arcand et al. (2012), namely the research aims to provide empirical evidences for “too much finance” hypothesis by identifying the threshold, above which the financial development has a “diminishing effect” on economic growth. However, in contrast to the research conducted by Arcand et al. (2012), this paper looks at heterogeneous response of countries in providing credit and more recent time period. The aim of this paper is to show that not only the level of credit is related to economic development, but also the structure of the credit. The research paper hypothesize that the use of credit, household or non- financial corporate credit, matters for the economic effectiveness. This is especially relevant within current circumstances and economic slow down induced by coronavirus.

Furthermore, this paper is closely related to research conducted by Bezemer et al. (2016) since it makes credit stock- flow distinction. However, in contrast to this paper, a larger time period is reviewed and the non- monotone relationship between household and non- financial corporate credit is considered. The main argument behind the inversed U-shaped relationship is the fact that households and non- financial corporations are private sectors. If the aggregate measure of private credit follows the non- monotone relationship, it is assumed that the effects of disaggregate measures of the private credit on economic growth can be diminishing as well.

dynamic nature of the finance- growth nexus and endogeneity. The methodology aims to provide reliable and credible evidences to answer the main research question:

“Does the credit structure matter for economic growth?”

Indeed by looking at the findings, a non- monotone relationship between financial development measured by private credit-to-GDP ratio and growth rate of GDP is observed. According to the threshold regression model for fixed effect panel data the expected threshold is at 85 percent of private credit to GDP. This result is close to the one obtained by Arcand et al. (2012). The non- monotone relationship between private credit and economic growth is observed across all countries. However, the magnitude differs, more specifically low- income economies experience largest positive as well as diminishing effect on economic growth. Low- income countries might experience the most pronounced “diminishing effect” due to lack of qualitative regulatory institution, which could efficiently allocate the resources. However, this conclusion should be further analyzed.

The result on credit structure provides less conclusive and significant results, more specifically the findings supports a non- monotone relation between household credit stock and economic growth. Household credit can positively affect growth through increase in human capital accumulation (De Gregorio, 1996); however, economic growth can slow down by increasing debt-to- burden ratio and stimulating volatility above the threshold. By analyzing the credit flows, it can be concluded that the “short term liquidity” injections of non- financial corporate credit have a positive effect on economic growth. However, the excessive flows of funds generate “diminishing” returns. During the “booming” phase, the excessive credit is a result of lowering loan standards and lack of supervision, which stimulate inefficient allocation of resources. The results are robust after controlling for business cycle fluctuations and quality of the government indicator provided by the ICRG.

A short time period as well as limited amount of countries engaged in the analysis of credit structure and its impact on economic growth is considered to be one of the main limitations of the research. The list of countries mostly includes high-income economies and as a consequence the relationship between household, non-financial corporate credit and economic growth is difficult to apply for low, and middle-income countries.

2. Literature Review

The link between financial development and economic growth has been widely discussed by economists such as Bagehot (1873) and Schumpeter (1911). According to Schumpeter (1911), the role provided by financial intermediaries is essential for economic development and technological innovation through creative destruction (King and Levine, 1993). Furthermore, Schumpeter acknowledged a close tie between the financial system and economic growth and the importance of injecting liquidity into the economy by providing credit for entrepreneurship and innovation.

2. 1. Financial development and economic growth.

The effect of financial development on economic performance has received a lot of attention at the empirical level suggesting that a well-developed financial system accelerates economic growth. Goldsmith (1969) was the first to acknowledge the positive link between the size of the financial system and economic development through efficiency improvements of the banking system. However, at that point he did not provide any attempts to establish the existence of causal link going from financial depth to economic growth. From the 1990s onwards, the economists started to explore the causal link between finance and economy. Based on the cross-country analysis over the 1960-1989 period, King and Levine (1993) identified that higher levels of financial development are positively correlated with economic growth and physical capital accumulation after controlling for country and policy characteristics. Furthermore, they stated a positive link between financial development and long-run growth. Levine et al. (2000) highlights more evidences of causality, which were not addressed by King and Levine (1993). For instance Levine et al. (2000) addresses the causality issues, biases induced by simultaneity, unobserved country-specific effects and omitted variable biases by using new data and different econometric practices. Based on the insights provided by LaPorta et al. (1997), Levine et al. (2000) uses legal origin as an instrumental variable to control for simultaneity. According to LaPorta et al. (1997) legal origin has an impact on legal and regulatory environment and explains the difference of financial intermediary development. The results favor the growth-enhancing view of financial intermediation. Moreover, legal origin and contract enforcement influence a country's ability to provide better financial services.

between financial development and economic growth for all countries and have demonstrated that patterns vary across different economies. It is argued that financial development follows economic growth due to increased demand for financial services by enterprises, namely the need to store money as banks’ deposits stimulated financial development. Joan Robinson (1952) was the first to acknowledge that enterprises lead while the financial sector follows. Demetriades and Hussein (1996) cited Adams (1819), who claimed that financial intermediaries harm the “morality, tranquility, and even wealth” of the nation. One of the explanations is that higher returns from more efficient resource allocation decrease saving rate and slow down economic growth as a consequence. Furthermore, the restrictions on financial system, namely reserve requirement, interest rate ceiling slow down economic growth.

More recent research shows a non-monotone relationship between financial development and economic growth, namely at some point a high ratio of credit to GDP may slow down economic growth rather than boost the economy. While the empirical evidence of non- monotone relationship is more recent, the idea of non- monotone relationship is hardly new. Minsky (1974) and Kindleberger (1978) explicitly described financial instability, volatility and crises. Minsky moment is defined as a point when excessive leverage created by financial sector during the long period of steady prosperity, optimism collapses and financial system as well as the whole economy enters the recession phase.

Rajan (2005) suggested that developments in the financial sector led to an expansion of risks. The excessive risks come from the originate- to- distribute model, namely lenders make loans with the intention to sell it to other institutions. The changes, which took place in the financial sector, have changed the nature of the typical financial transaction by allowing broader participation and risk sharing in the economy by transforming illiquid assets into liquid liabilities. Even though the improvements in the financial sector provided more opportunities and better access to funds for households and business, the presence of a whole range of intermediaries, whose excessive desire for risks expands during the booming phase, leads to a bust. During the bust phase investors and speculators are faced with margin calls and want to liquidate at any price.

on economic growth co-exists with financial fragility in the short-run. The theoretical explanation was proposed by Dell’Ariccia and Marquez (2004) suggesting that financial liberalization is as a phase of excessive demand for credit. During the “boom” bankers do not have an incentive to screen new projects, which request funding and therefore the risks increase. An expansion of credit portfolio leads to over-lending, increasing number of non-performing loans and financial fragility as a consequence. Therefore, in the short-run, financial deepening leads to volatility while produces output growth in the long- run. Rajan (1994) explained that proper banking supervision is required in order to control for inefficient lending. In the short-run, poor development of supervisory capacity leads to credit expansion, booms and busts. As institutional and supervision quality develops, booms and busts are gradually replaced by stable financial depth. In contrast to Loayza and Ranciere (2006), who only looked at financial sector size, Beck et al. (2012) disentangle the size versus intermediation activities. This paper supports that intermediation activities have a positive link with growth and reduction of volatility. In contrast to previous results, the size of the financial sector is not associated with long-run growth and volatility by controlling for intermediation activities.

2.2. Financial Development, economic growth and different use of credit in the economy

Schumpeter highlighted that economic development requires additional liquidity in the form of credit to stimulate production and innovation. Therefore, banks act as an intermediation and source of additional liquidity in the economy by creating money. Banks create new money when they make loans. By providing loans, the asset side of the balance sheet as well as liabilities side increases as new deposits are created. But the use of credit determines whether the economy's GDP rises or not. Based on Schumpeter’s theory of development two types of credit are identified, namely productive and unproductive. Productive credit aims to support innovative ideas, which drive capitalist development while unproductive one stimulates booming phase and speculation. During the booming phase, optimistic expectations based on past experience of success lead financial intermediaries to soften lending standards; to create new financial instruments, e.g. securitization in order to be able to provide more loans, to validate riskier projects, and to reduce margins of safety. Therefore, the booming phase can lead to euphoria, feeling of stability and prolonged prosperity. However, according to Minsky (1974) “Stability is destabilizing” and can lead to a bust, when the ability to finance debt declines and the borrowers are required to sell their assets to meet payments, which in turn puts a downward pressure on prices. Most of the theoretical models within finance- economic growth nexus focus on financial development solely from the firm’s perspective (Beck, 2012). While the effect of household credit on economic growth is less conclusive. Jappeli and Pagano (1994) suggested that by lowering liquidity constraints on household, the savings rate decreases and economic growth slows down. Furthermore, household credit leads to increase in debt- burden ratio if it is not used in income- generating activities or even stimulate volatility and likelihood of the crisis. De Gregorio (1996), on the other hand, found a positive effect of household credit through increase in human capital accumulation.

research conducted by Beck et al. (2012) enterprise credit has a positive and statistically significant impact on GDP per capita while household credit not. The research conducted by Sassi & Gasmi (2014) confirms the conclusion derived by Beck et al. (2012), namely enterprise credit enhance economic growth through capital accumulation, productivity growth and efficient resource allocation. Household credit tends to negatively affect economic growth in line with the previous empirical evidence. In contrast to Beck et al. (2012), this paper aims to explain heterogeneous response across countries due to disparity of the composition of credit across countries.

The research conducted by Bezemer et al. (2016) further suggests rethinking the relation between banks and economic growth nexus, more precisely distinguishing between growth effects of stocks and flows of credit. Furthermore, the research disaggregates credits into two broad categories, namely “nonfinancial” credit (credit to nonfinancial business and consumption credit) and “asset market” credit (mortgage loan and credit to financial businesses). According to Bezemer et al. (2016) the correlation between banks’ credit stocks and output was not significantly different from zero during the 1990s and 2000s across 50 countries. This paper cited the following explanations, namely Wachtel (2011) proposed that credit/ GDP ratio could represent financial fragility rather than financial deepening. Beck et al. (2012) highlighted the increasing share of household credit in total credit. Since household credit does not have a positive or statistically significant impact on GDP per capita, the correlation between banks’ credit stocks and output was not significantly different from zero. Cecchetti and Kharroubi (2013) represented evidence of the importance of human capital, which moved from skill-intensive industries to the financial sector during the credit boom. As a result, the financial sector develops faster than the real economy. Taking into account the distinction of flows of credit, Bezemer et al. (2016) shows a positive link between economic growth and credit flows to nonfinancial institutions, but not for mortgages and assets market based on the data for 46 economies over 1990- 2011. Furthermore, new trends in bank lending are no longer beneficial for growth and can potentially lead to crisis. Therefore, the role of banks in promoting economic growth should be questioned, namely use of credit is essential in explaining economic performance.

threshold model for corporate and household debt separately for 18 OECD countries from 1980 to 2010.

2.3. Financial development, economic growth and Country Heterogeneity.

Arestis and Demetriades (1997) highlights that there is no one-size-fits-all relationship between financial development and growth suggesting that institutional factors, legal origin and income level as well as other country specific characteristics may have different impacts on financial- economic growth nexus. Ram (1999) supports that the results based on a country's homogeneous response are ambiguous and uncertain. Ram (1999) shows a significant heterogeneity across countries, more precisely high-growth economies gain more from financial development than the rest of the sample. In contrast to Ram (1999), Demetriades and Law (2006) find that the impact of financial development on economic growth has its largest effect when the financial system is embedded within a sound institutional framework. Low-income and middle- income countries do not observe economic growth if the financial system is not coupled with well-developed institutional factors. High- income countries benefit from development of financial system and high quality institutions. However, the coefficients are smaller than the ones for middle-income countries. Even though the results derived from different econometric models discussed above are inconsistent and blurred, the importance of countries' heterogeneity, time period is essential to address as different sub-samples might have a different response.

Bezemer et al. (2016) highlights that on average the total credit-to- GDP ratio increased from 75% to 120 % over the period 1990- 2011 for both developed and emerging economies. However, the composition of credit differs between countries based on their income level, more precisely the share of nonfinancial credit is negatively correlated with income level. Rioja and Valev (2004) further acknowledge a different finance- economic growth relationship for developed and emerging economies.

the reasons, which can partially explain a negative relationship between financial development and economic growth, is the use of the credit. The excessive credit if not used in income- generating activities leads to increase of debt- burden ratio and volatility as a consequence.

The credit structure, on the other hand, is related to the economic development as well. While literature mostly focuses on enterprise credit, the financial sector in highly developed economies largely provides household credit. Due to the fact that overall empirical evidence suggests a weak relationship between household credit and economic growth, the finance– economic growth nexus is different in those countries with a bigger share of household credit. The enterprise credit, on the other hand, has a positive effect on output growth. Moreover, credit flows- stocks distinction suggested by Bezemer et al. (2016) is important for economic growth outcomes due to the fact that flows can stimulate economic growth through “short-term liquidity injections” while high levels of credit stocks can theoretically create instability and volatility in the economy.

By analyzing previous empirical literature, it was concluded that the non- monotone relationship between household and non- financial corporate credit stocks and flows has not been properly analyzed. Even though, previous empirical findings looked at the linear relationship, the inversed U-shaped relationship was not considered. The research paper hypothesize that the relationship between household and non- financial corporate credit follows the non- monotone relationship identified for the private credit. Due to the fact that households and non- financial corporations are primary private sector, there are evidences to believe that the inversed U- shaped relationship exists. Furthermore, Cecchetti et al. (2011) has shown the threshold, above which the effect of household and non- financial corporate credit diminishes. However, the model did not account for endogeneity. Based on the more recent dataset, the research paper aims to prove non- monotone relationship between private credit and economic growth across all country groups based on the income level. Lastly, the inversed U- shaped relationship between disaggregate measure of household, non- financial corporate credit stocks/ flows and economic growth will be analyzed.

3.

Methodology

Arcand et al. (2012), Bezemer et al. (2016) and Beck et al. (2012). These papers are a good benchmark within most recent empirical evidences. However, it also deviates in a few ways described in this section.

Most of the literature that examines the financial-economic nexus uses the real growth rate of GDP per capita in US dollars as a proxy for economic development. The growth rate of real GDP is considered to be a good proxy due to the fact that it accounts for countries’ population and population growth. Furthermore, it takes under consideration inflation and deflation; as a consequence we observe an increase of output due to increase in volumes rather than price increase. Following the methodology described by Arcand et al. (2012), credit to the private sector is used to quantify the financial depth. Levine (2005) suggests that all empirical methods and proxies have their problems; credit to the private sector is not an exception. Theoretically, growth is observed through reduction in transaction costs and by easing information about firms and corporate governance. However, empirically these financial functions are difficult to measure. Arcand et al. (2012) acknowledged the imperfection of credit to the private sector variable and argues that at this stage it remains the best indicator to capture financial depth available for large cross-country analysis. Levine et al. (2000) uses three different indicators of financial intermediary development, namely liquid liabilities, ratio of commercial bank assets divided by commercial banks plus central banks assets and private credit as a percentage of GDP. According to the methodology used, private credit is a preferred indicator to measure the size of the financial sector. The exclusion of credit issued by the monetary authority and government agencies is the biggest advantage of using private credit variable. The central bank’s interventions do not allow to solely focus on the banks’ effectiveness in researching firms, allocating savings, providing corporate control, and reducing transaction costs. Even though private credit variable does not directly measure the transaction costs and the amelioration of information, Levine et al. (2000) interprets that higher levels of private credit indicator represents higher levels of financial services and consequently greater financial intermediation. Based on the literature review section, a quadratic form of private credit variable is used in order to account for non-monotone relationship.

distinction between the uses of the credit is essential due to the fact that they aim to support different economic transactions. Furthermore, in line with the methodology described by Bezemer et al. (2016) both stocks as well as flows of nonfinancial corporate credit are considered. The main argument behind using both stocks and flows of the credit is the fact that credit stocks are debt stocks, which may lead to instability and slow down economic growth as a consequence. Therefore, the increase of credit stock theoretically follows the inversed U- shaped relationship proposed for private credit and economic growth. Since households and non- financial corporations are private sectors, the quadratic term is included in order to control for non- monotone relationship between household and non- financial corporate credit stocks. Based on the following formula the credit stocks are calculated:

Stock i, t (Household or non-financial corporate) = (Measure of credit in billion US

dollar)i,t / GDPi, t (1)

where i denotes country, t denotes time.

Credit flows, on the other hand, are “short- term liquidity effect” on GDP, which aim to stimulate the ability to fund expenditures. By taking loans at the financial intermediary, the amount of deposits increases on bank’s liability side. These deposits cause more transactions between economic agents and stimulate GDP growth as a consequence (Bezemer et al. 2016). The transactions used in a productive way, namely to stimulate production of good and services, enhance total factor productivity will lead to output growth. The transactions in asset market, on the other hand, inflate prices rather than stimulate economic growth. Based on the following formula the credit flows are calculated:

Flows i, t (Household or non-financial corporate) = (CI,t - CI,t -1)/ GDPi, t-1 (2)

, _{where i denotes country, t denotes time and C denotes to measure of household or}

non-financial corporate credit in billion dollars.

Furthermore, the quadratic term of credit flows is included since excessive credit flows can be a signal of disproportional growth of loans and defaults in a period of low policy rates, relaxing loan standards, which lead to “diminishing effects” on growth effectiveness.

size. The last set of control variables possesses information regarding political stability. Therefore, the standard for literature on financial depth and economic growth is to control for the following variables: the initial level of real GDP per capita, trade openness, inflation, enrollment in secondary education to control for human stock, the general government consumption as a percentage of GDP. The list of variables included in the empirical model could be found in Appendix A and it is agreed within the previous researchers, e.g. Levine et al. (2000), Arcand et al. (2012) as well as more recent studies by Karagiannis & Kvedaras (2016), Cave et al. (2019). The exclusion of those variables might lead to omitted- variables bias (Levine et al., 2000).

3.2. Empirical model

By analyzing the existing literature, it was concluded that the “vanishing effect” of the financial development, countries’ heterogeneity, simultaneity, different use of credit and credit stock- flows distinction should be taken under consideration. Firstly, the objective of the research paper is to reassess the inverse U-shaped relationship between financial development and economic growth. This is done in order to support evidences of the non- monotone relationship based on more recent time period. Secondly, the paper aims to provide empirical evidence of impact of household’s and non-financial corporate’s credit on the economy. Based on the findings confirming the inversed U- shaped relationship of private credit on the economy, household and non- financial corporate credit is assumed to follow non-monotone relationship as well since households and non- financial corporations are private sector. Firstly, the estimated model for impact of total private credit to economic growth is:

g(i, t) = + 1g(i, t-1) + 2PC(i, t) + 3PC2(i, t) +  C(i, t) + i + t +(i, t) (3)

, where i and t stand for country and time period respectively.

g(i, t) is the real growth rate of GDP per capita of country i in year t. Based on the

histogram shown in Appendix C, a normal distribution of real GDP growth variable is observed. Therefore, no transformations are required. The financial development variable is measured by log of PC private credit variable as percentage of GDP, while PC2 _{is used in order to capture non-monotonic relationships suggested in the}

literature. The explanatory variables C include control variables such as inflation, years of schooling, openness to trade, government expenditure as a percentage of GDP, initial level of real GDP per capita and institutional quality. Furthermore, the following baseline model includes unobserved country-specific time-invariant effects

researches financial development tends to have positive impact on financial development up to a point, above which the impact has a “diminishing effect”. In line with the previous empirical results, it is expected that 2 > 0 while 3 < 0 and the

coefficients support the “vanishing effect”.

The second step aims to disaggregate total private credit into two categories, namely household credit and nonfinancial corporate credit as percentage of GDP. Secondly, the research paper will make the distinction between the effect of credit stocks and flows as suggested by Bezemer et al. (2016). Previous researches mostly looked at linear relationship between these variables, e.g. Beck et al. (2012), Bezemer et al. (2016). This paper deviates from previous empirical findings due to the fact that it is assumed that household and non- financial corporate credits follow the inversed U-shaped relationship confirmed for private credit. Therefore, the baseline model includes the quadratic term of both household as well as non- financial corporate credit.

g(i, t) = + 1g(i, t-1) + 2NCD(i, t) + 3NCD2(i, t) + 4HH(i, t) + 5HH2(i, t) + C(i, t) + (i,)

+ t +  (i, t), where i and t stand for country and time period

respectively. (4)

2NCD(i, t) captures the effect of nonfinancial corporate debt while 4HH(i, t) looks at

the household credit. Both non- financial corporate as well as household credits are logarithmically transformed. Furthermore, the following baseline model includes unobserved country-specific time-invariant effects, set of control variables described in model (3), time effects and an error term with zero mean. Firstly, the stocks of credit will be included into the model and secondly the flows of household and non-financial corporate credit will be included in the fourth model.

The final step is to control for both stocks and flows of the household and non- financial corporate credit simultaneously due to the fact that the effect of stocks or flows separately can be overestimated. Both stocks and flows of credit include the quadratic term in order to account for “diminishing effect” of excessive credit in the economy. Therefore, the fifth baseline model is as follows:

g(i, t) = + 1g(i, t-1) + S(i, t) + S2(i, t) + F(i, t) + F2(i, t) + C(i, t) +  (i,) + t +  (i, t)

(5)

In the robustness check, the three- year periods are computed in order to capture business cycle fluctuations in line with the research conducted by Bezemer et al. (2016). In contrast to yearly observation, the estimated three- year period possesses significant loss of observations. Furthermore, an index of institutional quality is included in the robustness check.

instruments for the first- difference equations and lagged differences as instruments for the level equation. Since financial development measured by credit variables are endogenous to growth, these variables are instrumented by their lags in levels and in differences in the GMM system model. The lagged variables are internal instruments since they are obtained from the existing econometric model. By combining system of equations in differences and in levels, the empirical method aims to provide the unbiased and reliable coefficients. In order to test for overall validity of the instruments and the absence of serial correlation of the residuals, the Hansen test for over-identifying restrictions and test for second order serial correlations AR (2) is respectively performed. Due to the presence of heteroskedasticity and serial correlation within the data, the two-step system GMM estimator proposed by Arellano and Bond (1991) is used. In order to correct for small sample bias and to obtain robust standard errors, the Windmeijer (2005) adjustment is applied.

4. Data description

The analysis is based on a country-level data over the period 1991 to 2018 for 124 countries. The time period was chosen based on the data availability as well as due to a collapse of the Soviet Union in 1991, which resulted in emergence of fifteen independent republics. In order to obtain a balanced panel dataset, it was decided to focus on the following period 1991- 2018. The data regarding economic development proxied by annual real growth rate of GDP per capita and measured in percentage points was obtained from the World Bank database. The financial development variable is proxied by three different variables, namely private credit as a percentage of GDP, stocks and flows of household credit and nonfinancial corporate credit. The following table 4.1 will provide the basic descriptive statistics of the chosen variables. Credit to private sector as percentage of GDP was obtained from the World Bank database over the period of 1991- 2018 for 124 countries (the full list of countries can be found in Appendix B), while household credit and nonfinancial corporate credit was obtained from the Global Debt Database over the same period, but due to data availability, the analysis is limited to 43 countries. Credit flows were calculated using the formula described in section 3.1.

normally distributed around the means. Furthermore, by looking at the raw data, the hyperinflation immediately stands out. After the collapse of the Soviet Union, newly emerging countries have experienced a hyperinflation. It was decided not to remove those data points since they represent economic conditions in those countries at that time. However, the variable was logarithmically transformed in order to normally distribute the data and account for outliers.

Based on the descriptive statistics private credit reaches approximately 309 percent of the GDP, while household and non-financial corporate credits reach 177 percent and 190 percent respectively. However, all the variables were logarithmically transformed to deal with outliers and to normally distribute the observations around the mean. Due to the fact that private credit and household credit variables had originally extremely low level of credit- to-GDP ratio, namely below 1 percent of the GDP, the negative observations are shown in the table. However, by looking at the Appendix C, the observations are normally distributed around the mean. Even though, the GMM estimation does not require satisfying the assumptions of normality of the errors, the OLS and FE models fail to provide consistent coefficients when the assumption is violated. The credit flows variables are normally distributed and therefore no transformations are required. Appendix C provides the histograms for all the variables included in the model in order to ensure that the assumptions of normality is not violated.

Table 4.1. Descriptive Statistics

Variable Unit Obs. Mean Std.Dev. Min Max GDP growth Percentage points 3583 3.544 3.986 -8.933 19.675 Private credit % of GDP in log 3471 3.553 1.006 -.711 5.733 Household credit stock % of GDP in log 1203 3.525 1.079 -1.949 5.159 Non-financial corporate

credit (NFC) stock

% of GDP in log 1203 4.2 .633 2.215 5.244 Household credit flow % of lagged GDP 1157 2.936 6.105 -13.78 37.492 Non-financial corporate

credit (NFC) flow % of lagged GDP 1157 4.485 8.967 -19.446 38.525 Inflation In log 3583 1.441 1.11 -4.791 4.226 Openness to Trade In log 3583 4.325 .57 2.439 6.093 Initial GDP per capita In log 3583 8.391 1.59 4.631 11.685 Government Effectiveness Index 3583 .224 .975 -1.88 2.44 Years of schooling Years 3583 8.06 2.989 1.2 13.7 Government expenditure % of GDP 3583 16.037 6.025 3.208 76.222 ICRG Index 3291 67.533 12.275 14.25 96.083 Low Dummy variable 3583 .086 .280 0 1 Middle Dummy variable 3583 .515 .499 0 1 High Dummy variable 3583 .399 .490 0 1

Figure 4.1. Evolution of Private Credit over Figure 4.2a. Evolution of Private Credit over the the period 1991-2018. period 1991-2018 in high-income economies.

Figure 4.2b. Evolution of Private Credit over Figure 4.2c. Evolution of Private Credit over the period 1991-2018 in low-income economies. the period 1991-2018 in middle-income economies

As it was mentioned in the literature review section, household credit and credit to non-financial businesses are two principal categories of private credit on a bank's balance sheet. By splitting private credit into two subcategories, the following patterns have been obtained on Figure 4.3a and 4.3b. Both figures are based on the balanced data for 43 countries over the period 1991- 2018 due to data availability. In comparison to household credit, the non-financial corporate credit on figure 4.3b reacts more to the external conditions due to the fact that more up and downs could be observed. On the other hand, surprisingly increase in household credit is more stable and gradual over the period. The non- financial corporate credit has experienced a sharp decrease during financial crisis 2008 while managed to recover and surpass the level before the financial crisis by reaching approximately 85 percent of the GDP. Household credit, on the other hand, has been gradually increasing from the beginning of 90s. However, after the financial crisis on average the level of household credit remains stable at approximately 58 percent of the GDP.

Figure 4.3a. Evolution of Household Credit Figure 4.3b. Evolution of Non-financial over the period 1991-2018. corporate credit over the period 1991-2018.

Netherlands, where the household credit prevails, the overall pattern in credit structure can be seen in table 4.2. Non- financial corporate credit prevails in all country groups. However, the share of household credit increases with the level of countries’ economic development. In low-income economies the household credit remains relatively small. The low share of household credit in low- and middle- income countries reflect volatile macro- economic conditions in these economies since it creates uncertainty about future income. Therefore, households are uncertain about future sustained economic growth rates, inflation, interest rate and as a consequence the demand for loans is weak.

Table 4.2. Credit structure by country group.

Country Obs. Credit to GDP

Household Non-financial corporate By income: Low- income Middle- income 84 252 8.573 24.637 28.037 46.252 High- income 868 56.502 83.632

Before implementing econometric regressions, it is crucial to look at the correlation between variables as they can signal regarding multicollinearity issues within the model. By looking at the correlation matrix Table 4.3, the negative correlation between all three different types of credit and GDP growth within the time period is observed. This is in line with the findings suggested by Bezemer et al. (2016). However, the negative correlation of non-financial corporate credit and GDP growth contradicts a significant body of the empirical literature. Credit flows, on the other hand, have a positive correlation with GDP growth supporting the view that credit flows are perceived as “short- term liquidity injections” to stimulate economic growth, while credit stocks are debt levels, which can create volatility.

development and it significantly contributes to economic growth. Most of the literature regarding openness to trade and economic development support a growth-enhancing view through productivity improvement and efficient allocation of resources. Therefore, a positive correlation between openness to trade and GDP growth is expected. The government effectiveness has a positive correlation with economic growth through improvements in transparency and credibility in government commitment to implement policies. The “years of schooling” variable is positively and significantly correlated with GDP per capita. A positive correlation goes through ability to obtain a higher level of education, to participate in skill-intensive industries, which are mostly specialized on innovation, R&D. A negative correlation of inflation with all the variables induced in the model is observed due to the fact that economists tend to perceive inflation as an obstacle for credit expansion and economic development.

Table 4.3. Matrix of correlations

Variables (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (1) Real GDP growth 1.000 (2) Private Credit -0.137 1.000 (3) Household credit stock -0.268 0.682 1.000 (4) NFC stock -0.171 0.542 0.719 1.000 (5) Household credit flow 0.122 0.219 0.257 0.148 1.000 (6) NFC flow 0.143 0.126 0.122 0.225 0.748 1.000 (7) Inflation -0.114 -0.396 -0.464 -0.419 -0.014 -0.042 1.000 (8) Openness to Trade 0.068 0.174 0.329 0.420 0.117 0.188 -0.265 1.000 (9) Initial GDP -0.341 0.487 0.607 0.502 0.215 0.179 -0.544 0.332 1.000 (10) Gov. Effectiveness 0.194 0.516 0.688 0.647 0.241 0.178 -0.479 0.433 0.799 1.000 (11) Years of schooling 0.273 0.416 0.542 0.373 0.167 0.069 -0.418 0.243 0.753 0.639 1.000 (12) Gov. expenditure -0.400 0.136 0.296 0.314 0.017 0.001 -0.258 -0.020 0.491 0.381 0.458 1.000 1. (13) ICRG -0.156 0.439 0.585 0.553 0.265 0.205 -0.436 0.366 0.742 0.650 0.576 0.333 1.00

5. Empirical results

In this section we firstly present estimation results for private credit and its impact on economic development. Secondly, we disaggregate private credit into household and non- financial corporate credit. Lastly, the credit stocks- flows distinction will be considered. The estimation results are discussed based on different econometric methods.

Firstly, the effect of aggregate private credit on economic growth is analyzed. Following the methodology described by Arcand et al. (2012) the OLS regression model is applied and the results are shown on table 5.1.1. Even though, the OLS model possesses drawbacks since it does not account for endogeneity, omitted variable bias, according to Arcand et al. (2012) it is a useful tool to analyze the data in a transparent way. Indeed, the simplicity of OLS model attracts the widespread use of the technique since it is a useful approach to look at the relationship between variables. However, the estimated findings require support from more advanced methods such as fixed effect model and GMM.

Table 5.1.1: OLS Regression results for private credit

(1)

Full sample (2) Low-income (3) Middle-income (4) High- income

Private Credit 2.592*** 2.671*** 2.598*** 1.781*** (0.441) (0.444) (0.485) (0.570) Private Credit2 -0.407*** -0.414*** -0.476*** -0.254*** (0.064) (0.063) (0.068) (0.090) Inflation -0.298*** -0.273*** -0.276*** -0.305*** (0.074) (0.075) (0.074) (0.074) Openness to Trade 0.628*** 0.641*** 0.612*** 0.604*** (0.119) (0.119) (0.118) (0.119) Initial GDP -0.194* -0.156 -0.096 -0.134 (0.106) (0.108) (0.109) (0.111) Gov. Effectiveness 0.348** 0.322** 0.435*** 0.468*** (0.140) (0.142) (0.154) (0.158) Years of schooling -0.135*** -0.129*** -0.130*** -0.136*** (0.038) (0.038) (0.038) (0.038) Gov. expenditure -0.132*** -0.132*** -0.136*** -0.135*** (0.015) (0.015) (0.015) (0.015) Low#c. Private Credit 0.794*

(0.425) Low#c. Private Credit2 -0.215*

(0.128)

Middle#c. Private Credit -0.930***

(0.226)

Middle#c. Private Credit 2 0.255***

(0.055)

High#c. Private Credit 0.663**

(0.286)

High#c. Private Credit 2 -0.191*** (0.066) _cons 2.405** 1.722 2.628** 3.126*** (1.087) (1.164) (1.109) (1.197)

Obs. 3244 3244 3244 3244

R-squared 0.082 0.083 0.088 0.086 Robust Standard errors are in parenthesis *** p<0.01, ** p<0.05, * p<0.1

Notes: This table presents the results using equation (3) using OLS estimation method. Column

(1) presents the results for the full sample, while column (2) to (4) include the dummy variable for low-, middle, and high- income economies respectively. The dependent variable is the real growth of GDP per capita. Dummy variable low stands for low- income economies; middle for middle- income countries and high for high- income countries respectively.

significant impact of private credit up to a point, above which an additional percentage increase leads to a decline in economic growth. By looking at column (2) to (4) the same statistically significant patterns can be observed across all country groups; however, the magnitude varies by including dummy variables of low-, middle, and high-income countries respectively. All the dummies in column (2) to (4) take the value of 1 if country is low-, middle, and high- income respectively.

Therefore, in order to obtain a true effect of private credit in different country groups, it is necessary to sum up the coefficients of private credit and its interaction term with income group. The same holds true for squared term of private credit. The coefficient for linear and squared term of private credit reflects the effect when the dummy variable takes the value of zero. More specifically, it can be concluded that private credit has the highest positive effect on economic growth in low- income economies. However, above a certain point the “diminishing effect” on economic growth is larger for low-income economies due to the fact that a percent increase in private credit tends to have a greater negative effect on the economy. In line with the previous empirical studies, the coefficients for openness to trade and the quality of government measured by government effectiveness index have a positive and statistically significant signs. Inflation and government expenditure, on the other hand, have negative coefficients suggesting a negative relationship with economic growth. In contrast to findings obtained by Arcand et al. (2012), education has a negative coefficient across all country groups. The sample of countries and different time period can explain the difference in findings. Cross-sectional data analysis improves the descriptive statistics, but it has major drawbacks since it does not account for unobserved country specific characteristics.

Table 5.1.2: FE and system GMM regression results for private credit (1) Full sample (2) Low- income (3) Middle- income (4) High- income (5) Full sample (6) Low- income (7) Middle- income (8) High- income

Fixed effect model System GMM

Private Credit 4.100*** 3.805** 4.500*** 3.484*** 2.599** 2.015*** 2.478*** 2.367 (1.216) (0.590) (0.698) (0.551) (0.842) (0.977) (0.367) (0.665) Private Credit2 -0.749*** -0.709*** -0.862*** -0.604*** -1.658* -1.772*** -1.907*** 0.331 (0.176) (0.225) (0.099) (0.097) (0.877) (0.468) (0.551) (1.127) Inflation -0.302** -0.306** -0.283*** -0.281*** -1.300*** -1.406*** -0.781*** -0.390 (0.126) (0.124) (0.081) (0.081) (0.414) (0.203) (0.187) (0.471) Gov. expenditure -0.124** -0.125** -0.124*** -0.129*** -0.004 -0.036 -0.090 -0.118 (0.051) (0.051) (0.021) (0.021) (0.135) (0.074) (0.072) (0.180) Openness to Trade -0.237 -0.265 -0.251 -0.065 -1.267 -1.193 -1.118 -2.394 (0.528) (0.524) (0.311) (0.311) (1.469) (0.874) (0.920) (2.053) Initial GDP -0.881*** -0.867*** -0.886*** -0.867*** -8.134*** -8.286*** -8.885*** -13.442*** (0.215) (0.212) (0.134) (0.134) (2.097) (1.069) (1.166) (2.823) Years of schooling -0.115 -0.106 -0.136 -0.127 1.071** 1.031*** 0.959*** 0.848 (0.222) (0.220) (0.121) (0.121) (0.443) (0.224) (0.313) (0.614) Gov. Effectiveness 0.558 0.569 0.489* 0.505* 8.324*** 8.600*** 7.804*** 8.222** (0.555) (0.558) (0.284) (0.283) (2.910) (1.342) (1.355) (3.780) 1.Low#c. Private Credit 0.411 0.877*

(0.310) (0.366)

1.Low#c. Private Credit2 0.072 0.334

(0.400) (0.372)

1.Middle#c. Private Credit -1.684* -1.468**

(1.001) (0.612)

1.Middle#c. Private Credit

2 0.363** 0.973*

(0.159) (0.556)

1.High#c. Private Credit -1.103** 1.956***

(0.697) (0.604)

1.High#c. Private Credit 2 0.461 -0.279***

(0.291) (0.164) GDP growth (t-1) 0.204*** 0.219*** 0.176*** 0.151* (0.076) (0.035) (0.035) (0.089) _cons 9.466** 9.825** 10.534*** 14.026*** 0.000 0.000 13.102*** 10.463*** (4.303) (4.559) (2.136) (2.520) (0.000) (0.000) (3.935) (4.821) Obs. 3243 3243 3243 3243 3243 3243 3243 3243 R- squared 0.216 0.174 0.175 0.177 Time FE TH-1 Yes 85.352

Yes Yes Yes Yes Yes Yes Yes

AR (1) 0.000 0.000 0.000 0.000

AR (2) 0.612 0.626 0.363 0.380

OID 0.237 0.326 0.070 0.457

Robust Standard errors are in parenthesis *** p<0.01, ** p<0.05, * p<0.1

Notes: This table presents the results using equation (3) by applying FE and GMM estimation methods. Column (1) and (4)

presents the results for the full sample, while column (2) to (4) and (6) to (8) include the dummy variable for low-, middle, and high- income economies respectively. The dependent variable is the real growth of GDP per capita. Dummy variable low stands for low- income economies; middle for middle- income countries and high for high- income countries respectively. TH-1 stands for panel threshold regression model, which allows accounting for non-monotone relationship. Fixed effect model and GMM model includes time dummies. Furthermore, the p- values for first and second Arellano- Bond serial correlations tests and OID- over-identification Hansen test are reported.

Another point that stands out is the fact that the coefficients for linear and quadratic term of private credit become smaller by controlling for the endogeneity. The biased coefficients in OLS and FE models do not equal to the true parameter as expected since the independent variables are correlated with the error term. In column (5), an increase in one standard deviation of private credit leads to 0.656 increase in real GDP growth in the full sample. Due to the fact that the average GDP growth is 3.544, the effect of financial development on economic outcomes is large. The calculations of the effect size are based on the calculations included in the paper written by Bezemer et al. (2016). Furthermore, the effect is large across all country groups.

By looking at the control variables, the results are similar to whose obtained in cross- sectional regressions, more precisely inflation tends to have negative effect on economic growth even when the dynamic nature of the relationship is analyzed. In line with previous literature, the coefficients of government effectiveness are positive and statistically significant. Government expenditure and openness to trade, on the other hand, has a negative effect. Theoretically government expenditure can have negative effect on economic development in case of misallocation of resources. However, the results obtained for openness to trade are contradicting to those obtained by Arcand et al. (2012). In contrast to the results obtained in OLS model, the effect of years of schooling have a positive and statistically significant effect on economic growth.

5.2. Household credit, non-financial corporate credit and economic growth

This section focuses on disaggregate measures of financial development on economic growth, more precisely in line with the previous literature the effect of household as well as non-financial corporate credit is considered. Furthermore, this paper applies the methodology suggested by Bezemer et al. (2016), namely not only the credit structure matters for growth effectiveness, but also the distinction between credit stocks and flows. The main argument behind this is that credit flows are perceived as ”short- term liquidity” to stimulate growth by increasing the ability to fund investments. The credit stocks, on the other hand, are stocks of debt, which can have negative effect on growth at high levels. Knowing the relationship between private credit and GDP growth, this paper deviates from Bezemer et al. (2016) since it captures non- monotone relationship between household, non-financial corporate credit and economic growth by adding the quadratic term. Due to the fact that private credit is an aggregate measure of total claims on the non-financial private sector, which includes both households as well as non-financial corporations, there is a reason to assume that disaggregate measure of household credit and non-financial corporate credit follow the same inversed U- shaped relationship. The analysis is conducted for 43 economies and the sample mostly consists out of high- income countries over the period 1991- 2018. Therefore, the distinction between low-, middle-, and high- income countries are not taken into consideration. The list of countries included can be found in Appendix B.

provide any significant coefficients for non- financial corporate credit. Furthermore, the effect of household credit stocks on economic growth is larger in GMM model. More specifically, the increase in one standard deviation of household credit stock leads to 0.778 increase standard deviation of the GDP growth. Since the average growth in the sample is 3.986, the effect is large.

Table 5.2.1: Regression results household and non-financial corporate credit stocks

(1) (2) (3) (4) (5) (6)

OLS OLS FE FE GMM GMM

Household credit stock

Non-financial corporate credit stock Inflation Openness to Trade Initial GDP Gov. effectiveness Years of schooling Gov. expenditure

Household credit stock 2

Non-financial corporate credit stock 2 GDP growth (t-1) _Cons Obs. R-squared Time FE AR (1) AR (2) OID -0.301** 0.572* 0.922* 1.925*** -0.007** 2.874** (0.144) (0.316) (0.469) (0.545) (0.003) (1.303) -0.167 3.080* -1.978*** 0.113 0.002 -1.456 (0.241) (1.647) (0.463) (3.511) (0.003) (3.576) -0.476*** -0.419*** -0.318* -0.310 -0.485** -0.209 (0.105) (0.104) (0.188) (0.190) (0.217) (0.329) 0.501*** 0.610*** 0.784 0.722 0.281 0.471 (0.166) (0.168) (1.056) (1.008) (0.189) (0.280) -0.887*** -0.818*** -0.245 -0.485 -1.198*** -1.202*** (0.154) (0.158) (0.758) (0.825) (0.200) (0.298) 0.757*** 0.788*** 0.054 -0.149 0.700** 0.440 (0.217) (0.223) (0.576) (0.648) (0.285) (0.508) -0.050 -0.027 0.143 0.106 0.133* 0.207** (0.058) (0.059) (0.159) (0.145) (0.071) (0.095) -0.189*** -0.200*** -0.513*** -0.488*** -0.063** -0.066* (0.022) (0.023) (0.131) (0.123) (0.027) (0.037) -0.186*** -0.275** -0.438** (0.053) (0.112) (0.183) -0.393* -0.230 0.158 (0.201) (0.452) (0.431) 0.479*** 0.438*** (0.061) (0.070) 14.793*** 6.421* 15.256 13.205 0.000 0.000 (1.503) (3.701) (10.013) (8.878) (0.000) (0.000) 1130 1130 1130 1130 1130 1130 0.185 0.202 0.316 0.333

Yes Yes Yes

0.000 0.000 Yes 0.391

0.264 0.406 0.253

The GMM system is our preferred model due to the fact that it account for endogeneity within the model, the Hansen test reported in the table shows that the instruments used are valid and model is not over- identified. Furthermore, the Arellano- Bond test for serial autocorrelation states the absence of the second order correlation since we fail to reject the null hypothesis. The post- GMM test provides evidences that the estimated coefficients are reliable and efficient in GMM model. In contrast to previous empirical research conducted by Beck et al. (2012), Sassi & Gasmi (2014), who identified positive effect of enterprise credit, this paper does not provide any significant evidences for positive effect. However, it supports a non-

Robust Standard errors are in parenthesis *** p<0.01, ** p<0.05, * p<0.1

Notes: This table presents the results using equation (4) by applying OLS, FE and GMM estimation

monotone relation between household credit stock and economic growth. Indeed, in line with the previous findings, household credit can positively affect growth through increase in human capital accumulation (De Gregorio, 1996). However, at some point economic growth can slow down by increasing debt-to- burden ratio and stimulating volatility. By controlling for endogeneity, reverse causality within the model, the significant coefficients of non- financial corporate credit vanish. In line with the findings suggested by Bezemer et al. (2016), non- financial corporate credit stocks become insignificant once the endogeneity is controlled. These imply that non- financial corporate lending responds weaker to real GDP growth and the growth effects from productive investments funded by bank’s loan are larger.

Table 5.2.2: Regression results for household and non-financial corporate credit flows

(1) (2) (3) (6) (7) (8) OLS OLS FE FE GMM GMM Household (HH) credit flow 0.048** (0.021) 0.058* (0.032) 0.037* (0.021) 0.064** (0.028) -0.105** (0.053) -0.086 (0.292)

Non-finan. Corp. cred.

Flow (NFC) 0.050*** (0.015) 0.089*** (0.022) 0.022 0.014) 0.054*** (0.019) -0.013 (0.051) 0.254* (0.153) Inflation -0.507*** (0.101) -0.503*** (0.100) -0.412*** (0.105) -0.386*** (0.104) -0.241 (0.245) -0.533** (0.262) Openness to Trade 0.264* 0.357** 0.629 0.404 0.323 0.687*** (0.158) (0.160) (0.526) (0.519) (0.217) (0.258) Initial GDP -1.208*** -1.149*** -0.279 -0.356 -0.990*** -0.743*** (0.153) (0.153) (0.365) (0.359) (0.230) (0.246) Gov. Effectiveness 0.506*** 0.529*** 0.451 0.227 0.799*** 0.408 (0.186) (0.186) (0.408) (0.404) (0.281) (0.332) Years of schooling 0.018 0.004 0.108 0.136 0.105 0.075 (0.057) (0.057) (0.134) (0.132) (0.070) (0.067) Gov. expenditure -0.178*** -0.180*** -0.529*** -0.536*** -0.074*** -0.092*** (0.022) (0.022) (0.058) (0.057) (0.026) (0.031) HH flow2 -0.002 (0.001) -0.003** (0.001) -0.005 (0.014) NFC flow2 -0.002*** -0.002*** -0.011** (0.001) (0.001) (0.005) GDP growth (t-1) 0.500*** (0.042) 0.401*** (0.058) _cons 16.188*** 15.479*** 10.775** 12.647*** 0.000 0.000 (1.345) (1.353) (4.842) (4.775) (0.000) (0.000) Obs. 1083 1083 1083 1083 1043 1043 Pseudo R2 _{0.231 0.245 0.338 0.361}

Time FE Yes Yes Yes Yes

AR (1) 0.000 0.000

AR (2)) 0.234 0.310

OID 0.166 0.653

Robust Standard errors are in parenthesis *** p<0.01, ** p<0.05, * p<0.1

Notes: This table presents the results using equation (4) by applying OLS, FE and GMM estimation

methods. Column (1) and (3) and (5) presents the results for linear effect of household and non-financial corporate credit flows on economic growth, while column (2) and (4) and (6) includes the quadratic term. The dependent variable is the real growth of GDP per capita. Fixed effect model and GMM model includes time dummies. Furthermore, the p- values for first and second Arellano- Bond serial correlations tests and OID- over-identification Hansen test are reported.

is a short-term liquidity injection, which stimulates economic growth. However, based on the results supported in all three econometric models, the credit to non- financial corporations has a “diminishing effect” on output growth.

Table 5.2.3: Regression results for household and non-financial corporate credit stocks and flows

(1) OLS (2) OLS (3) FE (4) FE (5) GMM (6) GMM HH credit stock -0.452*** 0.457 0.891*** 1.908*** 0.145 6.319*** (0.146) (0.302) (0.225) (0.330) (0.392) (2.179) HH credit flow 0.059*** 0.077** 0.024 0.065** 0.246 0.058 (0.023) (0.032) (0.021) (0.028) (0.212) (0.418) NFC credit stock -0.155 2.535 -2.068 0.004 -1.016 3.486 (0.249) (1.626) (0.374) (1.919) (0.638) (0.454) NFC credit flow 0.047*** 0.071*** 0.034** 0.057*** 0.338** 0.454* (0.016) (0.022) (0.014) (0.019) (0.135) (0.271) Inflation -0.608*** -0.534*** -0.402*** -0.375*** -1.242*** -1.618** (0.103) (0.102) (0.105) (0.102) (0.331) (0.694) Openness to Trade 0.330** 0.479*** 0.252 0.136 0.281 0.945 (0.162) (0.164) (0.524) (0.513) (0.272) (0.702) Initial GDP -1.185*** -1.062*** -0.968** -1.258*** -1.949*** -3.112*** (0.153) (0.156) (0.381) (0.377) (0.321) (0.894) Gov. Effectiveness 0.904*** 0.894*** 0.162 0.196 1.002*** 2.157*** (0.212) (0.216) (0.407) (0.403) (0.355) (0.778) Years of schooling 0.007 0.022 0.093 0.069 0.272*** 0.565*** (0.057) (0.057) (0.132) (0.129) (0.091) (0.151) Gov. expenditure -0.176*** -0.188*** -0.584*** -0.547*** -0.060** -0.052 (0.022) (0.022) (0.058) (0.057) (0.025) (0.072) HH credit stock 2 -0.198*** -0.300*** -1.061** (0.053) (0.066) (0.427) HH credit flow 2 -0.001 -0.003** -0.007 (0.001) (0.001) (0.021) NFC credit stock2 -0.297 -0.199 -1.034 (0.201) (0.247) (0.677) NFC credit flow 2 -0.002** -0.002*** -0.005* (0.001) (0.001) (0.006) GDP growth (t-1) 0.380*** 0.265*** (0.051) (0.068) _cons 17.653*** 9.478*** 25.609*** 24.079*** 23.972*** 22.115*** (1.527) (3.606) (5.495) (5.900) (4.807) (39.362) Obs. 1083 1083 1083 1083 1043 1043 R- squared 0.243 0.273 0.358 0.398

Time FE Yes Yes 0.000 0.000 AR (1) AR (2) OID 0.761 0.342 0.175 0.117

Robust Standard errors are in parenthesis *** p<0.01, ** p<0.05, * p<0.1

Notes: This table presents the results using equation (5) by applying OLS, FE and GMM estimation methods. Column

(1) and (3) and (5) presents the results for linear effect of household and non-financial corporate credit stocks and flows on economic growth, while column (2) and (4) and (6) includes the quadratic term. The dependent variable is the real growth of GDP per capita. Fixed effect model and GMM model includes time dummies. Furthermore, the p- values for first and second AR serial correlations tests and OID- over-identification Hansen test are reported.

of non- financial corporate credit flows is small since an increase of one standard deviations of credit flow leads to 0.040 increase of economic growth’ standard deviation. By comparing he average growth rate, the effect is small while “the diminishing effect” is minor.

The following table 5.2.3 includes the stock and the flow effect at the same time into the regression model. Without including both stocks and flows into the model, the effect of financial development on economic growth is overestimated. The table provides the findings for three empirical methods, namely OLS, fixed-effect and GMM model.

Columns (1), (3) and (5) include linear effects while the remaining columns capture non-monotone relationship between financial development and economic growth. Indeed, in line with the previous findings, the household credit stock has a positive effect on economic growth up to a point, above which the effect turns negative. This pattern is observed across all empirical models once the quadratic term is included. Furthermore, GMM model fails to provide any significant results for household credit flows in contrast to FE model, which suggests a “diminishing effect”. The same pattern has been observed in table 5.2.2. The effect of non- financial corporate credit stock on economic growth is negative; however, once the quadratic term is included, the inverse U-shaped relationship is observed. In line with the table 5.2.1, the model fails to provide any significant results. Non- financial corporate credit flows, on the other hand, provide statistically significant results supporting the inverse U- shaped relationship. Furthermore, the growth effect of statistically significant results is larger by controlling for stock and flow effect.

5.3. Robustness check

A number of robustness checks have been performed in order to make sure that the findings are robust. Firstly, the quality of government indicator provided by the ICRG is used as a control variable. The quality and institutional level in the economy has a beneficial effect on economic growth since countries with better institutions can allocate resources in more productive way. According to Demetriades and Law (2006), the financial development has its largest effect in countries embedded with better institutional framework.