Yorben Blok
0312614
June 2015
NON LINEARITY OF FINANCE
AND ECONOMIC GROWTH
Do we go up or down?
Abstract
Since the theoretical outlay of Schumpeter (1911, 1934) both primary and secondary functions of credit are theoretically founded. Subsequent empirical work focused and found empirical support for the positive (primary) function: providing credit for innovation and economic growth. The financial crisis however made clear there are negative (secondary) effects of credit as well and more recent empirical work suggests a non-linear relationship. Since these are at first sight conflicting views, this thesis researches theoretically the positive and negative effects of credit as potential explanations of non-linearity. Empirically this thesis researches the finance-growth relationship over several time frames and finds support for both linearity and non-linearity. Non-linearity of the finance growth relation is found in more recent data and alinear finance-growth relationship is found in the earlier years. Furthermore there are considerable issues with the chosen variables, specifications used and countries included, as well as a strong influence of the last ten years that include the financial crisis.
Master Thesis Economics
Supervisor: dr. Ward Romp
Second reader: dr. Christian Stoltenberg
MSc Economics: Monetary Policy and Banking
Amsterdam School of Economics
University of Amsterdam
Statement of Originality
This document is written by Yorben Blok 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 source other than those mentioned in the text and it 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.
1
Index
1. Introduction ... 2
2. Financial development, credit and growth ... 1
2.1 Finance growth nexus – Primary effects ... 3
2.1.1 Empirical research on the finance-growth nexus ... 5
2.2 Finance growth nexus – Secondary effects ... 8
2.2.1 Dysfunctional financial intermediation ... 10
2.2.2 Financial instability and the credit cycle... 11
2.2.3 Non-intermediation and differentiation of credit ... 12
2.2.4 A dampening effect ... 13
3. Non linearity of credit-growth relationship ... 13
3.1 Other causes of non-linearity? ... 14
4. Hypotheses, methodology and data ... 16
4.1 Hypotheses ... 16
4.2 Methodology ... 17
4.3 Data ... 19
5. Results ... 24
5.1 OLS Cross Country ... 24
5.2 Instrument Variable Regression ... 27
5.3 Credit variables (U-test) ... 30
5.3.1 High and Low income countries ... 31
5.4 Threshold regressions ... 36
5.4.1 High and Low income countries ... 36
6. Discussion and Conclusion ... 40
6.1 Discussion ... 40
6.2 Conclusion ... 41
2
1. Introduction
Since the ancient Babylonians’ (2500BC) usage of a clay tablets (shubati) to denote a debt, the uses of credit are well known. The shubati, which roughly translates into “received”, was officiated by an official, denoted the important information and made trade possible beyond the obvious borders of direct barter (Smithin, 2000). Since Schumpeter’s work the importance of finance and credit was also theoretically founded. According to Schumpeter (1934, p. 101): “From this it follows that in real life total credit must be greater than it could be if there were only fully covered credit. The credit structure projects … beyond the existing commodity basis.”. Rephrased it means that growth in the credit stock beyond real production, is necessary for economic development to be realized. It was this new combination of means of production and credit that would be the fundamental phenomena of economic development (Schumpeter, 1934 p. 74). The theoretical work that followed, stood on the shoulders of Schumpeter and emphasised the primary, positive effects of financial development and credit availability. Subsequently most of the empirical work done, and specifically the seminal work and well titled article of King and Levine (1993) “Schumpeter might be right”, empirically proved the positive (linear) finance-growth effect. However, as made abundantly clear to the world and economic profession in the wake of the financial collapse of 2007, there is a negative (secondary) effect to credit that works counter-productive to economic growth. Bezemer’s also well titled article “Schumpeter might be right again” points towards this negative effect which was also part of the intellectual legacy of Schumpeter.
Recent empirical work builds on both premises and shows that the supposed positive relationship between finance and growth is not stable over time (Rousseau and Wachtel, 2011) while others show a non-linear relationship for the last 50 years (Arcand et al., 2015 and Law and Singh, 2014). To find out how these at first sight conflicting finds coexist, this thesis researches the growth relationship on the most recent dataset available. The main research question is: is the finance-growth relationship non-linear? In the process of answering the main research question, the research infers the finance-growth relationship over time, the sensitivity of the specification, for High and Low income countries and the effect of the financial crisis. Theoretically it explores how both positive and negative effects of credit to economic growth arises. It contributes to the empirical literature by using a more recent dataset and splitting the dataset both over time and countries to infer if non-linearity is of all times and all places, or more of a recent phenomenon specifically for High income countries.
1
2. Financial development, credit and growth
Since the work of Robert Solow in the 1950s, growth theory viewed and to a major extent views, economic growth as the result of innovation, human capital and physical accumulation. Therefore the research focused on the use of technology, capital and labour resources but the financial system was largely ignored (Wachtel, 2001). Broadly speaking the neoclassical theory assumes that financial systems function efficiently, where financial factors are often abstracted from the (macroeconomic) analyses. A significant oversight according to Ang (2008) who remarks that policies designed for economic development that ignore improvements (and I use improvements loosely) within the financial system, could lead to a crisis-prone financial system. Borio (2014) remarks that for most of the post-war period the theory on financial cycles and booms and busts only featured a role outside the mainstream. He mentions that financial factors disappeared from the macroeconomic radar, finance became a veil and could be ignored when seeking to understand business fluctuations.
This disappearance from the macroeconomic radar is especially a surprise given the vast academic attention on the finance-growth nexus in the 20th century. The notable early work on finance
and growth started with Schumpeter (1911) which was further developed by Gurley and Shaw (1955), Goldsmith (1969, as cited in Ang, 2008) and Hicks (1969, as cited in Ang, 2008) during the 50’s and 60’s. They supposed that the creation of more financial institutions, financial products and services, would generate a positive effect on the saving-investment process and thereby economic growth. This was dubbed the “financial structuralist view” (Ang, 2008, p. 540). However Schumpeter’s’ view and that of his followers had little impact due to the dominant Keynesian paradigm of financial repression and it was not presented in a formal manner.
Within the 50’s and 60’s credit however was at its heyday and at the “centre stage” of monetary policy making due to the nature of the policy objectives (Borio, 2004). The restrictions on interest rates and balance sheet quantities were instrumental in allowing the financing of the government sector while limiting upward pressure on interest rates. In this context Borio and Lowe (2004) argue that credit had a salient place in policy implication. Furthermore to illustrate the thinking of that time on private credit Borio and Lowe (2004, p. 4) refer to the Radcliffe Report of 1959 in the UK: “The report played down the role of money on the grounds of high substitutability with other assets, favoured a broader, if not very precise, notion of “liquidity” and stressed the importance of the ability of economic agents to finance their expenditures.” The report was used to justify the use of direct controls or suasion of credit extension, especially in the context of a rapid revival of the demand for private domestic credit.
In the 70’s the rise of monetary targeting came at the expense of both credit in policymaking and financial repression. In this context credit “fell by the wayside” and any special attention was
2 redirected to monetary aggregates. The few countries that had variants of credit aggregate variables or direct controls on credit, were the exception, not the rule. Alongside the growing dominance of Monetarism in monetary and political policy, the role of financial intermediaries and markets also changed. Two authors that favoured the finance-growth nexus argued financial liberalization benefits, but through different channels. McKinnon (1973, as cited in Ang, 2008) assumed that investment in an economy is mostly self-financed from savings build up in bank deposits. This was dubbed the “complementarity hypothesis” which referred to the complementary role between money and physical capital. Another channel, put forward by Shaw (1973 as cited in Ang, 2008), emphasised the intermediary role of financial institutions that promoted investment and raise output growth through borrowing and lending, dubbed the “debt-intermediation hypothesis”. Fry (1978), in an empirical test to validate the above theories, found for ten Asian economies that the transmission mechanism is found in the debt intermediation channel and finds little backing for the complementarity hypotheses. Fry (1978) however supported both McKinnon and Shaw that a higher level of financial development would lead to increased output which led to an emphasis on financial liberalisation in the 70’s and 80’s.
In the early 80’s there was some criticism on financial liberalisation by a group of neo-structural economist such as van Wijnbergen (1982, 1983), Taylor (1982 in Ang, 2008) and Buffie (1984). They argued that curb markets, non-institution credit markets, were more efficient in the intermediation process between savers and investors. Any raise in the bank deposit rate would induce loanable funds away from these efficient curb markets to commercial banks, discouraging investment and growth. The neo-structuralists therefore claimed that in the presences of efficient curb markets, financial liberalization was unlikely to raise growth. However at the end of the 80’s authors such as Fry (1988 in Ang, 2008) argued that curb markets are not necessarily as competitive and efficient as commercial banks. Subsequently Owen and Solis-Fallas (1989) show that the relative intermediation efficiency significantly influences the outcome of portfolio effects and argued that a perfectly efficient intermediation curb market seems highly unrealistic.
In the 90’s more formal modelling led to endogenous growth models (Bencivenga and Smith, 1991; Greenwood and Jovanovic, 1989; King and Levine, 1993; Pagano, 1993; Saint-Paul, 1992). All these models supported that financial development reduces informational frictions and improves resource allocation efficiency. However in the literature in the 90’s there was an increasing awareness of credit market frictions and, especially around the dotcom crisis, that financial liberalisation made booms and busts in credit and asset prices more likely as well (Bernanke et al., 1996, 1999; Borio et al., 2001, 2003; Borio and Lowe, 2002; Borio and White, 2004). Furthermore Borio and Lowe (2004) argue that the environment maintained by the central bank would lead to a slower emergence of price pressures normally associated with unsustainable expansion. Instead symptoms would emerge in excessive credit and asset price growth and subsequently (overstretched) balance sheets that would
3 make the economy more vulnerable to a slowdown or contraction. Bezemer and Grydaki (2014) empirically show that during the Great Moderation this was likely the case.
One thing is clear from the above analysis: the theoretical foundations that supported financial liberalisation and development apparently went hand in hand with a gradual disappearing of credit from the policy radar (Schularick and Taylor, 2012). However this is not to imply that the problem is that no attention was paid to the financial system, the references above as well as the next chapter proves the opposite. The main problem is the abstraction of the financial world from real (macro)economic analyses, due to the view of money and the liability side of financial institutions as a passive by-product. Therefore integrating credit in the mainstream framework has proven to be difficult, which arguably led to an unawareness of the problematic build-up of credit (Cecchetti et al., 2011).
2.1 Finance growth nexus – Primary effects
In 1911 Schumpeter already argued that the services provided by financial intermediaries were essential for technological innovation and economic development (King and Levine, 1993). He viewed banks as ephors1 of the capitalist economy that controlled, selected and financed entrepreneurs and
saw that as banks primary function (Festré and Nasica, 2009). The main argument for financial liberalization was therefore the positive effects of financial development and the availability of credit for innovation and economic growth. However describing credit as if credit by itself would have a primary or secondary effect is nonsense. A concise description of the financial development process is therefore needed, since both effects of credit hinge on the interaction within the intermediation of the financial market (supply and demand). This will also allow the research to infer which of the financial intermediation arguments might contribute to the primary effect, a positive effect on economic growth, or the secondary effect, a negative effect on economic growth.
The survey articles by Ang (2008), Pagano (1993) and Levine (1997, 2005) identify four main problems following from information and transaction costs. Firstly there needs to be an efficient way of matching agents with entrepreneurs and surpluses with shortages. In the absence of financial intermediaries, this process would be costly and very time consuming. Secondly, when there is no common information between borrowers and lenders, financial contracts usually come with considerable agency costs to monitor the investment project. Thirdly there is little incentive for borrowers to disclose information about the investment project and efforts to obtain such information are usually costly. Lastly, since lenders cannot distinguish between honest and dishonest borrowers prior to issuing the loan, or even during the maturity of, there is a considerable premium that increases
1 An Ephor is an official title for an elected magistrate of Sparta who exercised supervisory power over
4 costs which subsequently discourages honest borrowers. The result of these inefficiencies are that without proper information transfer, the functioning of credit markets will be poor.
Following Schumpeter, the modern literature reviews five solutions for market frictions alleviated by financial intermediaries. With Schumpeter’s’ slightly different definitions in parentheses these are: Allocating resources (evaluating projects), mobilizing savings, reducing risks (managing risks), facilitating transactions and exercising corporate control (monitoring managers). First of allocating resources or evaluating projects refers to the ability of a financial intermediary to evaluate investment projects and enable the entrepreneur to borrow at lower rates and at easier terms (Tobin and Brainard, 1963). Evaluating the associated risk and return of an investment opportunity leads to a flow of funds to those project where the marginal product of capital is highest. This will lead to an improvement of quality in investments and have an expansionary effect on the economy. Secondly the mobilizing of savings ensures that the savings and investment decision of households are coordinated (Wicksell, 1935, as cited in Ang, 2008). Financial intermediaries can pool savings from many households and make the aggregate available for lending which increases liquidity. Hence as the financial system expands, more savings will be deposited and more funds will be available for investment projects. Furthermore as Pagano (1993) remarks, any decrease in the fraction of savings absorbed due to inefficiencies in the market, will be lower in the presence of efficient financial intermediaries. Thirdly efficient financial intermediaries reduce or manage risk by diversifying the portfolio of investment opportunities. The advantage of a large lender and borrower base leads to higher liquidity, better risk sharing and proper matching of different maturity periods. (Diamond and Dybvig, 1983). Fourth, financial intermediaries facilitate transaction by offering credit facilities and guaranteeing payment. Due to economies of scale they can manage and invest funds at lower costs. It avoids the hassle for the small individual lender to evaluate every borrower and vice versa. Gurley and Shaw (1960, as cited in Ang, 2008) contend that this is the primary function of financial intermediaries, transforming primary securities into indirect securities. Lastly it is suggested that financial intermediaries contribute to lower monitoring costs by exercising corporate control. Monitoring firms as individual investor is costly, which could otherwise lead to scarcity of funds and discourage borrowers (Bernanke and Gertler, 1989; Diamond, 1984).
There are however two unresolved issues that remain unclear. The first issue is the direction of the causal relationship between financial development and growth. Patrick (1966) contends that the relationship might change over the course of development. The first stage he calls the “supply-leading” hypothesis, where the creation of financial institutions and the supply of their financial assets leads to access of funds by growth sectors. The second stage he calls “demand-following” in which the financial system responses to the demand for services by investors and savers. The second issue is that financial development can have an effect on the quality of financial intermediation (efficiency) or the
5 quantity of funds available (savings rate or savings invested). It is also highly likely that the two aforementioned issues interact, since endogenous growth models of the 90’s show reciprocal interaction between financial development and economic growth (Pagano, 1993). Furthermore it does not necessarily have to be actual economic growth, the expectation or anticipation of future economic activity might induce financial markets to develop as well (Levine, 2005).
To sketch a simple example of how quantity and quality effects and endogeneity issues can go hand in hand, take the birth and growth of the subprime market. This both increased quality of financial intermediation, risk and return was matched between lenders and borrowers and increased quantity, due to the (by the state) enforced percentage of savings directed to mortgages (Ashton, 2009). Eventually the increased efficiency in originating mortgages probably reduced the associated costs and left a larger part of savings for additional mortgage investment. Secondly with regard to endogeneity, the start of the subprime mortgage sector created funds for borrowers who were previously unable to attain a mortgage (supply-leading). However the anticipation of the subprime mortgage market growth induced more companies to specialise in originating subprime mortgages (demand-following).
2.1.1 Empirical research on the finance-growth nexus
The aforementioned issues were extensively researched by Gurley and Shaw (1955), Patrick (1966), Goldsmith (1969, as cited in Pagano, 1993), Wallich (1969), McKinnon (1973, as cited in Pagano, 1993), Shaw (1973, as cited in Pagano, 1993) and others who produced evidence that financial development and economic growth correlated for different sets of countries. However, it was until the 90’s that the empirical literature burgeoned. King and Levines’ (1993) article, using the Barro Lee growth regression (Barro, 1991), was the starting point for most of the empirical work that followed afterwards. To avoid an epistle of epic proportions not all of the literature will be reviewed here in detail. For a full summary of the empirical work, the survey article of Ang (2008) is an excellent source.
The first difficulty already encountered by King and Levine (1993), is to define financial development empirically and link it to theory. As Edwards (1996) puts forward, defining appropriate measure of financial development is one of the (many) challenges faced by finance-growth researchers. King and Levine’s (1993) solution is to use four different indicators that are all designed to measure the services provided by financial intermediaries. The first is liquid liabilities to GDP (M3) which is the traditional measure of financial depth. The second indicator is measuring the importance of deposit banks to the central bank and the last two indicators are two measurements of non-financial Domestic Credit to the Private sector (DCP) compared to total credit and compared to GDP. King and Levine (1993) argue that the DCP indicator is the best proxy for financial intermediation due to measuring actual levels of credit. Their results indicate that all four financial development indicators enter with a positive and significant coefficient when the dependent variable is economic growth.
6 Furthermore, using a pooled cross country time series with initial values, King and Levine (1993) find that these indicators are good predictors of long-run growth. According to the authors this implicates that financial development does not just follow growth but is leading.
Most of the cross country empirical work use similar M3 and DCP indicators to research the effects of financial development on economic growth. For instance Levine and Zervos (1998) find that both stock market liquidity and development of the banking sector positively affects real per capita GDP. Deidda and Fattouh (2002) find that higher level of financial development are positively related to higher growth rates for high-income countries. McCaig and Stengos (2005) also find a strong positive effects for both DCP and M3 as explanatory variables. Additional research that finds similar positive relations between financial development is also done while controlling for other influences on economic growth. For instance Atje and Jovanovic (1993) find that stock markets have a positive effect on economic growth. Other papers such as Demirguc-Kunt and Maksimovic (1998, 2002) introduce legal and/or regulatory systems in their analysis and find that a well-developed legal system contributes considerably to both financial development and economic growth (Levine, 1998, 1999, 2002). Not all cross country papers find a positive significant relation between financial development and economic growth. For instance Atje and Jovanovic (1993) find a positive relation for stock markets on economic growth, but no such effect is found when bank based indicators are included. Harris (1997) finds only little support for the stock market relationship, especially for less developed countries, while Levine (2002) only finds a positive relation for the overall level of financial development. When it is classified as bank-based or market-based, the relationship disappears.
Most of the research conducted with time series data exclusively focuses on finding evidence on the supply driven and/or demand following hypothesis from Patrick (1966). Using granger-type causality tests and vector autoregressive (VAR) procedures they examine the causality of the finance-growth relationship. One of the first attempts to find a finance-growth-enhancing impact of financial development on economic growth was conducted by Gupta (1984 in Ang 2008). His results support the supply driven hypothesis of Patrick (1966) that causality runs from financial development to growth. These results are echoed by the majority of work done on time series for a variety of or individual countries (Bell and Rousseau, 2001; Choe and Moosa, 1999; Christopoulos and Tsionas, 2004; Demetriades and Luintel, 1996; Luintel and Kahn, 1999; Neusser and Kugler, 1998; Rousseau and Sylla, 2005; Rousseau and Vuthipadadorn, 2005; Rousseau and Wachtel, 1998; Xu, 2000). Arestis, Demetriade and Luintel (2001) find additional support for the finance led hypothesis of growth, but emphasis the size of the relationship. They find that the banking sector development effect on growth is substantially larger than that of the stock market. There is only limited research that supports the notion that the direction of causality runs both ways as found by Jung (1986). Demetriades and Hussein (1996) find a bi-direction of causality as well, which especially significant for developing countries.
7 There is however no evidence that finds no relationship at all between financial development and economic growth. (Ang, 2008, Levine, 2005)
The only article that finds a truly (weak) negative correlation between financial development and economic growth for individual countries is Ram (1999) and when the analysis is performed on grouped countries by growth rates, the same pattern emerges. Arestis and Demetriades (1997) research individual countries and find substantial variation among developed and developing countries, which highlights the limitations of cross country studies. They refer to three country specific differences that could cause this variation. Firstly, dissimilar financial systems (in different countries) may have a different effect on economic growth. Secondly, differencing financial policies might have an influence on the relationship. Thirdly, even with identical financial systems, the institutions within that system may still differ in their effectiveness/efficiency.
Ang (2008), Levine (2005) and other authors have considerable critique on both cross country as time series estimation. Time series are often short and results highly susceptible to lags chosen and the inclusion of trend terms. Limitations in cross country estimations are the causation of the variable financial development on economic growth, possible reverse causality is mostly not taken into account, which Demetriades and Hussein (1996) argue was already a shortcoming in the seminal paper by King and Levine (1993). Further critique mentions averaging data out over long periods may introduce contemporaneous correlation and using a large variety of cross country data arguably looks at short rather than long term economic behaviour. Furthermore grouping countries together can produce results which are at best ambiguous and fragile and highly susceptible to the control variables, time period and econometric techniques employed.
Some of the critique is alleviated by researching panel data such as conducted by Beck et al. (2000), Beck and Levine (2004), Benhabib and Spiegel (2000), Gregorio and Guidotti (1995), Levine (1999), Levine et al. (2000), Rousseau and Wachtel (2000) and Rioja and Valev (2004a). They all find consistent results that the measures employed for financial development have a positive impact on growth. However Wachtel (2001, 2003) argues that the econometric techniques used for panel data run into two econometric problems. First, similar to the critique above, there may be simultaneity or reverse causality between the financial development and economic development variable. To deal with this most researchers use the initial values of the independent variables as instruments with Instrumental Variable (IV) estimation (Beck et al., 2000; Rioja and Valev, 2004a, 2004b; Rousseau and Wachtel, 2000, 2002; Loayza and Ranciere, 2002). The second problem according to Wachtel (2001, 2003) is similar to the cross country limitations: the unobserved country-specific influence. The solution usually sought by econometricians is to implement a fixed effects variable into the regression. Wachtel (2001, 2003) however shows that by including a fixed effects variable there is considerable colinearity between fixed effects and the phenomenon under investigation which leads to coefficients
8 highly sensitive to the fixed effects term. Benhabib and Spiegel (2000) also show that adding fixed effects leads to coefficient instability and a loss of significance on the financial development coefficients.
Another critical element that is broadly shared is the choice of financial depth/development indicators, which may not fully reflect the functions of financial intermediaries (Levine, 2005). For instance Gregorio and Guidotti (1995) mention that monetary aggregates could be very poor indicators where it reflects only money as medium of exchange, not the development of the financial sector. Subsequently highly aggregated credit measures do not necessarily indicate a highly developed financial system, nor does it differentiate between credit supplied to entrepreneurs, household or financial companies (Ang, 2008; Bezemer, 2014; Bezemer and Grydaki, 2014). The problem with disaggregated data however, is either the smaller time frame or limited amount of countries for which it is available. This is also one of the last points of critique mentioned in the literature: the lack of long term, comparable, high quality (non-aggregated) data (Thiel, 2001). Commonly the reason why most of the empirical research is conducted with broad measure of financial development such as the DCP indicator and monetary aggregates.
Even though Levine (1997 pp.708-709, 2005) discusses the same forms of critique as Ang (2008) and Wachtel (2001, 2003), he adds that the extensive body of research is quite compelling. Most of the research supports the finance-growth nexus as a first-order relationship. However recent history, most notably the credit crisis, surprised the economic professions and warrants a more thorough look at financial development and the secondary effects of credit.
2.2 Finance growth nexus – Secondary effects
The secondary effect of credit according to Schumpeter’s’ (1911) hinges on the second fundamental aspect of banking: the fact that that system itself is also an industry of innovation and could embark upon commercial and industrial enterprise themselves (Festré and Nasica, 2009). For instance the increase in financial and real estate credit could be a result of this innovative (profit seeking) behaviour. Surprisingly references to either Schumpeter’ (1911) secondary effects, negative effects of credit or critique on financial development (growth) are hardly found in the more recent academic (empirical) work (Bezemer, 2014). Early adaptors however, such as Gurley and Shaw (1955), devote a critical part of their paper on the economic (real) and financial world. They thereby make a similar separation as the real and monetary analysis of Schumpeter (1911) and mention the possibility or even necessity of financial policy and control. Two forms of critique in Gurley and Shaw (1955, pp. 536-537) stand out: “As principal administrators of the payments mechanism, they (commercial banks) must be solvent beyond doubt. As financial intermediaries, on the other hand, the commercial banks must underwrite the growth process with venture finance, buying securities from spending units that are taking the risk
9 of innovation and expansion. In one role, the banks must minimize risk; in the other role, risk-taking is necessary and proper.” and “A monetary authority which is tempted to stay within the bounds of its traditional controls … may find itself more and more out of touch with credit developments.”. The first quote highlights the misalignment of goals for commercial banks and the second quote hints at the supervisory role of central banks on financial stability.
The main lines of critique in more recent work are summarised by Ang (2008) as: negative influence of banks, irrelevance of finance, destabilizing effects of stock markets, and financial crises. For instance Morck et al. (2000) put forward that excessive corporate control by bankers could encourage risk-averse behaviour which could actually hinder growth. Lucas (1988) argues that economists tend to overemphasise the role of financial factors for economic growth. Stiglitz (1994, 2000) has substantial reservations that financial liberalisation in general is beneficial for economic growth. In In an earlier period it was Keynes (1936) who argued that stock markets produce too much speculative activities that would reduce deposits and investment. Subsequently the Kindleberger framework (2011) and financial instability hypothesis of Minsky (1977) both argue that the financial system will become unstable during the financial cycle resulting in financial crises. Stiglitz (2000) argues that the increased frequency of financial crises is closely associated with liberalization of the financial sector and argues for new forms of repression.
The first financial crises that reverberated through the academic literature was arguably the dotcom crisis in 2000. Literature such as Borio’s (2004) article: “should credit come back from the wilderness?” and the massive body on money bubbles driving asset prices (Brunnermeier and Oehmke, 2012) showed a renewed interest. However the emphasis was more on inflated asset prices in conjunction with lose money than on credit. Even after the bursting of the bubble there was considerably reservation to implement measures at booms. For instance Borio (2004, 2005) argues that more awareness to credit indicators is a good idea, but no actual action should be taken. The longstanding view was: it is better to deal with the bust than to try to prevent the boom (Dell’Ariccia et al., 2012b). After the financial crises of 2007 a more profound realisation that something was missing dawned on most economist. To quote former chairman of the FED Alan Greenspan (2008) in his testimony before the Committee of Government Oversight and Reform he professed to “shocked disbelief” while watching his “whole intellectual edifice collapse in the summer of 2007” (Bezemer, 2010, p.677). Suggesting however that the financial crisis was not foreseen in academics is not completely correct. Bezemer (2011) refers to several authors who saw the financial crises coming. Common aspect of all of these authors: they are not mainstream and mainly stand on the shoulders of (selectively disregarded) giants or heterodox theory on credit and the financial (credit) cycle. In most of the post crisis research it therefore becomes clear that older academic work picked up on financial issues that the recent pre-crisis work in general did not.
10 Broadly speaking the same lines of inquiry with regard to quality vs. quantity and endogeneity is revisited with the notable exception that credit can also have a negative (secondary) effect. Even though considerable effort has been put into identifying what went wrong in the aftermath, it is not easy to find a singularly defined problem or channel. The relative simple subprime example above already sketched the complexity a priori to financial development. Also ex ante finding “the culprit” is easier said than done. Simply put it was not just the quantity that led to the collapse of the subprime market but also the quality in the intermediation.
2.2.1 Dysfunctional financial intermediation
Negative effects of financial development on economic growth can directly be linked to inefficient or dysfunctional financial intermediation. The main problem with financial intermediation is namely the assumptions that are made beforehand on efficiency, rationality and equilibria. Specifically on (in)efficiency, an oversized financial sector could results in misallocation of resources and instability (Bolton et al., 2011 and Cahuc and Challe, 2012). In that regard the neo-structuralist were on the right track by emphasising the relative efficiency of intermediaries that is crucial to the outcome of the intermediation process. The first function, mobilizing savings is arguably still fulfilled by commercial (deposit) banks and (highly) regulated by central banks. Questions with regard to continuity, similar to the stock market empirics, could be warranted if this will remain the case in the future, but in all likeliness, not much has changed on the mobilizing savings function of financial intermediaries2.
The second function, facilitating transactions, has arguably changed in the last years. For instance a reservation is made when it comes to opaque borrowers (individuals) and small companies (entrepreneurs). There is some literature that shows that small companies have considerable difficulty to attain external finance (Beck and Demirguc-Kunt, 2006 and Beck et al., 2006, 2008). Furthermore, since profits are made by facilitating transactions, the ease at which both a company and an individual can get a (household) loan, changes considerably during booms and busts. A decrease in lending standards as described by Dell’Ariccia and Marquez (2006) and Dell’Ariccia et al. (2012a) is shown by clear evidence: there is a negative relationship between new loan demand and lending standards.
The third channel, a reduction in risk due to intermediation is quite contrary to what happened in the last two decades (Aikman et al., 2014; Rajan, 2005). Financial liberalization and innovation such as securitisation, credit default swaps and an ever increasing interconnectedness and opaqueness of the financial system and structure, contributed to an increase in risk rather than a reduction (Gorton, 2012, 2013). Moreover the risk of securitised mortgages was mislabelled by credit rating agencies (Ashcraft et al., 2011). Keys et al. (2010) argue that specifically securitization was responsible for the
2 Investment alternatives such as credit unions and crowdfunding might take savings and subsequent
investment decision away from deposit banks and shift to an individual-to-individual basis. Arguably the informational frictions that play a role here are less important in the information age.
11 aforementioned lack of screening, poor underwriting standards and higher default rates. This lead to lower quality loans and lower quality borrowers, which also increases risk (Dell’Ariccia et al. 2012a). Furthermore easy credit increases leverage ratios, which makes households and corporates more vulnerable to shocks. The hypothesised result of diversified portfolios reducing risk, eventually led to similar exposure in mortgages, systematic risk and subsequent collapse (Baele et al., 2007).
The fourth function, allocating resources to the most promising project, interacts with the other functions and consequently suffers from similar problems. There is literature that would suggest that credit does not go where the most promising projects is, but where most of the profit for the financial intermediary is made. An aspect of profit seeking was found in the build-up to the financial crisis with the originate to distribute model (Mian and Sufi, 2008). Bajari (2013) finds that the originators of mortgages did not typically hold them but sold them off to investors. An argument could therefore be made that in booms credit is created not for economic growth, but for more profits, were excessive risk-taking is rewarded and the subsequent marginal productivity of credit turns negative (Baele et al., 2007, Laeven and Levine, 2007, 2009 and Demirguc-Kunt and Huizinga 2010). The decrease in lending standards is one conducive element on the firm level. On a state level it was looser regulation and financial liberalization by the authorities that made it possible. Institutions that could (purely) originate subprime mortgages became legal and savings were purposefully directed to the mortgage sector (Chomsisengphet and Pennington-Cross, 2006; Ely, 2009).
Lastly exercising corporate control can also depend on economic circumstances. There is some literature that supports the view that there is both a slackness in corporate control and a higher occurrence of fraud during booms (Ferreira and Matos, 2012; Tarraf, 2011). Easier credit also means additional measures must be taken similar to the free cash flow problem that stems from excess debt (Jensen, 1986). Furthermore one could argue that the internal corporate control of commercial banks also lacked focus in the good times with an emphasis on profit and insignificance of risk (Crotty, 2009; Mehran et al., 2011). Especially the “too big to fail” argument and the implicit insurance of a bailout contributed to a lesser standard of corporate governance (Arcand et al., 2015).
2.2.2 Financial instability and the credit cycle
Minsky was one of the first to describe how a stable financial system becomes unstable (The Minksy moment). Minsky (1977) identifies that the problem is the accumulation of insolvent debt and identifies three types of financial postures that contribute to insolvency. The first posture is “Hedge finance” in which borrowers can meet all debt payments, interest and principal. The second posture is “Speculative finance” in which borrowers can meet their interest payments but have to roll over on their debt to pay back the principal. The final posture is “Ponzi finance” when borrowers can neither
12 pay back the interest or the principal. With Ponzi finance borrowers rely solely on the rising price of the underlying assets to (re)finance their debt.
In a similar vein Kindleberger (2011) starts with a displacement that triggers economic expansion. The subsequent positive economic prospects leads to optimism and the increased availability of credit, which is almost always possible due to changes in institutional arrangements. An increase in the availability of credit leads to economic growth which in turn leads to even greater optimism. Eventually the credit expansion makes speculative manias to gather speed and ever increasing optimism that leads to euphoria. According to Kindleberger (2011) euphoria is a mild form of irrationality but can eventually lead to manias which emphasise irrationally. When euphoria keeps increasing it might evolve into overtrading up until the point that a contradictory event leads to a slow-down. This eventually leads to financial distress and subsequent panic in which there is a rush to sell off less liquid assets in favour of more liquid ones (money). The two authors are very similar to Keynes, unrealistic expectations, asset speculation and irrationality, will likely result in over-leveraged situations with serious consequences for economic growth. The most important aspect that Keynes, Minsky and Kindleberger have in common is irrational human behaviour which is subsequently the main difference with mainstream economics. Mainstream neo classical economic theory assumes rationality and the subsequent outcome of individual optimization which leads to equilibrium of markets (Bezemer, 2011). So neo classical theory, even in dysfunctional financial intermediation, finds new arguments in which the behaviour is individually rational, but socially unwanted.
The work of Minsky and Kindleberger is nowadays explored further with extensive work on credit cycles. For instance Dell’Ariccia et al. (2012a) that show decreasing lending standards as mechanism of the credit cycle. Schularick and Taylor (2012) show empirically that leverage increases partially because of monetary policy responses to crises and that subsequent credit booms are a good predictor of financial crises. Jordá et al. (2013) shows that the financial factors contribute to a great extent to the modern business cycle and that financial crisis recessions are more credit-intensive and tend to be followed by deeper recessions and slower recoveries. Lastly from a policy perspective Aikman et al. (2010) and Dell’Ariccia et al. (2012a, 2012b) focus on how to contribute to macro-finance stability and the curbing of the credit market to reduce the negative effects of credit.
2.2.3 Non-intermediation and differentiation of credit
Two additional topics are described in recent literature which also contribute to a negative effect of financial development on economic growth: the functional differentiation of credit and non-intermediation services. The functional differentiation of credit can contribute negatively based on the literal interpretation of Schumpeter’s’ (1911) secondary effects. Recent literature empirically shows that the distribution of credit shifts from the non-financial sector to the real estate and financial sectors
13 and in the process loosens the link between credit growth and economic growth (Bezemer and Grydaki, 2014). Beck et al. (2012), Bezemer and Grydaki (2014) and Büyükkarabacak and Valev (2010) find a positive effects of non-financial credit on economic growth and a negative effect of credit on economic growth to the financial and real estate sector.
Work by Beck et al. (2014) shows considerable growth of non-intermediation services in the financial sector which arguably does not contribute to economic growth. Demirguc-Kunt and Huizinga (2010) and Baele et al. (2007) find that a higher share of non-interest income, more non-intermediation services, contributes positively to the value of a bank while subsequently increasing the systematic risk. This suggests that financial intermediaries are not just in search for higher profits within intermediation, but also shift to non-intermediation. In novel empirical work Beck et al. (2014) disentangle intermediation and non-intermediation services. They find that intermediation does contribute to growth and reduces volatility but any other expansion of the financial sector has no long-run effect on real sector outcomes.
2.2.4 A dampening effect
The first post financial crisis article that empirically finds no finance-growth relation is Rousseau and Wachtel (2011). By doing the same estimation as King and Levine (1993) but with a dataset from 1960 - 2004 Rousseau and Wachtel (2011) find a dampening of the finance effect on economic growth. By splitting the sample into two periods 1960-1989 and 1990 – 2004 they find similar results to King and Levine (1993) for the early period, but no positive effects for the latter one. The authors try to find an explanation for this result by incorporating the higher frequency of financial crises and find that after controlling for financial episodes with time dummies that the positive finance-growth relation found in the entire period turns positive after controlling for financial crises. They are unsure however on the exact workings, Rousseau and Wachtel (2011, p. 287): “The question of how these countries acquired large financial sectors and how they may have served as engines of growth, however, remains imperfectly understood.” Which channel of the reviewed literature is therefore responsible for the dampening is hard to say. Rousseau and Wachtels’ (2011) study is therefore a stark reminder that the link between finance and growth is more complex than the simple relationship suggests.
3. Non linearity of credit-growth relationship
More empirical literature after the financial crisis embraced the idea of limits to the benefits of finance. Research surfaced that specifically looks at negative effects of credit after it reaches a certain threshold with respect to GDP. Several articles find that at higher levels of credit to GDP the relationship between finance and growth turns negative supposing a non-linearity in the finance-growth relation. (Arcand et al., 2015; Cecchetti and Kharroubi, 2012; Law and Singh, 2014).
14 Cecchetti and Kharroubi (2012) show by using DCP to GDP (% of GDP) and the same indicator squared that there is indeed a non-linear relationship between private credit and economic growth. Using these coefficients to compute the peak of the parabola they find point estimates of roughly 90-100% to GDP. In economic terms they thereby suggest that if the USA would half its private credit to GDP from 200% to 100% it would yield a productivity gain of more than 150 basis points. Arcand et al. (2015) use a similar approach to test for non-linearity. They however employ the U-test by Lind and Mehlum (2010) to verify the inverse-U relationship of finance and growth. The U-test confirms the inverse-U relation found and estimates the peak at a range of 60%-120% of GDP. Arcand et al. (2015, p. 124) own calculations to narrow the range down show an interesting pattern over time. As more recent data is included the thresholds comes down. For the 1960-1995 period the estimation of the peak is at 144%, but only 90% in the 1960-2010 period.
Law and Singh (2014) use a more recent dataset of the World Bank with regard to the financial data with 87 countries from 1970 until 2010. They include in their estimation different control variables than the ones employed by the two previous articles such as: investment, institutions and population growth. Besides their research focusses solely on threshold effects hence no actual comparison can be made with the previous IV work. The thresholds they find are however very much in line with the previous work. At roughly 90% of GDP the finance-growth relation turns negative.
One last thing that is apparent in all the empirical work is the changing coefficients in size and significance of the control variables. Dependent on estimation approach, time frame and countries included there is considerable difference in the size and significance of the control variables. For instance, human capital is insignificant in Law and Singh (2014), not included in Cecchetti and Kharroubi (2012) and changes from insignificant to significant in Arcand et al. (2015) dependent on estimation technique and time frame.
3.1 Other causes of non-linearity?
Before starting with the empirical estimation this last chapter is devoted to discussing some remaining academic work on the non-linearity of credit. Instead of using credit to GDP as threshold, this literature supposes non-linearity through control variables such as: initial GDP, stage of development, financial liberalisation, regulation, laws and institutions, the macro-environment and differentiation of credit. Summarised the stage of a countries development besides financial development is the main interest. Deidda and Fattouh (2002) split their sample in High and Low income countries and find that financial depth as measured by liquid liabilities (M3) only contributes to economic growth in High income countries. Rioja and Valev (2004a, 2004b) use a similar threshold specification but use private credit indicators. They find a similar positive effect of finance on economic growth for High income countries and, in some cases more pronounced, for the Middle income countries. The Low income countries
15 however have no positive effect from finance on growth. Shen and Lee (2006) find that this is best explained by controlling for the (legal) environment. They use control indicators such as: crises, currency dummies, creditor protection and corruption to show that those effects destroy the finance-growth relationship, especially in Low or less developed countries. In a similar vein Ergungor (2006) finds that when he controls for the structure of the financial system there is a (contingent) non-linear relationship between growth and finance. Rousseau and Wachtel (2002) show that the effect of finance can be non-linear in very high inflation environments. All these authors therefore suggest that the positive financial development effect is largest in Middle income countries that have the economical and judicial environment to support it.
In a more recent article by Huang and Lin (2009) countries are separated in Low and High income groups. They find a more pronounced positive effect of finance on growth in Low income countries which, given the other literatures, is likely due to the inclusion of the Middle income countries in the Low income dataset. More importantly however is that Shen and Lee (2006) and Huang and Lin (2009) use a threshold variable to split the dataset in High and Low income countries, but find a non-linear estimation for the finance growth relationship within those datasets more likely than the linear one. Not much of the research pays any specific attention to how credit contribute to the non-linear relation in High and Low income countries. However common sense combined with the literature above would suggest that the avenues causing non-linearity might differ completely. Low income or less developed countries are far more likely prone to dysfunctional intermediation and (endogenous) instability effects, than a widespread shift of credit from non-financial to financial and household credit and non-intermediation services. Furthermore the absence of judicial systems, regulation and sound financial structure will leave more room for credit to get out of hand at an earlier stage of credit to GDP than in developed or High income countries. On the other hand High income countries could suffer more from endogenous financial sector growth, non-intermediation but, also on a larger scale but less noticeable, innovative (dysfunctional) intermediation. Furthermore the size of the financial sector that can eventually harm the economy is much larger in High income countries than in Low income countries it is therefore interesting to see if non-linearity is found for both High and Low incomes countries.
Lastly Cecchetti et al. (2011) combine the non-linearity literature with the functional differentiation and find that all types of credit display non-linearity. The only difference is the peak, where the relationship between corporate debt and growth turns negative at roughly 90% of GDP, household and government debt do so at roughly 85%.
16
4. Hypotheses, methodology and data
4.1 Hypotheses
Since the empirical work in the next chapter is based upon Rousseau and Wachtel (2010), Law and Singh (2014) and Arcand et al. (2015), the first expectation is to find similar results to these authors. Summarised: a dampening of the finance-growth relationship over time. Furthermore by including the years of the financial crises up until 2014, there is strong expectation that in the last data period there could even be a significant negative effect of finance on growth in the linear specification.
Following Arcand et al. (2015) estimations will be conducted to see if a non-linear relation is found. The hypothesis is that there is a non-linear relationship between finance and growth for the entire period. Separating the dataset into three time frames will show if the non-linear relationship is also found over multiple timeframes. Since most of the empirical research found a positive finance-growth relationship on older datasets, they all start around the 60s or 70s and nothing is over 1995 with the majority under the 1990s, chances are the finance-growth relationship has changed with the increasing occurrence of financial crises as is suggested in more recent literature. The hypothesis is therefore that the inverse-U relationship is found over the complete period but less so over the earlier time frame. The reasoning behind this is that the primary effects are dominant in the earlier years, while the secondary effects (negative) effect are more pronounced in the later years. The secondary effects can include a decrease in the marginal efficiency of credit (quality), the domestic credit to the private sector (DCP) to GDP levels in the later period being considerably higher (quantity) and the more common recurrences of financial crises.
The non-linear specification is followed through by using credit indicators as threshold variable. The expectation is that thresholds are found mostly for the full period and the later period but less so for the early period due to similar arguments as above. Furthermore similar thresholds as Law and Singh (2014) are to be expected for the full dataset. Total credit (TC) is included in the first specification to confirm the different behaviour of credit if public credit is included. Law and Singh (2014) also conduct thresholds estimations on Total Private Credit and results are expected to be similar to theirs.
Lastly, by following the literature that separates countries, the dataset is split into Low and High income countries. Non-linear specification and threshold estimation will be conducted which will first and foremost show a much lower threshold in low income countries. Furthermore the expectation is that the non-linear specification is a better fit for High income countries than for low since the marginal efficiency of credit is probably higher and DCP levels are left of the peak compared to High income countries.
17
4.2 Methodology
The starting point of the empirical analysis is the King and Levine’s version of the Barro growth regression:
𝐺𝑟𝑜𝑤𝑡ℎ𝑖,𝑡 = 𝛼0+ 𝛼1𝐶𝑟𝑒𝑑𝑖𝑡𝑖,𝑡+ 𝛽𝑋𝑖,𝑡+ 𝑢𝑖,𝑡 (1) Where 𝛼0 is the constant, 𝐺𝑟𝑜𝑤𝑡ℎ𝑖,𝑡 is the real GDP growth rate, 𝐶𝑟𝑒𝑑𝑖𝑡𝑖,𝑡 is a financial development
measure usually related to GDP and 𝑋𝑖,𝑡 is a set of baseline explanatory variables that have shown to
be robust determinants of growth and 𝜀𝑖𝑡 is the error term. The 𝑋𝑖,𝑡 set contains the log of initial real
per capita GDP, which should control for convergence across countries and over time and the log of average years of secondary schooling which should capture the human capital effects. Furthermore the ratio of trade to GDP is included (import plus exports) and the ratio of government final expenditure to GDP.
The above growth regression is estimated in three different versions. First a basic OLS growth regression on pure cross country data is done to infer the effect of credit to GDP ratios on the average real economic growth rate. Following Rousseau and Wachtel (2011) multiple time frames will be chosen to infer if credit effects have changed over time. The OLS growth regression with pure cross section data is:
1
𝑛∑ 𝐺𝑟𝑜𝑤𝑡ℎ𝑖 𝑛
𝑖=0 = 𝛼0+ 𝛼1𝐶𝑟𝑒𝑑𝑖𝑡𝑖+ 𝛽𝑋𝑖+ 𝜀𝑖 (2)
Where 𝐺𝑟𝑜𝑤𝑡ℎ𝑖 is the real average growth rate for the time period chosen, 𝛼0 is the constant and
both 𝐶𝑟𝑒𝑑𝑖𝑡𝑖 and 𝑋𝑖 are the initial values from the start of the cross-section for all explanatory
variables in the regression. This specification should reduce any simultaneity bias that might result from the stimulus of economic growth on financial development.
The motivation for using domestic (private) credit to GDP ratios is twofold, one motivation is with regard to being a good indicator for the level of financial intermediation and readily available as thoroughly explained in the above literature. Second, it is in concurrence with most of the empirical literature that looks at similar effects of credit to GDP ratios on economic growth.
After the OLS cross country regressions a similar empirical exercise is conducted with panel data. In line with the empirical growth literature the dataset is averaged over five-year data periods3.
This will enable the study of long term effects and smooth the business cycle effects. An Instrumental Variable regression is estimated similar to (1). A vector of instruments is added which will contain all initial values of the right hand side variables. Additionally for every time period a dummy is included. To control for non-linearity in the basic specification, the log of the level of credit to GDP is taken ln (𝐶𝑟𝑒𝑑𝑖𝑡𝑖). Subsequently preliminary tests to infer non-linearity between 𝐶𝑟𝑒𝑑𝑖𝑡 and 𝐺𝑟𝑜𝑤𝑡ℎ is
18 done by replacing the log of the credit variable with the actual level of credit to the private sector (𝐶𝑟𝑒𝑑𝑖𝑡𝑖) and a quadratic term of this variable (𝐶𝑟𝑒𝑑𝑖𝑡𝑖2). Arcand et al. (2015) namely suggest that the
assumption of a linear relationship could be a misspecification of the finance-growth relation which will result in a downwards bias over time. Since Lind and Mehlum (2011) show that using a level and quadratic term of this level are not sufficient conditions to test for the presence of a non-monotonic relationship, there U-test is applied to test for the presence of an inverted U shape. Lind and Mehlum (2011) suggest to test whether the relationship is decreasing at low levels and increasing at high level with a minimum or monotonic relation in the interval between the high and low level, or vice versa for an inverse U relationship. Given a model of the form:
𝐺𝑟𝑜𝑤𝑡ℎ𝑖,𝑡 = 𝛽1𝐶𝑟𝑒𝑑𝑖𝑡𝑖,𝑡+ 𝛽2𝐶𝑟𝑒𝑑𝑖𝑡𝑖,𝑡2 + 𝛾𝑋
𝑖,𝑡+ 𝜀𝑖,𝑡 (3)
it is necessary to formulate the following joint null hypotheses:
𝐻0∶ ( 𝛽1+ 2𝛽2𝐶𝑟𝑒𝑑𝑖𝑡𝑀𝐼𝑁 ≤ 0) ∪ ( 𝛽1+ 2𝛽2𝐶𝑟𝑒𝑑𝑖𝑡𝑀𝐴𝑋≥ 0) (4) against the alternative:
𝐻1∶ ( 𝛽1+ 2𝛽2𝐶𝑟𝑒𝑑𝑖𝑡𝑀𝐼𝑁 > 0) ∩ ( 𝛽1+ 2𝛽2𝐶𝑟𝑒𝑑𝑖𝑡𝑀𝐴𝑋< 0) (5)
where 𝐶𝑟𝑒𝑑𝑖𝑡𝑀𝐼𝑁 and 𝐶𝑟𝑒𝑑𝑖𝑡𝑀𝐴𝑋 are the minimum and maximum values of the credit variable and 𝛽1
and 𝛽2 are the slope estimates at the low and high level respectively. The test described is non-trivial
because of the presence of inequality constraints. Lind and Mehlum (2011) use Sasabuchi’s (1980) likelihood ratio approach to build a test for the joint hypotheses given by equations (4) and (5).
Finally the threshold model introduced by Hansen (1999) will look if there is a break in the relationship between the ratio of credit to GDP with regard to economic growth. The following one-threshold equation with varying intercept and coefficient is estimated:
𝐺𝑟𝑜𝑤𝑡ℎ𝑖,𝑡 = { 𝛼0+ 𝛽0𝐶𝑟𝑒𝑑𝑖𝑡𝑖,𝑡+ 𝛾𝑋𝑖,𝑡+ 𝜀𝑖,𝑡, 𝜆 < 𝐶𝑟𝑒𝑑𝑖𝑡𝑖,𝑡 𝛼1+ 𝛽1𝐶𝑟𝑒𝑑𝑖𝑡
𝑖,𝑡+ 𝛾𝑋𝑖,𝑡 + 𝜀𝑖,𝑡, 𝜆 ≥ 𝐶𝑟𝑒𝑑𝑖𝑡𝑖,𝑡 (6)
where the threshold level 𝜆 is estimated simultaneously with the other parameters. The conditionality is dependent upon 𝐶𝑟𝑒𝑑𝑖𝑡𝑖𝑡 which is either smaller or larger/equal to 𝜆 so that the regression is able
to estimate the slopes of 𝛼0 and 𝛼1 separately and thereby show the effect of credit on growth below and above that threshold4. Following Hansen (1999) this thesis takes three steps to estimate the
specification coefficients. First an estimation of the basic regression is run and its residual sum of squares for testing the significance of the thresholds is stored. Secondly for an entire range of values of 𝐶𝑟𝑒𝑑𝑖𝑡𝑖𝑡, thresholds are fitted and additional regressions are estimated with OLS. Finally the
4 I would like to thank Ward Romp for sharing his STATA code for the threshold tests and the underlying
19 estimator of the threshold 𝜆 that is selected is the one associated with the smallest sum of squared residuals.
To test for the statistical significance of the threshold the Hansen’ (1999) F-test of no threshold effect is used which infers 𝐻0∶ 𝛽0= 𝛽1. The F-test as suggested by Hansen:
𝐹1 = (𝑆0− 𝑆1(𝜆̂))/𝜎̂2 (7)
Furthermore the bootstrap implementation of Hansen (1999) is used to test the hypothesis 𝐻0∶ 𝜆 =
𝜆0, the specification with and without a threshold. The bootstrap procedure treats the regressors and threshold variable as given and holds them fixed in repeated bootstrap samples. The regressions residuals are grouped by individual and treated as the empirical distribution to be used for bootstrapping. Subsequently a sample of sufficient size is drawn from the empirical distribution and these errors are used to create a bootstrap sample. Using the bootstrap sample the model is estimated with and without a threshold and the bootstrap value of the likelihood ratio statistic is calculated. This procedure is repeated a large number of times and the percentage of draws for which the simulated statistic exceeds the actual is calculated. This resulting figure is the bootstrap estimate of the asymptotic p-value. The null of no threshold is rejected if the p-value is smaller than 0.05. When a significant threshold effect is found, the best way to form confidence intervals for 𝜆 is to form the so called “no-rejection region” using the likelihood ration statistic for test on 𝜆. The likelihood ratio test is to reject for large values of the threshold:
𝐿𝑅1(𝜆) = (𝑆1(𝜆) − 𝑆1(𝜆̂))/𝜎̂2 (8)
Subsequently a 95% confidence interval can easily be constructed using the same likelihood ratio:
Γ = {𝜆 ∶ 𝐿𝑅1(𝜆) ≤ 𝐶(𝛼)} (9)
where 𝐶(𝛼) is the 95th percentile of the asymptotic distribution of the likelihood ratio statistic 𝐿𝑅
1(𝜆).
4.3 Data
The starting point for the data is the World Development Indicator (WDI) database of which the most recent one (March 2015) is downloaded. Quite surprisingly a lot of data seems to be missing from this dataset with regard to the 1960 – 1970 period. Both 2007 editions of the WDI Database, either one of which is used by Rousseau and Wachtel (2011), misses data, specifically secondary school enrolment data and many liquid liabilities entries (M3). Following Law and Singh (2014) two similar but more completely available variables are imported from different data streams. The first is liquid liability (M3) data from the financial structure database by Beck et al. (2000) as indicator for financial depth. The second is average years of secondary education from the Barro and Lee (2013) dataset. Since all datasets have been previously used in the literature the comparability between this empirical exercise and the ones conducted in the referred literature is considerable.
20 Table 1 shows the descriptive statistics of the dataset, while Figure 1 through 3 show the level of private credit to GDP and per capita GDP growth respectively. The first indication that GDP growth and credit growth do not necessarily move together is shown in the figures. Where there is considerable volatility in GDP growth, there is only a positive trend with regard to Domestic credit to the Private sector (DCP). Even in 2009, where a considerable drop in GDP growth is visible, there is only a small correction in DCP that already picks up again in 2012-2013. This shows that economies increased debt levels resulting in an ever more indebted world.
Table 1. Descriptive Statistics
Unit of Measurement
Obs. Mean Std. Dev. Min Max Dependent Variable
Real GDP Growth per capita % 864 1,820 2,825 -11,511 16,273
Financial Development
M3 % of GDP 864 48,051 44,494 0,000 370,426
M3 - M1 % of GDP 740 41,047 43,645 0,004 370,426 Total Domestic Credit % of GDP
in logs 821 3,858 0,770 1,024 5,844
Domestic Credit to the Private Sector
% of GDP
in logs 835 3,468 0,906 0,652 5,367
Domestic Credit to the Private sector provided by Banks % of GDP in logs 835 3,403 0,884 0,652 5,327 Control Variables Initial GDP % of GDP in logs 831 8,113 1,616 5,161 11,309
Initial Schooling Average years
in logs 855 0,432 0,872 -3,507 1,923
Government Expenditure % of GDP 864 14,729 6,418 0,000 48,062 Trade % of GDP 864 72,749 54,365 0,000 447,083
N= 95. T= 1970-2013. List of Countries in the Appendix (p. i-v)
Most of the literature finds, but not always explicitly mentions, considerable differences between countries. For instance the debt levels of High and Low income countries as displayed in Tables A2.1 and A2.2 (Appendix, p. vi) and Figures 1 through 3, are considerably different. By comparison, out of the 32 high income countries only four are considerably below 100% private domestic credit to GDP (Saudi-Arabia, Kuwait, Trinidad and Tobago and Barbados) and out of the 64 low income countries only five are above 100% private domestic credit to GDP (Thailand, South-Africa, Malaysia, Chili, Mauritius). This is a strong argument that coincides with the literature reviewed above to split the group in a later stage of the threshold estimation.
21 Figure 1. Level of Total Domestic Credit to the Private Sector to GDP (High Income Countries)
0% 20% 40% 60% 80% 100% 120% 140% 160% 180% 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013
TOTAL DCP TO GDP (HIGH INCOME)
United States United Kingdom Trinidad and Tobago Switzerland Sweden Spain Singapore Saudi Arabia
Portugal Norway New Zealand Netherlands Malta Luxembourg Kuwait Korea, Rep.
Japan Italy Israel Ireland Iceland Hong Kong Greece Germany
22 Figure 2. Level of Total Domestic Credit to the Private Sector to GDP (Low Income Countries)
0% 10% 20% 30% 40% 50% 60% 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013
TOTAL DCP TO GDP (LOW INCOME)
Zimbabwe Zambia Venezuela, RB Uruguay Turkey Tunisia Togo
Thailand Syrian Arab Republic Sudan Sri Lanka South Africa Sierra Leone Senegal
Rwanda Philippines Peru Paraguay Papua New Guinea Panama Pakistan
Niger Nicaragua Nepal Morocco Mexico Mauritius Mali
Malaysia Malawi Lesotho Latvia Kenya Jordan Jamaica
Iran Indonesia India Hungary Honduras Haiti Guyana
Guatemala Ghana Gambia, The Gabon Fiji El Salvador Egypt
Ecuador Dominican Republic Cote d'Ivoire Costa Rica Congo, Rep. Colombia Chile
Central African Republic Cameroon Brazil Bolivia Barbados Bangladesh Argentina
23 Figure 3. GDP Growth High and Low income countries
-6% -4% -2% 0% 2% 4% 6% 8% 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013
GDP Growth (High income)
-4% -2% 0% 2% 4% 6% 8% 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013
