P ROBABILITY C HARACTERISTICS OF B ANKING S YSTEMS AND B ANKING C RISIS

Hele tekst

(1)CHARACTERISTICS OF BANKING SYSTEMS AND BANKING CRISIS PROBABILITY ABSTRACT This paper investigates how characteristics of banking systems affect the likelihood that a systemic banking crisis will occur. Using data on 124 countries from 1980 to 2008, this paper finds that the profitability of a banking system, leverage ratio of banks, cost to income ratio, bank credit to deposit ratio, and banking supervision are quite robustly related to the probability of banking crises. Countries that have banking systems with high profitability are less likely to experience a systemic banking crisis. The same holds for countries with banking systems that have a comparatively lower leverage ratio, cost to income ratio, or bank credit to deposit ratio. Last but not least, countries with more and better banking supervision are also less prone to a systemic banking crisis.. JEL –classifications: G21, G01, G28, E58 Keywords: Banking crisis, Banking System, Policy. Master Thesis University of Groningen Student number: 1760157 October 2012 Author: Marilinda A. Croes Supervisor: Prof. dr. Jakob de Haan. Corresponding author: Marilinda A. Croes, Faculty of Economics and Business, University of Groningen, The Netherlands. E-mail: marilinda.croes@gmail.com . The author would like to thank Jakob de Haan and Robert Vermeulen for their helpful comments and suggestions..

(2) 1 I.. INTRODUCTION. The 2007/2008 financial crisis has once more demonstrated that current financial systems are not indestructible. Whether the financial system is categorized as being bank-based, market-based or a hybrid, they have all suffered from the crises, although there were significant differences. The starting year of the banking crisis in the United States was 2007 and one year later a banking crisis started in Germany. The U.S. is typically categorized as a market-based system, while Germany is typically categorized as a bank-based system, yet in both countries the government had to bail out or nationalize banks at some point because of liquidity problems that turned into solvency problems. This paper focuses on banking systems as a whole (per country) as opposed to studying individual banks. The word ‘systemic’ already indicates that a crisis of this nature is not idiosyncratic, thus if affects a great part of the banking system. Summer (2003, p.44) simply defines systemic risk (in the context of this paper) as ‘the problem of simultaneous failure of many banks’. The problem is that when many banks fail simultaneously, the spillover effects to the real economy can become colossal. For long researchers have studied the causes of banking crises by looking mostly into macroeconomic conditions (Klomp, 2010). While these conditions are important, nowadays we must not just focus on macroeconomic factors, but also study the internal banking system. More specifically, we must study which characteristics of banking systems matter for financial stability because these are aspects regulators could possibly steer. This paper is certainly not the first one to recognize that characteristics of the banking system matters for (systemic) banking issues. Demirgüç-Kunt and Detragiache (1998) added structural characteristics of the banking sector to their analyses. In addition, they added structural characteristics of the economic environment in general. These scholars discovered that banking crises tend to emerge when the macroeconomic environment is weak. More specifically, low GDP growth is significantly correlated with increased risk to the banking sector. Demirgüç-Kunt and Detragiache (1998) observed that financial liberalization substantially increased the opportunities for risk taking in many countries. Furthermore, they added that this financial liberalization should be combined with a well-designed and effective system of prudential regulation and supervision. If not, bank managers will be given a gateway to take in riskier projects (because they offer higher returns). This moral hazard issue could possibly lead to a banking crisis. Klomp (2010) confirms that banking crises are commenced by deregulatory measures that lead to overly rapid credit expansion. Also, Beck et al. (2006) studied a bank system characteristic, namely concentration of banks. The authors find that banking concentration is negatively associated with bank system fragility. We use the research paper of Beck et al. (2006) as our main reference point for our model specification..

(3) 2 The main research question of this paper is: Do characteristics of a banking system have an effect on the probability that a banking crisis may occur? In addition, this paper studies, if there is any, the interaction effect between the banking system characteristics. Boyd et al. (2004) concur that a monopolistic banking system will always result in a higher (lower) crisis probability given that the rate of inflation is below (above) some threshold. We do believe that the authors have a point in arguing that we need to consider interaction effects. Thus, the second question we will address is: If characteristics of a banking system have an effect on the probability that a banking crisis may occur, will it be a direct effect or an indirect effect? In sum, this paper attempts to shed some light on the issue of which characteristics a stable banking system ought to have by testing a logit model of macroeconomic conditions and banking system characteristics on the probability that a banking crisis occurs. This topic is especially relevant in light of the financial and banking crises that have occurred recently. Before continuing this research it is important to give a clear definition of a (systemic) banking crisis and review the literature relating to this topic. This will be done in section IIa. In section IIb the macroeconomic characteristics used in the existing literature will be briefly explained. Section IIc will report the relevant bank system characteristics that are important to consider for this paper and why. Furthermore, section IId will consider the theories on possible interaction effects. Section III presents the methodology: in section IIIa the econometric model and assumptions will be explained, while section IIIb will explain the data used. Section 4 will offer the results, and finally section 5 will conclude.. II.. LITERATURE REVIEW a. BANKING CRISIS. Laeven and Valencia (2008) define a systemic banking crisis as an event where a “country’s corporate and financial sectors experience a large number of defaults and financial institutions and corporations face great difficulties repaying contracts on time. This situation may be accompanied by depressed asset prices (such as equity and real estate prices) on the heels of run-ups before the crisis, sharp increases in real interest rates, and a slowdown or reversal in capital flows. In some cases, the crisis is triggered by depositor runs on banks, though in most cases it is a general realization that systemically important financial institutions are in distress.”(p.5).

(4) 3 Because our dataset on systemic banking crises comes from Laeven and Valencia (2008) and Laeven and Valencia (2010), this paper will primarily use their definition, which is less specific, and keep the definition provided by Demirgüç-Kunt and Detragiache (1998) in mind.1 To identify a systemic banking crisis Laeven and Valencia (2008, 2010) identify the starting year of systemic banking crises around the world where the starting year for this paper will be 1980 because of the availability of the other variables used for this research. Banking crisis is a dummy variable, which takes on the value of one if there is a systemic crisis in a country in a given year, and the value of zero otherwise. In addition, to closely follow the specification of Beck et al. (2006), we create a second dummy variable for banking crises to account for the duration of crises periods by giving each year of the duration of a crisis a value of one. b. MACROECONOMIC VARIABLES Bandt and Hartmann (2000) establish that the probability of banking crises occurring directly after lending boom periods is higher than during quiet periods and also that the size of the boom seemed to increase banking crises probabilities. The authors conclude that banking crises in history have been related to macroeconomic fluctuations and other aggregate or regional shocks. The literature summarized in table (3) in appendix A uses several other macroeconomic variables in their specification. Below we discuss these macroeconomic variables: why are they important, how do we measure them, and what are our expectations regarding the impact of these variables?2 We refer the reader to table (4) in appendix B for a more detailed explanation of how the variables are computed and the sources of the variables. GDP. “It is intuitively obvious that a decline in national income or wealth will lead to a reduction in the quality of bank portfolios” (Gavin and Hausmann, 1998 p.9). In the literature there are different measures of GDP that are used. The first one is real GDP growth. This measure is used because it captures booms and busts cycles in economies. Demirgüç-Kunt and Detragiache (1998) found that low GDP growth, significantly increases the likelihood of a systemic banking crises. Table (3) shows other studies that have drawn the. Demirgüç-Kunt and Detragiache (1998) establish that at least one of the following four conditions must for an episode of distress to be classified as a crisis: 1. The ratio of nonperforming assets to total assets in the banking system exceeded 10 percent. 2. The cost of rescue operation was at least 2 percent of GDP. 3. Banking sector problems resulted in a large scale nationalization of banks. 4. Extensive banks runs took place or emergency measures such as deposit freezes, prolonged bank holiday or generalized deposit guarantees were enacted by the government in response to the crisis. The 11th Geneva report on the world economy as well as Beck et al. (2006) use the above definition of Demirguc-Kunt and Detraiche (1998) to identify systemic banking crises. 2 Please note that some variables are only used for replication purposes. We only intend to use statistically significant variables in our basic regression model after which we add the bank system characteristics variables one by one to this basic model. 1.

(5) 4 same or different conclusion. Economies that grow strong are regarded as safer and therefore government bonds will have lower interest rates to be paid out. The opposite holds when growth is slow. Moreover, GDP growth seems to have an immediate effect on banking crisis, as results become insignificant when lags are added for this variable (Demirgüç-Kunt and Detragiache, 1998). A second measure used in the literature is Real GDP per capita. Beck et al. (2006) use this variable to control for the level of development. The authors conclude that countries in crisis grow slower. This is different from arguing that low GDP growth causes crises. In our replication model we use both GDP growth and GDP per capita. Real interest rate. Interest rates bear a risk because of a mismatch between the maturities of assets and liabilities. It is common for banks to finance their long-term assets with short-term loans, which needs to be refinanced until the asset matures. In normal times this mechanism works because banks borrow from each other on the interbank market. However, in times of high interest rates (due to risk), refinancing becomes expensive and banks which rely heavily on this refinancing mechanism could run into serious liquidity problems, which can turn into insolvency of the bank. Demirgüç-Kunt and Detragiache (1998) confirm that excessively high real interest rates significantly increase the likelihood that a systemic banking crisis may occur. In our preliminary analyses we observe that this variable has many outliers and that these outliers coincide with the periods before the start of banking crises. Inflation. High inflation has also been found to increase the probability of a systemic banking crisis by Demirgüç-Kunt and Detragiache (1998). Investors need to be compensated for the loss of value due to inflation by having the nominal interest rate increased. On the extreme side, hyperinflation makes investors uncertain about what the net present value of their investment is worth. Extreme high nominal interest rates to compensate for inflation will be unsustainable. With this knowledge, investors render it better to pull out their deposits/money. However, Beck et al. (2006) conclude that countries in crisis have higher inflation compared to countries that are not in crisis. When examining our data closely it is noticeable that inflation is not always very high in the pre-crisis period, sometimes inflation increases or even explodes during a crisis, which confirms the finding of Beck et al. (2006). Still there are remarkable differences between countries experiencing a systemic banking crisis. Be that as is it may, inflation definitely seems to be correlated with systemic banking crisis, however, whether inflation is an indicator that can predict a banking crisis is highly debatable. To deal with hyperinflation we decide to transform inflation by following the method of Dreher et al. (2007). We do not apply this method only for the replication regressions, as we strictly follow Beck et al. (2006). In table (4) appendix B it can be seen how we construct this transformation..

(6) 5 Depreciation. To measure the vulnerability of the banking system to sudden capital outflows triggered by foreign exchange risk, Beck et al. (2006) use the exchange rate depreciation rate and the ratio of M2 to foreign exchange reserves. The authors find that countries in crisis have higher depreciation rates compared to countries that are not in a crisis. Obviously, if a country is in crisis, there will be less demand for its currency and people who do hold the local currency might want to sell. In other words, this theory proposes that a banking crisis affects the depreciation rate and not the other way around. If a currency is depreciating and we notice that the reserves are depleting (because they are being used to defend the currency), investors will know that a central bank cannot keep this up forever. This combination can lead to bank runs. Yet, only Klomp (2010) and Kolulainen and Lukkarila (2003) find a significant and positive relationship between depreciation and banking crisis probability (see table (3)). Money and quasi money (M2) to reserves ratio (total). All studies use the ratio of M2 to foreign exchange reserves while we use the total reserve ratio. The difference between the two variables is in the denominator. The total reserves include foreign exchange reserves as well as holdings of monetary gold, special drawing rights and reserves at the IMF. We use the ratio of M2 to total reserves for two reasons. First, data availability and second, the fact that when there is a systemic banking crisis in a country it is not only foreign exchange reserves that are exhausted. Most studies find a significant positive relationship between M2 to foreign exchange reserves and banking crisis probability. Growth rate of credit. Lending booms (observed by credit growth) create financial vulnerability. Most studies use the growth rate of credit as opposed to using just the (domestic) credit to private sector. Beck et al. (2006) use credit growth lagged for two periods as a control variable they argue that high rates of credit expansion may finance an asset price bubble which may cause a crisis when it bursts. Before continuing it is important to take into account where the domestic credit that is provided to the private sector comes from. Banks obtain capital from the domestic as well as the international market and use this credit to finance the private sector. The latter is referred to as indirect cross-border financing of domestic credit (Advjiev et al. (2012). The more capital flows in, the more credit could be provided. International credit is attracted by comparatively higher domestic interest rates. In pre-crisis periods it is expected that the domestic credit to the private sector would be increasing. This can be so for a number of reasons. First of all, when a domestic economy is growing, company perspectives are temporarily improved in such a way that companies can take out more loans and banks are more willing to lend. As the economy is growing it is comparatively easier for banks to lend long to companies and finance these loans by borrowing short on the capital market (domestic or international). However, when the booming period comes to a halt (due to any reason leading to mistrust), refinancing becomes difficult for banks as liquidity starts to dry up..

(7) 6 Second, real interest rates attract foreign capital to local banks and financial institutions. This foreign capital is usually extended for a short-term period. Banks and other financial institutions use this capital to expand credit to the private sector. Another possibility is that this foreign capital would flow directly to the private sector (not having banks or financial institutions as intermediaries; however our focus remains solely on the indirect channel of international capital flows). Almost all studies in table 3 find a significant positive relationship between the growth rate of credit and banking crisis probability, which confirms the theories. We also expect to find positive relationship between the growth rate of credit and banking crisis probability. Government Budget Balance. This variable is defined as revenues (excluding grants) minus expenses divided by GDP. Demirguc-Kunt and Detraiche (1998) include this variable because they believe it is an indication of a government’s reluctance to restructure fragile banking systems. The government budget balance was computed by using data from the World Development Indicators (WDI). We took the “Revenue, excluding grants (local currency unit(LCU))” minus the “Expense (LCU)”, and divided by the “GDP (LCU)” times 100%. Revenue as defined by the WDI is cash receipts from taxes, social contributions, and other revenues such as fines, fees, rent, and income from property or sales. Expenses are defined as cash payments for operating activities of the government in providing goods and services. It includes compensation of employees (such as wages and salaries), interest and subsidies, grants, social benefits, and other expenses such as rent and dividends. We expect that the lower the balance (expenses>revenues), the more chance there could be of a banking crisis. As mentioned before, a government that spends more than it makes has to borrow and this puts upward pressure on the interest rates. Institutional Quality. The 11th Geneva report on the world economy concluded that a weak legal system that allows fraud to go unpunished, increases the probability of a banking crisis. In addition, Demirgüç-Kunt and Detragiache (1998) report a negative relationship between the quality of institutions and banking crisis probability. This variable was also used in Klomp (2010) but his results were inconsistent. In the pooled regression of Klomp (2010) there are no significant results but for developing countries the paper found negative significant results. This is a plausible outcome as developed countries already have sound institutions so the marginal effect in improvements would be really small. This leads us to believe that better institutional quality lowers the probability of a systemic banking crisis. Law and order is used as a measure of institutional quality. We have taken this data from the International Country Risk Guide Group and this is the proxy that was also used in Klomp (2010) and Demirguc-Kunt and Detraiche (1998)..

(8) 7 c.. BANKING SYSTEM CHARACTERISTICS. There are numerous banking system characteristics, but not all are quantifiable and, not all of them are relevant with respect to banking crisis probability. The existing literature helps to identify a few, yet this paper aims to test for more than what has been studied previously3. Also, the existing literature studies one or some banking system characteristics, but we will test all of them. We use lags of one year for all the variables to rule out reverse causality. Concentration. According to Allen and Gale (2004), a banking sector with many banks, which is less concentrated, is more susceptible to a banking crisis than a concentrated banking sector with a few banks. The reasoning behind this argument is that concentrated banking systems have more market power and so profits are comparatively high. If used properly, these profits could be used as a “buffer” during bad times. In addition, when there are only a few banks in a banking system, supervision becomes easier and more effective and thus reducing the risk of contagion, which should reduce the probability of a systemic crisis (Beck et al., 2006). A different theory suggests that concentration actually increases the likelihood of a systemic banking crisis for a number of reasons. First of all, when there are only a few banks present in the system, policy makers will be more concerned with individual bank failures. The banks then believe that they will be bailed out if necessary. This gives bank managers incentive to engage in riskier projects (moral hazard). Beck et al. (2006) find results that support the first line of argument. The authors conclude that “concentration”4 of the banking system is negatively associated with banking system fragility when controlling for regulations and institutions associated with greater banking system stability. This paper aims to use a banking concentration variable; not to proxy for competition5, but as a characteristic of a bank system. We measure concentration as the assets of three largest banks as a share of assets of all commercial banks6, see table (4) in appendix B. It is important to note that we chose to use a time-varying concentration variable because we do notice that the variable changes over time. In our study we discover that concentration is not as important as previous literature may have suggested. There are numerous reasons why we find different results. First of all, Beck et al. (2006) 3 The. reader should keep in mind that this research is using the paper of Beck et al. (2006) as the main reference point. Concentration is calculated as the fraction of assets held by the three largest banks in each country, averaged over the sample period. 5 A Better measure for banking system competition is the so-called Lerner-index. Bikker and Haaf (2002) use the Lernerindex and H-statistic which they derive from Panzar and Rosse (1987). The Lerner index is used to calculate market power of banks; the lower the market power, the more competition there is in the banking market. The Rosse-Panzar H-statistic will not be used in this paper because the measure relies on the assumption that markets are in the long-run equilibrium at each point in time when the data are observed (Goddard and Wilson 2009). This paper is studying crises and the state of crisis is not an equilibrium point, but rather a disequilibrium point. 6 A major difference between this paper and the study of Beck et al. (2006) is that we do not take the average of concentration for the years for which data is available and then extrapolate this for our entire sample period. 4.

(9) 8 extrapolate the average of concentration over their sample period. They claim to use this method because of the coverage issue, namely that over time more banks are added to the database and this may change the concentration value. Concentration indeed changes over time. Without extrapolating the variable Beck et al. (2006) would not have had enough observations to conduct their study. However, our sample period is larger and allows us to keep the time-varying concentration as it is. Economic Freedom. Restricting a bank in its activities (financial or non-financial) hinders its options to diversify its portfolio, which consequently makes a bank more sensitive to macroeconomic fluctuations. However, liberalization also has its caveats. Another way to check in how far banks are either free or restricted in their activities is to look at institutional factors. “Economic freedom” is a composite of ten indicators ranking institutions and policies in the areas of trade, government finances, government interventions, monetary policy, capital flows and foreign investment, banking and finance, wages and prices, property rights, regulation, and black market activity7 (Beck et al., 2006). High scores of this index imply that there are institutions and policies in place that foster competition and economic freedom. As mentioned before, this helps banks to diversify their portfolios as well as engage in risky activities. Beck et al. (2006) find that economic freedom reduces the probability of a (systemic) banking crisis. Net interest margin. The “Net interest margin” has previously been used in Barth et al. (2004) to measure market power. The reasoning is that as the degree of competition in a banking system decrease, the markup between the interest rates banks charge their customers and their own interest expenses becomes larger. We also include this variable because as was previously mentioned, concentration is not the best measure for competition and we consider the net interest margin as a better proxy for competition. The net interest margin is defined as the accounting value of a bank’s net interest revenue as a share of its interest bearing (total earning) assets. We expect that a higher margin could either lead to less systemic banking crisis provided that part of the extra mark-up is used as a buffer for bad times. Credit market regulation. The credit market regulation variable looks at regulation of credit. It partly constitutes of indices of ownership of banks, foreign bank competition, private sector credit and interest rate controls/negative real interest rates8. We take this as a proxy for financial regulation. When financial regulations were relaxed (after the efficient market paradigm became popular in the 1970’s and especially after the Glass-Steagall act in 1999 in the U.S.), universal banks were allowed to become fully involved in the financial markets (De Grauwe, 2008). Demirgüç-Kunt and Detragiache (1998) observed that financial liberalization substantially increased the opportunities for risk taking in many countries. With greater risk come potentially larger profits, but also greater losses. In any case, increased risk increases volatility of the banking system. To obtain a complete description of the variable and how it is computed we refer the reader to the data source: Economic Freedom of the World: 2011 annual report, www.freetheworld.com . 8 To obtain a complete description of the variable and how it is computed we refer the reader to the data source: Economic Freedom of the World: 2011 annual report, www.freetheworld.com . 7.

(10) 9 Regulation can prevent banks from engaging in other activities in the financial market such as insurance. The upside is that when a negative shock hits the insurance market, the banks will have minimum exposure to this sector and the bank’s balance sheet will remain intact. The downside is that bank has less scope to diversify its portfolio. Private ownership. La Porta et al. (2000) find that government ownership of banks is large and common all around the world and that it is exceptionally significant in countries with low levels of per capita income, underdeveloped financial systems, interventionist and inefficient governments, and poor protection of property rights. They also find that government ownership of banks is related to slower financial development and lower growth of per capita income9. Barth et al. (1999) find that a higher degree of government ownership of banks tends to be associated with higher fragility of financial systems. Goodhart (2004) makes an even stronger statement about this finding by stating that the presence of any non-profit-maximizing banking entities could make financial systems weaker. Private ownership is also in de index of “credit market regulation”. We take it out because we wish to test its separate effect. What this variable measures exactly is the percentage of bank deposits held by privately owned banks10. Countries with larger shares of privately held deposits get a higher rating. When the percentage of privately held deposits in a country accumulates to 95%-100%, a rating of 10 is given. Between 75%-95%, a rating of 8 is given. Private deposits between 40%-75% are assigned a rating of 4, 10%-40% a rating of 2, and 10% or less a rating of zero. We expect to find a negative relationship between private ownership and banking crisis probability. Central Bank Independence. A central bank plays an important role in banking systems, in some more than in others and this depends on its independence and power. Central Bank Independence (CBI) has not generally been used in studies relating to banking crisis probabilities, with the exception of Klomp (2010) who used it only in his pooled estimations and found no significant results. A reason for this could be that CBI is correlated with inflation and therefore these two variables should not be used in the same model specification. Klomp and De Haan (2009) concluded that CBI has a significant negative relationship with financial instability (caused by political independence). Therefore, we also expect to find a negative relationship between CBI and banking crisis probability. Deposit Insurance Scheme. Deposit insurance can work in two ways according to theory. First of all, it can increase bank stability by reducing the information effect11. Secondly, deposit insurance can decrease bank stability by inducing banks to engage in riskier activities.. In particular with lower growth of productivity rather than slower factor accumulation. obtain a complete description of the variable and how it is computed we refer the reader to the data source: Economic Freedom of the World: 2011 annual report, www.freetheworld.com . 11 The information effect entails that when one unknown bank has solvency problems, depositors start distrusting banks in general and pull out their capital because they are unable to point out the insolvent bank. 9. 10 To.

(11) 10 Beck et al. (2006) include an index of Moral Hazard caused by deposit insurance generosity because Demirgüç-Kunt and Detriache (2002) found that moral hazard is significantly related to financial fragility. Beck et al. (2006) find a positive significant relationship between banking crisis probability and their Moral Hazard Index12. Klomp (2010) only includes deposit insurance in his model and fails to find significant results and the same holds for Herrero and Rio (2003). Demirgüç-Kunt et al. (2005) created a database for deposit insurance in which they identify the years in which a deposit insurance scheme was enacted or revised. The years for which countries have no deposit insurance scheme in place are assigned these years a value of zero and when there is a deposit insurance scheme in place we assign a value of one. Now turning to the characteristics that have not yet been used in the context of banking crisis probability, but that this paper deem important to add to the model. Bank (liquid) reserves to assets ratio. This measure tells us how liquid banks in a country are in aggregate. During the (current) crisis liquidity has dried up in the international financial markets. As a result, even banks that were solvent became insolvent because they had to sell their assets at decreasing prices to attain liquidity. There is no doubt that liquidity is an issue. However, the question remains whether liquidity can serve to predict crisis. Profitability. Another characteristic of the banking system is its profitability. To proxy for this we will use “average return on assets” and “average return on equity” calculated as net income/total assets and net income/total equity respectively. If a bank system is profitable in terms of return on assets and/or return on equity, the bank has less incentive to engage in risky activities to generate higher returns. We expect higher profitability to be related to lower crisis probability. Bank Assets to GDP ratio. Iceland and Ireland have banking sector larger than their GDP. These are cases where banks or the banking sector are too big to save. A major difference between these two countries is that Ireland had the ECB as a lender of last resort while Iceland did not have a lender of last resort (Buiter and Sibert, 2008). The question is, does it matter if the banking sector is bigger than a country’s economy when we try to predict a systemic banking crisis? Or do other aspects or the banking system matter more? Furthermore, by also using the concentration measure mentioned previously, we hope to control for distributional differences in terms of bank sizes to some extent (seeing that we are using country aggregates).. We would have liked to include the Moral Hazard Index in our model. However, it was insufficiently clear how Beck et al. (2006) computed this variable and it was not included in their dataset on the World Bank website. We decided to use only the deposit insurance instead.. 12.

(12) 11 Leverage ratio. If a bank prefers issuing debt as a form of financing, it puts itself at risk. During normal times banks usually finance their long-term assets with short-term loans. As a proxy for the leverage ratio we use bank assets to bank deposit ratio. A second measure we use for the leverage ratio is the total financial market assets to total financial markets deposit ratio. The expectation is that the more leverage a bank has, the more fragile it is. It is natural to assume the same for the aggregate of banks. The more fragile banks are in general, the greater the probability could be that a systemic banking crisis will occur. Cost to income ratio. The cost to income ratio is an efficiency measure (Tripe, 1998). The lower this ratio is for a bank system, the more efficient the bank system is. The definition in our dataset for this ratio is the total costs as a share of total income of all commercial banks. We expect to find a positive sign for this variable. The higher the costs compared to income for a bank, the more fragile the bank will be as it is operating inefficiently, losing potential money. Investors will take notice and bring their funds to another bank or financial institutions, which can generate higher returns. Bank credit to bank deposit ratio. Before or during crisis an increase in this ratio does not necessarily reflect an increase in credit, but mostly reflects a decrease in deposits. This could be related to either lower interest rates, where lower interest rates could induce people/invertors to deposit less in banks or pull out their cash and deposit or invest it somewhere else with higher returns. Moreover, interest rates could just seem low due to increased perceived risk, in other words investors might not feel that the interest rate is high enough to cover the higher risk. The definition we use for the bank credit to bank deposit ratio is the ratio of claims on the private sector to deposit money banks (Beck and DemirgüçKunt, 2009). A more general definition is “the proportion of loans generated by banks from the deposits received”. It is also a measure of bank efficiency. It is unclear what kind of output should be expected. Do deposits diminish before a crisis or just during a crisis? Interest rate control. This variable measures to what extent interest rates are either controlled or liberalized. Our source for this variable explains it as follows. “In the most restrictive case the government specifies both lending and deposit rates by fiat, or equivalently, sets ceilings or floors tight enough to be binding in most circumstances. An intermediate regime allows interest rates to fluctuate within a band. Interest rates are considered fully liberalized when all ceilings, floors or bands are eliminated” (Abdul et al., 2008, p.5). The most restrictive case is appointed a value of zero, and the most liberal case a value of three. Controlled interest rates guarantees depositors and borrowers the minimum and maximum cost each party would bear when engaging in a transaction. It also matters whether interest rates are fixed in a contract over a course of time, or whether they are allowed to fluctuate. In the former case government protection is not needed. This variable is also taken upon in the credit market regulation index. Like private ownership, we wish to take it out to isolate its effect on banking crisis probability. The index for economic freedom on the other hand, is an overall.

(13) 12 index of Area’s one to five in the Economic Freedom of the World database. Area 5 encompasses credit market regulation, labor market and business regulation. Banking Supervision. As mentioned before Demirgüç-Kunt and Detragiache (1998) argue that if financial liberalization is a fact in a country, it should be combined with a well-designed and effective system of prudential regulation and supervision to reduce the severity of moral hazard by making it more intricate for bank managers to engage in riskier projects. The banking supervision is a variable between zero and three. Zero means that there is little to no supervision, 3 is when there is strong supervision. The data is taken from Abdul’s et al. (2008) financial reform dataset. d. INTERACTION EFFECTS Boyd et al. (2004) concur that a monopolistic banking system will always result in a higher crisis probability given that the rate of inflation is above some threshold, which is not specified. Schaek et al. (2009) find that competitive banking systems are less susceptible to experience a systemic crisis, however they do not account for the interaction effect of inflation, which could be crucial and could possibly change how economists view a monopolistic banking system versus a competitive banking system, the latter may not always be better. Whether inflation actually matters is of extreme importance for policy makers given that they know in which type of financial system they are operating. If the statement made by Boyd et al. (2004) turns out to be true, than depending on what the threshold is and assuming it is above 2% for the Euro Area, then the Eurosystem is following the right policy as to maintaining a low inflation rate. This line of thought has intrigued us to think of different interaction terms that have not yet been tested in the literature. After careful consideration we have decided to that it is interesting to test whether there is an interaction effect between bank concentration and bank size. While most studies focus on bank size as the size of an individual bank, we focus on bank size on a country level. When a country has a large banking system, and bank concentration is large, we would expect the problem of the “too big to fail” banks to be more pronounced. The size of the system encompasses a substantive part of GDP, which makes it more costly for government to bail out banks. Adding to this that there may only be a few banks in the system, it would seem that government will have to bail out the banks when they run into liquidity or solvency problems. A second interaction effect we would like to consider is that between bank concentration and banking supervision. The reasoning behind this is that we are intrigued by the suggestion of Beck et al. (2006) that the more banks there are in a banking system, the more tedious it becomes to closely monitor these banks by supervisory authorities..

(14) 13. III.. METHODOLOGY a. ECONOMETRIC SPECIFICATIONS. For the econometric specification this paper will firstly follow Beck et al. (2006)’s specification of a binary choice model. We will use the logit model, because the dependent variable only takes the value of one if a banking crisis has occurred and zero otherwise. Banking Crisisi , t = β 0 + β1 X i , t + ε i , t. (1.1). β 0 is the intercept coefficient. Χ is for now the vector for all the variables in the model. ε is the error term and i, t are for denoting country and time respectively. =

(15) . 1 with probability p 0 with probability 1-p. The log-likelihood function looks like this: Ln L =. ∑ ∑{P(i, t ) ln[ F (β ' X (i, t ))] + (1 − P(i, t )) ln[1 − F ( β ' X (i, t ))]}. (1.2). t =1,..., T t =1,..., n. A logit model’s coefficient magnitudes are not readily interpretable, however this is easily dealt with by calculating marginal effects. Marginal effect: (∂p ∂Χ j ) = Λ (Χ ' β )[1 − Λ ( Χ ' β )]β j. (1.3). Probability:. ( p = Pr[C = 1 | Χ ]) = Λ ( Χ' β ) =. e Χ 'β 1 + e Χ 'β. (1.4). Where Λ(.) is the logistic cumulative distribution function with. Λ(a ) = ea (1 + e a ) = 1 (1 + e− a ) (1.5) For all the bank system characteristics variables, Law and order, and Government budget balance we take the lag for one year to take care of reverse causality problems. So our simplified model would look like this: Banking Crisisi , t = β 0 + β1 X i , t + β 2 Z i , t −1 + ε i , t. (1.6). Where X is now the vector for only the macro economic variables and Z is the vector for the bank system characteristics, Law and order, and Government budget balance. Furthermore, this paper tests Boyd’s et al. (2004) argument that banking characteristics have an effect on the probability of an occurrence of a banking crisis through the macroeconomic conditions. More specifically, they contend that given that inflation is not higher than a threshold (yet to be determined),.

(16) 14 monopolistic banking systems always increase the probability of a banking crisis. Because of the complexity of a banking crisis it is interesting to test as well whether other banking characteristics have an effect on banking crisis probability given the macroeconomic conditions. Model 2: Ci ,t = β 0 + β 1∀ i ,t + β 2 Υi ,t + β 3 Ζ i ,t + β 4 ∀ i ,t * Ζ i ,t + β 5 Υi ,t * Ζ i ,t + ε i ,t. (2.1). This is a general model and when empirically testing it will be determined which macroeconomic conditions are important for which banking characteristics, in other words we need statistical proof and economic sense. It is not enough to find significant results if there is no economic reasoning behind the relationship. b. DATA Appendix B table (4) provides detailed information regarding the variable description descriptive statistics and sources. Furthermore, table (5) in appendix B. reports the pairwise correlation matrix for the full sample and will make it clear why sometimes we cannot add variables together. IV.. RESULTS. As our starting point we initially tried to replicate column (2) in table (2) of Beck et al. (2006). Further details of this replication can be found in appendix E. Table (8) in appendix E excludes crisis duration periods from the model. We basically tried to figure out why bank concentration is not always as significant as Beck et al. (2006) claim it to be. From this replication we decided which variables we were going to keep in our basic model. We focused on the variables that were consistently significant in relation to banking crisis probability. We observed that GDP growth is the most robust systemic crisis indicator followed by the real interest rate. Inflation did not seem to be important in Appendix E, because the raw data contains so many outliers that coincide with crisis periods. We decided to follow Dreher et al. (2007) by transforming the inflation first and then adding the variable to our model relating to tables (1) and (6). Table (1) starts with the basic model consisting of GDP growth, real interest rate and transformed inflation. Step-wise we add the bank system characteristics and test their individual effect on banking crisis probability. Table (1) omits the crisis duration period. After testing the individual effects we added all the significant variables together in two different models, columns (20) and (21). We could not add all the significant variables in one model because of correlation issues. See table (5) for further details. The only difference between table (1) and table (6) is that we also include the full crisis period in the latter. When we add all the significant variables together in columns (20) and (21), we look at the.

(17) 15 significance of the variables in table (1). This makes all columns in table (6) directly comparable to those in table (1). Table (7) differs from table (1) only because it has a different more expanded basic model which we base on Beck et al. (2006). We do this to test the sensitivity of our model when we add more variables that were used in the literature. See table (3). Last but not least, in table (9) we test if there is an interaction effect between bank concentration and bank assets to GDP ratio. The latter is a proxy for bank size. In addition, we test whether there is an interaction effect between bank concentration and banking supervision. In tables (1) (6) (7) and (9) we lag all the banking system characteristics variables, the Law and order variable, and the Government budget variable with one year. We find that GDP growth is a robust predictor of systemic banking crisis, as we expected. Low GDP growth increases the probability of a systemic banking crisis. Beck et al. (2006) mentioned that countries in crisis grow at a slower rate, so we also lagged GDP growth by one year and this did not alter the results. The real interest rates appear to have a positive relationship with banking crisis probabilities. This was in line with our expectations because refinancing becomes expensive when interest rates rise, making banks prone to liquidity problems. We confirm the findings by Demirgüç-Kunt and Detragiache (1998) and Beck et al. (2006) that higher inflation increases banking crisis probability, regardless of the transformation. The results for GDP growth, real interest rates and inflation (transformed) do not significantly change in table (6) for the full crisis episodes; neither do they change in table (7) where we apply a different basic model. Economic freedom is significant in column (1) of tables (1) and (6) (initial and full crisis), but not in table (7). The variables ceases to be significant when we apply a different basic model and/or when we add the variable in a regression with multiple banking system characteristics. The sign is consistently negative, supporting the view that less economic freedom would lead to more crisis because banks are more constrained in diversifying their portfolios. Beck et al. (2006) also draw the same conclusion. Even so, we would like to stress that this result should be interpreted carefully as it is not completely robust. The liquidity ratio only becomes significant when we use the full crisis period, in table (6). However, when we use the more elaborate model of table (7) with the full crisis period the variable is not significant anymore although it always has a negative coefficient (results available upon request). We believe that there may be some indication that the liquidity ratio matters. However, for many countries data is not available so that we cannot draw a general conclusion..

(18) 16 Profitability of a banking system is a rather robust crisis predictor in our models. To proxy for profitability we use return on equity and return on assets. The return on equity is significant with a negative sign in columns (3) and (21) of tables (1) and (6). When we expand the model in table (7) the variable ceases to be significant. On the other hand, the return on assets is robust throughout with a negative coefficient in columns (4) and (21) of tables (1), (6) and (7). These findings were in line with our expectations. We conclude that the return on assets is a good crisis predictor. Banking systems that are not profitable should be of great concern to supervisory authorities. The net interest margin appears to have a negative relationship with banking crisis probability confirming the theory that a higher margin could be used as a buffer for bad times. However, this result is not completely robust as the variable only appears to be individually significant in column (5) of table (1). When we only use the initial crisis year in table (1), we fail to find any evidence that credit market regulation and private ownership are relevant in relation to banking crisis probability. However, when we use the full crisis period in table (6) columns (6) and (7), the variables do become significant with a negative sign. That is in line with Goodhart’s (2004) argument that any non-profit-maximizing banking entity can make financial systems weaker. Moreover, an increase in credit market regulation can be seen as preventing banks from engaging in further risky activities during crisis and thus preventing them from diversifying their portfolios. This is of course at odds with the previous finding for the variable “economic freedom”. What this means is that banks are better off constrained on one end, the credit market, but not in general. The economic freedom variable includes many “freedom” dimensions other than credit market regulation such as: the government, the legal structure, regulation of business and labor, access to sound money and international trade. The leverage ratio of banks and the leverage ratio of the total financial system seem to have a positive significant relationship with banking crisis. The former is significant in column (11) of tables (1), (6) and (7). The latter is only significant in column (12) of table (1). Both variables lose their statistical significance when we add the significant banking system characteristics variables together in one regression (column (21) of tables (1), (6) and (7)). The reason we do not include the leverage ratio of the total financial system in either columns (20) nor (21) is because this variable does not have many observations. We conclude here that a banking system is more fragile and more prone to a systemic crisis if it has too much leverage in aggregate. Supervisors should not only monitor this ratio per bank, but also for the system as a whole. Another very robust variable is the bank credit to deposit ratio. The variable has a significant positive relationship with banking crisis probability in columns (13) and (20) of tables (1), (6) and (7). This means that a high ratio of credit to deposits is a good crisis predictor. We can now answer our previous question that deposits diminish before as well during crisis. It is also possible that more loans are.

(19) 17 generated out of the deposits received by banks prior to crisis and that after crisis deposits actually diminish. In any case, we believe that this variable should also be closely monitored by supervisory authorities. The cost to income ratio also has a positive relationship with dependent variable in column (14) of tables (1), (6) and (7). This variable is thus only significant when we add it individually to the model. Nonetheless, supervisors should monitor this ratio more closely and also think twice before salvaging a bank that is not internally efficient. Bandt and Hartmann (2000) concur that some bank failures are desirable if the concerned bank is inefficient. Supervisors should however keep in mind that the cost to income ratio does not directly point to the source of the problem, but is an indication that there is a problem. Our results show that banking supervision matters and it is robust no matter how our model changes. The sign is negative and significant in columns (17), (20) and (21) of tables (1), (6) and (7). The results imply that supervisory authorities such as central banks are good for crises prevention. If supervision is carried out properly, banks will have less scope for excessive risk taking. The government budget variable is negative and significant only in column (18) of table (6). This leads us to conclude that this variable is not robust. Although it may indicate that the government budget may matter during a crisis as opposed to predicting crisis episodes. Note that this variable is defined as revenues minus expenses. There could be numerous reasons why the revenues could decline and the expenses could rise. Whatever the reasons, governments should keep their balances in check. There are some variables that are not significant at all in either tables (1), (6) nor (7). Thus, we fail to find evidence that these variables matter in the context of this paper. These variables are: central bank independence (CBI), bank assets, total assets, deposit insurance, interest rate control and law and order. This does not entail that these variables are not important. Our finding that CBI, deposit insurance and law and order are not significant is in line with the results of Klomp (2010). See table (3) in appendix A. DemirgüçKunt and Detragiache (1998) on the other hand find the moral hazard index to be significant and negative. However, the deposit insurance forms just one aspect of the moral hazard index. We fail to reject that interest rate controls have no effect on the probability of a banking crisis. However, during a full crisis period our results demonstrate that the credit market regulation, a variable composed using the interest rate controls, is significant. As for bank assets and total financial system assets, we posed the question of whether the size of the banking sector or financial sector has an influence on banking crisis probability. Our results show that on its own the size of the banking system does not seem to be a good crisis predictor. From our replication in table (8) we concluded that bank concentration was also not robustly significant on its own. We decided to test whether there is an interaction effect between bank concentration and bank size. As the size of the banking system increases, the degree of concentration of the banks may become.

(20) 18 more important. We test this interaction in table (9) appendix F and the results report a significant negative relationship. Even though solely this regression does not let us interpret the exact relationship between bank size and bank concentration, it does give scope for future research. From graph 1 in appendix F we notice that the interaction effect is slightly significant, however we cannot interpret the relationship when the line falls below zero as this would imply negative probabilities which is absurd. For the second interaction effect in table (9) between bank concentration and banking supervision we find no significant relationship. This is confirmed in graph 2. Our motive for adding this interaction effect is our belief that it could be true that the more banks there are in a banking system, the more tedious it becomes for supervisory authorities to closely monitor these banks (Beck et al., 2006). V.. CONCLUSION. This paper investigates how characteristics of banking systems affect the likelihood that a banking crisis will occur. Using data on 124 countries from 1980 to 2008, this paper finds that a number of banking system characteristics are robustly related to the probability of systemic banking crisis. We started our empirical analysis by replicating column (2) in table (2) of Beck et al. (2006) and from there decided what our basic model should consist of. We followed the method of Beck et al. (2006) by using a logit model and including only the initial crisis year episodes. Later we also include the full crisis period as a robustness check for our model. In addition, we expand our basic model to see whether this changes our initial results. Our results show that the profitability of a banking system, leverage ratio of banks, cost to income ratio, bank credit to deposit ratio, and banking supervision are the most important banking system characteristics that supervisors should monitor in relation to banking crisis probability. Countries which have banking systems with high profitability are less likely to experience a systemic banking crisis. Supervisory authorities should thus not only look at liquidity and solvency of banks, but also at their profitability. A low(er) profitability could signal inefficiency. Regardless, lower profitability sends negative signals to investors and the consequences hereof should be of concern to supervisory authorities. In addition, countries which have banking systems with comparatively lower leverage ratio, cost to income ratio, and bank credit to deposit ratio are less likely to experience a systemic banking crisis. Furthermore, countries with more and better banking supervision are also less prone to a systemic banking crisis. Moreover, we are able to confirm that GDP growth, real interest rates and to some lesser extent inflation are good predictors of crisis. Supervisory authorities should also closely monitor the leverage ratios of banks as an increase in this ratio also increases the probability that a systemic banking crisis can occur. The same holds for the cost.

(21) 19 to income ratio and bank credit to deposit ratio. The former is also positively related to banking crisis probability indicating that a banking system that spends more than it earns is bound to run into problems. The latter, also positively related to banking crisis probability, indicates that if too many loans are being generated based on deposits received, banks in aggregate will have too little liquidity to back up these loans making the system fragile and thus more prone to crisis. Furthermore, our attempts to find an interaction effect between bank concentration and bank size, and an interaction effect between bank concentration and banking supervision has not been very successful. Nevertheless, there are still some indications that future studies could find interesting results from studying these or other interaction effects in relation to banking crisis probability. So how should a sound banking system look like? In other words, which characteristics should a banking system have where the chances of crisis are theoretically minimized in the light of this paper? In our view, a sound banking system would have a low leverage ratio, low cost to income ratio, good banking supervision and moderate bank credit to deposits ratio. There should be constraints on how banks operate on the credit market, but banks should still have freedom on other markets to help spread their portfolio risk. Certainly, our models are not complete even though we manage to do predict crisis reasonably well. Systemic banking crisis episodes are very complex and elements outside of the banking system can also influence crises. Recently, Le et al. (2012) claim that productivity and labor supply also play an important role in predicting crisis. Many more studies can be conducted in this field, and it should be done, as these works are more than relevant for policy makers..

(22) 20 Table 1. Initial Crises VARIABLES GDPG Real interest rate Transformed inflation Economic freedom. (1) Crisis -0.0038*** (0.0007) 0.0005** (0.0002) 0.0111 (0.0245) -0.0062* (0.0033). Liquidity ratio. (2) Crisis -0.0011 (0.0007) 3.98e-05 (6.18e-05) 0.0515 (0.0377). (3) Crisis -0.0039*** (0.0007) 0.0005*** (0.0002) 0.0508* (0.0268). (4) Crisis -0.0038*** (0.0008) 0.0005*** (0.0002) 0.0331 (0.0298). (5) Crisis -0.0041*** (0.0008) 0.0006** (0.0003) 0.0725* (0.0420). (6) Crisis -0.0037*** (0.0007) 0.0004* (0.0002) 0.0091 (0.0241). -0.0433 (0.0266). Return on equity. -0.0392*** (0.0115). Return on assets. -0.205*** (0.0782). Net interest margin. -0.200* (0.114). Credit market regulation. -0.0021 (0.0016). Private ownership No. of countries No. of crises Observations % Crises correct % Correct R-squared (pseudo) Log likelihood. VARIABLES GDPG Real interest rate Transformed inflation CBI. -0.0013 (0.0011) 104 73 2075 41.10 84.82 0.0771 -291.7. 77 14 524 57.14 89.89 0.431 -36.71. (8) Crisis -0.0021** (0.0010) 0.0005* (0.0003) 0.0539* (0.0295) -0.0171 (0.0181). (9) Crisis -0.00412*** (0.0006) 0.0004** (0.0002) 0.0295 (0.0234). Bank assets. 111 44 1459 45.45 89.58 0.106 -176.4. (10) Crisis -0.0032*** (0.0009) 0.0005*** (0.0002) 0.0157 (0.0354). 111 44 1459 38.64 89.79 0.103 -177.1. (11) Crisis -0.0040*** (0.0006) 0.0004** (0.0002) 0.0271 (0.0226). 111 42 1421 33.33 90.85 0.115 -167.6. (12) Crisis -0.0030*** (0.0008) 0.0005*** (0.0002) 0.0133 (0.0314). 115 77 2174 36.36 89.79 0.0626 -312.0. (13) Crisis -0.0030*** (0.0007) 0.0004** (0.0002) 0.0624*** (0.0197). 111 71 2019 42.25 86.58 0.0772 -283.7. (14) Crisis -0.0037*** (0.0008) 0.0004** (0.0002) 0.0271 (0.0283). 0.0031 (0.0089). Total assets. 0.0073 (0.0113). Leverage bank. 0.0071** (0.0033). Leverage total. 0.0227** (0.0114). Bank credit/deposit. 0.0215*** (0.0050). Cost income No. of countries No. of crises Observations % Crises correct % Correct R-squared (pseudo) Log likelihood. (7) Crisis -0.0040*** (0.0007) 0.0005** (0.0002) 0.0080 (0.0245). 0.0283** (0.0134) 108 44 1548 36.36 93.15 0.0603 -188.0. 112 69 2115 40.58 87.90 0.0776 -280.4. 36 16 603 37.50 90.38 0.120 -64.96. 112 69 2114 42.50 87.61 0.0807 -279.5. 36 16 603 43.75 88.72 0.136 -63.79. 120 83 2340 46.93 85.90 0.0771 -331.0. 111 43 1449 41.86 90.13 0.0994 -174.4.

(23) 21. VARIABLES GDPG Real interest rate Transformed inflation Deposit insurance. (15) Crisis -0.0025*** (0.0008) 0.0003* (0.0002) 0.0492** (0.0232) -0.0088 (0.0077). Interest rate control. (16) Crisis -0.0043*** (0.0009) 0.0006*** (0.0002) 0.0317 (0.0251). (17) Crisis -0.0033*** (0.000783) 0.0004*** (0.0001) 0.0080 (0.0207). (18) Crisis -0.0038*** (0.001) 0.0001 (0.00021) 0.0254 (0.0247). (19) Crisis -0.0029*** (0.0010) 0.0004* (0.0002) 0.0511* (0.0277). (20) Crisis -0.0022*** (0.0007) 0.0004** (0.0002) 0.0056 (0.0326). (21) Crisis -0.0016*** (0.0006) 0.0002** (0.0001) 0.0133 (0.0197). -0.0110*** (0.0036). -0.0103*** (0.0034). -0.0050 (0.0035). Banking supervision. -0.0183*** (0.0036). Government budget. 0.0007 (0.0006). Law and order. 0.0003 (0.0027). Bank credit deposit. 0.0164** (0.0067) -0.0096 (0.0107) -0.0031 (0.0030) -0.0641** (0.0308). Cost income Economic freedom Return on assets Return on equity Net interest margin. -0.0662 (0.0523). Credit market regulation Leverage bank No. of countries No. of crises Observations % Crises correct % Correct R-squared (pseudo) Log likelihood. 120 65 1932 32.31 89.03 0.0574 -268.0. 85 57 1539 43.86 85.51 0.0826 -223.7. 85 57 1539 61.40 77.97 0.121 -214.2. 98 28 910 28.57 90.00 0.0707 -116.2. 109 65 1851 33.85 89.25 0.0421 -269.69. 73 26 881 57.69 86.15 0.2739 -85.10. -0.0045 (0.0064). -0.0241** (0.0109) -0.0229 (0.0441) -0.0000 (0.0007) 0.0094 (0.0058) 75 24 846 62.50 88.06 0.3206 -74.16. The models are estimated by logit. The logit probability model estimated is Banking Crisisi,t = β0 + β1 GDPGi,t + β2 Real interest ratei,t + β3 Transformed inflationi,t + β4 Zi,t + ɛi,t. The dependent variable is a crisis dummy that takes on the value of one at the start of a systemic banking crisis and zero otherwise. The data for systemic banking crises is taken from Leaven and Valencia (2008, 2010). Z is a vector of bank system characteristics, Law and order, and Government budget. The bank system characteristics are added stepwise to test their individual effect. In columns (20) and (21) we pool only the significant variables (with the exception of Leverage bank because then we are left with too little observations to make any interpretations). We do this in two separate columns because of correlation between some of the independent variables. To assess our model fit we apply a cut-off point of 0.05. In these models we drop the following outliers: Nicaragua 1988 due to excessive real interest rate and inflation, a systemic banking crisis followed in 1990; Slovenia 1991 dues to high real interest rate. A banking crisis followed in 1992. All the bank system characteristics, Law and order, and Government budget are lagged for 1 period so that we do not have reverse causality problems. Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1..

(24) 22 APPENDIX A: RESEARCH COMPARISON Table 2. Bank system Characteristics Literature Beck,. Bank System. DemirgucCharacteristics. Kunt et al. (2006). Variables Systemic crisis Concentration. Klomp. Klomp. (2010). (2010) not. pooled. pooled. Demirguc-. Barth, Caprio. Garcia. Kunt and. Herrero and. Detragiache. Del Rio. (1998). et al (2004). Cihak (2007). Glick and. Eichengree Domac and n and. Hutchison. (2003). Arteta. (2000). (2002). Hutchinson. Komulainen. Martinez-. and McDill. and Lukkarila. Peria (2003). (1999). (2003). Tanveer and De Haan (2008). -. Economic Freedom. +. Bank (liquid) reserves to assets. ns. ns. -. ns. -. ratio Bank ROE Bank ROA Net interest margin Credit market regulation Private ownership Central bank independence˚ Bank assets to GDP ratio Total financial system assets to GDP ratio Leverage ratio. ns. ns. Bank assets/Bank deposits Bank credit to deposit ratio Cost to income ratio Deposit Insurance. ns. -. Interest rate controls Banking supervision. * 1980-2005. Most studies use Moral Hazard Index, which includes deposit insurance, interest rate controls, and other measures that are not available to us.. Table 3. Macro and Financial (other) variables Literature Other (Macro and Financial) Variables Real GDP Per Capita Real GDP Growth Terms of trade change Real Interest rate Inflation. Beck, DemirgucKunt et al. (2006) -. Klomp. Klomp. (2010). (2010) not. DemirgucGarcia Cihak (2007) Kunt and Herrero and Detragiache Del Rio (1998) (2003) ns. pooled. pooled. -. -. -. -. ns. ns. ns. s. +. +. +. +. ns ns. +. +. +. +. Depreciation. ns. ns. +. ns. M2/reserves*. +. Domestic Credit to the Private Sector Growth Rate of Credit Government Budget Balance Law&Order (Institutional quality). +. +. +. ns. +. +. +. +. +. ns. ns. -. ns. Glick and Hutchison (2000). -. ns. ns. ns. -. ns + ns. ns ns. ns ns. ns. ns. ns. +. ns. ns ns. +. s/ns. ns. ns. ns. ns. +. s/ns. ns. +. ns/s ns. ns. +. ns. -. Eichengreen Domac and Hutchinson Komulainen Tanveer and and and Arteta Martinez- and McDill De Haan Lukkarila (2002) Peria (2003) (1999) (2008) (2003) s. -. *In this paper we use M2/total reserves, however most papers use M2/foreign exchange reserves.. + +. +. ns ns. +.

(25) 23. Appendix B. Variable description, Descriptive statistics and sources Table 4. Variable description Variable:. Description:. Mean:. Std. Dev.:. Obs.:. Source:. Crisis (initial). Initial crisis year=1, crisis duration years are excluded, all other years=0. 0.0351. 0.1840. 3392. Crisis (full). Initial crisis year=1, crisis duration years=1, all other years=0. 0.0890. 0.2848. 3596. GDPG*. Gross domestic product growth %. 3.3390. 5.3565. 3403. Laeven and Valencia (2010) Laeven and Valencia (2010) WDI (2011). Terms of trade change. Terms of trade change is the change in the net barter terms of trade index. The latter is calculated as the percentage ratio of the export unit value indexes to the import unit value indexes, measured relative to the base year 2000. Real interest rate is the lending interest rate adjusted for inflation as measured by the GDP deflator.. 0.4545. 11.52. 2622. WDI (2011). 6.1183. 14.1463. 2612. WDI (2011). Inflation as measured by the consumer price index. Transformed inflation = ((Inflation/100)/(1+(Inflation/100))). 43.6496 0.1155. 343.296 0.1644. 3388 3391. WDI (2011) Calculated. Money and quasi money comprise the sum of currency outside banks, demand deposits other than those of the central government, and the time, savings, and foreign currency deposits of resident sectors other than the central government. This definition is frequently called M2; it corresponds to lines 34 and 35 in the International Monetary Fund's (IMF) International Financial Statistics (IFS). Total reserves comprise holdings of monetary gold, special drawing rights, reserves of IMF members held by the IMF, and holdings of foreign exchange under the control of monetary authorities. The gold component of these reserves is valued at year-end (December 31) London prices. Depreciation of the nominal exchange rate. The nominal exchange rate is the official exchange rate and it refers to the exchange rate determined by national authorities or to the rate determined in the legally sanctioned exchange market. It is calculated as an annual average based on monthly averages (local currency units relative to the U.S. dollar). Growth of credit from the domestic to private sector lagged by two periods. Domestic credit to private sector refers to financial resources provided to the private sector, such as through loans, purchases of nonequity securities, and trade credits and other accounts receivable, that establish a claim for repayment. For some countries these claims include credit to public enterprises. GDP per capita is gross domestic product divided by midyear population.. 11.8481. 48.1195. 3032. WDI (2011). 117.67. 4684.74. 3160. WDI (2011). 3.3752. 25.4239. 2960. WDI (2011). 6732.424. 11139.5. 3360. WDI (2011) Financial structure database (2009) Financial structure database (2009) Economic Freedom of the World (2011). Real interest rate* Inflation Transformed inflation* M2/reserves. Depreciation 13. Credit growth. GDP per capita WB, annual (1988-08) WB, average (1988-08) Economic freedom Credit market regulation Private ownership Liquidity ratio. Return on assets Return on equity Net interest margin Bank assets. Total assets. Concentration - calculated as the fraction of assets held by the three largest banks in each country. Time varying for the period 1988-2008. Concentration - calculated as the fraction of assets held by the three largest banks in each country. Averaged over the time period 1988 to 2008. http://www.freetheworld.com/2011/2011/Dataset.xls.. 0.7012. 0.2039. 1966. 0.7213. 0.1653. 3593. 6.0127. 1.2460. 2740. http://www.freetheworld.com/2011/2011/Dataset.xls.. 6.4601. 2.5692. 2894. http://www.freetheworld.com/2011/2011/Dataset.xls.. 5.4444. 3.6558. 2707. Ratio of bank liquid reserves to bank assets is the ratio of domestic currency holdings and deposits with the monetary authorities to claims on other governments, nonfinancial public enterprises, the private sector, and other banking institutions. Average Return on Assets (Net Income/Total Assets). 0.1613. 0.1593. 643. 0.0121. 0.0255. 1855. Average Return on Assets (Net Income/Total Equity). 0.1292. 0.2079. 1855. The accounting value of a bank’s net interest revenue as a share of its interest bearing (total earning) assets. Claims on domestic real nonfinancial sector by deposit money banks as a share of GDP, calculated using the following deflation method: {(0.5)*[Ft/P_et + Ft-1/P_et-1]}/[GDPt/P_at] where F is deposit money bank claims, P_e is end-of period CPI, and P_a is average annual CPI. Other assets + bank assets. Other assets = claims on domestic real nonfinancial sector by other financial institutions as a share of GDP, calculated using the following deflation method: {(0.5)*[Ft/P_et + Ft1/P_et-1]}/[GDPt/P_at] where F is other financial institutions' claims, P_e. 0.0533. 0.0367. 1799. 0.4848. 0.4053. 2798. 0.6642. 0.4543. 777. Economic Freedom of the World (2011) Economic Freedom of the World (2011) WDI (2011). Financial structure database (2009) Financial structure database (2009) Financial structure database (2009) Financial structure database (2009) Financial structure database (2009). 13 We also ran the regressions for depreciation based on the real effective exchange rate (for which we have less countries). However, it did not affect our conclusion that concentration is not robust. Regressions are available upon request..

(26) 24. Bank credit deposit Bank deposits. Total deposits. Cost income. is end-of period CPI, and P_a is average annual CPI. Private credit by deposit money banks as a share of demand, time and saving deposits in deposit money banks. Demand, time and saving deposits in deposit money banks as a share of GDP, calculated using the following deflation method: {(0.5)*[Ft/P_et + Ft-1/P_et-1]}/[GDPt/P_at] where F is demand and time and saving deposits, P_e is end-of period CPI, and P_a is average annual CPI. Other deposits + bank deposits. Other deposits = Demand, time and saving deposits in deposit money banks and other financial institutions as a share of GDP, calculated using the following deflation method: {(0.5)*[Ft/P_et + Ft-1/P_et-1]}/[GDPt/P_at] where F is demand and time and saving deposits, P_e is end-of period CPI, and P_a is average annual CPI. Total costs as a share of total income of all commercial banks.. 0.9423. 0.4535. 3117. Financial structure database (2009) Financial structure database (2009). 0.4140. 0.3849. 2796. 0.8383. 0.7748. 2796. Financial structure database (2009). 0.6730. 0.2056. 1842. Financial structure database (2009) Arnone et al.(2007) and Acemoglu, Johnson, Querubin, and Robinson (2008) "A New Database of Financial Reforms" (2008) IMF "A New Database of Financial Reforms" (2008) IMF "Deposit insurance around the world: a comprehensive database" DemirgucKunt, Karacaovali et al. (2005) WDI( 2011). CBI. Central bank independence: Average of economic and political independence. Measuring from 0 (full dependent) to 1 (full independent).. 0.4953. 0.2291. 1961. Interest rate control. Read the cited source for more details on computation. The variable varies between zero and three. Zero= no controls up to three the most substantive controls. Read the cited source for more details on computation. The variable varies between zero and three. Zero for no supervision, and 3 for great supervision. The database mentioned in the next column provides information on which year a deposit insurance scheme was enacted (if ever) for each country. The years for which countries have no deposit insurance scheme in place we give a value of zero and one when there is a deposit insurance scheme in place.. 2.0916. 1.2119. 2074. 0.9423. 0.4535. 2074. 0.3618. 0.4806. 2728. ((Revenue, excluding grants (current LCU)- Expense (current LCU) )/ GDP (current LCU))*100%. -0.2695. 8.1583. 1145. Law and Order are assessed separately, with each sub-component comprising zero to three points. The Law sub-component is an assessment of the strength and impartiality of the legal system, while the Order subcomponent is an assessment of popular observance of the law. Thus, a country can enjoy a high rating – 3 – in terms of its judicial system, but a low rating – 1 – if it suffers from a very high crime rate of if the law is routinely ignored without effective sanction (for example, widespread illegal strikes). Bank assets to bank deposits ratio.. 3.7288. 1.4832. 2492. International Country Risk Guide (2009). 1.2057. 0.4750. 2796. Calculated. Total financial assets to total financial deposits ratio.. 0.7695. 0.2977. 776. Calculated. Banking supervision Deposit insurance. Government budget balance Law and order. Leverage bank Leverage total. * Variables marked with a star are the only ones kept in the basic model for this paper after careful consideration. We use the full sample for each variable to compute the descriptive statistics..

No results found