THE STRUCTURE OF FINANCING AND ITS EFFECTS ON
THE REAL ECONOMY THROUGH SYSTEMIC RISK
AUTHOR
BATS, JOOST V.
STUDENT NUMBER
11131799
SUPERVISOR
DHR. PROF. DR. HOUBEN, AERDT C.F.J.
UNIVERSITY OF AMSTERDAM
MSC ECONOMICS: MONETARY POLICY, BANKING AND REGULATION JULY 2016
KEYWORDS
2 ABSTRACT
In light of the debate on the superiority of bank-based versus market-based financing, this thesis explores the relationship between financial structure and systemic risk. Linear regression models are estimated over a panel of 19 European countries and the United States. The linear estimations show that more bank activity is associated with more systemic risk, whereas more market activity is associated with less systemic risk. Additionally, threshold regression models are estimated over a panel of 12 European countries and the United States. The threshold estimations suggest that some degree of bank-based financing is superior to no bank-based financing and that more market-based financing is always better. This implies that Europe’s financing is not optimally structured. The European financial structure should become less bank-based through both more market-based financing and less bank-based financing.
STATEMENT OF ORIGINALITY
This thesis is written by Joost V. Bats who declares to take full responsibility for the contents of this thesis. I declare that the text and the work presented in this thesis is original and that no source other than those referred to in the text and those in section XII have been used in creating it. The faculty of Economics and Business of the University of Amsterdam is responsible solely for the supervision of the completion of the work, not for the contents.
3
TABLE OF CONTENTS
I. INTRODUCTION 4
II. THE FINANCIAL SYSTEM AND THE REAL ECONOMY 5
III. BANK-BASED VS. MARKET-BASED FINANCING: THEORY 8
IV. BANK-BASED VS. MARKET-BASED FINANCING: EMPIRICAL
LITERATURE 11
A. EMPIRICAL LITERATURE UNTIL 2008 12
B. EMPIRICAL LITERATURE AFTER 2008 13
C. CONCLUSIONS FROM THE LITERATURE 15
V. SYSTEMIC RISK AND THE FINANCIAL SYSTEM 15
VI. METHODOLOGY 16
VII. DESCRIPTIVE DATA 20
VIII. RESULTS 23
A. FIXED EFFECTS PANEL REGRESSIONS MODEL (1) 23
B. STRUCTURAL BREAK MODELS (2) AND (3) 26
IX. DISCUSSION 29
X. LIMITATIONS BY PATH DEPENDENCE 31
XI. CONCLUSION 33
XII. REFERENCES 34
4
I.
INTRODUCTION
The European banking system has grown significantly in comparison to European capital markets and European economic production. As a result, the European financial structure has become relatively bank-based. The United States on the other hand, contains a more market-based financial structure. They therefore channel their savings into investments differently. The financing in Europe consists mostly of financial institutions that execute intermediation on their balance sheet. These financial institutions lend primarily in the form of loans, often through close relationships with their clients. In contrast, the United States channels their savings more through markets. These markets serve as a platform where equity and debt securities are distributed and traded.
Both bank-based and market-based financing have different advantages to the real economy. One advantage of banks is that clients are served in a more stable manner through the business cycle. This is due to banks’ supply of credit via relationships and trust. Contrary to banks, markets’ advantage stems from their continuous incentives to improve performance. Markets do this via the provision of better risk management tools and better allocation of capital. In light of these differences, a century old debate exists on what financial structure better serves the real economy.
To assess financial structure’s effect on the real economy, many researchers have studied its effect on economic growth. Remarkably, the literature published before 2008 does not favor one particular financial structure. The literature published after 2008 however, shows that once business cycles are accompanied with 2008’s financial crisis, market-based financial structures outperform bank-based financial structures. The stability of the financial system therefore proves to be of significant importance.
The stability of the financial system is largely influenced by systemic risk. Systemic risk is the risk that a triggering event, such as the failure of a large financial institution, will break down financial markets and impair the real economy. Hence, more systemic risk worsens economic activity. Since the literature shows that the instability of the financial system affects banks’ performance, it is important to assess whether financial structure has an effect on this instability. Therefore, this thesis empirically studies financial structure’s effect on the real economy; not by looking at economic growth, but by looking at systemic risk. The research question is the following: what financial structure minimizes systemic risk negatively influencing the real economy?
5 Although Langfield and Pagano (2015) also test financial structure’s effect on systemic risk, this thesis does not distinguish between a housing market crisis and a stock market crisis. Instead, the analysis accounts for the recent banking crisis. Additionally, this thesis assesses whether financial structure’s effect on systemic risk changes below and above a certain threshold, thereby allowing for a structural break in the data. This is done by estimating Hansen’s (1999) threshold regression model over a panel of European countries and the United States.
The rest of this thesis consists of 11 sections. Section II presents the three transmission channels through which the real economy is affected when the financial system turns unstable. Section III then gives an overview on the advantages and disadvantages of both bank-based and market-based financing. Thereafter, section IV presents the empirical literature on financial structure and the real economy. This section separates the literature that was published before 2008 from the literature that was published after 2008. Section V then proves that financial system’s stability is greatly affected by systemic risk. The methodology is presented in section VI whereafter the description of the data is given in section VII. The empirical results are subsequently shown in section VIII and discussed in section IX. Additionally, section X explains the results’ limitations by path dependence. Although the results suggest that financial structures should change, past decisions and practices largely drive current and future situations. Finally, the conclusion is given in section XI.
II.
THE FINANCIAL SYSTEM AND THE REAL ECONOMY
In the view of an economist, studying the financial system is especially interesting when this system has an effect on the real economy. Therefore, this thesis begins with addressing the transmission channels between the financial and real sectors of the economy, ultimately showing that studying any effects on the financial system is of economic interest. The theoretical literature has highlighted three transmission channels between the real and financial sector [Basel Committee on Banking Supervision (2011)]. These channels are discussed in the following paragraphs.
The first represents the borrower balance sheet channel. This channel addresses the limited ability of lenders to determine borrowers’ risk and solvency, to monitor borrowers’ investments, and to enforce repayment of debt. This inability causes lenders to charge higher interest rates and to demand collateral for borrowing [Basel Committee on Banking Supervision (2011)]. For this channel, two broad classes of borrower balance sheet models
6 exist that help to explain the effects on the real economy. In the first class, borrowers face an external finance premium. This premium increases once a borrower’s net worth goes down (once a borrower’s creditworthiness decreases). Any shock affecting net worth will change the borrower’s cost of financing so that aggregate demand is affected [Bernanke and Gertler (1989)]. Because net worth is affected by shocks to the real sector, the external finance premium propagates shocks to the real economy and amplifies business cycle fluctuations. The borrower balance sheet channel therefore acts as a financial accelerator. The second class of the borrower balance sheet channel presents a dual role for assets in the economy. In this model, assets are both used to produce goods and services, and to provide collateral for loans. Subsequently, any financial shock that lowers asset prices will cause lenders to demand more collateral. The increase in demand then lowers production and spending [Kiyotaki and Moore (1997)].
Regarding these theoretical models, the borrower balance sheet channel carries much empirical support. Quantitative literature can be found on the effect of credit-market shocks to the macroeconomy [Gilchrist et al. (2009), Mody and Taylor (2003), Stock and Watson (2001)]. More specifically, these studies have found that increases in corporate credit spreads can cause large and persistent contractions in economic activity.
The second transmission channel runs via the bank balance sheet. This channel is therefore more applicable in relatively bank-based financial structures as it focuses on changes in credit coming from banks. The bank balance sheet channel can be divided into two components, each explaining their effects on the real economy. The first is the traditional bank lending channel. Following negative monetary changes, both money supply and money demand decrease. However, as the monetary shock also changes the asset composition on a bank’s balance sheet, the bank supply of credit declines by even more [Bernanke and Blinder (1988)]. This causes economic activity to fall and the real economy to be negatively affected. The traditional bank lending channel assumes that banks hold no capital and are fully funded by external liabilities. Therefore, the above model cannot analyze the effects coming from changes in banks’ capital levels. For this reason, the second bank balance sheet component studies the changes in banks’ capital levels and is called the bank capital channel. Bank capital may affect economic activity due to capital’s effect on the creditworthiness of banks [Stein (1998)]. Relatively well capitalized banks can continue to extend credit when adverse selection shocks occur, so that banks with higher capital ratios are superior in terms of their credit provision to the economy. This leads better capitalized banks to attract funding at a
7 lower cost, implying that a decrease in bank capital increases the cost of funds faced by banks and borrowers. This ultimately hinders the real economy by a decrease in loans. Another reason why bank capital may have an effect on the real sector is due to regulatory requirements. When financial institutions contain insufficient holdings of capital, bank lending will be constrained [Van den Heuvel (2002)].
Empirically, both components of the bank balance sheet channel have both been supported and rejected by the literature. By looking at movements in bank lending following a monetary policy shock, Kashyap et al. (1993, 1996) find support for the traditional bank balance sheet channel. However, when taking into account the size of firms to whom these loans are granted, the support is rejected [Oliner and Rudebusch (1995)]. The bank capital channel has both been supported [Hancock and Wilcox (1994), Brinkmann and Horvitz (1995)] as rejected [Berger and Udell (1994)] when looking at whether capital requirements have affected the supply of loans; thereby hampering economic activity.
The last channel that accounts for the transmission of shocks moving from the financial to the real sector is the liquidity channel. This channel has gained relevance during the recent financial crisis where insufficient liquidity influenced banks’ supply of credit and hence, economic activity [Basel Committee on Banking Supervision (2011)]. This influence may in some cases occur by enhancing the other transmission channels. For example, Kashyap and Stein (2000) showed that by introducing the liquidity structure of banks’ balance sheets into the framework of the bank lending channel, the effect of monetary policy shocks on bank lending increased. In many other cases, liquidity proves to be its own additional transmission channel where shocks in the financial system have an effect on the real sector of the economy. One example stems from the liquidity mismatch on a bank’s balance sheet. Since banks finance illiquid assets with short-term debt, they may be forced to prematurely stop profitable loans when depositors unexpectedly demand payments [Diamond and Rajan (2005)]. This may then cause significant losses leading banks to withhold future lending. A more recent debate however, distinguishes between funding liquidity and market liquidity. Where funding liquidity represents the bank’s ability to receive immediate funding, market liquidity represents the ease at which institutions can trade assets [Brunnermeier and Pedersen (2007)]. Diamond and Rajan (2009) show that the presence of both funding and market liquidity can induce even healthy banks to hold back their future supply of loans. More specifically, due to the anticipation of distressed banks selling illiquid assets at heavily discounted prices, healthy
8 banks set aside liquid funds as to make use of profitable investment opportunities in the future. Therefore, one may find high demand for funding liquidity in times of illiquid markets.
Regarding the liquidity channel, the literature provides much empirical support for the third transmission channel. Drehmann and Nikolaou (2009) find strong evidence for the presence of an inverse relation between market liquidity and demand for funding liquidity; especially during distressed conditions. Furthermore, stock prices have presented themselves to be positively correlated with market liquidity; therefore resulting in discounted prices when the market turns illiquid [Chordia et al. (2000)].
To summarize the above transmission channels, figure 1 gives an overview on their theory and corresponding mechanisms.
Figure 1: Transmission channels between the real and financial sector
Overall, the literature has been clear in presenting the general conclusion that shocks in the financial system may have significant consequences for the real economy. Having explained the transmissions of these shocks via three different channels, this thesis continues on the workings of the financial system by discussing the structure of financing in today’s economy.
III.
BANK-BASED VS. MARKET-BASED FINANCING: THEORY
Although all financial systems contain a mix of both bank- and market-based financing, the financial structure – the actual mixture of bank- and market-based intermediation – is
9 different across the world. Where in Europe financing mainly consists of financial institutions who execute intermediation on their balance sheet, the United States channel their savings more through markets in which equity and debt securities are distributed and traded. In light of these differences, a debate has risen questioning which type of financing is better in terms of the real economy [Allen and Gale (2000)].
For over half a century, financial economists have discussed the comparative advantages of both bank- and market-based intermediation [Goldsmith (1969), Boot and Thakor (1997), Allen and Gale (2000), Demirgüç-Kunt and Levine (2001c)]. Generally, banks prove to be a better remedy for informational asymmetry because of advantages in collecting and processing information, while markets have far less incentives to acquire information due to free-riding by other market participants [Boot (2000), Stiglitz (1985)]. Therefore, banks are able to reduce adverse selection ex ante and moral hazard ex post such that a bank-based structure may better allocate resources and promote economic development. Also, since markets publicly provide their information in the market, market-investors are discouraged from allocating resources to research firms [Stiglitz (1985)]. Banks on the other hand, can privatize information since they form long-term relationships with firms, ultimately fostering growth. Proponents of the bank-based view additionally argue that liquid markets weaken the incentives for investors to exert corporate control [Bhide (1993)]. Since liquid markets lower the costs of selling for investors, market development may discourage corporate control and may eventually hinder economic growth. Lastly, bank-based financial structures may minimize external investors’ reluctance to finance industrial expansions that foster growth [Gerschenkron (1962), Rajan and Zingales (1998)]. This is because especially in countries with weak contract enforcement, powerful banks are better able to force firms to repay their debts. Hence, a bank-based financial structure provides investors with more certainty regarding their loans.
Contrary to bank-based intermediation, proponents of market-based finance argue that well-functioning markets are better able to foster growth due to better risk management tools [Levine and Zervos (1998), Obstfeld (1994)]. Also, where bank-based proponents believe that information privatization is superior, the market-based view places more value on markets’ public provision of information since it strengthens research firms’ incentives to participate [Holmstrom and Tirole (1993)]. Research firms readily profit from information by trading in big markets therefore enhancing economic growth. Furthermore, Jensen and Murphy (1991) explain that markets improve corporate governance by smoother takeovers and that
10 managerial compensation better matches firm performance in case of market-based economies. This respectively adds to economic growth.
Besides markets’ own benefits, market-based proponents additionally address problems that are present in the case of bank-based structures. Strong banks can extract sizeable shares of their borrowers’ profits due to their informational advantage, therefore hampering borrowers’ incentives to invest [Rajan (1992)]. Rajan (1992) subsequently showed that this problem can be mitigated in case of access to markets, therefore boosting economic growth. Furthermore, in case of powerful banks, corporate governance may be hindered as banks team up with firm managers against other creditors [Wenger and Kaserer (1998)]. This type of collusion between banks and firm managers is not present in the case of market-based structures since markets do not base their financing on relationships. Lastly, although bank-based proponents argue that banks may punish defaulting borrowers by simply refusing further credit, Dewatripont and Maskin (1995) point out that this threat is not credible. Namely, once a default has occurred, the lender’s costs are sunk. The bank will therefore still want to provide a loan when the defaulting borrower has a new project with a positive net present value (calculated by incorporating new default risks). For this reason, banks may not necessarily be considered better in mitigating the moral hazard problem.
Even though the debate between bank-based and market-based proponents is extensive and presents concise and clear arguments, there is an additional view that does not favor one of the two financial structures. The financial services view says that different financial structures add to the real economy by the same degree, as long as these structures provide sound financial services [Merton and Bodie (1995)]. Therefore, the problem does not relate to whether a financial structure should be market- or bank-based, but whether these financial services are actually provided. The financial services view places its focus on how to create better functioning banks and markets, and presents no preference in the debate between bank-based and market-bank-based financing.
Overall, the debate sheds light on many arguments favoring both bank-based and market-based financing when looking at economic performance. To give a brief overview, table 1 summarizes the advantages and disadvantages of both bank-based and market-based financial structures.
11 Table 1: The advantages and disadvantages of bank-based and market-based financing
FINANCIAL STRUCTURE
ADVANTAGES DISADVANTAGES
Bank-based
Collecting and processing information: reduce adverse selection ex ante and moral hazard ex post
Privatize information due to long-term relationships More stable supply of credit
through the business cycle due to relationships and trust Enforcement of debt
repayments: contract enforcement
Hampers borrowers’ investments due to profit extractions by banks
Hinders corporate governance due to teaming up of banks with firm managers
Defaulting borrowers keep receiving bank credit as default costs are sunk
Market-based
Continuous incentives to improve performance
Better risk management tools Public provision of information
strengthens research firms’ participation
Improved corporate governance Managerial compensation better
matches firm performance
Informational asymmetry: acquire less information
Discouragement from allocating resources to research firms due to public provision of information Less corporate control due to
liquid markets
Despite the advantages and disadvantages, the debate on bank-based versus market-based financing has proven ambiguous regarding the superiority of the two types of financial structures. In light of this unpredictability, there exists a varying set of literature which empirically tries to assess whether one financial structure outperforms the other in terms of economic growth. This empirical literature is presented in the next section.
IV.
BANK-BASED VS. MARKET-BASED FINANCING:
EMPIRICAL LITERATURE
The empirical literature on the relation between the structure of financing and the real economy is large and diverse. Nonetheless, one finds a changing trend in conclusions over time. More specifically, once the data covers 2008’s financial crisis, results change. The literature therefore shows that the instability of the financial system changes financial structure’s effect on economic performance. In light of this, the empirical literature until 2008 is presented first, whereas the post-2008 literature is presented second.
12 A. EMPIRICAL LITERATURE UNTIL 2008
Having published “Financial Structure and Development” in 1969, Goldsmith (1969) was one of the first to contribute to the empirical literature of financial structure’s effect on the real economy. More specifically, he examines financial structure’s effect on economic growth by conducting a cross-country study for the period 1864-1914. His cross-country sample consists of Germany, Japan, the United Kingdom and the United States. In his analysis, Germany and Japan are considered relatively bank-based whereas the United Kingdom and the United States are considered relatively market-based. Goldsmith (1969) is able to distinguish between bank-based and market-based financial structures. However, he argues that comparisons between the four countries cannot serve as evidence since no significant differences are present in the countries’ growth rates. Also, since the comparisons were based on only four countries, and as no more countries could be compared due to data limitations, the findings could not be extended to different countries. Accordingly, he does not find support for the superiority of either bank-based or market-based financial structures in terms of economic development, nor is he able to explain whether the structure of financing really matters for economic growth.
More recently, Levine (2002) supports Goldsmith’s (1969) argument by concluding that no evidence can be drawn from results based on only four countries containing similar growth rates. Therefore, and due to better data accessibility, Levine (2002) examines a panel of 48 countries for the time period 1980-1995. Nevertheless, the results appear insignificant, thus leading him to conclude that a country’s financial structure does not affect economic growth. Similarly, looking at 44 industrial and developing countries for the period 1986-1993, Demirgüç-Kunt and Levine (1996) find that the distinction between bank-based and market-based financing seems to be of no importance when tested on economic performance. This unimportance is subsequently supported by several other studies, ultimately concluding that only the overall provision of financial services matters for the real economy (services provided by banks and markets together), not the financial structure itself [Demirgüc-Kunt and Maksimovic (2002), Beck and Levine (2002)]. For example, Beck and Levine (2002) test whether industries that rely heavily on external finance, are fostered to grow due to the structure of financing. Using a panel with 42 countries and 36 industries, they fail to find any significant results; the financial structure proves not to have an effect on industrial growth. Instead, it is the development of the financial system that significantly affects growth.
13 Contrary to the above conclusions, Levine and Zervos (1998) find significant results for the relation between countries’ financial structure and economic growth. By conducting a cross-country analysis for 47 countries over the time period from 1976 to 1993, they conclude that market-based structures provide different financial services from bank-based structures, and that both these different services enhance economic growth. In a more recent study, Beck and Levine (2004) support the above findings. Using a panel data set for the period 1976-1998, they find that both markets and banks positively affect economic growth. Furthermore, the World Bank (2001) finds similar conclusions in which the development of banks and markets both boost economic growth. They additionally argue that banks and markets complement each other in their effects on the real economy.
Despite these significant results, none of the above studies find evidence for the superiority of one specific financial structure; that being either bank-based or market-based. All they find is that both types of financial structures positively add to the real economy. Hence, the overall consensus in the literature is that there is no type of financial structure outperforming the other. Continuing on this consensus, Arestis et al. (2001) empirically show that both banks and stock markets support economic growth. However, contrary to the above literature, the positive effect of banks is more powerful. They therefore provide evidence in favor of the superiority of bank-based financial structures. Nonetheless, their analysis is only based on Germany, Japan, United States, United Kingdom, and France. Following Goldsmith (1969) and Levine (2002), this subsample of similar countries might be too small to draw broad conclusions on favoring a financial structure.
Therefore, the literature until 2008 presents results not favoring one type of financial structure over the other. These conclusions change however, when looking at more recent studies. The conclusions of these studies are presented in the next subsection.
B. EMPIRICAL LITERATURE AFTER 2008
Much of Levine’s work has shown not to favor a particular financial structure. However, his results have changed once data was covered from 1990 to 2011, thereby including 2008’s financial crisis. Levine et al. (2015) show that stronger shareholder protection laws mitigate the negative effects of a banking crisis on profitability, employment, and investment. Following the view that stronger shareholder protection laws provide the legal infrastructure for markets to function as an alternative source of financing, market-based economies outperform bank-based systems by lessening a banking crisis’ impact on the real economy.
14 They hence conclude that markets work as a “spare tire” during banking crises. Banks extend credit to their clients even during business cycle downturns. This is because bank-based financing is more based on relationships and trust. However, when faith in banks falls during a banking crisis, lenders and borrowers will choose markets as an alternative source of financing.
In line with Levine et al. (2015), Gambacorta et al. (2014) support markets’ positive role during a financial crisis. They look at financial structure’s effect on economic downturns, thereby making a distinction between normal downturns and downturns accompanied with a financial crisis. The results show that for normal downturns, bank-based economies are favored since healthy banks work as buffers during negative shocks. Yet, when these normal downturns are accompanied with a financial crisis, they find that the effect on GDP growth is three times worse for bank-based economies than for market-based economies. Furthermore, they conclude that both banks and markets foster economic growth, but that the former carries the largest effect. Also, by allowing for non-linearity, they find that beyond a certain point of bank and market activity, the positive effect for both banks and markets is no longer present. Overall, their results suggest that for long-term growth, both banks and markets are important. However, in times of a financial crisis, market-based systems are favored over bank-based systems.
Contrary to Gambacorta et al. (2014), Pagano et al. (2014) show that market-based economies are favored not only during a financial crisis, but also over a period from 1989 to 2011; a period not only including downturns. Even when they shorten their data period to 1989-2007; thereby leaving out the financial crisis, the results support a financial structure which is market-based. They explain that due to their highly leveraged nature, banks overextend and misallocate credit in financial upturns and ration credit in financial downturns. Markets are therefore favored, since they do not face capital requirements and leverage ratios. Their results suggest that market-based financial structures provide credit more efficiently. Following Pagano et al. (2014): “Based on these results, if Germany’s financial structure had followed that of the US over the past 20 years, the level of Germany’s GDP would now be approximately 2% higher”.
The same result is found by Langfield and Pagano (2015), concluding that bank-based financial structures lower economic growth. To also test financial structure’s effect on economic growth during a financial crisis, crisis dummies are included. Different from other studies however, two types of crises are classified; one being the housing market crisis, the
15 other being the stock market crisis. Although no effect is found during a stock market crisis, they conclude that in countries with bank-based financial structures, the housing market crisis exerts a more negative effect on economic growth than in countries with market-based financial structures. Following Langfield and Pagano (2015): “Consequently, when house prices drop, banks are constrained in their ability to provide new funding to profitable projects”. Additionally, they test whether bank-based countries increase systemic risk. The results show that a more bank-based financial structure increases systemic risk both during a housing market crisis as a stock market crisis. No significant results are found however, when testing the effect outside a housing market crisis.
C. CONCLUSIONS FROM THE LITERATURE
With the exception of Pagano et al. (2014), the existing literature shows that in normal situations, with business cycles but without a financial crisis, neither bank-based financial structures nor market-based financial structures appear superior in terms of the real economy. However, once these business cycles are accompanied with a crisis affecting bank balance sheets, market-based financial structures outperform bank-based financial structures. This is consistent with the second transmission channel from section II, where changes in a bank’s assets can lower a bank’s supply of credit; thereby transferring the crisis’s negative shocks to the real economy via banks’ balance sheets. Therefore, since the bank balance sheet transmission channel does not relate to markets, the empirical literature has given coherent evidence in favor of market-based structures during a financial crisis; a crisis severely affecting the value of banks’ assets.
Since the literature shows that a financial crisis negatively influences banks’ effect on the real economy, instability in the financial system has presented itself to be an important element regarding financial structure’s effect on economic performance. More specifically, the literature shows that instability in the financial system hampers banks’ performance compared to markets. A cause of this instability is systemic risk. Hence, the next section briefly discusses financial system’s exposure to systemic risk.
V.
SYSTEMIC RISK AND THE FINANCIAL SYSTEM
Following Bullard et al. (2009): “Systemic risk is the risk that a triggering event, such as the failure of a large financial firm, will seriously impair financial markets and harm the broader economy”. Several reasons explain why the financial system is vulnerable to systemic risk. The first reason draws on large commercial banks trading with each other through many
16 markets and systems; leading these institutions to be highly interconnected. Due to settlement risk; where one party may default after the other party has delivered its financial service, this interconnectedness can lead to large losses for financial institutions. More specifically, losses for one bank may cause losses for another. Subsequently, these losses can lead to a sharp decline in the supply of credit by a collective of banks; thereby hampering the real economy.
A second reason is that financial institutions and banks are highly leveraged. When times are good – that is, when markets are rising – highly leveraged firms can extract high returns on their equity. However, when times are bad – that is, when assets’ value is decreasing – highly leveraged firms are suddenly required to raise capital in order not to fail. Since the financial system is highly leveraged by itself, declining markets combined with a serious degree of systemic risk, can lead to a collapse of the financial system. Ultimately, this could trigger a financial crisis.
A third reason explaining financial system’s vulnerability to systemic risk is that in general, financial institutions face an asset-liability maturity mismatch. Long-term illiquid assets are financed with short-term debt, making these institutions extra vulnerable to interest rate and liquidity shocks. For example, if lenders suddenly refuse to purchase commercial paper from securities firms, these firms or institutions could be forced into bankruptcy. These bankruptcies can downsize the creditworthiness of the market’s borrower so that other institutions will also face additional funding costs.
Section II has shown the channels through which shocks move from the financial system to the real economy. Accordingly, systemic risk’s influence on the financial system highlights the important causing link between systemic risk and the real economy. A substantial amount of systemic risk proves to cause sufficient instability in the financial system, eventually holding back economic performance. In the words of Bullard et al. (2009): “the economy could benefit from reforms that reduce systemic risks, such as the creation of an improved regime for resolving failures of large financial firms”. Part of this regime should require higher capital buffers, especially for systemically relevant financial institutions.
VI.
METHODOLOGY
The methodology is based on a benchmark statistical model drawing on the relation between financial structure and systemic risk:
17 To account for the financial structure of country “i” at time “t”, 𝐵𝐴𝑁𝐾𝑖,𝑡 indicates the degree
of banks in a financial structure, and 𝑀𝐴𝑅𝐾𝐸𝑇𝑖,𝑡 indicates the degree of markets in a financial
structure. The indicator used for 𝐵𝐴𝑁𝐾𝑖,𝑡 is defined as the ratio of bank credit to GDP. For
𝑀𝐴𝑅𝐾𝐸𝑇𝑖,𝑡, this thesis distinguishes between two separate indicators. The first is defined as the
logarithm of the ratio of stock market capitalization to GDP. The second is defined as the logarithm of the turnover ratio. The second ratio represents total shares’ value traded divided by average market capitalization. Subsequently the higher 𝐵𝐴𝑁𝐾𝑖,𝑡, the more a financial
system is bank-dependent; the higher 𝑀𝐴𝑅𝐾𝐸𝑇𝑖,𝑡, the more a financial system is
market-dependent.
To check whether 𝐵𝐴𝑁𝐾𝑖,𝑡 and 𝑀𝐴𝑅𝐾𝐸𝑇𝑖,𝑡 are strongly related, correlations have been
estimated between bank credit to GDP and the logarithm of the stock market capitalization to GDP, and bank credit to GDP and the logarithm of the turnover ratio. The correlation of the former equals 0.0949. The correlation of the latter equals -0.1667. The two independent variables therefore seem to create no multi-collinearity issue on the right hand side of the equation. Furthermore, the appendix presents time-plots for bank credit to GDP, stock market capitalization to GDP, and the turnover ratio to GDP on al 20 countries. The time-plots show that there is sufficient time variation in the three indicator variables so that 𝛼1and𝛼2can be
identified in fixed effects panel regressions.
The indicators for 𝑀𝐴𝑅𝐾𝐸𝑇𝑖,𝑡 and for 𝐵𝐴𝑁𝐾𝑖,𝑡 follow from earlier studies [Gambacorta et
al. (2014), Beck and Levine (2004), and Levine (2002)]. It should be noted that the indicators for 𝑀𝐴𝑅𝐾𝐸𝑇𝑖,𝑡 do not reflect any activity in the bond market. Nonetheless, this thesis’ analysis
builds on the existing empirical literature with data selection open for equal comparison and discussion. The degree of markets in a financial structure is therefore only indicated by stock market activity. Additionally, Gambacorta et al. (2014) conduct a robustness check in which the degree of market activity also consists of bond market capitalization. They find that when they include bond market capitalization, their results are very similar.
The effects of the financial structure’s indicators are tested on the dependent variable
𝑆𝑅𝐼𝑆𝐾𝑖,𝑡; a measure for the quantity of a country’s systemic risk per unit of asset. The
calculation of systemic risk follows the work from Acharya et al. (2010) and measures the euro-amount of equity capital that a financial institution needs to raise in case of a 40% broad market index decline during a 6 month time period. More specifically, financial institutions must meet a particular capital requirement to stay solvent. When the broad market index falls, the capital requirement might no longer be met, since a financial institution’s equity has
18 declined in value. Therefore, the systemic risk value presented by 𝑆𝑅𝐼𝑆𝐾𝑖,𝑡 equals the amount
that a financial institution falls short to meet the capital requirement after the 40% broad market index decline. For this study, the European capital requirement is set at 5.5%; whereas the American capital requirement is set at 8%. The reason for this difference is due to differences in accounting principles between the institutions from which the microdata is obtained: European institutions follow the International Financial Reporting Standards; American institutions follow the Generally Accepted Accounting Principles. If the capital requirement for European institutions would be set to 8%, these institutions would be overcapitalized compared to American firms; therefore causing a bias to the United States.
The euro-amount of equity capital found after the 40% broad market index decline is then divided by the financial institution’s euro-amount of assets as to take away any size effects. This consequently enables the different institutions’ systemic risk values to be compared with each other. Furthermore, following Acharya et al. (2010), negative 𝑆𝑅𝐼𝑆𝐾𝑖,𝑡 values (equaling
negative equity shortfalls) are set to zero since these values do not add to systemic risk. To control for 𝑀𝐴𝑅𝐾𝐸𝑇𝑖,𝑡’s effect during a banking crisis, 𝐶𝑅𝐼𝑆𝐼𝑆𝑖,𝑡 represents a dummy
variable that contains a value of 1 in case of a banking crisis and 0 in case of no banking crisis. 𝛼3presents the separate effect of a banking crisis on systemic risk whereas 𝛼4presents
the effect of market activity during a banking crisis on systemic risk.
Following Langfield and Pagano (2015), two control variables are added. These variables are represented by 𝑋𝑖,𝑡−1 and are lagged by one year to mitigate the endogeneity problem. The
first control is a country’s bank size (assets of the three largest commercial banks as a share of total commercial bank assets). Since larger banks tend to be more interconnected with other banks, and since larger banks have less stable funding structure and conduct more market-based activities, bigger banks induce more systemic risk [Langfield et al. (2014), Laeven et al. (2014)]. The second control is the first control relative to GDP per capita. A country’s bank size relative to GDP per capita shows the importance of large individual banks with respect to the economy. The larger a bank’s relative size to GDP per capita, the larger its share of deposits and provision of loans to the real economy. Due to this greater importance, larger banks are more likely to receive public support; thereby creating more systemic risk via moral hazard [Afonso et al. (2014)].
Similarly to Langfield and Pagano (2015), this study controls for time-invariant effects that differ across countries by including country fixed effects represented by 𝜀𝑖. Differently
19 however, the analysis does not control for time fixed effects. All countries in the panel are hit at the same time by a banking crisis. Including both the crisis dummy and time fixed effects would therefore give parameter identification problems. Lastly, the error term is represented by 𝑢𝑖,𝑡.
To find whether financial structure’s effect differs after a certain point, two threshold models are constructed following Hansen (1999). To find an optimum threshold, the first model detects a break in the relationship between the ratio of bank credit to GDP and systemic risk. The second model detects a break in the relationship between the logarithm of the ratio of stock market capitalization to GDP and systemic risk.
𝑆𝑅𝐼𝑆𝐾𝑖,𝑡= {𝛼𝛼01+ 𝛼11𝐵𝐴𝑁𝐾𝑖,𝑡+ 𝛼21𝑀𝐴𝑅𝐾𝐸𝑇𝑖,𝑡+ 𝛽𝑋𝑖,𝑡+ 𝑢𝑖,𝑡+ 𝜀𝑖, 𝜆 ≤ 𝐵𝐴𝑁𝐾𝑖,𝑡
02+ 𝛼12𝐵𝐴𝑁𝐾𝑖,𝑡+ 𝛼22𝑀𝐴𝑅𝐾𝐸𝑇𝑖,𝑡+ 𝛽𝑋𝑖,𝑡+ 𝑢𝑖,𝑡+ 𝜀𝑖, 𝜆 > 𝐵𝐴𝑁𝐾𝑖,𝑡 (2) 𝑆𝑅𝐼𝑆𝐾𝑖,𝑡= {
𝛼01+ 𝛼11𝐵𝐴𝑁𝐾𝑖,𝑡+ 𝛼21𝑀𝐴𝑅𝐾𝐸𝑇𝑖,𝑡+ 𝛽𝑋𝑖,𝑡+ 𝑢𝑖,𝑡+ 𝜀𝑖, 𝜆 ≤ 𝑀𝐴𝑅𝐾𝐸𝑇𝑖,𝑡
𝛼02+ 𝛼12𝐵𝐴𝑁𝐾𝑖,𝑡+ 𝛼22𝑀𝐴𝑅𝐾𝐸𝑇𝑖,𝑡+ 𝛽𝑋𝑖,𝑡+ 𝑢𝑖,𝑡+ 𝜀𝑖, 𝜆 > 𝑀𝐴𝑅𝐾𝐸𝑇𝑖,𝑡 (3)
Here, the indicator variables 𝐵𝐴𝑁𝐾𝑖,𝑡 and 𝑀𝐴𝑅𝐾𝐸𝑇𝑖,𝑡 are either smaller or larger/equal to a
given threshold 𝜆. The slopes of 𝛼01, 𝛼11, 𝛼21, and𝛼02, 𝛼12, 𝛼22are estimated separately so to
show the effect below and above an estimated threshold. The threshold level is found by estimating regressions for a range of values of 𝐵𝐴𝑁𝐾𝑖,𝑡 and 𝑀𝐴𝑅𝐾𝐸𝑇𝑖,𝑡 with fitted values for
the threshold 𝜆. The threshold value present in the regression with the smallest sum of squared residuals is selected. It should be noted that models (2) and (3) are regressed without time-fixed effects.
Hansen’s (1999) F-test of no threshold effect is used to test the significance of the chosen threshold value 𝜆 for both indicator variables. The following two constraints are tested:
𝐻0 : 𝛼11= 𝛼12 (4)
𝐻0 : 𝛼21= 𝛼22 (5)
where under 𝐻0 the threshold value 𝜆 is not identified. To compare the goodness of fit of the
two models (a model where 𝜆 is identified and one where it is not) the likelihood ratio test of
𝐻0 is based on:
𝐹1= (𝑆0− 𝑆1(𝜆̂))/𝜎̂2 (6)
where 𝑆0equals the sum of squared errors under the null hypothesis of no threshold. Since 𝐹1
generally appears to depend upon moments of the sample, its critical values are not valid. Therefore, Hansen’s (1996) bootstrap procedure is used, since p-values constructed from a bootstrap are asymptotically valid. For the bootstrap procedure, the regressors and the
20 threshold variable (𝐵𝐴𝑁𝐾𝑖,𝑡 and 𝑀𝐴𝑅𝐾𝐸𝑇𝑖,𝑡 ) are treated as given and their values are held
fixed repeatedly in the bootstrap samples [Hansen (1996)]. The regression residuals are grouped by individual and are treated as the empirical distribution to be used for bootstrapping. A sample is drawn from the empirical distribution where these errors are used to create a bootstrap sample. Using this bootstrap sample, the model has been estimated with and without an existing threshold after which the bootstrap value of the likelihood ratio statistic 𝐹1is calculated. This procedure is consequently repeated a large number of times. The
percentage of draws for which the simulated 𝐹1 value exceeds the actual is calculated. The
resulting value is the bootstrap estimate of the asymptotic p-value. The null hypothesis [(5) & (6)] is rejected if the p-value is smaller than 0.05.
VII. DESCRIPTIVE DATA
The analysis relies on two different data sources. The first source provides data for the variable 𝑆𝑅𝐼𝑆𝐾𝑖,𝑡. Both the calculation and the provision of financial institutions’ assets are
done by New York University’s Volatility Laboratory. The second data source provides data for all independent variables. The ratio of bank credit to GDP, the stock market capitalization to GDP, and the stock market turnover ratio are obtained from the World Bank’s Global Financial Development Database. Additionally, the control variables bank size and the ratio of bank size to GDP per capita are also collected from the Global Financial Development Database. Since the 𝑆𝑅𝐼𝑆𝐾𝑖,𝑡 values start from 2000, the panel covers a timespan from 2000 to
2013 with yearly observations for all variables. Furthermore, to distinguish between different financial structures, this thesis focuses on the difference between Europe and the United States. The panel therefore consists of the following 20 countries: Austria, Belgium, Denmark, Finland, France, Germany, Greece, Ireland, Italy, Luxembourg, Netherlands, Norway, Poland, Portugal, Sweden, Switzerland, Turkey, the United Kingdom, and the United States.
Table 2 gives a brief summary on the statistics of the data. Both systemic risk and bank credit show with 280 observations to have no missing values. Other variables have missing values. Nonetheless, the missing values for the variable bank size only appear in the data of Greece.
21 Table 2: Descriptive statistics
VARIABLES
UNIT OF
MEASUREMENT OBS MEAN
STD
DEV MIN MAX
Dependent variable
Systemic risk % of assets 280 1.599 1.559 0.000 5.357 Financial structure
Bank credit % of GDP 280 111.932 50.702 11.554 220.000 Market capitalization % of GDP 257 69.274 51.038 0.340 265.112 Turnover ratio % of avg. market 260 96.297 60.033 0.140 394.317
capitalization
Control variables
Bank size % of bank assets 275 60.455 25.632 21.093 100.000 Bank size/GDP per
capita % of GDP 275 0.002 0.003 0.000 0.016
To show the difference in financial structures between Europe and the United States, figure 2 and 3 present time-plots on the ratio of bank credit to GDP and the ratio of stock market capitalization to GDP for Europe and the United States. The time-plots for Europe present the average of the 19 European countries. Figure 2 shows that Europe provides about twice as much bank credit to GDP compared to the United States. Figure 3 on the other hand shows that the total value of listed shares in a stock market as a percentage of GDP is about double compared to Europe. The data therefore indicates Europe’s financial structure to be more bank-based and the United States’ financial structure to be more market-based.
Figure 2: Bank credit to GDP for Europe and the United States
0 20 40 60 80 100 120 140 2000 2002 2004 2006 2008 2010 2012 Europe United States
22 Figure 3: Stock market capitalization to GDP for Europe and the United States
Subsequently, the question this thesis tries to answer is whether these financial structures affect a country’s systemic risk differently. Hence, figure 4 presents the time-plot on systemic risk as a percentage of financial institutions’ assets for the average of European countries and the United States. The data shows that systemic risk is almost always at a higher level for the United States than for Europe. Even more striking is that in 2008 the United States faces an increase in systemic risk much larger than Europe. Nonetheless, one should pay attention to the change in trends after 2008’s financial crisis. Although Europe remains around the same level of systemic risk from 2008 to 2013, the United States presents a clear decrease in its systemic risk values. Figure 5 therefore presents systemic risk’s decreasing trend lines for both Europe and the United States from 2008 to 2013.
Figure 4: Systemic risk for Europe and the United States 0 20 40 60 80 100 120 140 160 180 2000 2002 2004 2006 2008 2010 2012 Europe United States 0 1 2 3 4 5 6 2000 2002 2004 2006 2008 2010 2012 SYSTEMIC RISK Europe United States
23 Figure 5: Systemic risk trend lines for Europe and the United States after 2008
Figure 5 shows that in Europe, systemic risk is almost constant. The United States however, decrease their systemic risk level from almost 5 to 2 per unit of asset. Therefore, the data shows that the financial crisis has affected Europe and the United States differently with respect to systemic risk. The next section tests whether this difference is related to the financial structure.
VIII. RESULTS
This section presents the results from the fixed effects panel regression model for (1) and the results from the structural break models for (2) and (3). The discussion of these results is presented in the following section. For the fixed effects panel regression model both the logarithm of the ratio of stock market capitalization to GDP and the logarithm of the turnover ratio are used as indicators for the variable 𝑀𝐴𝑅𝐾𝐸𝑇𝑖,𝑡. For the structural break models only
the logarithm of the ratio of stock market capitalization to GDP is used as an indicator for
𝑀𝐴𝑅𝐾𝐸𝑇𝑖,𝑡.
A. FIXED EFFECTS PANEL REGRESSIONS MODEL (1)
Table 2 presents the panel regression estimations model for (1) using the logarithm of the stock market capitalization to GDP as an indicator for 𝑀𝐴𝑅𝐾𝐸𝑇𝑖,𝑡. The regressions include
country-fixed effects to control for unobserved effects that vary over time but not across countries. Additionally, all regressions test positive for serial correlation. The model therefore includes standard errors robust to clustering and heteroskedasticity. These tests are also given in table 2. 0 1 2 3 4 5 6 2008 2009 2010 2011 2012 2013 SYSTEMIC RISK
Trend Lines from 2008
Linear (Europe) Linear (United States)
24 Table 2: Fixed effects panel regression model
REGRESSORS I II III IV
Bank credit 0.0388*** 0.0193*** 0.0197***
(0.0039) (0.0042) (0.0044)
Market capitalization (log) -0.0134** -0.0046** -0.0045*** (0.0059) (0.0018) (0.0016)
Banking crisis 0.0132*** 0.0134***
(0.003) (0.003)
Market capitalization (log) -0.0028 -0.0030*
during banking crisis (0.0018) (0.0017)
Bank size (1-year lag) -0.0062
(0.0049)
Bank size/GDP per capita (1-year lag) 0.8040**
(0.3036)
constant -0.0165*** -0.0002 -0.0137*** -0.0125**
(0.0048) (0.0029) (0.004) (0.0048)
Time fixed effects No No No No
Country fixed effects Yes Yes Yes Yes
R-sqr 0.515 0.705 0.705 0.710
N 280 257 235 229
Serial correlation test 1 0.000 0.000 0.000 0.000
The dependent variable is systemic risk per unit of asset. Standard errors, robust to clustering and heteroskedasticity, are given in the parentheses. Significance levels: * p<0.1, ** p<0.05, *** p<0.01
1Reports p-values from the Wooldridge test for the null hypothesis of no first-order serial correlation.
The results indicate that increases in bank activity are associated with more systemic risk and increases in market activity are associated with less systemic risk. Although controlling for a banking crisis brings the coefficient of both independent variables closer to zero, the effect remains statistically different from zero at the 5% significance level for the logarithm of market capitalization and at the 1% significance level for bank credit. This thesis therefore finds evidence for financial structure’s effect on systemic risk outside a banking crisis. As expected however, the presence of a banking crisis alone increases systemic risk as well. This positive association is statistically different from zero at the 1% significance level. To look at financial structure’s effect on systemic risk during a banking crisis, the fourth regressor presents an interaction between the dummy for a banking crisis and stock market capitalization to GDP. It shows that even during a banking crisis, higher market activity is associated with lower systemic risk. This effect however, only becomes significant at the 10% level when controlling for bank size and bank size relative to GDP. It should be noted that including the control variables changes the independent variables’ coefficients and standard
25 errors arbitrarily little. Financial structure’s effect on systemic risk therefore proves not to be conditional on the size of banks in a country. Nonetheless, Stock market capitalization’s negative effect does become significant at the 1% level when controlling for bank size and bank size relative to GDP.
Similar to studies like Beck and Levine (2004) and Gambacorta et al. (2014), table 3 presents the panel regression estimations model using the logarithm of the stock turnover ratio as an indicator for the degree of market-based financing. This model includes country-fixed effects, and standard errors robust to clustering and heteroskedasticity for the same reasons as in table 2.
Table 3: Fixed effect panel regression model
REGRESSORS I II III IV
Bank credit 0.0388*** 0.0180*** 0.0183***
(0.0039) (0.0045) (0.0046)
Turnover ratio (log) -0.0031 0.0006 0.001
(0.0033) (0.0021) (0.0019)
Banking crisis 0.0162*** 0.0165***
(0.0025) (0.0024)
Turnover ratio (log) -0.0015* -0.0016*
during banking crisis (0.0008) (0.0008)
Bank size (1-year lag) -0.0058
(0.0042)
Bank size/GDP per capita (1-year lag) 0.7765*
(0.3805)
constant -0.0165*** 0.0038** -0.0096* -0.0083
(0.0048) (0.0017) (0.0049) (0.0052)
Time fixed effects No No No No
Country fixed effects Yes Yes Yes Yes
R-sqr 0.515 0.008 0.694 0.700
N 280 260 240 234
Serial correlation test 1 0.000 0.000 0.000 0.000
The dependent variable is systemic risk per unit of asset. Standard errors, robust to clustering and heteroskedasticity, are given in the parentheses. Significance levels: * p<0.1, ** p<0.05, *** p<0.01
1Reports p-values from the Wooldridge test for the null hypothesis of no first-order serial correlation.
Although bank credit still enters table 3 significantly at the 1% level, the stock turnover ratio has no significant effect on systemic risk. Interestingly, controlling for a banking crisis lowers turnover ratio’s standard errors and changes its coefficient’s sign. Turnover ratio’s effect remains insignificant however, thereby presenting zero effect on the dependent
26 variable. On the other hand, the stock turnover ratio shows to have a negative effect on systemic risk once interacted with the banking crisis. This effect is significant at the 10% level (but very close to the 5% level). Again, adding the control variables does arbitrarily little to the significance of bank credit and the banking crisis dummy.
Overall, table 3 again concludes that more bank activity is associated with more systemic risk. More market activity however, only decreases systemic risk during a banking crisis. Table 3 provides no association between market activity and systemic risk outside a banking crisis.
B. SRUCTURAL BREAK MODELS (2) AND (3)
Table 4 shows the results for the structural break models (2) and (3) using robust standard errors for heteroskedasticity. Since the detection of a structural break in the data requires a sample of countries that are somewhat similar, Hansen’s (1999) structural break model is regressed on a subsample 13 countries. These countries are: Austria, Belgium, Finland, France, Germany, Ireland, Italy, Netherlands, Portugal, Spain, Switzerland, Turkey, and the United States. Countries like Luxembourg have very different financial structures from the above subsample. Including these countries would therefore give inaccurate generalized results. Additionally, the timespan has been shortened from 2013 to 2012 since the logarithm of stock market capitalization presents data only until 2012.
Table 4 presents 1 threshold for model (2) and 1 threshold for model (3). These threshold values represent percentages given in decimals. In both models, the null hypothesis of no threshold effect is rejected using Hansen’s (1999) F-test. Subsequently, the null hypotheses (4) and (5) are rejected using Hansen’s (1996) bootstrap procedure with a bootstrap sample of 300. The threshold values are therefore statistically significant.
The slopes and constants are estimated separately for when bank credit and market capitalization are above and below/equal to its threshold value. To give a clear overview on the slopes and constants from table 4, table 5 presents the slopes’ independent variable and corresponding threshold assumption specified in the methodology section earlier.
27 Table 4: Structural break model financial structure to systemic risk
MODEL (2) MODEL (3)
Bank credit Market capitalization (log)
Threshold (λ) 1 1.2548 -0.1194 95% Confidence interval [1.1867;1.2930] [-0.3341;-.0990] SLOPES α11 0.0089 0.0193*** (0.0071) (0.0032) α21 -0.0028 -0.0033 (0.0024) (0.0023) α12 0.0373*** -0.0058* (0.0064) (0.0034) α22 -0.0084*** -0.0243*** (0.0028) (0.0048) F-Test 2 43.1173 31.7847 Bootstrap P-value 3 0.0000 0.0000 CONSTANTS α01 0.0243*** 0.0121*** (0.0060) (0.0053) α02 -0.0508*** 0.0303*** (0.0101) (0.0046) N<λ 115 123 N>λ 54 46
The dependent variable is systemic risk per unit of asset. Standard errors, robust to heteroskedasticity, are given in parentheses. Significance levels: * p<0.1, ** p<0.05, *** p<0.01.1 The threshold is a percentage presented in decimals: bank credit’s threshold equals 125.48%, market capitalization’s antilog equals 88.75%.2 Reports F-statistic from Hansen’s (1999) F-test for the null hypothesis of no threshold effect.3 Reports P-value from Hansen’s (1996) bootstrap procedure for the null hypotheses of (4) and (5).
Table 5: threshold slopes’ independent variable and corresponding threshold assumption
SLOPES INDEPENDENT VARIABLE THRESHOLD ASSUMPTION (2) THRESHOLD ASSUMPTION (3) α11 𝐵𝐴𝑁𝐾𝑖,𝑡 𝜆 ≤ 𝐵𝐴𝑁𝐾𝑖,𝑡 𝜆 ≤ 𝑀𝐴𝑅𝐾𝐸𝑇𝑖,𝑡 α21 𝑀𝐴𝑅𝐾𝐸𝑇𝑖,𝑡 𝜆 ≤ 𝐵𝐴𝑁𝐾𝑖,𝑡 𝜆 ≤ 𝑀𝐴𝑅𝐾𝐸𝑇𝑖,𝑡 α12 𝐵𝐴𝑁𝐾𝑖,𝑡 𝜆 > 𝐵𝐴𝑁𝐾𝑖,𝑡 𝜆 > 𝑀𝐴𝑅𝐾𝐸𝑇𝑖,𝑡 α22 𝑀𝐴𝑅𝐾𝐸𝑇𝑖,𝑡 𝜆 > 𝐵𝐴𝑁𝐾𝑖,𝑡 𝜆 > 𝑀𝐴𝑅𝐾𝐸𝑇𝑖,𝑡 α01 𝐶𝑂𝑁𝑆𝑇𝐴𝑁𝑇 𝜆 ≤ 𝐵𝐴𝑁𝐾𝑖,𝑡 𝜆 ≤ 𝑀𝐴𝑅𝐾𝐸𝑇𝑖,𝑡 α02 𝐶𝑂𝑁𝑆𝑇𝐴𝑁𝑇 𝜆 > 𝐵𝐴𝑁𝐾𝑖,𝑡 𝜆 > 𝑀𝐴𝑅𝐾𝐸𝑇𝑖,𝑡
28 The estimation of model (2) shows that below/equal to 125% of GDP, bank credit’s effect on systemic risk equals 0, since the coefficient is not statistically significant at the 10% level. Above 125% however, bank credit’s effect on systemic risk equals 0.0373. This is statistically significant at the 1% level. A structural break is therefore detected, providing evidence for bank credit’s positive effect on systemic risk only above 125% of bank credit to GDP. On the other hand, the estimation of model (3) proves that market capitalization’s negative effect on systemic risk is only present after a certain degree of market-based financing. Table 4 shows that below/equal to 89% (the antilog of -0.1194), market capitalization presents no statistically significant effect on systemic risk. Above 89% however, the effect equals -0.0243. This is statistically significant at the 1% level. The results therefore suggest that although some degree of bank-based financing will not harm economic performance via systemic risk, a more market-based type of financing is favored.
Market capitalization’s slopes are also estimated when looking at bank credit’s threshold. Below/equal to 125%, market capitalization shows no statistically significant effect on systemic risk. Above the threshold however, its effect on systemic risk is negative and statistically significant at the 1% level. The results therefore show that market capitalization’s effect only turns negative when a financial structure contains a certain degree of banks. However, since bank credit’s positive effect on systemic risk is greater than market capitalization’s negative effect, the results suggest bank credit to GDP to remain below/equal to 125%. Table 4 shows an even more interesting story when looking at model (3). Below/equal to market capitalization’s threshold, bank credit’s effect on systemic risk is positive and statistically significant at the 1% level. Above the threshold however, its effect turns negative. This effect shows to be statistically significant at the 10% level. Bank credit’s effect on systemic risk therefore turns negative once financial structure’s degree of market-based financing is large enough.
Overall, table 4 shows that an increase in the degree of market-based financing is always favored since its effect on systemic risk turns negative above its threshold. Additionally, the estimations show that even bank credit’s effect on systemic risk turns negative when market capitalization is above 89%. Furthermore, bank credit to GDP should not increase above 125% since bank credit’s positive effect on systemic risk outweighs market capitalization’s negative effect. Any value below 125% will not influence systemic risk however, since bank credit’s negative effect is statistically insignificant.
29
IX.
DISCUSSION
This section presents the discussion and interpretation on the previous section. The results strongly reject that financial structure is unimportant for systemic risk. By separating financial structure’s indicators into bank credit to GDP and stock market capitalization, this thesis finds evidence for financial structure’s influence on systemic risk. The fixed effects panel regression models show that a higher degree of banks in the financial structure leads to more systemic risk and a higher degree of markets in the financial structure leads to less systemic risk. The results additionally show that even during a banking crisis, more market activity is associated with less systemic risk.
An explanation for banks’ positive contribution to systemic risk stems from the greater pro-cyclicality of bank lending compared to market financing. This is due to their high leverage. When upturns lead to rising asset prices, the value of firm equity increases which allows banks to expand credit. This increases asset prices further leading banks to expand their credit even more [Bernanke and Gertler (1989)]. The opposite occurs during downturns when asset prices go down. Banks start to deleverage, thereby prompting further decreases in asset prices. Bank-based countries are therefore more sensitive to the presence of financial cycles, causing higher values of systemic risk. In contrast, markets match their financing to risk via better risk management tools. Additionally, markets’ public provision of information leads to more transparency, ultimately explaining markets’ negative effect on systemic risk.
In line with the above reasoning, Langfield and Pagano (2015) derive similar results. They find that bank-based countries face greater systemic risk by testing the effect of a bank-market ratio on systemic risk. One particular drawback from Langfield and Pagano (2015) however, is that they use one market ratio to indicate the financial structure. Although their bank-market ratio presents a positive effect on systemic risk, it does not indicate whether this positive effect stems from an increase in banks or a decrease in markets. Instead, by separating bank credit and stock market capitalization as indicators for financial structure, this thesis’ results show that financial structure’s positive contribution to systemic risk comes from both an increase in banks and a decrease in markets.
Nonetheless, when the degree of markets in a financial structure is indicated by the stock turnover ratio instead of the stock market capitalization, results change. Only during a banking crisis, the stock turnover ratio presents a negative effect on systemic risk. Outside a banking crisis, the indicator does not significantly affect systemic risk. The stock turnover
