PREDICTING BANKING CRISES: THE ROLE OF BANK STOCK IN
ADVANCED ECONOMIES
July 7
th, 2016
Name:
Douglas Konadu
Student Number:
10342745
Course:
MSc Business Economics, Specialization Finance
I
for the contents of this document.
I declare that the text and the work presented in this document is original and that no
sources other than those mentioned in the text and its references have been used in
creating it.
The Faculty of Economics and Business is responsible solely for the supervision of
completion of the work, not for the contents.
II Abstract
This study explores the predictive capabilities of bank stock return for banking crises. In a sample of 14 advanced OECD countries, bank stock is found to be a significant predictor of banking crisis, with an average marginal effect of about 22 percent increase in crisis likelihood resulting from a percentage increase in average bank stock return. A credit boom episode increases crisis likelihood by about 78 percent. The results in this study also most macroeconomic and banking sector variables that have been widely accepted as the best crisis predictors to be insignificant after inclusion of bank stock return in logistic specifications. Bank stock is also found to be a better predictor of banking crises than stock market return. The results are robust to several controls and alternative specification methods.
III
I Introduction ...1
II Literature Review ...3
2.1 What causes a banking crisis?...4
2.1.1. On the nature of banking crises ...5
2.2 Predictive capabilities of stocks and the role of financial institutions ...6
2.2.1. On the role of banks ...7
2.2.2. Stocks as leading indicators ...9
III Data ... 11
3.1 Banking crisis variable ... 11
3.1.1 Dating and measurement of banking crisis... 12
3.1.2 Crisis frequency and descriptives ... 14
3.2 Bank stock variable... 15
3.2.1 Bank stock descriptives ... 15
3.3 Macroeconomic and other explanatory variables ... 17
IV Methodology ... 18
4.1 Popular models in crisis prediction literature ... 18
4.1.1. The signals approach ... 19
4.1.2. The logistic approach ... 19
4.1.3. Linear Probability Model and the Lagged Return Model ... 21
4.2 Hypotheses ... 22
V Findings ... 23
5.1 Lagged Return Model ... 23
5.2 LPM and Logistic Regressions ... 24
5.3 Robustness checks ... 29
5.3.1 Country Level Regressions ... 30
5.3.2 Market return as a predictor... 32
VI Conclusion ... 34
Bibliography... 35
1
I
Introduction
The 2008 financial crisis was arguably the most severe financial crisis since the Great Depression of the 1930s. Prior to this there had been a number of financial crises in several countries, both developed and emerging, usually followed by a recession and a slowdown in economic activities. Failure of the financial system is most common to these usually devastating crises. Most failures of banks and other financial institutions during financial crises are systemic and reflects dramatic changes in the financial environment, as explained by Kindleberger and Aliber (2011). These financial crises and resulting bank failures are accompanied by slowdown in rates of growth of the real economy and costly increases in unemployment. Laeven and Valencia (2012), for example, find that banking crises episodes leads to a 23% cumulative output loss as well as significant fiscal costs. Bussiere and Fratscher (2006) argue that financial crises that initiated in a certain country can spread to other markets in other countries as well, and this contagion effect can lead to systemic repercussions for the international financial system as a whole. The most recent financial crisis that had its genesis in the United States and later spread to the rest of the world is a perfect example of this phenomenon.
The enormous run-up in U.S. equity prices preceding the crisis was consistently denied to be a bubble by many financial analysts, but rather attributed to financial innovation, inclusive of sub-prime mortgages (Reinhart & Rogoff, 2008). Naturally, every bubble has to burst some time, and this particular one finally burst in late 2007, leading to the collapse and near collapse of many financial institutions worldwide.
In view of the high costs that accompany financial crises, there is a need for forecasting, or at the very least a need to signal an impending crises ahead of time. Policymakers can then decide on the necessary course of action that could potentially help avoid or limit the severity of such an episode. There is an extensive amount of literature covering financial crises forecasting. The vast majority of this, as will be discussed in later parts of this study, focuses on a common set of explanatory variables to forecast financial crises with varying success pertaining to the predictive capabilities of the models employed.
At the heart of this is the relative unpredictability of financial crises a priori, which poses a challenging issue for economists and policymakers alike. Crises could be caused by various variables, whose behaviour do not necessarily follow a recognizable pattern that could be diffused into warning signals. It is especially difficult for the reason that financial crises are usually preceded by periods of economic and asset price booms (Kindleberger and Aliber, 2011), which seem to disguise the any impending problems of systemic distress. Moreover, not all asset price booms end up in financial distress, even though financial cycle peaks have been shown to be closely associated with financial crises1.
Inretrospect it seems relatively straightforward to identify the factors that caused a particular crisis. Multiple studies have identified the growth rate of credit (Taylor and Schularick, 2012), capital outflows, banking liquidity (Demirgüç-Kunt & Detragiache, 1997), bank balance sheet data like the leverage ratio and the banking sector liquidity ratio (Barrell et al., 2010), but also real estate price growth as the determinants of financial crises. Among the most used variables are macroeconomic variables such as real GDP growth and domestic real credit growth, with the domestic credit growth proving to contain the most predictive capabilities in crisis forecasting. In spite of the extensive amount of literature covering financial crisis prediction with prime focus on the role of credit growth, evidence
2 that conclusions on emerging-markets can be extended to advanced economies is yet to attain a consensus position (Taylor and Schularick, 2012).
Furthermore, as argued by Barrell et al. (2010), the triggers of a crisis depends on the type of economy and the nature of the banking system. The nature of the banking system in particular, can be argued to play an important role in the occurrence and relative severity of a financial crisis. From this it becomes apparent that forecasting models can be improved tremendously when the role of the banking system is accounted for in the forecasting models. The role of the banking sector, as also recognized by Barrell et al. (2010), has been almost completely ignored in crises prediction literature. This may explain the varying conclusions drawn by various empirical literature on the subject.
The varying types of variables used in forecasting models also suggests a uniqueness to crises which makes it difficult to make accurate predictions, and generalizations even less so. The challenge thus lies in identifying a factor common to the bulk of financial crises in its behavioral patterns, as well as possessing over predictive capabilities with a relatively high degree of accuracy, as well as recognizing the role of banks in the occurrence of crises. The aim of this study is to access the validity of just such a tool for forecasting of financial crises, namely bank stock prices (further to named bank stock).
Stock prices in general contain information on expectations of future economic events. Reasoned from the efficient market hypothesis, stock prices fully reflect all available information of an asset, such that the price of a stock reflects the unbiased estimates of the underlying values of an asset (Basu, 1977). However, a typical characteristic of crises is the deviation of prices from fundamentals, such that prices no longer reflect expectations of the values of underlying fundamentals (Hau and Lai, 2013). Crises tend to share striking similarities in among other things the run-up of asset prices in the period preceding a crisis2, in particular stock and property prices. This entails that a careful analysis of
the developments around stock prices should reveal information on a potential crisis. This can significantly improve the results of crises forecasting. Bank stock may be the best instrument in this respect. The motivation for this stems from the role of banks in the occurrence of financial crises, along with the predictive capabilities of stocks. Considering bank stock as opposed to the stock market in general may in turn help reduce the noise contained in stocks, thereby improving on the accuracy of predictions, as will become apparent throughout the rest of this paper. Further motivation for this choice is contained in Section II.
The contribution of this study is in twofold. Firstly, the application of a novel variable that possesses over significant predictive power in the prediction of financial crises, as well as accounting for the role of banks in the occurrence and propagation of crises. Recognition of the role of banks, as stated earlier, can help improve forecast results. This has mainly been ignored in crisis prediction literature. Secondly, the methodology employed differs from the previous literature on the subject, as crisis is observed on the bank level, which allows to access the probability of a bank facing a banking crisis based on the economic conditions facing the country in which it is based. Country level regressions are estimated as well to ensure results can be compared to literature on this subject. Section IV provides detailed information on the methodology.
Data used in this study covers fourteen OECD countries (see Table 3) over 1980-2011. Advanced countries are used subject to data availability, especially of bank stocks, and relative comparability of banking system. Detailed descriptions of the data can be found in Section III. Along with bank stock, macroeconomic variables are used in the regressions to estimate the contributions of
3 these variables for crisis probability. Logistic and linear probability models are used to estimate crisis probabilities. Various robustness tests are carried out to test the robustness of the results obtained.
In the sections that follow, the choice for bank stock as the main explanatory variable is motivated. The literature review in Section II gives an overview of literature on financial crises prediction, as well as show the predictive capabilities of stocks in general. Further explored in this section are the general causes of financial crises and the role banks play in their occurrence. Section III shares information on the data used and presents descriptive statistics. Section IV explores the econometric methodology used to obtain the results. The main findings are analyzed in section 5. The final sections contains the conclusions and appendixes.
II Literature Review
Three forms of financial crises are identified in the crisis literature; banking, currency and twin crisis. A twin crisis is defined as the simultaneous occurrence of a banking and currency crisis within a specified period. Mishkin et al. (2003) defines a financial crisis as a disruption to financial markets in which adverse selection and moral hazard problems become increasingly worse, such that financial markets are not able to perform their primary task of channeling funds efficiently for the best investment opportunities. This state of market inefficiency is usually preceded by a period of credit boom, where there is overlending of credit. According to Kindleberger and Aliber (2011) there have been four waves of financial crises since the early 1970s, each of which was preceded by a wave of credit bubbles. Typical of these crises was the cross-border influx of money leading to non-sustainable increases in the prices of stocks and real estate. An inevitable implosion of the bubble usually follows, leading to a full blown financial crisis.
Mishkin et al. (2003) further identifies four factors that cause financial crises through the channel of increasing asymmetric information problems, being: (i) deterioration of financial-sector balance sheets, (ii) increases in interest rates, (iii) increases in uncertainty, and (iv) deterioration of nonfinancial balance sheets due to changes in asset prices. These are factors that are prevalent in the core business of financial institutions. It is also worth noting that these are fundamental factors that can be expected to be reflected in the stock prices of financial institutions.
This section explores the causes and effects of financial crises, the role of financial institutions in the occurrence of crises and the predictive capabilities of stocks as seen in the financial crisis literature. The focus in the rest of this paper lies on banking crisis. The assumption is that banks play a greater role in the occurrence of banking crises, as the initial stage of financial crises is depicted by the deterioration in financial and nonfinancial balance sheets (Mishkin et al., 2003). Moreover, empirical evidence shows that banking crises typically precede currency crises (Kaminsky and Reinhart, 1999), such that the detection of possible distress in the banking sector offers greater potential in mitigating the effects of financial crises.
Before diving into the subject, an overview of relevant literature on banking crisis prediction is presented in Table 1 to offer a short summary of what is known on the subject. As already specified, the literature on financial crisis prediction is least to say, extensive. The variables employed to determine the causal effects of crisis differ from paper to paper as can be seen in the table. Further exploration of the literature listed follows in sub-sections 2.1 and 2.2.
4 Table 1: List of literature on financial crises predictions and variables used
Author(s) Variables employed. Methodology Conclusion
Sach et al. (1995). Real exchange rate, lending boom, current account, M2, capital inflows, short-term capital inflows.
Simple OLS. Overvalued exchange rate and lending booms are conditions for crisis.
Kaminsky and Reinhart (1999).
M2, domestic credit, real interest rate, lending-deposit rate, bank deposits, exports, imports, terms of trade, stock returns, etc.
Signals approach methodology.
Financial liberalization activates boom-bust cycle, fueled by credit through capital inflows to cause banking and currency crisis.
Borio and Lowe (2002).
Credit, asset prices and investment. Signals approach methodology.
Credit gap is best indicator in crisis prediction.
Demirgüç-Kunt & Detragiache (2005).
GDP growth, terms of trade, exchange rate change, real interest rate, inflation, fiscal surplus, M2, credit growth, etc.
Multivariate logit model. Macroeconomic environment plays important role in occurrence of banking crisis.
Davis and Karim (2008)
GDP growth, terms of trade, depreciation, real interest rate, inflation, fiscal surplus, M2, credit growth, private credit, deposit insurance.
Multivariate logit and binary recursive tree.
Logit model is marginally better in prediction of sub-prime crisis.
Barrell et al. (2010). Bank capital and liquidity variables (liquidity ratio and unweighted capital adequacy ratio) and real property price growth, inclusive of macroeconomic variables as in Demirgüç-Kunt & Detragiache (2005).
Cumulative logistic distribution.
Bank capital and liquidity variables and property price growth significantly impact banking crisis probabilities.
El-Shagi et al. (2013) Replicates Kaminsky and Reinhart. Signals approach with an evaluation of significance based on bootstrap approach.
Deficit, real interest rate and real interest rate differential best predictors of banking crises.
Taylor and Schularick (2012)
Credit, M2, bank assets, bank loans, bank assets, GDP growth, Inflation
Logistic regressions. Credit is an important predictor of banking crises.
2.1 What causes a banking crisis?
Demirgüç-Kunt & Detragiache (1997) explore the determinants of banking crises. Banks, they argue, perform the primary task of liquidity transformation, where short term liabilities like short-term deposits are transformed to short and long-term assets like loans to businesses and consumers. Banks, like any other type of firms, thus become insolvent and ultimately fail when the value of their assets fall short of the value of their liabilities (Demirgüç-Kunt & Detragiache, 1997). This introduces banks to credit risk that cannot be completely eliminated without severely curtailing the role of banks, even in the presence of collateral. It is precisely the issue of asymmetric information and moral hazard that induces banks to the forms of risk, among which credit risk, that tend to propagate financial crises (Mishkin et al., 2003). As the authors argue, when banks’ balance sheets deteriorate such that they have to cut back on lending (raising capital is considered too expensive), this can cause economic contraction, and potential contagion in the absence of government safety nets can lead to a collapse of the banking system. Asymmetric information is cited as one of the main reasons why banks in emerging markets play a more important role in the financial systems and perhaps contribute more to financial crises as compared to the banks in advanced economies. Banks in emerging markets have
5 relatively more difficulties in gathering information on borrower characteristics. Assessment of counterparty risk is therefore less accurate.
Empirical evidence from Demirgüç-Kunt and Detragiache (1997) show crisis to occur in deteriorated macroeconomic environments. The likelihood of a crisis seems to be higher in periods of low GDP growth, high real interest rates and high inflation. Higher real interest rates is theorized to stem from adverse selection that emerges from a setting of asymmetric information problems in markets (Mishkin et al., 2003). As interest rates rise, the more prudent borrowers may find it unwise to borrow. Borrowers with the riskiest projects are those willing to pay high rates to fund highly risky projects, such that the likelihood of defaults which induces banks to credit risk increases significantly, thereby increasing the probability of systemic banking distress.
Further identified by Demirgüç-Kunt and Detragiache (1997) as causes of banking crises is sudden capital outflows that may be the result of a lower domestic interest rate. Banking crises are also more likely to occur when a larger portion of credit goes to the private sector, which may be indicative of a link between financial liberalization and banking sector instability. Interestingly, they find the presence of deposit insurance in a country to increase the probability of banking crises. This finding underpins the unintended moral hazard that accompanies deposit insurance, as deposit insurance creates incentives for excessive risk-taking by bank managers (Demirgüç-Kunt and Detragiache, 1997).
In countries where banking supervision is frail, banking crises may be caused by widespread systematic looting by banks (Demirgüç-Kunt and Detragiache, 1997). They explain that bank managers in such a country may not only invest in projects that are too risky, but also in projects that are surely to fail where they can divert funds for personal profit. Such practices can introduce systemic risks in the banking system and subsequently subject the financial sector to systemic fragility. As such, a weak legal system and weak banking supervision also increases the probability of a crisis. This is not only limited to emerging market or developing economies. In fact, Akerlof and Romer (1993) and Kane (1989) make the claim that this form of looting behaviour was at the heart of the Savings and Loans crisis of the 80s and 90s in the United States.
The causes of banking crises cited above seem to stem from two directions. On the one hand there are the macroeconomic conditions that influences the operations of banks and subsequently increase crisis probability, and on the other hand structural banking sector characteristics that play a significant role in crisis occurrence.
2.1.1. On the nature of banking crises
To further explore the causes of financial crises there is a need to distinguish between financial crises and recessions. The analysis on this subject by Drehmann et al. (2012), who researches the characteristics of financial crises, identify a financial cycle that is closely related financial crises, yet significantly different from the business cycle. Moreover, their findings show that the financial cycle leads the business cycle.
The financial cycle is defined in terms of credit and property prices. The two variables co-vary closely with each other. This underlines the importance of credit in the financing of construction and purchase of property (Borio, 2014). Equity prices in general, is less correlated with the aforementioned variables. A feature of the financial cycle is that it has a much lower frequency than that of the business cycle (Drehmann et al., 2012). The business cycle, with an average frequency of about 1 to 8 years, is
6 significantly different from the average financial cycle frequency of 16 years3. The figure below from
Drehmann et al. (2012) shows this difference for the United States. The figure illustrates the high correlation between the financial cycle (blue curve) and financial crises (orange bars). Crises indeed coincide with peaks of the financial cycle, mostly followed by recessions (grey bars). Recessions, which occur at a more frequent basis, seems to be more correlated with the business cycle (red curve). This depicts the fundamental difference between financial crises and recessions as well as illustrate the different dynamics that play a significant role in crises occurrence.
Figure 1: The Business and Financial Cycle
Note: Financial cycle is measured by the combined behaviour of credit, credit to GDP ratio and house prices. Source: Drehmann et al. (2012).
From Borio (2014), financial liberalisation is named to weaken financing constraints and alter perceptions of risk and value. The financial regime in place thus influences the financial cycle. Consequently, major positive supply side developments provide fuel for financial booms and subsequently raise growth potential, thereby offering increased opportunities for credit and asset price booms. Bernanke and Gertler (1989) illustrate the consequences of this cycle. High asset values in booms following a period of credit expansion induces lenders to ease credit terms which further prolongs the expansion. After losses are incurred the reverse of the situation manifests, credit is rationed and interest rates begin to rise. The boom, as result, gradually develops into a bust, with a full-blown financial crisis as a result. Credit thus plays a central role in the occurrence of financial crises. The role of banks comes into play when crises are put in this perspective. Another finding from Borio (2014), that is highly relevant for this study, is that the financial cycle permits the identification of the risks of future financial crises in real time and with a good lead, with credit again being an important part of this.
2.2 Predictive capabilities of stocks and the role of financial institutions
As Eichengreen and Arteta (2002) put it, the conventional answer to the question: Could a sudden decline in stock prices indicate an economic downturn like that of the Great Depression is “yes but only if allowed to engulf the banking system.” The instability of the banking system is that which distinguishes economic crises from ordinary recessions. In fact, as the authors notice, the instability of
7 the banking system is a way of understanding the more pronounced business cycles in the later parts of the 20th century.
Throughout this document, the argument has been made that banks play a highly significant role in the occurrence of banking crises. Remarkably, the role of banks has not been made explicit in the banking crisis prediction literature. A notable exception is Barrell et al. (2010), who include bank balance sheet variables as predictors and find these variables to be significant predictors. The next subsection summarizes the literature on the role of banks in the propagation of banking crises and how bank stock could be used as an early signaling device for crisis.
2.2.1. On the role of banks
A number of studies have been conducted on the effects of banks on economic growth, of which most conclude a strong effect of the financial sector on the economy. King and Levine (1993) for example, find banking sector variables in the form of the percentage of credit allocated by banks to private firms, the importance of banks relative to the central bank and the size of the financial sector relative to GDP to be strongly correlated to economic growth. King and Levine (1993) argue their findings to be in line with the view that financial services stimulate economic growth. In fact, the size and level of significance of the coefficients in their findings is suggestive of a causal relationship between financial development and growth, although causality is not explicitly concluded in their analysis. Another interesting finding in their paper is that financial development seems to lead economic growth. Consequently, shocks to financial development, of which banks play a central role, can be expected to affect economic growth, both positively and negatively.
A follow up to this study by Levine and Zervos (1998) suggest that since measures of stock market liquidity and banking development positively and significantly enter into growth regressions, banks provided different financial services from those provided by stock markets. There is therefore a need to understand the relationship between the financial system and long-run economic growth, as this is an integral part of the growth process. The role of the banking sector on the growth of the economy is well emphasized in this study. This finding is in stark contrast with the ‘money view’ that expresses the role of banks in credit creation to be of no importance (Taylor and Schularick, 2012).
The findings of King and Levine are confirmed in the paper of Drehmann et al (2014), who also find the financial cycle to lead the business cycle. Borio (2014) follows on these findings and show the peaks of the financial cycle to be closely related with systemic banking crises. In his sample of seven industrialized countries, all the financial crises that originated domestically occur at, or close to the peak of the financial cycle. The reverse of the phenomena is also true, that most financial cycle peaks coincide with banking crises.
The Nordic crisis of 1991-1992 that affected Norway, Sweden and Finland, illustrates the role of banks in the occurrence of financial crises due to shocks in financial development. The crucial role the banks played can be seen in Table 2. Erling Steigum4 notes that financial deregulation in the lead
up to the banking crisis changed the competitive environment of the credit market, releasing an aggressive form of competition for market shares in the credit market. This led to an expansionary lending behavior of banks, creating strong incentives to originate new loans at the expense of a breakdown of internal control and credit evaluation. The different banking systems in the three nations led to different results in terms of the severity of the crisis experienced by each nation. Norwegian
8 banks, Steigum argues, were poorly capitalized at the moment of deregulation since the regulatory regime in place meant the banks could replace equity with subordinated loan capital. Deregulation had further reduced capital requirement from 10 to 6.5 percent in 1985. Surprising, he finds, is the absence of any form of worry from top management of banks about the risks involved in the aggressive growth strategies their banks had adopted.
Table 2: Bank loan market shares of commercial banks in Norway, Sweden and Finland (loans in percent of total year-end assets)
Year Norway Sweden Finland
1980 56.5 66.3 55.4
1985 57.8 71.7 58.8
1990 59.3 72.9 66.6
1995 58.8 93.2 69.8
Source: The Great Financial Crisis in Finland and Sweden: The Nordic Experience of Financial Liberalization
This is in line with the view of Stiglitz (1972) that the amount of risk that bank managers choose to take is likely to exceed that which is socially optimal because of limited liability (Stiglitz, 1972). A combination of limited liability and deposit insurance may have contributed significantly to the risk-seeking behaviour of banks. Indeed, many of the banks that adopted an aggressive growth strategy needed support from the deposit insurance scheme once the crisis emerged. Almost all of the banks in Norway, Sweden and Finland went bankrupt when the financial crisis inevitably unfolded (Kindleberger and Aliber, 2011). The data of Laeven and Valencia (2012) shows a 69.6 percent loss of output (in percent of GDP) in Finland. This was 32.9 and 5.1 percent for Sweden and Norway respectively.
On the subject of deposit insurance and market discipline, Demirgüç-Kunt and Huizinga (2004) find a negative relation between the two. In their view, financial safety nets in the form of deposit insurance which is intended to reduce vulnerability of the banking system, have had the opposite effect. Kane (1989) accredit the Savings and Loans crisis mentioned earlier in its entirety to the deposit insurance scheme existent at the time. He recognizes that such schemes cut the link between the riskiness of the assets of an institution (a bank in this case) and its capacity to raise funds from these assets. Moreover, as he explains, deposit insurance is not insurance at all in the strict sense of the word. Guarantees are not written against a specified set of risks that could potentially destroy the financial viability of a bank, risks that cannot be calculated upfront. It is essentially an unconditional guarantee from the state to repay a particular class of the banks debts since the contract does not limit the set of unfavourable events to which the bank is exposed. This induces banks to take on socially inappropriate forms of risk. This excessive risk taking by banks has been the root cause of many of the bank failures according to Demirgüç-Kunt and Huizinga (2004).
A case study by Temin and Voth (2004) of the South Sea bubble of 1720 shows that Hoare’s Bank knowingly rode the bubble and profited significantly from it. The study also highlights that limitations to arbitrage and short-selling constraints cannot explain the bubble of 1720. In the rational bubbles literature, entry of noise traders makes it optimal for informed traders (who could be banks in this case) to hold on to stocks that they know are overvalued. This thus encourages informed traders to ride the bubble and not attack the bubble (Temin and Voth, 2004). Abreu and Brunnermeier
9 (2003) argue that informed traders despite their awareness of the inevitable collapse of the bubble, would like to ride it for as long as it continues to grow and generate high returns. Ultimately the idea is to exit the market precisely before it crashes, although market timing proves difficult to realize. As every trader tackles the timing differently, there is a lack of synchronized exit strategies, such that the bubble is permitted to grow until it finally collapses when sufficient traders exit.
Thakor (2015) provides an interesting perspective on the psychological dimensions that fuel the repeated occurrence of financial crises. In his view, crises are caused by agents’ beliefs that outcomes are influenced by the a priori unknown skills of banks. As a consequence, good outcomes lead to an upward revision in beliefs about bankers’ skills, wherefore a long sequence of good outcomes lead all agents (banks, investors and regulators alike) to believe that banks are capable of managing their risks. As a result, banks and investors underestimate the true risk of high-risk products. A side effect of this is that other institutions underestimate these risks as well, and invest in the high-risk products. Essentially, as Thakor further explains, high-risk is mispriced because high-risk-management capabilities of banks is overestimated, and increased investment of banks’ loans provides increased liquidity for high-risk assets. Ultimately, investors learn about the true risk in highly risky products, resulting in the dry up of liquidity and eventually in a crisis.
From the arguments provided in this subsection concerning the role of banks in the propagation of banking crisis, it becomes apparent that this role should be explicitly accounted for in the banking crisis prediction. A less crucial role of the banking sector means less costly banking crises (Taylor and Schularick, 2012). One way of doing this is the inclusion of bank stock in the prediction models. Section 2.2.2 explores the reason for this choice.
2.2.2. Stocks as leading indicators
Estrella and Mishkin (1998) find that stock prices are particularly useful in the prediction of recessions (in the U.S.). The authors examine out-of-sample performance of, among other financial variables, stock prices as a predictor of U.S. recessions and find stock prices to contain useful information for predictive purposes. They conclude that among the many variables observed, stock prices is among the most useful financial variables and can be used to supplement forecast models as a reliable check of more extensive forecasts.
Since banks are central in the occurrence of financial crises, it is reasonable to assume that developments around bank stock prices may contain information of looming financial crises. Reinhart and Rogoff (2008) find that the run-up in equity and housing prices, which serve as the best leading indicator for financial crises, closely tracks the average of earlier crises. In their examination of 18 post-war banking crises in industrialized countries, they find qualitative and quantitative similarities between the considered crises. The authors hence argue that the 2007-2008 financial crisis is not unique. This suggests that equity prices may contain information for forecasting financial crises if crises indeed are not unique. Roy and Kemme (2011) in their research into commonalities in the run-up to financial crises also find asset bubbles to be the most common precursors.
One could argue that if crises have similar characteristics, then they should be predictable and the materialization of any crisis can only be attributed to their self-fulfilling nature. Kaminsky and Reinhart (1999) reject this view. In their analysis of a multitude of financial crises the authors identify a number of weak and deteriorating economic fundamentals (including stock prices) that contribute to crises across several countries (both emerging and developing nations). Financial liberalization or increased access to international capital markets appears to activate the boom-bust cycle that
10 eventually ends in a financial crisis in their view. Recognizing the important role banks play in the occurrence of financial crises, they advise for strong banking regulation and supervision.
Davis and Karim (2008) make use of the binary recursive tree (BRT) and logit models to predict the sub-prime crisis. Making use of macroeconomic variables (see Table 1) as in Demirgüç-Kunt and Detragiache (2005) and Kaminsky and Reinhart (1999), the results show only a marginal predictability of the crisis. Adaptation of the models for the specific features of advanced countries to include aspects of the securities market instability is suggested by the authors as a method of making the models a useful supplement for macroprudential analysis. A variable with the highest potential for this purpose is argued in this study to be bank stocks. This is also recognized by Kaminsky and Reinhart (1999) in their analysis concerning twin crisis in emerging markets. They find that an important number of banks are not traded publicly in developing countries, such that bank stock cannot be used as an indicator for banking crisis. In this study, this problem is circumvented since only advanced OECD countries are included in the sample. Even so, admittedly, many of the nations in the sample have very few banks that are publicly traded, with the U.S. being an exception.
A question that often arises when using stock prices in crisis prediction is; if investors are aware that stock prices could indicate an imminent crisis, that is, contain information on fundamentals, should this then not be priced? Pricing this information entails that stock prices consistently represent an accurate valuation of underlying fundamentals, much in the spirit of the efficient market hypothesis. The answer to this question is not straightforward. As Brunnermeier and Nagel (2004) explain, there are two opposing views on this. Proponents of the efficient markets hypothesis (example Friedman, 1953 and Fama, 1965) argue that speculative activity should eliminate risk-free arbitrage opportunities and other forms of mispricing. This would therefore ensure that prices reflect fundamentals at all times. However, literature on the limits to arbitrage show that various factors may constrain5 arbitrageurs from taking advantage of arbitrage opportunities, which allows for the
persistence of mispricing. Other models indicate that rational investors might find it optimal to “ride bubbles” for a while before attacking them, thereby further destabilizing the bubble as has been shown by Temin and Voth (2004) and Brunnermeier and Abreu (2003).
Until recently, the consensus has been that asset price developments should not have any far reaching influences for monetary policy, except up to the degree that it should influence the inflation forecast of the central bank (Taylor and Schularick, 2012). Taylor and Schularick (2012) recognize that recent events have highlighted the need for better assessment of asset bubbles. The argument put forth is that if asset price booms relax collateral constraints which in turn fuel more lending and higher asset prices, rising assets prices could be indicative of elevated risks to financial stability. To disentangle the effect of asset price booms on crisis probability, the authors include the stock price data (in the form of stock market indices) in their regressions. The log of real stock prices proves to be a significant predictor of crisis.
The above analysis strongly suggests that stock prices may contain relevant information for the prediction of financial crises. Borio and Lowe (2002) observe that despite the importance of asset price developments, they have received little attention in empirical literature on the determinants of financial crises. This study aims to achieve this through the inclusion of bank stock, which is argued to be a better predictor. Stock prices in the form of market indices may be influenced by general economic trends which may shroud the influence of banks.
5Beber and Pagano (2013) for example, find that short-selling bans constrained arbitrageurs and hence slowed price
discovery in stock markets in almost all countries (with U.S. being an exception) who introduced a short-selling ban during the 2007-09 sub-prime crisis.
11
III Data
The objective of this study is to test the hypothesis that bank stock predicts banking crises, or at least improves prediction results. As motivated above, balance sheet variables of financial institutions (in particular banks) contain information for the prediction of banking crises (Barrell et al., 2010). Developments around these variables should be reflected in the stock prices of banks, such that stock prices should reveal information of an impending crisis, as is implied by the efficient market hypothesis. This can be achieved through the channel of decomposing the developments around bank stock in the lead up to a financial crisis. This chapter explores the data that is used in this thesis. The data is constructed from several databases and span the 1980-2011 period.
3.1 Banking crisis variable
Following Barrell et al. (2010), the banking crisis dependent variable is defined as a binary dummy that takes the value of 1 when a banking crisis took place in a particular year and 0 otherwise. Data on banking crisis is retrieved from the IMF Financial Crisis Episodes database which covers 147 financial crises between 1970 and 2011, and the World Bank database of banking crises. In their publication: “Systemic banking crises database: An update”, Laeven and Valencia (2012) update their database on systemic banking crises from 2008. The updated database contains data on recent crises episodes, including data on the recent financial crises. Most of the systemic banking crisis in the sample, 62.5 percent to be precise, is dated in the 2007-2008 period, at the height of the sub-prime crisis (see Table 3 below). The data also contains information on output loss, fiscal costs and monetary expansion among others. Another important variable in the database is the credit boom dummy, which measures whether a crisis was preceded by a credit boom.
An important factor is the definition and measurement of the banking crisis variable. A systemic banking crisis, according to the authors, is the situation where there are significant signs of financial distress in the banking system and significant banking policy intervention measures in response to significant losses in the banking system. The first year in which these conditions are met is then considered the year in which the crisis became systemic. Based on this definition 147 crises episodes are measured across 162 countries.
Of the 34 OECD nations in the sample, 31 experienced at least one episode of systemic financial crisis between 1980 and 2011, which is the sample period chosen in this paper. Only Australia, Canada and New Zealand did not experience a financial crisis in this period. A sample of 14 OECD countries (see Table 3 below) are selected on basis of relative comparability of the nations in terms of the macro-economic conditions and the nature of the banking sector.
This thesis differs from Barrell et al. (2008) and other financial crisis prediction literature in terms of the nature of the crises included in the analysis, which typically has focused their analysis on systemic crises only. Barrel et al. (2010) for example, terminate their estimation before the sub-prime episode to control for possible endogeneity problems that may arise from the effects from ongoing crises on the explanatory variables. In this thesis however, non-systemic crises are also included in estimations in order to highlight the full measure of the effect of banks in the occurrence of banking crises in all forms. To mitigate the effects that ongoing crises may have on the explanatory variables, variables are lagged by one year.
A total of 16 systemic and 99 non-systemic crises episodes are recorded for the period across the fourteen countries in the sample. The World Bank database of banking crises provides data on
non-12 systemic crises over the chosen sample period. Joining the two databases on banking crises provides the matrix of crises as given in Table 4. A period of crisis is marked 1, with the bold observations marking systemic crisis as defined by Laeven and Valencia (2012). End dates for the Japanese crisis that started in 1991 and the UK crisis of the 1980s and 1990s are not given. Instead, for Japan, the end date is given to be 2002, this being the last year of data in the database. For the UK, the end date is given as 1999, following the description of the crisis to have occurred in the 1980s and 1990s. The credit boom dummy specifies whether there is a credit boom preceding a particular crisis and only applies to the systemic crises. From Dell’Ariccia et al. (2012), the credit boom variable is defined as the years during which the deviation of credit-to-GDP ratio to its trend is greater than 1.5 times its historical standard deviation and its annual rate of growth is beyond 10 percent, or in the years during which the yearly growth rate of credit-to-GDP ratio exceeds 20 percent. Only 5 of the 16 systemic crises in the sample were preceded by a credit boom. Surprisingly perhaps, the 2007 banking crisis of the US was not preceded by a credit boom, at least not according to the formal definition of a credit boom as giving by Laeven and Valencia (2012) and Dell’Ariccia et al. (2012). Unlike previous literature on the subject, this study includes the credit boom in the form of a dummy in order to measure the marginal increase in crisis probability resulting from a country experiencing a credit boom period.
Table 3: Dating of Systemic Crises Country Date of systemic crisis
Belgium 2008 Denmark 2008 Finland 1991 France 2008 Germany 2008 Italy 2008 Japan 1997 Netherlands 2008 Norway 1991 Portugal 2008 Spain 2008 Sweden 1991, 2008 United Kingdom 2007 United States 1988, 2007
3.1.1 Dating and measurement of banking crisis
In identification of the starting dates of systemic crises Laeven and Valencia (2012) identify two conditions6 that must be met;
Significant signs of financial distress in the banking system (as indicated by significant bank runs, losses in the banking system, and/or bank liquidations).
Significant banking policy intervention measures in response to significant losses in the banking system.
13 Table 4: List of Systemic and Non-systemic Banking Crises.
BE DE FI FR GE IT JP NL NW PT SP SW UK US 1985 0 0 0 0 0 0 0 0 0 0 1 0 1 0 1986 0 0 0 0 0 0 0 0 0 0 0 0 1 0 1987 0 1 0 0 0 0 0 0 0 0 0 0 1 0 1988 0 1 0 0 0 0 0 0 0 0 0 0 1 1 1989 0 1 0 0 0 0 0 0 0 0 0 0 1 1 1990 0 1 0 0 0 1 0 0 1 0 0 0 1 1 1991 0 1 1 0 0 1 1 0 1 0 0 1 1 1 1992 0 1 1 0 0 1 1 0 1 0 0 1 1 0 1993 0 0 1 0 0 1 1 0 1 0 0 1 1 0 1994 0 0 1 1 0 1 1 0 0 0 0 1 1 0 1995 0 0 1 1 0 1 1 0 0 0 0 0 1 0 1996 0 0 0 0 0 0 1 0 0 0 0 0 1 0 1997 0 0 0 0 0 0 1 0 0 0 0 0 1 0 1998 0 0 0 0 0 0 1 0 0 0 0 0 1 0 1999 0 0 0 0 0 0 1 0 0 0 0 0 1 0 2000 0 0 0 0 0 0 1 0 0 0 0 0 0 0 2001 0 0 0 0 0 0 1 0 0 0 0 0 0 0 2002 0 0 0 0 0 0 1 0 0 0 0 0 0 0 2003 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2004 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2005 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2006 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2007 0 0 0 0 0 0 0 0 0 0 0 0 1 1 2008 1 1 0 1 1 1 0 1 0 1 1 1 1 1 2009 1 1 0 1 1 1 0 1 0 1 1 1 1 1 2010 1 1 0 1 1 1 0 1 0 1 1 1 1 1 2011 1 1 0 1 1 1 0 1 0 1 1 1 1 1 Credit Boom 1 0 1 0 0 0 0 0 0 0 1 17 1 0
Note: BG – Belgium, DE – Denmark, FI – Finland, FR – France, GE – Germany, IT – Italy, JP – Japan, NL – Netherlands, NW – Norway, PT – Portugal, SP – Spain, SW – Sweden, UK– United Kingdom, US – USA. Only the period of 1985-2011 is shown in the table.
Significant policy interventions in the banking sector must meet at least three of the following six measures are used; (i) extensive liquidity support, which is considered extensive when the ratio of central bank claims on the financial sector to foreign liabilities exceeds 5 percent and more than twice its pre-crisis level, (ii) bank restructuring costs of at least 3 percent of GDP are incurred, (iii) significant bank nationalizations, (iv) significant government guarantees to financial sector are put in place, (v) financial sector asset purchases of at least 5 percent of GDP, and (vi) deposit freezes and/or bank holidays. The first year in which both conditions are met is considered to be the year in which the crisis is systemic (see Table 3 for dates of systemic crises). An important source of variation to identify the effects of a systemic crisis is the differences in the starting dates of crises across the countries in the sample. In absence of high frequency data that capture the period in which a crisis begins, Kaminsky and Reinhart (1999) also employ this methodology in dating banking crises. The authors recognize that banking crises are usually initiated by deterioration in asset quality, such that asset price changes or
14 large increases in bankruptcies or NPL’s could be used to mark the beginning of the dates. However, as already mentioned, stock market data was not available to serve this purpose in their analysis.
Barrell et al. (2010) who also use the above mentioned databases for their analyses on banking crisis, identify that end dates of the crises are subjectively chosen to some extent, thereby introducing potential endogeneity problems in the estimation. This is especially relevant for crises that are still ongoing at the end of the sample period. The solution the authors use to mitigate this is the termination of the estimation results before the sub-prime crisis. Contrary to their analysis, in this study, the end dates for crises ongoing as of 2011 (then end year of the data) are terminated in that year. Another issue identified concerning dating of the crises is the timing in terms of annual dummies. A crisis starting in December 2000 has the value of 1, representing the entire year of 2000, and 0 in 2001. The authors contend with one year duration of crises as further analysis show robust results even after relaxation of this assumption.
3.1.2 Crisis frequency and descriptives
Table 5 below presents the frequency of crises and the unconditional probability of the occurrence of a crisis per decade in the sample. Results from 2007 to 2011, the period of the most recent financial crisis, is included in the table as most crises (40 percent) are concentrated in this 5 year period. The unconditional probability of a crisis occurring in a particular period is simply the total number of crises in that period divided by the number of observations of that period.
Table 5: Crisis Frequency and Unconditional Probability
Period Crisis frequency Crisis frequency (percent) Unconditional probability of a crisis (percent) 1980-2011 114 28.85 25.7 80s 21 19.61 15.0 90s 45 33.40 32.1 00s 49 37.51 29.2 2007-2011 46 82.84 65.7
Note: 00s include 2011, representing 12 years.
There are a total of 114 banking crises occurring between 1980 and 2011, which represents a crisis frequency of 28.85 percent, and a frequency of 3.57 percent of frequency for systemic crises, which is slightly above the 3.2 percent in Barrell et al. (2010). This slight difference in systemic crises frequency comes from the extension of the sample period to include the sub-prime episode of 2008. A clear pattern of increasing probability of the occurrence of banking crises emerges from Table 5. Over the entire sample period, the probability that a banking crisis will occur in any giving year is 25.7 percent. In the 80s however, the probability of a crisis is 15 percent, increasing to 32 percent in the 90s and dropping only slightly to 29 percent in the 12 years that followed. The high frequency of crises in the 80s and 90s have been largely due to the UK and Japanese crises in these decades. This highlights an uneven distribution of crises over the periods prior to the sub-prime financial crisis. Almost all of the 14 nations in the sample experienced banking crisis episodes in the sub-prime period. The probability of a sovereign in the sample experiencing a crisis in a year in this period is 65.7 percent, highlighting the severity of the sub-prime episode and perhaps the importance of the level of contagion in the banking system.
15
3.2 Bank stock variable
Data on banks is retrieved from the Bankscope and Datastream databases. Bankscope provides a list of publicly listed banks (both active and non-active) in the countries in the sample. All banks with an IPO date beyond the end of the sample period are dropped. Bankscope data on stock prices goes only as far back as 2003. Therefore, the ISIN numbers of the banks are run through Datastream, which provides stock price data from 1973. From this a final sample of 1078 banks across the fourteen countries remain, of which the majority are US banks (see Table A in the Appendix for information on number of banks per country). Not all banks were listed at the beginning of the sample period. This entails that the estimation suffers from missing data issues, especially in the initial periods of the estimation. This also means that nations with too few banks may have too few observations in the years where banks were not listed to estimate significant effects for these countries. However, the estimation makes use of the variation in the available data to estimate the effects, such that the effects estimated are representative of the bank stock price conditions within the sample period.
Datastream provides yearly stock price data measured as the closing price at year end. This price however, does not fully reflect the development of the stock prices throughout the year. As a result, daily stock prices is retrieved for all banks that remain in the sample. Yearly returns are generated by calculating a yearly average from daily prices and calculating the log changes in the yearly prices. This procedure takes the stock price movements over the entire year into account and thus controls for any seasonal effects that may be present in bank stock prices.
3.2.1 Bank stock descriptives
Figure 1 presents the average bank stock return of the banks in the sample. The light grey bars indicate systemic banking crises in 1991 and 2008 respectively. The systemic crises of 1991 refers to the Nordic crisis mentioned in subsection 2.2.1. Notable is that bank stock return tend to rise significantly in the period before a systemic banking crisis. This is consistent with the description given by Kindleberger and Aliber (2011), describing an episode of surging stock prices in the preiod preceding both crises highlighted here. Eleven of the fourteen nations experienced a systemic crisis in the 2007-2008 sub-prime episode, and the figure consistently illustrates rising stock returns in the period preceding this crisis as well, albeit this may be dominated by U.S. banks.
Naturally, stock prices drop once crises begin. Figure A of the Appendix provide similar graphs for the individual countries. The trend is the same for all countries, namely, rising stock prices before crises become systemic, thereby illustrating that the effect that can be seen in Figure 1 is not dominated by U.S. banks. In almost all cases, the peak of these increases tend to be associated with the systemic crises, whose occurrence is followed by sharp drops in stock returns.
It is worth noting that not all increases in bank stock results in systemic banking crisis.For example, the rise in US and German bank stock returns between 1996 and 2000 is not associated with any particular banking crisis, systemic or otherwise. However, the period coincides with the Dotcom bubble of the late 90s. Evidence show that institutional traders (including banks) were amongst the largest buyers of stocks of tech firms8. This induced these firms to profit from the rise of internet stock
and likewise the fall. This is important to note, primarily because stock prices are subject to changes in the economy at large. Even so, the main takeaway from the graph is the suggestive evidence that increases in bank stock tend to be associated with systemic banking crisis. From Table 6 of the summary
16 statistics, the average bank stock return is 0.16 percent with a staggering volatility of about 36 percent, much higher and more volaltile than the average stock market return of -4 percent and corresponding volatility of 25 percent. This also shows that bank stock return and the stock market return are not necessarily interchangeable. The market return variable is included in the regressions to formally test this relation. Figure B of the Appendix plots the average market return, which seems to show a comparable trend to market return. The two variables are also highly correlated with a correlation coefficient of 0.537. Market return is calculated from the log change in market capitalization of all listed domestic firms per country. The data is retrieved from the World Development Indicators (WDI) database of the World Bank.
Figure 1: Average Bank Stock Return
Table 6 also contrasts the pre-Dotcom bubble period of 1980-1999 with the 2000-2011 period. Bank stock return of the latter period exceeds that of the former greatly at 4.62 percent compared to -9.77 percent. Even though both periods are not directly comparable due to the different number of banks in the different periods, this vast difference in bank stock return shows the high profitability of banks in the 2000-2011 period. Furthermore, this period is charectirized by financial innovation in the banking sector in the form of securitization, which has been termed to be one of the most important causes of the sub-prime finacial crises (Kindleberger and Aliber, 2011). This result emphasizes the usefulness of bank stock as a predictor of systemic banking distress.
17
3.3 Macroeconomic and other explanatory variables
To align this study with previous research on the subject, several macroeconomic and banking sector variables that have been found to be correlated with banking crises are included in the regressions. Crises have been found to occur in periods of low GDP growth, high inflation and high real interest rates (Demirgüç-Kunt and Detragiache, 2005 and El-Shagi et al., 2013). The M2 money multiplier, which serves as a proxy for capital inflows (Demirgüç-Kunt and Detragiache, 1997) and in some studies a proxy for increase in credit in the economy (Barrell et al., 2010) has been found to be correlated to crises as well. The ratio of government deficit to GDP has also been found to be a significant predictor of crises as well, especially in periods when the deficit is significantly large (Demirgüç-Kunt and Detragiache, 2005). The house price index is also included since it has been argued to significantly impact crisis likelihood (Barrell et al., 2010). Further identified to be an important crises predictor is the amount of credit to private sector by banks (Laeven and Valencia, 2012 and Borio and Lowe, 2002). Table 6: Summary Statistics
Overall Sample 1980-1999 2000-2010
Variable Mean S.D. N Mean S.D. N Mean S.D. N
Return 0.002 0.358 15,344 -0.0977 0.270 4,756 0.0462 0.383 10,588 Market Return -0.040 0.251 32,432 -0.073 0.236 19,852 0.011 0.264 12,580 Inflation 3.667 3.19 29,443 4.5700 3.561 18,663 2.1034 1.388 10,780 Interest 5.109 2.24 29,475 5.840 2.003 19,394 3.7019 1.979 10,081 Credit 70.891 38.43 29,828 65.7728 36.825 19,362 80.3581 39.522 10,466 Deficit -3.028 3.71 17,892 -2.5914 2.163 8,345 -3.4096 4.625 9,547 M2 33.414 88.64 34,291 35.6133 95.065 21,451 29.7393 76.592 12,840 GDP growth 2.501 2.17 30,184 2.8664 2.009 19,404 1.8422 2.289 10,780 HPI 116.92 60.40 34,469 81.8139 21.476 21,533 175.352 59.047 12,936 HHI 5.437 164.82 34,496 8.5833 208.42 21,560 0.1930 0.545 12,936
Notes: Return denotes bank stock return, market return the stock market return, interest the real rate of interest, credit the credit to private sector by banks ratio to GDP, HPI the house price index and HHI the Hirschmann-Herfindahl Index (constructed in this study). All values are lagged by one period with the exception of HHI. Mean denotes the average, S.D. the standard deviation and N the number of observations. Return and Market return are measured in percentages, whilst inflation, interest, credit, deficit, M2, GDP growth are measured in percentage points. HPI is measured as an index, with 2005 as base year.
Real GDP growth data is retrieved from the International Financial Statistics (IFS) database of the IMF. The real interest rate, inflation (as measured by consumer prices), credit and deficit are retrieved from the WDI database. Credit is measured as the weighted average domestic credit to private sector by depository institutions (except the central banks) such as through loans, trade credits, purchase of nonequity securities, etc. From Table 6, the average credit to GDP ratio has an average of about 71 percent. This figure is higher in the 2000-2010 period, highlighting the increase of credit in this period, and perhaps the subsequent credit booms that preceded the 2007-2008 crises.
M2 is measured as the ratio of the M2 money multiplier to GDP. For the Euro Area countries, neither the World Bank nor IMF provide data on the M2 money multiplier for the individual countries. As such, the M2 variable is allocated to the Euro Area countries on basis of each country’s GDP. Euro
18 Area M2 data is retrieved from the Statistical Data Warehouse database of the European Central Bank. For the non-Euro Area countries, M2 data is retrieved from the IFS database. GDP data for all countries comes from the IFS database as well. Comparing the M2 variable in the two comparison periods in Table 6, there seems to be a decrease in the variable moving from the 1980-1999 to the 2000-2011 period. In the same period, credit to GDP and the real growth rate of GDP were substantially lower. The two seemingly opposing effects may be indicative of the cross border influx of money that had increased the amount of credit that banks could lend to the private sector which may have contributed to the instability of the financial sector as narrated by Kindleberger and Aliber (2011) and Reinhart and Rogoff (2008).
The house price index (HPI) data is retrieved from the Bank for International Settlements (BIS) database. The quarterly residential property price index data provided by the BIS is collapsed to yearly indexes with 2005 as base year. The average HPI of about 117 is dominated by the enormous increase in property price growth in the 2000-2011 period. This period naturally coincides with the rise in the growth rate of credit, which seems to confirm the findings of Borio (2014) that the covariance between credit and property price growth underpins the importance of credit for the purchase of property.
To measure the effect bank concentration on crisis, a variable (HHI) in the spirit of the Hirschman-Herfindahl index (Hirschman 1964) is included in the analysis. This variable is estimated by dividing the total assets of individual banks in the sample by the total assets of all banks in the country (both asset types retrieved from the Bankscope database) to obtain the market shares for the banks. This market share value is then squared summed to obtain an index for each country. An increase in this variable shows an increase in banking sector concentration of listed firms in the country. The low number of banks in the sample for the 1980-2011 period has produced a much higher HHI value for this period in comparison with the 2000-2011 period, where most of the bank data is concentrated, as can be seen in Table 6.
This study also follows the analysis by Barrell et al. (2008), arguing deposit insurance schemes and GDP per capita among other variables to be irrelevant as they are broadly comparable amongst the OECD countries in the sample. As such these variables are excluded from the regressions.
IV Methodology
This section explores the methodology employed in analyzing the predictive capabilities of bank stock for banking crisis. Table 1 presents an overview of literature on financial crises and the methodology used in estimation of predictive properties of the variables used. This chapter introduces a brief discussion of these models and describes the methodology this thesis uses in answering the main question. The hypotheses and robustness measures are also discussed.
4.1 Popular models in crisis prediction literature
Literature on financial crisis prediction employ varying models as has become apparent in Table 1. The IMF, as well as several financial institutions (including the Bundesbank), has developed early warning system (EWS) models with the aim of anticipating potential financial crises in individual nations (Bussiere and Fratscher, 2006), underlining the significance of early signaling and detection of financial crises. Most research in the prediction of financial crises literature hence make use of such EWS models. These models typically make use of relevant economic variables, some of which have been
19 extensively discussed in Section III, in detection of economic vulnerabilities that may lead to a crisis. EWS models come in different forms, ranging from signal extraction models (for example Kaminsky & Reinhart, 1999 and El-Shagi et al., 2013) to panel logit models (an example is Davis & Karim, 2008 and Barrell et al., 2010).
4.1.1. The signals approach
Kaminsky and Reinhart (1999) develop the signals approach in their analysis concerning twin crises. This methodology observes the behavior of each of the relevant variables in the model during the 24 months preceding a currency crisis. For a banking crisis this period is 12 months. In the case of a currency crisis, a variable in this model correctly signals an impending crisis any time it crosses a particular pre-specified threshold and a crisis occurs within 24 months after the signal is given. For banking crisis, a good signal is given either 12 months before the crisis or within 12 months after the beginning of the crisis. In any other case the signal is deemed to be a false alarm. The different signaling windows for the 2 crises is attributed to the different timing of the peaks of either crises type. The quality of the signal is assessed through evaluation of the type I (the probability of rejecting the null hypothesis when it is true) and type II errors (not rejecting the null hypothesis when it is false) along with the conditional probability of the occurrence of a crisis after a signal has been issued.
Appreciation of the real exchange rate, equity prices and the money multiplier are found to have the lowest noise-to-signal ratio and the highest probability of predicting a crisis. These variables, however, predict with a large type I error, failing to give a signal in about 79 percent of the observations during the 24 months preceding a crisis. Type II error, on the other hand, is at a much lower 8 to 9 percent.
In their application of the signals approach to three crisis types, El-Shagi et al. (2013) confirm the results of Kaminsky and Reinhart (2013), and further argue its usefulness and success in crisis prediction. The authors also emphasize the advantages of the signals approach. In particular, the signals approach provides both a composite crisis indicator and offers a more detailed picture of the individual variables. It enables policy makers to identify relevant areas pertaining to the variables where action can be taken if needed.
With the intended approach of this thesis, the signals approach seems to be the best method of obtaining the required results. In particular, it offers the ability to observe the behavior of the relevant variables in isolation. This is especially fitting for this analysis as the added value of including bank stock in the crisis prediction models is the aim of this research. However, as Demirgüç-Kunt and Detragiache (2005) also note, the signals approach fails to aggregate the information from the relevant variables such that important information may be lost in the process. Demirgüç-Kunt and Detragiache (2000) propose a different approach, which according to the authors, remedies some of the problems in the signals approach, namely the multivariate logit model. This model has also been employed by Davis and Karim (2008), Demirgüç-Kunt and Detragiache (1998), Demirgüç-Kunt and Detragiache (2005) and Taylor and Schularick (2012). Due to the ease of application, this study also employs the logit methodology, which is explained in the next subsection, in evaluating banking crisis prediction.
4.1.2. The logistic approach
Barrell et al. (2010) and Demirgüç-Kunt and Detragiache (2005) use the cumulative logistic distribution as their estimator in analyzing banking crises. The regression model specified is;
