The impact of financial frictions during banking crises

The impact of financial frictions during banking crises

Abstract

This thesis focuses at the impact of financial frictions during banking crises. Having empirical insights on these frictions will add value to knowledge on their quantitative importance. An assessment of these features aims at the need for a more sophisticated incorporation of the financial economy in macroeconomic forecasting models. These models do not address the impact of financial shocks and frictions simultaneously. A central focus is on the role of balance sheet mechanisms which are fundamental in financial frictions theory. This study finds that the capital asset ratio, funding from other financial intermediaries and loan loss provisioning moderate the impact of banking crises. An inclusion of these items can add value to forecasting models that assess the effect of frictions after the occurrence of a financial shock. Moreover, an amelioration of these components will improve the ability of the banking system to absorb shocks which is a main goal of the recent Basel III agreement.

Keywords: macroeconomic forecasting models, financial frictions, banking crises, credit channel, financial accelerator

Christiaan van der Weerd December, 2010

Table of content

Acknowledgments 2

1.Introduction 3

1.1 Background; mainstream equilibrium models and the inclusion of the financial sector 4 1.2 Channels between the financial and real economy: implications of the recognition

of financial frictions 6

1.3 Inside the banklending channel; the business of banking 8

2. Hypotheses 10

2.1 Capital asset ratio 11

2.2 Loan loss provisioning 12

2.3Interbank funding 13

2.4Non performing loans 14

3. Data 15

4. Methodology 17

4.1 Capital asset ratio 18

4.2 Loan loss provisioning 19

4.3 Interbank funding 20

4.4 Non-performing loans 20

5. Results 21

5.1 The effect of non-performing loans (NPL), loan loss provisions (LLP), interbank

funding (IBF) and capital asset ratio (CAR) on the growth of banklending 23

5.2 Estimation procedure 25

5.3 Moderating effect of the capital asset ratio during banking crises 25 5.4 Moderating effect of interbank funding during banking crises 28 5.5 Moderating effect of non-performing loans during banking crises 29 5.6 Moderating effect of loan loss provisions during banking crises 31

5.7 Robustness 34

6. Limitations 35

7. Conclusion 36

List of literature 39

Appendix A: Variables, definitions and sources 42

Appendix B: Systemic banking crises 43

Appendix C: Specification tests 44

Acknowledgments

First of all I would like to thank dr. Bezemer for good supervision of my thesis. I appreciated it very much to have room to develop my ideas whereas mr. Bezemer gave clear, good and to the point evaluations. Having a sincere interest for financial economics I was very glad to read a very interesting article from mr. Bezemer on macroeconomic forecasting models and the incorporation of the financial economy. This raised my enthusiasm and I am glad that I could write my thesis with his supervision. I enjoyed working on a theme I personally consider to be very relevant and I hope I can develop further within this field in my working life. Furthermore, I would like to thank prof. dr. Dietzenbacher for helpful comments on my research methodology and prof. dr. Koetter for co-assessment. Most important, because my life is, luckily, not solely in my hands I want to thank my heavenly Father for giving me every opportunity I got so far. It is my sincere belief that You kept an eye on me during my studies in good and in bad times.

1. Introduction

Nobel prize laureate in economics 2008 Paul Krugman interestingly asked himself how economists could get it so wrong in predicting the global economic crisis that started in 2007-8 (Krugman, 2009). Krugman points at the poor predictive performance of the economic field and the neglect of possibilities of major failures in the economy. In his view there has been a general disbelief among economists that the market mechanism in the economy would lead to outcomes like the global financial and economic crisis that started in 2007-8. Nowadays, several economists argue that the state of knowledge on the macroeconomy should be revised and refreshed. Leading economic paradigms have severe difficulties in explaining past events which led economists to conclude that this academic knowledge has got too much separated from reality. Especially the vulnerability of the real economy to shocks from the financial sector seems to have surprised the majority of policymakers and economists. Research and debate on this seems to be more justified as ever as societies, governments, people and businesses bear large economic and social costs in crises like the one that started in 2007-8. In the end the real economy should be fully able to benefit form a well-functioning financial economy that brings it on a steady growth path without major disruptions with its severe implications for the whole society.

This thesis explains why macroeconomic forecasting models incorporate the financial sector only very limited. With a very minimalistic incorporation of the financial sector, it is difficult to estimate the effect of a financial shock to the real economy. Theory on the role of frictions in financial markets point at the importance of balance sheet mechanisms. I will examine whether balance sheets of banking systems contain predictive power with respect to the impact of banking crises. This will shed light on events within the banklending channel and the financial accelerator within this channel.

The banklending channel, identified and examined by current FED chairman Bernanke and others in the eighties, focuses on the role banks and banklending plays in the process of monetary transmission instead of a sole look at interest rates. The strength of this mechanism is dependent on two features. First, it claims that an economy consists of bank dependent borrowers who have limited or no access to financial markets. The behavior of banks is of special importance to these agents. Second, the availability of loans to these agents is dependent on the overall health of the banking sector (Gilchrist and Zakrajsek, 2008; Hubbard, 2005). More recently, scholars pointed out that there is a financial accelerator mechanism within the banklending channel because of the existence of the interbank market (Van Den Heuvel, 2002). Therefore, this thesis builds on ideas on: (a) the health of the banking sector as one of the determinants of the strength of the banklending channel and (b) at the impact of the financial accelerator within this banklending channel.

occurrence of a banking crisis. Scholars point at the need for empirical insights on the role of financial frictions. Gilchrist and Zakrasjek (2008:1) postulate this by claiming that: “channels (between financial and real sectors) are well-understood from a theoretical perspective, assessing their quantitative implications remains a considerable challenge for macroeconomists”. Adding (Gilchrist and Zakrasjek, 2008:19) that they are “not aware of empirical work that seeks to estimate simultaneously the key parameters of the financial accelerator mechanism along with shocks to the financial sector”, I will try to give some empirical insights while being aware of the difficulties within this particular field.

The research question upon which this thesis is based is as follows: Does the strength of the balance sheet of the banking system contain any predictive power for a fruitful assessment of expected effects after a banking crisis? I will specifically focus on the intermediating effect of balance sheet components because of its prominence in theory on financial frictions (Bernanke and Gertler, 1995). If these components have an intermediating effect on the impact of a banking crisis, they can be considered as a predictor of the expected decline in banklending. Also, if balance sheet components I investigate moderate the effect of banking crises, they can be considered as an improvement of the ability to absorb financial shocks. This ability is the projected and desired effect of the recent Basel III accords (Blundell-Wignall and Atkinson, 2010). The thesis has the following structure. I will discuss the current macroeconomic forecasting models to show their shortcomings towards financial vulnerabilities and will evaluate why this is the case. After this, I will focus on linkages between the financial and real economy to discuss the channels through which financial vulnerabilities affect the real economy and the role of financial frictions in these linkages. Subsequently, I will do empirical research on transmission within the banklending channel and will try to find whether predictions derived from theory on financial frictions contain any predictive power.

1.1 Background; mainstream equilibrium models and the inclusion of the financial sector

economic variables and respond to deviations from a predetermined inflation target. This implies that when inflation outruns economic growth, the target interest rates will rise, to slow down money demand and output in the short run. Based on the idea that money is neutral in the long run (Hubbard, 2005), the dominance of real sector variables is a plausible assumption. However, the complexity and the functioning of financial markets, and its impact on short run output, are not reflected with it. This seems to be in line with the idea of Krugman (2009) who claims that economist were ‘mistaking beauty for truth’. The ‘beauty’ of a perfect financial system that is fully in service of the real economy is most of the time not the ‘true’ world we live in.

Scholars claim, therefore, that financial features should be modeled in a more accurate way (Den Haan, 2009; Tovar, 2008; Bezemer, 2009). Forecasting models underestimate the impact of financial disruptions because of theoretical assumptions such as the efficient market hypothesis (De Grauwe, 2008; Krugman, 2009). Several scholars point out that this hypothesis is an unrealistic assumption for an assessment of financial markets (Buiter, 2009; Krugman, 2009; Shiller, 2008). This hypothesis applies rational expectations to the pricing of assets (Malkiel, 2003). The theory predicts that when traders and investors use all available information in forming expectations of future rates of return, the equilibrium price of the assets equals the market’s optimal forecast of fundamental value (Hubbard, 2005). This prediction, in turn, provides market participants guidance in their financial and economic decisions. According to the classic work of Fama (1970: 416):” the evidence in support of the efficient markets model is extensive and (somewhat uniquely in economics) contradictory evidence is sparse”. A main and fundamental critique on this theory is that finance theorists built their evidence on comparison based on relative asset prices (Krugman, 2009). This leaves open the possibility for a collective misjudgment of many market participants which is never reflected in relative prices.

If the market is not fully in line with fundamental values there might be over- or undervaluation. In that case bubbles will exist in financial markets. Banks who have marketable securities on their balance sheet will have to impair these assets on a mark to market basis, unless these assets are held to maturity. However, a considerable amount of assets are not held to maturity but traded in secondary markets which is reflected by the increase of the importance of securitized asset markets. Accounting standards demand that these asset have to be valuated at fair value (Robinson et al, 2009). In such a case, busts can negatively affect the financial position of banks. More importantly, these busts can have negative effects on the stability of the financial system. Buiter (2009) adds, by referring to forecasting models, that mathematical beauty cannot reflect the idea that the market can collectively misjudge fundamental values. Prices of assets do depend to a large extent on undetermined and uncertain future prospects. However, as Buiter (2009) argues, macroeconomic forecasting models assume that the influence of prospects in a distant future on valuation of assets is mathematically zero.

which complicates a good way of describing the dynamics of real-world data (Tovar, 2008; Bezemer, 2009; Buiter, 2009). Complexities and deviations from efficiency in financial markets cannot be grasped by a sole focus at interest rates. Therefore I will focus on theories that claim that frictions in financial markets play a vital role additionally to the role of interest rates. These theories are based on Bernanke (1983), Bernanke and Gertler (1989; 1995) and Van Den Heuvel (2002) who developed the so called credit channel and financial accelerator within the transmission of monetary policy.

1.2 Channels between the financial and real economy: implications of the recognition of financial frictions

After providing a general treatment on financial markets and their inclusion in macroeconomic forecasting models, I will now theorize at the channels that relate the financial and real sector. Gilchrist and Zakrasjek (2008) indicate three main channels as having an important influence in the transmission of financial market disruptions to the real economy. Firstly, a pullback in spending due to reductions in wealth. The balance sheets of households are important for their consumption decisions. Households can get into precautionary saving to overcome an expected deterioration of economic conditions (Gilchrist and Zakrasjek, 2008). Especially when there is a fall in market values in real estate markets the effect of household wealth is important (Case et al, 2005). Secondly, Gilchrist and Zakrasjek (2008) point at balance sheet mechanisms that lead to a widening of credit spreads which decrease the ability to obtain credit. Adverse financial events that decrease net worth of businesses aggravate the impact of financial frictions such as moral hazard and adverse selection. This increases monitoring costs, risk premia and, thereby, credit spreads. Thirdly, Gilchrist and Zakrasjek (2008) point at a direct effect for financial institutions on the ability to get access to intermediate credit. Here they point at the financial accelerator within the banklending channel which will be discussed in detail in this paragraph.

With the assumption of complete markets, without financial frictions, the financial system was underestimated as having an important impact on the real economy. The Modigliani-Miller paradigm predicted that balance sheets of firms and households do not have an influence on their spending decisions. In this theorem households base their spending decisions solely on permanent income whereas firms build on discounted present values of investment projects (Gilchrist and Zakrajsek, 2008). It was with the recognition of ideas on asymmetric information, incentives and principal-agent problems in financial markets that disruptions in these markets, and its effects on the real economy, could be explained better than with the traditional ‘cost of capital’ channel of monetary transmission (Bernanke and Blinder, 1992).

interest-sensitive components of aggregate demand. Real interest rates adjust to changes in money supply whereas aggregate demand reacts to changes in real interest rates (Hubbard, 2005). Scholars developed two additional mechanisms that could influence monetary policy: the so-called balance sheet channel and bank lending channel which together form the ‘broad’ credit channel of monetary transmission.

The balance sheet channel is related to the creditworthiness of borrowers. In the first place we can think of the balance sheet of firms. Firms can finance themselves internally and externally whereas external financing is more expensive due to monitoring costs (Bernanke, 2007). When the creditworthiness of a firm, which is reflected by its balance sheet, improves, it can borrow externally more easily. External finance is related to information problems and information costs which should be paid by the firm as a premium. This is called the external finance premium which is defined as the difference between the cost to a borrower of raising funds externally and the opportunity cost of internal funds (Bernanke and Gertler, 1989). When there is a monetary contraction, the net worth of firms with floating interest rate loans falls as their debt burdens rise. Because of the lower net worth, monitoring costs increase because of increased problems with respect to moral hazard. Consequently, the external finance premium rises which increases the interest rate on loans. Subsequently, moral hazard problems become even more prominent which implicates that a initial monetary change has a multiplier effect. This is a phenomenon known as the ‘financial accelerator’ (Bernanke and Gertler, 1995). Because of this multiplier effect, the financial accelerator and the external finance premium is difficult to validate empirically. At least, we can say that the external finance premium is inversely related to the strength of balance sheet of firms. A lower capacity to pledge collateral discourages credit extension as the borrower is expected to be less aligned with the incentives of the lender (Bernanke, 2007). In the case of a monetary expansion and an interest rate fall on loans, the financial accelerator will have the opposite effect. An improvement in the cash, liquidity and net worth position of the firm lowers the external finance premium which amplifies the initial effect (Bernanke and Gertler, 1995). This phenomenon causes procyclicality of financial- and credit conditions.

availability of bank loans is an additional effect above the ‘traditional’ effect of an increase in spending and investment via the money channel after a interest rate fall.

Interestingly, the balance sheet channel can also be of relevance for banks themselves just as for firms (Van den Heuvel, 2002). There is a financial accelerator within the banklending channel which is known as the ‘narrow’ credit channel (Gilchrist and Zakarsjek, 2008). Banks also face an external finance premium as different sources of funding have different costs (Bernanke, 2007). Whereas in the past there were restrictions on the funding of banks, nowadays capital markets are more easily accessible to banks. Nondeposit funding has increasingly become a source of funding which is relatively more expensive than traditional deposit funding because of higher credit risks associated with this type of funding (Stein, 1998; Speight and Parkinson, 2003). The costs and availability of these nondeposit funds depend on the perceived creditworthiness of the specific institution (Van den Heuvel, 2002). Just as with other firms, a decrease in net worth and concerns about credit quality of bank’s assets will increase the external finance premium. As this premium increases, bank-dependent borrowers might face a decrease in loan availability and an increase in interest rates on available loans (Hubbard, 2005). Bank capital and the structure of the funding of the bank, becomes an important determinant of credit supply then (Van den Heuvel, 2002; Stein, 1998). Hypotheses are based on these building blocks: the existence of a banklending channel and the existence of a financial accelerator within this channel.

Scholars, are aware of these blindspots in forecasting models and this has lead to attempts to incorporate more financial features than only interest rates. The newest DSGE models, therefore, try to incorporate market frictions that lead to disequilibria. Christiano, Motto and Rostagno (2007), for example, estimate a DSGE model with financial shocks. Queijo van Heideken (2008), from the Sveriges Riksbank (Swedish central bank), and Christensen and Dib (2008) test a DSGE model with a financial accelerator mechanism. Also Graeve (2008) finds a better fit of macroeconomic data with incorporation of the financial accelerator in a DSGE model. Nevertheless, Gilchrist and Zakrasjek (2008:19) acknowledge that ‘to date, we are aware of no empirical work that seeks to estimate simultaneously the key parameters of the financial accelerator mechanism along with the shocks to the financial sector’. So, that means that we have some empirical knowledge on the financial accelerator and financial shocks, but we cannot model all these features simultaneously. Attempts to incorporate a financial accelerator and financial shocks are good developments, but they are still in an early phase. Generally speaking, there is progress to be made within this field.

1.3 Inside the banklending channel; the business of banking

asymmetry that increases external finance premia. Therefore, I will provide insight on bank balance sheets.

The most important balance sheet items are divided into three main items which together make up the accounting equation: asset = liabilities + equity capital (Robinson et al, 2009). Within the class of assets three items are of importance: reserves, loans and securities (Hubbard, 2005). Generally, banks can allocate their funds between these three general balance sheet items. Reserves, including cash, are held for regulatory purposes and for the provision of the withdrawal of deposits. Banks also have marketable securities in their portfolio which are relatively liquid and have low default risk. These securities provide banks with funds with good liquidity characteristics which implicates that in times of a withdrawal of funds these securities are an additional buffer above the reserves. Advantage of these securities is that they contain some return whereas reserves at the central bank are not interest-bearing. These marketable securities are therefore also known as secondary reserves. Lastly, banks hold loans to firms and households on their balance sheet. These loans have higher credit risks than marketable securities, are relatively illiquid and need to be monitored intensively (Hubbard, 2005).

There are mainly two important items on the liability side of banks: deposits and funds from the capital market (e.g. funds from other banks; interbank funding). Generally speaking, banks have more loan opportunities than they have deposits and they therefore finance advanced loans by a subsequent raising of funds from other banks. This type of funding is not covered by deposit insurance from the central bank. The remainder of the assets minus the liabilities is the bank’s net worth: equity capital brought in by shareholders plus retained earnings from the past. Net worth is important as it is a buffer against losses and impairments. If asset writedowns are large, it can result in a operating loss which will affect the equity capital. Moreover, the net worth of the institution is important for the perceived creditworthiness. Low net worth increases asymmetrical information problems, the external finance premium and so the costs of funds (Van Den Heuvel, 2002).

Banks, furthermore, have the incentive to maximize their return on equity (ROE) to maximize shareholder wealth and firm value. The return on equity is dependent on bank’s return on assets (ROA) and their leverage. The following function represent this idea (Hubbard, 2005):

Return on equity = Net after-tax profit / Bank equity capital =

(Net after tax profit / Bank assets) * (Bank assets / Bank equity capital ) = Return on equity = Return on assets * leverage

more interbank and deposit funds. These funds can be used for lending (except the reserves that should to be in line with regulatory requirements) or investment. With a positive interest spread this will, ceteris paribus, lead to an increase in profits given the amount of equity the bank has. However, this also decreases the net worth of the bank. Having the accounting equation in mind (A = L + E) an asset impairment will lower the equity capital of a particular institution (Shilller, 2008). If the amount of equity capital is not sufficient the bank will get into default because it has not sufficient resources to cover its impairments.

Because of the fact that banks have a systemic impact on the economy, the government has a regulatory authority which acts in the interest of the stability of the financial system. A stable financial system is served by well capitalized banks. If asset impairments are high to the extent that a bank cannot pay its depositors, and it does not have sufficient reserves at the central bank, the regulatory authority can claim that the bank is in bankruptcy. In general, the probability for bankruptcy will be lower the higher the level of equity capital, reserves and marketable securities (Hubbard, 2005). All this implicates that the leverage of a bank will be bounded by the regulatory constraints and their profit maximizing behavior.

In sum, the most important features of the business of banking that are relevant for my investigation are:

(a) the role of equity capital as determinant of balance sheet strength and perceived creditworthiness

(b) the structure of funding which determines the sensitivity to the financial accelerator

(c) the role of asset impairments as having an impact on net worth and perceived creditworthiness1

The following section will incorporate these ideas into hypotheses. The leverage of the bank will be represented by the capital asset ratio. The asset impairments will be evaluated by the level of non-performing loans. Interbank funding is used as a proxy for the funding of banks. Lastly, the level of loan loss provisions is related to the buffer or cushion a bank has to mitigate impairments.

2. Hypotheses

Because there is an abundance of evidence on the presence of a banklending channel and the financial accelerator within this channel (Kashyap et al; 1993; Gertler and Gilchrist, 1994; Calomiris and Himmelberg, 1995; 2000; Kishan and Opiela, 2006; Cetorelli and Goldberg, 2008) I will try to find insights on the effect of banking crises based on these concepts. Based on the prominence of the balance sheet I can test whether these concepts contain any predictive power. Specific ratios deducted from the balance sheet will represent the strength, opportunities and

1

constraints banks face with respect to their lending behaviour. This strength, and the expected influence of these balance sheet items, are deducted from studies that evaluate their general relatedness with banklending and not specifically their relatedness to banking crises. If the value for the specific balance sheet item, during the occurrence of the banking crisis, moderates the impact of a banking crisis, it is to a certain extent a predictor of the effect of a banking crisis. The stronger the moderating effect, the larger the predictive value of the balance sheet item. An evaluation, then, of the balance sheets throughout the banking system sheds light about the expected drop in banklending.

On the other hand, it can be the case that, statistically, this effect is small. A banking crisis as such is then related to low lending whereas different values of the specific balance sheet item does not influence the impact of the banking crisis. An investigation of the balance sheet of the banking system would, in that case, not provide any information on expected drops in banklending. To put it within the larger framework: a forecaster can either neglect the level of the balance sheet of the banking system during the crisis and focus at other determinants. Alternatively, it can put effort on finding out the expected effects within the banklending channel by investigating balance sheet items.

I first, shortly, hypothesize what the predicted effect of a banking crisis as such is before I get to the moderating effects. Banking crises are related to deteriorating asset values of banks. Necessary impairments will affect net income and can decrease the equity capital a bank has. If the equity capital decreases banks will get closer to the minimum amount of required capital. Very important here is the implication of the relative high leverage of banks because an asset impairment of 1 percent can deplete the equity capital (in percentage terms) by a multiple of this. The closer the equity capital gets to the minimum required capital, the less opportunities banks have to increase their asset base by attracting new funds. In sum, because banking crisis are related to deteriorating asset values and because deteriorating asset values are related to low lending, we can say that banking crises lead to low lending (Van den Heuvel, 2002; Zimmerman and Yuan,1999). Moreover, banks might shift their portfolio from loans to marketable securities to keep the balance sheet more liquid (Hubbard, 2005; Van den Heuvel, 2002). Based on this I derive my first hypothesis.

1.Banking crises will impact banklending negatively.

2.1 Capital asset ratio

Heuvel (2002) who focuses at the role of capital within the ‘narrow’ credit channel. When a bank’s capital is not much higher than the requirements of the regulathory authority, the cost of funds for banks will be higher because of higher monitoring costs other institutions incur (Kishan and Opiela, 2000; 2006). Empirical evidence confirms this idea: Gambacorta and Mistrulli (2004), for example, find evidence that the credit supply of well-capitalized Italian banks tend to be less pro-cyclical than less-capitalized banks. These well-capitalized banks can better absorb temporary financial difficulties and preserve long-term lending relationships (Gambacorta and Mistrulli, 2004: 438). Other scholars focus on the same relationship in the US and find similar results (Furfine, 2000; Hancock et al, 1995). Bank leverage is, furthermore, a very good predictor of bank failures. This is empirically confirmed by Estrella, Park and Peristiani (2000) in an analysis of US banks from 1988-1993. All these studies provide empirical evidence that bank capital ratios and lending are linked. Bernanke and Lown (1991) point at a second way: a high level of the capital asset ratio leads to above average growth in subsequent years. A bank, in that case, has excess capital which decreases its leverage. Subsequently, the bank can attract additional funds to increase its leverage which will, other things equal, lead to an increase in lending. A banking system with above average equity level can (a) better absorb the effect of a banking crisis and (b) will have lower costs of funds based on the ideas on the ‘narrow’ credit channel. Based on this I derive my second hypothesis:

2. The capital asset ratio of the banking system moderates the effect of a banking crisis positively.

2.2 Loan loss provisioning

A second item that improves the strength of the balance sheet is the provisioning for loans that perform badly. Deteriorating loan portfolios can be mitigated by means of reserved funds for perceived loan losses. Laeven and Majnoni (2003) observe that scholars have far more attention for capital regulation than for loan loss provisioning. Nevertheless, capital regulation and loan loss provisioning are both related to the stability of individual banks. Laeven and Majnoni (2003) find empirical evidence for an undesirable negative relationship between loan loss provision and GDP growth. This implicates that banks provision too less in times of economic growth. Subsequently, banks have to provision more during economic downturns which dampens loan growth and, thereby, GDP growth. Bikker and Metzemakers (2005) find the same result and confirm that loan loss provisioning in the US, Europe and Japan contributes to the pro-cyclicality of the business cycle albeit to a different extent. Meanwhile, non-G10 countries also under-provision which causes shortages of loan loss reserves during crises and a subsequent fall in credit supply (Cavallo and Majnoni, 2001).

banking crisis) provides banks with a cushion to overcome asset impairments. With this buffer banks can mitigate the impact of impairments as they can impair the funds that were specially arranged for this purpose. On the other hand, if a bank increases its provisioning during a crisis it can be a sign that it did not provision enough in years of good financial conditions. This implicates that it uses funds for provisioning instead of lending or investing it in marketable securities. Provisioning, then, is a substitute for other activities on the asset side of banks. Based on these ideas I derive my third hypothesis:

3. a. Loan loss provisioning during a crisis will moderate the effect on lending negatively. 3. b. Loan loss provisioning before a crisis will moderate the effect on lending positively.

2.3 Interbank funding

Now I turn to the ‘narrow’ credit channel. It is the mechanism Gilchrist and Zakrasjek (2008) mentioned as having an important influence in the transmission between the financial and the real sector: the direct effect of financial institutions to get access to intermediate credit. It is important for scholars to know the importance of the interbank market as a channel of contagion during crises (Gilchrist, Ortiz and Zakrajsek, 2009). Researchers at central banks use counterfactual simulations to estimate the danger of contagion via this channel (Upper, 2007). However, as Upper (2007:14) recognizes: “there is still a long way to go until they become an integral part of the toolbox of any authority responsible for financial stability”. Scenarios mostly studied till now by empirical research, is on the failure of single banks. It is precisely the effect of common shocks to the financial system that is of utmost importance. So far, little is know about this (Upper, 2007). Based on the ideas of Van den Heuvel (2002) and Bernanke (2007) on external premia in the interbank market, I expect that a banking sector which is relatively more dependent on non-deposit funding will face a higher external premium in times of banking crises. For the reason that an initial shock (e.g. a banking crisis) increases monitoring costs, the external finance premium will rise. This will aggravate asymmetrical information problems in the banking system and so the external finance premium increases further. This process continues and cost of funds increase continuously. Banks can shift their portfolio from loans to other institutions to other more safe asset classes, for example government bonds. This constrains lending and funding of other institutions. Subsequently, this will lead to an increase in the interest rate charged on loans which will, other things equal, decrease lending (Hubbard, 2005). Bernanke (2007) acknowledges that bank-dependent borrowers might face a decrease in loan availability and an increase in interest rates on available loans. The more interconnected a banking system is, the higher will be the impact of problems that appear in the interbank market. This leads to my fourth hypothesis:

2.4 Non-performing loans

Another indicator that determines the strength of the balance sheet of the banking system is the level of non-performing loans. Scholars see the level of non-performing loans as being important to the level of lending (Peek and Rosengren, 1997; Peek and Rosengren, 2000). In a study on the effect of a Japanese banking crisis on real activity in the U.S., scholars found a negative relationship between non-performing loans in the Japanese banking sector and lending in the US. This is caused by the presence of foreign branches of Japanese banks in the U.S. Another study finds a similar result: Sapienza (2004) finds that a higher level of non-performing loans leads to higher interest rates on loans. Higher levels of non-performing loans causes impairments that can decrease the equity capital. This increases the leverage of banks which leads to lower lending to let the leverage ratio be in line with capital requirements. The level of non-performing loans also determines the importance of the ‘narrow’ credit channel. The creditworthiness of the institution will deteriorate and other institutions can have doubts about the quality of assets. Doubts about quality of assets increase external premia and so the cost of funds (Van den Heuvel, 2002). Based on this I derive my fifth hypothesis:

5. The level of non-performing loans moderates the effect of banking crises negatively.

Graphical presentation of hypothesized relationships

3. Data

I will use a panel dataset with annual data from 1997-2009 on bank lending to the private sector and data on balance sheet indicators of the banking sector in 24 countries (see Appendix A). Data is on an aggregate country level to be able to include a sufficient number of banking crises. From this sample of 24 countries, 16 countries suffered from a banking crisis in one particular year (see Appendix B). This is ((16/297)*100)=5.8% of total observations on banklending. The panel dataset is unbalanced because there are some missing values in some time periods for some countries. This can be for any country in any given year which implicates that there is no structural gap within the dataset. Data on balance sheet indicators of the banking system is deducted from the IMF Financial Stability Reports of 2003, 2008 and 20102. I selected three reports as each report covers data for six subsequent years which implicates that these reports could cover the entire period between 1997-2009. More specifically, the Financial Soundness Indicators of this biannual report provide the data on the following indicators:

 Capital assets ratio: This figure is calculated in percentage terms by dividing banks’ capital by its total assets: ((capital/total assets)*100%). This ratio is presented for all countries in my dataset: 24. With some missing data in some time periods I have 292 observations for the capital asset ratio.

2

See appendix A for a complete overview of variables, definitions and sources.

Banklending channel

 Capital asset ratio

 Non performing

loans

 Loan loss provisions

 Interbank funding

Banklending

Banking system in

 Loan loss provisions to total loans: This figure is also presented in percentage terms: ((loan loss provisions/ total loans)*100%). This ratio is presented for 23 countries and has 258 observations.

 Non-performing loans to total loans: This figure is calculated by dividing the level of non-performing loans by total loans banks have on their balance sheet: ((non-non-performing loans/ total loans)*100%). This ratio is presented for all the countries (24) in my dataset and contains 292 observations.

With some exceptions, this data covers the balance sheets of foreign and domestic banks in a given country. Data on interbank funding comes from the OECD and is calculated as follows:

 Interbank funding to total liabilities. This figure is calculated by dividing the interbank funding by the level of total liabilities ((interbank funding/ total liabilities)*100%). This ratio is included for 15 countries in my dataset and has 175 observations.

Data on the dependent variable is on bank lending to the private sector. This comes from Thomson Financial Datastream:

 Thomson Financial Datastream provides quarterly or monthly data on bank lending from domestic and foreign banks in a particular country in local currencies (1997-2009). These figures are based on databases from national authorities, mostly central banks. To get annualized figures I calculated the yearly average to be able to compare the dependent variable with the independent variables which are based on annual data. There is data for 24 countries in the dataset which provides in total 274 observations.

4.Methodology

In this section I will focus on how scholars build models that explain banklending. Furthermore, I will present how I alter the models to answer my specific research question. The methodology I use is, amongst others, based on Bernanke and Lown (1991). In a paper on a credit crunch, and the 1990-1991 recession in the US, these scholars try to estimate the effect of a specific balance sheet component on bank lending. They estimate this with a model where loan growth (



L/L) is used as a proxy for bank lending. The specific model they tested can be seen in the next equation:

(



L/L)_199091 = - 0.182 + 2.733 (K/A)₁₉₈₉,

(0.067) (0.946) R2 = 0. 128

In this model (K/A) is the ratio of capital to bank assets. Rationale behind this model is that banks base their lending on a capital asset ratio target level. For example, a high (K/A) ratio in a certain year is positively related to loan growth in a subsequent year (



L/L). Moreover, loan growth is explained by absolute values of the independent variable (formulated as a ratio). A high (K/A) ratio implicates that banks have excess capital which they use to increase their asset base which would increase revenue. This, in turn, increases the denominator in the term (K/A) which brings this term in line with the predetermined target level. The target level, Bernanke and Lown argue, is not directly observable in this equation. Bernanke and Lown (1991), furthermore, argue that there is, apart of a correlation, a causal link between low capital asset ratios and low lending growth.

Another influential paper on the relationship between bank lending and capital asset ratios is from Peek and Rosengren (1995a). They estimate a similar model where lagged values of the capital asset ratio determine the change of the asset base in a certain year. Dependent on their research question, scholars add other terms within the model I provided above. We can think here, for example, of the influence of bank regulation on bank lending (Peek and Rosengren, 1995b). Others, Hancock and Wilcox (1994; 1997), look at the role of bank capital, the 1990-91 US credit crunch and its effect on the real estate market. Nevertheless, banklending is not only dependent on balance sheets mechanisms. Hancock et al (1995) investigate the effect of bank capital shocks and other important determinants on the holding of securities, loans and capital with help of the next equation (Hancock et al, 1995: 663):

In their estimated model they define different asset classes (Ai) of bank j at time t dependent on, again, the (t-m) lagged value of (A_{k ,j}) (where k is one of nine specific portfolio components of bank j) together with factors that approximate current and past expectations about economic conditions (

X

_n,_j,_tm). These are the most important explanatory variables whereas the others are control variables (



_i,j and T_t) for time trends such as, for example, disintermediation. Rationale behind this equation is that banks adjust the size and composition of their portfolio based on forecasted economic conditions (X). These can be approximated by the level of the federal funds rate and an index of leading macroeconomic indicators. For my own analysis, it is important to conclude that there are determinants of banklending not related to balance sheets. Therefore, I will control for these factors in my regression equation I will provide later.

So, with respect to my research question, I conclude from several important articles that bank lending is dependent on, amongst others, a target that banks define on the basis of the current state of their balance sheet. This idea can be best approximated with lagged values that constantly try to converge to the target level. I will build on this idea in my regression equations as well as I think it is a good proxy for bank behavior. Apart of the fact that other scholars use lagged values, the research question asks whether balance sheet items contain any predictive power for an expected drop in banklending. This implicates that I need to formulate differences in time periods between the independent and dependent variable because a forecast is normally meant for a future period.

4.1 Capital asset ratio

To estimate the intermediating effect of the capital asset ratio I will estimate an interaction model. Predictive power of balance sheet items during banking crisis can best be represented by interacting the relevant item with a banking crisis indicator variable. The hypotheses I test are conditional in nature. Conditional implicates in this case that the effect of a banking crisis is conditional on the value for the relevant balance sheet item. Important with interaction models is the inclusion of all constitutive terms (Brambor, 2006). Moreover, I need to control for country specific factors such as regulatory differences and openness. A fixed effects model with country fixed effects would, therefore, be appropriate3. These ideas lead to the following model:

(A)



BL/BL_it =



_1i +



₂C_it1+



₃CAR_it1+



₄(CAR_it1* C_it1) +



₅



BL/ BL_it1+



_it

BL_itrepresents bank lending to the private sector in country i at time t.



BL/BL_it, then, is the growth in bank lending in country i at time t. In all the models, the dependent variable is the

3_{In section 5 on the results the Hausman test also shows that a fixed effects model is appropriate. Moreover, I also test}

growth in banklending. The reason for this is that the absolute values of banklending in the dataset are very different in terms of currencies and absolute values. With help of the growth of bank lending, measurement differences are captured in the nominator and the denominator of the dependent variable



BL/BL_it= ((BL_it-BL_it1)/ BL_it1)*100). Therefore, these factors will cancel out. In this way, the comparability of the dependent variable improves and it can better be attributed to changes in independent variables.

C_it1, in equation (A-E), is the lagged value of a dummy for a systemic banking crisis in country i

in year t. This term is included because a systemic banking crisis might have its main effect on bank lending in the year after the crisis. Based on my hypotheses I expect that



₂<0,



₃>0 . The lag of the crisis dummy variable C_it1 is interacted with specific variables which can

moderate the effect on bank lending. The capital asset ratio was hypothesized to have a positive effect on banklending in a subsequent year. I expect that



₄>0 because high levels of the capital asset ratio can be used to dampen the effect of the banking crisis on banklending. Lastly, I only look at determinants within the banklending channel. Other determinants that influence banklending are monetary policies, regulatory constraints and macro-economic forecasts (Peek and Rosengren, 1995a). Moreover, the demand for loans also plays an important role in lending as well (Pazarbasioglu, 1996). To control for these factors I included a control variable which is the lagged value of the dependent variable. This term does control for before mentioned factors because these are present in the observations on banklending each year in each country. This implicates that when I include the observation on banklending from a previous year in the model, I will capture these factors within my model. Nevertheless, this control variable does not control for changes in these factors in time period t=-1 till period t=0.

4.2 Loan loss provisioning

In line with the previous model, the moderating effect of the level of loan loss provisions is estimated as follows:

(B)



BL/BL_it =



_1i +



₂C_it1+



₃ LLP_it1+



₄(C_it1* LLP_it1) +



₅



BL/ BL_it1+



_it

impact of provisioning before the crisis can be done with a minor adjustment to the previous model. In this model I expect that



₄>0:

(C)



BL/BL_it =



_1i +



₂C_it1+



₃ LLP_it1+



₄(C_it1* LLP_it2) +



₅



BL/ BL_it1+



_it

4.3 Interbank funding

Furthermore, the level of interbank funding during the crisis is used to determine its impact on bank lending:

(D)



BL/BL_it =



_1i +



₂C_it1+



₃ IBF_it1+



₄(C_it1 * IBF_it1) +



₅



BL/ BL_it1+



_it

In time of financial distress and illiquidity, interbank funding might be sensitive for contagion within the banking system (Speight and Parkinson, 2003; Gilchrist et al, 2008; Upper, 2007). Therefore it is expected that



₄<0. Theory on the effect of interbank funding on banklending without its relatedness to banking crises is harder to find. Nevertheless, Speight and Parkinson (2003) argue that banks attract non-deposit funds to increase their lending to, in the end, increase revenues. I expect



₃, therefore, to be positive (



₃>0). Furthermore, the effect of a banking crisis is still hypothesized as being negative (



₂<0).

4.4 Non-performing loans

Another interaction variable, this time for non-performing loans, is included in the next equation:

(E)



BL/BL_it =



_1i+



₂C_it1+



₃NPL_it1 +



₄( C_it1* NPL_it1) +



₅



BL/ BL_it1+



_it

I expect the level of non-performing loans to be related to the severity of the crisis. Just as in the investigation of Peek and Rosengren (2000), I expect the effect of NPL to be negative (



₃<0). A higher level of NPL during the crisis will negatively moderate its impact which means that I expect that



₄<0. The moderating effect, I am interested in, is analyzed by examination of the marginal effect of banking crises which is:

d (



BL/ BL_it)/d C_it1=



₂ +



₄(Z_it1)

banklending is now dependent on the balance sheet item in a previous time period. The difference in time periods represent the predictive idea of the balance sheet item.

The standard error of this marginal effect is normally not presented in statistical result tables. The standard error can be represented as follows (Brambor, 2006):



C

Y





= ((var (



₂)+ Z_it12var(



₄)+ 2 Z_it1cov (



₂



₄))

The standard error of the marginal effect of banking crises is not only dependent on the value of balance sheet item Z and the variances of the coefficients



₂ and



₄ but also on their covariance. Brambor (2006) acknowledges that this covariance is often negative which will lead to lower standard errors even if individual coefficients in the model were insignificant due to multicollinearity. The results of before mentioned models will be presented in the following section.

5. Results

First, the correlation table and the summary statistics will be evaluated. After that I will present scatterplots of banklending on the independent variables. Then I explain what tests I use to come up with appropriate models. In the appendices I examine all the specification tests I use and present their results. The following table (5.1) will present the correlations between the independent variables.

Table 5.1: Correlation table

BC1xNPL1 -0.0460 -0.0371 -0.0465 -0.0353 0.8751 0.2253 0.8251 0.8740 0.7884 1.0000 BC1xCAR1 0.0739 -0.0751 -0.0887 -0.0506 0.9416 0.1341 0.6731 0.6471 1.0000 BC1xLLP2 -0.0642 -0.0300 0.0042 -0.0235 0.7519 0.2521 0.9543 1.0000 BC1xLLP1 -0.0223 -0.0312 0.0183 -0.0045 0.7255 0.3995 1.0000 BC1xIBF1 0.0273 0.0086 0.0488 0.1091 0.1041 1.0000 L1. -0.0194 -0.0752 -0.0923 -0.0582 1.0000 BC L1. 0.0251 -0.1348 -0.0876 1.0000 IBF L1. 0.2215 0.9477 1.0000 LLP L1. 0.1618 1.0000 NPL L1. 1.0000 CAR

CAR NPL LLP IBF BC BC1xIBF1 BC1xLLP1 BC1xLLP2 BC1xCAR1 BC1xNPL1 L. L. L. L. L. (obs=117)

same year as the banking crisis. With correlation over time, within the LLP data, there will also be high correlations between the interacted terms. The next correlation table (5.2) shows whether there are high correlations over time.

Table 5.2: Correlation between independent variables in different time periods

L1. 0.1930 0.8681 -0.1273 0.8714 0.2585 0.9383 -0.1258 1.0000 LLP L1. -0.0102 -0.1199 -0.3215 -0.0903 0.0091 -0.1867 1.0000 IBF L1. 0.1142 0.9373 -0.1301 0.8871 0.2033 1.0000 NPL L1. 0.8924 0.1978 0.0067 0.2596 1.0000 CAR LLP 0.2149 0.9427 -0.0846 1.0000 IBF -0.0174 -0.1298 1.0000 NPL 0.1456 1.0000 CAR 1.0000

CAR NPL IBF LLP CAR NPL IBF LLP L. L. L. L. (obs=114)

There are high correlations between the same variables in different time periods. This can implicate that there are dynamic effects in the dataset pointing at autocorrelation over time. Therefore I use the Lagrange Multiplier test for each model I test. Subsequently, the summary statistics (table 5.3) of the main variables are evaluated:

Table 5.3: Summary statistics

dBanklending 274 10.0314 12.36893 -19.42943 97.8 BC1xNPL1 272 .4386029 2.999166 0 33.9 BC1xCAR1 271 .3232472 1.420672 0 10.3 BC1xLLP2 215 .123907 .6499393 0 5.9849 BC1xLLP1 237 .1787738 1.400074 0 19.7637 BC1xIBF1 158 .0954248 2.751715 -21.56039 17.47558 L1. BC 287 .0557491 .2298374 0 1 L1. IBF 157 7.990882 24.62925 -71.05075 170.5996 L1. LLP 237 2.802812 2.673783 .0784 19.7637 L1. NPL 272 4.239338 4.948379 .1 33.9 L1. 271 6.647601 2.463178 2.7 16.2 CAR

Variable Obs Mean Std. Dev. Min Max

5.1 The effect of non-performing loans (NPL), loan loss provisions (LLP), interbank funding (IBF) and capital asset ratio (CAR) on the growth of banklending

Before I get to the models different scatterplots and fitted regression lines will present the relationships between these variables and the growth in banklending. In these scatterplots, the growth in banklending is presented on the y-axis whereas the different variables are presented at the x-axis.

Graph 5.4: Scatterplot and regression line of dBanklending (Y) non-performing loans (X)

-5 0 0 5 0 1 0 0 % g ro w th i n b a n k le n d in g 0 10 20 30 40 non-performing loans

The figure shows that higher levels of non performing loans (x-axis) are related to lower levels of banklending growth (y-axis). This is in line with the idea that asset impairments can lower equity capital which constrains lending. The figure shows preliminary signs of heteroskedasticity with increasing variance at intermediate levels of the NPL variable.

Graph 5.5: Scatterplot and regression line of dBanklending (Y) on interbank funding (X)

(a) (b) -5 0 0 5 0 1 0 0 % g ro w th i n b a n k le n d in g 0 10 20 30 40 50 interbank funding -5 0 0 5 0 1 0 0 % g ro w th i n b a n k le n d in g -100 0 100 200

% growth in interbank funding

graph 5.5b to find out whether this relationship can be formulated in another way.4The effect of interbank funding growth is stronger and positively related to banklending. This represent the idea that banks have more loan opportunities than they have deposit funds (Hubbard, 2005). To be able to increase their lending they can make use of interbank funds to finance these loans (Speight and Parkinson, 2003).

Graph 5.6: Scatterplot and regression line of dBanklending (Y) on loan loss provisions (X)

-5 0 0 5 0 1 0 0 % g ro w th in b a n kl e n d in g 0 5 10 15 20

loan loss provisions

Banklending does not react strongly on differences in provisioning. Higher levels of provisioning are related to lower levels of growth. This observation is in line with the findings of Laeven and Majnoni (2003) who claim that provisioning offsets lending in the same time period. Variances are higher at intermediate levels which provides signs of heteroskedasticity.

Graph 5.7: Scatterplot and regression line of dBanklending (Y) on capital asset ratio (X)

-5 0 0 5 0 1 0 0 % g ro w th i n b a n k le n d in g 0 5 10 15

capital asset ratio

The capital asset ratio is positively related to banklending which is the predicted effect (Bernanke and Lown, 1991). Theorists claim that a high level of the capital asset ratio is a sign for banks to increase the asset base by increasing the loan portfolio to, in the end, increase profits and return on equity.

5.2 Estimation procedure

Panel data models can be tested with different kinds of estimation techniques. To find an appropriate model within my panel dataset, I will analyze my panel structure and examine the different models that are used within these data structures. My panel structure can be considered relatively long and wide with 26 countries and 13 year observations. The panel structure observes the cross-sectional units (i.e. countries) over time. These models come with different strands so it becomes important to choose the appropriate model. Generally, there are random effects models, fixed effects models and pooled regressions (Hill et al, 2008). The difference between them is in the extent of heterogeneity the model can explain. The choice for an appropriate model is dependent on the level of heterogeneity within my data and whether a specific estimation technique provides consistent, unbiased and efficient estimators (Hill et al, 2008). Furthermore, it is important to notice that panel data models have a time series component which might implicate that observations can correlate over time. Lastly, in some of the scatterplots there are preliminary sings of heteroskedasticity. Apart of these effects, there can be other misspecifications within the models such as omitted variable bias and endogeniety. All this implicates that each subsequent model will be tested for the presence of these effects with help of5:

 The Hausman test to decide between a random effects or a fixed effects model  The Breusch-Pagan test to test for the presence of heteroskedasticity

 The Lagrange Multiplier test to test for the presence of autocorrelation  The RESET(1) test to find signs of other misspecifications

5.3 Moderating effect of the capital asset ratio during banking crises

The first model estimates the intermediating effect of the capital asset ratio. The Breusch Pagan and the Lagrange Multiplier test pointed out that the model is heteroskedastic and not autocorrelated. The Hausman test shows that the random effects estimator is not consistent in large samples. There are, furthermore, no signs of misspecificantions based on the RESET(1) test. The following model is, therefore, a fixed effects model with robust standard errors:



BL/ BL_it = 7.51_i - 0.09CAR_it1 - 11.89C_it1+ 0.86 (C_it1* CAR_it1) +0.30



BL/ BL_it1+



_it (4.87) (0.67) (5.07)*** (0.69) (0.89)***

R2 = 0.4215 obs=248

The effect of banking crises is significant at the 1% level6as well as the control variable. The interacted term does not have a significant influence. The correlation table (5.1) shows a very high correlation between the interaction variable and the banking crisis variable (0.94). This will increase the standard errors and this will decrease the probability that the coefficients in the models are significant. It is important to notice that these standard errors, even if they are large, are the correct standard errors and not necessarily too large (Brambor, 2006). This implicates that my independent variables can be correlated to the extent that there is multicollinearity. Nevertheless, this would not necessarily impede my analysis as I am not interested in marginal effects of individual terms within my model but in the marginal effect of multiple parameters. The marginal effect of banking crises, moderated by the capital asset ratio, can be represented as follows:

d (



BL/ BL_it)/d C_it1= -11.89 + 0.86 CAR_it1

To find out whether this effect is significant, I calculate the standard error of this term. This standard error is not represented in the presentation of most statistical results. Brambor (2006) provides an equation for the standard error of the marginal effect which is in this particular case:



(d (



BL/ BL_it)/d C_it1)= (var (



₂)+CAR2var(



₄)+ 2CARcov (



₂



₄))

The standard error of the marginal effect is dependent on the value of CAR and therefore I calculate the standard error based on the mean value of CAR (6.65). This provides the following result:



d (



BL/ BL_it)/d C_it1= 1.89

Taking the marginal effect at the mean value of CAR (6.65) provides the following results for the t-test:

t =(-11.89 + (0.86*6.65))/ 1.89 = -3.27 Critical value = t(0.005, 242) = +-2.576

6

Standard errors are presented in parentheses: *** =significant at the 1% level

Based on this outcome the marginal effect of banking crises is significant at the 1% level at the mean value of CAR. Nevertheless, with different t-values over the range of CAR values, the 95% confidence interval will provide additional insights. Therefore, I present the line (-11.89 + 0.86 CAR_it1) in the following graph. The first lag of CAR is represented at the X-axis whereas the

effect of banking crises on banklending is represented at the Y-axis.

Graph 5.8: Effect of BC on banklending dependent on CAR (1) and 95% confidence interval.

-2 0 -1 0 0 1 0 2 0 E ff e c t o f B C 0 5 10 15 CAR(1)

The 95% confidence interval of the marginal effect becomes wide at high levels of the capital asset ratio. However, many observations of the capital asset ratio are close to the mean of CAR which can be seen in the low standard deviation of CAR (2.46). The following table shows how the values of CAR are distributed:

99% 15 16.2 Kurtosis 4.406961 95% 11.8 15.3 Skewness 1.165605 90% 9.8 15 Variance 6.067244 75% 8 14.7 Largest Std. Dev. 2.463178 50% 6.1 Mean 6.647601 25% 4.8 3.2 Sum of Wgt. 271 10% 4.2 3.1 Obs 271 5% 3.9 3 1% 3.1 2.7 Percentiles Smallest L.CAR

Within the interval between 5 and 10 the confidence interval is relatively close to the regression line. This relates to approximately 65% of all observations (90% - 25%).

banking crisis positively: an assessment of the capital asset ratio during the crisis will shed light on the expected drop in banklending.

5.4 Moderating effect of interbank funding during banking crises

The effect of banking crises and the moderating effect of the level of interbank funding are evaluated in the next model. The correlation (table 5.1) between the crisis variable and the interaction term (0.87) is high which provides signs of multicollinearity. The specification tests pointed out that the a fixed effects model is preferred. To overcome problems with respect to autocorrelation and heteroskedasticity the model is estimated with clustered standard errors. Reset(1) does not provide signs of misspecification of the model:



BL/ BL_it = 6.8_i - 0.11IBF_it1 + 0.11BC_it1- 0.30(BC_it1* IBF_it1) +0.48



BL/ BL_it1+ e_it (0.8)*** (0.18) (2.08) (0.11)** (0.78)***

R2 = 0.41 obs.=155

Within this model I am interested in the marginal effect banking crises moderated by dependency on interbank funding which can be represented as follows:

d (



BL/ BL_it)/d C_it1= 0.11 - 0.30 IBF_it1

Using the formula Brambor (2006) provides for the standard error of these terms I get the following t-value for this effect at the mean of IBF (19.9).

t=( 0.11 - (0.30*19.9))/ 2.32 = -2.52 critical value = t(0.025, 150) = +- 1.96

Graph 5.9: Effect of BC on banklending dependent on IBF (1) and 95% confidence interval -3 0 -2 0 -1 0 0 1 0 E ff e c t o f B C 0 10 20 30 40 50 IBF, L

The confidence interval becomes wider starting from values larger than 30 which is approximately 20 percent of all observations which can be seen in the summary statistics.

99% 48.04748 50.3576 Kurtosis 2.375389 95% 43.86863 48.04748 Skewness .1298868 90% 33.16987 47.47505 Variance 157.8482 75% 28.09057 46.62919Largest Std. Dev. 12.56377 50% 21.6365 Mean 19.90391 25% 8.756096 .0595421 Sum of Wgt. 173 10% 2.383717 .0450375 Obs 173 5% .3554396 .0212374 1% .0212374 .0173163 Percentiles Smallest IBF L.IBF

All this implicates that the effect of banking crises is negatively moderated by the growth in interbank funding. A 1% increase in interbank funding growth increases an initial effect by 0.3%. This outcome is in line with hypothesis 4: higher dependency on the interbank market affects banklending negatively during a banking crisis which can be caused by asymmetrical information problems and an increase in the external finance premium. This premium will be incorporated in interest rates on loans. Higher interest rates on loans will, ceteris paribus, lead to lower lending. An assessment of the dependency on the interbank market will provide insights on the expected drop in banklending.

5.5 Moderating effect of non-performing loans during banking crises

The subsequent model will estimate the moderating effect of non-performing loans on the effect of banking crises with inclusion of fixed effects:



BL/ BL_it = 9.47_i - 0.58NPL_it1 - 7.33BC_it1+ 0.35(BC_it1* NPL_it1) +0.26



BL/ BL_it1+ e_it (1.5)*** (0.25)** (2.61)*** (0.28) ` (0.09)*** R2 = 0.34

Within this model the marginal effect of banking crises can be represented as follows:

d (



BL/ BL_it)/d C_it1= -7.33 + 0.35 NPL_it1

Whereas the standard deviation (at the mean of NPL (4.24)) of this marginal effect is:



d (



BL/ BL_it)/d C_it1= 2.12

Which provides a t-value at the mean of NPL of:

t =(-7.33 + (0.35*4.24))/ 2.12 = - 2.76 critical value = t(0.005, 240) = -+ 2.58

This implicates that the marginal effect of banking crises is significant at the 1% level. The following graph represents the effect of banking crises with the 95% confidence interval.

Graph 5.10: Effect of NPL on the impact of a banking crisis and 95% confidence interval

-1 0 0 1 0 2 0 M a rg in a l e ff e ct o f B C 0 10 20 30 40 NPL(1)

The 95% confidence interval becomes wide for NPL values larger than 15. Approximately 5% of the observations fall within this range which is shown by the summary statistics.