Financial Crisis, Risk-taking and Early Warning Systems

UNIVERSITY OF SANTIAGO DE COMPOSTELA DEPARTMENT OF FINANCIAL ECONOMICS AND ACCOUNTING UNIVERSIDAD DE SANTIAGO DE COMPOSTELA DEPARTAMENTO DE ECONOMÍA FINANCIERA Y CONTABILIDAD Financial Crisis, Risk-taking and Early Warning Systems”. Crisis Financiera, Exposición al Riesgo y Sistema de Alerta Temprana Dildora Karimjonovna Ibragimova Santiago de Compostela 2014 UNIVERSIDAD DE SANTIAGO DE COMPOSTELA DEPARTAMENTO DE ECONOMÍA FINANCIERA Y CONTABILIDAD Financial Crisis, Risk-taking and Early Warning Systems”. Crisis Financiera, Exposición al Riesgo y Sistema de Alerta Temprana Tesis que, para la obtención del grado de doctor, presenta la licenciada Dña. Dildora Ibragimova, la cual fue realizada en el Departamento de Economía Financiera y Contabilidad de la Universidad de Santiago de Compostela, bajo la dirección de don Luis Alberto Otero González, profesores titulares de Economía Financiera y Contabilidad y don José Antonio Redondo López, Catedrático de Economía Financeira de la Universidad de Santiago de Compostela. Vº Bº D. Luis Alberto Otero González D. José Antonio Redondo López D. LUIS ALBERTO OTERO GONZÁLEZ, Profesor Titular del Departamento de Economía Financiera y Contabilidad y D. Jose Antonio Redondo López, Catedrático del Departamento de Economía Financiera y Contabilidad. CERTIFICAMOS que Dª Dildora Karimjonovna Ibragimova ha realizado bajo nuestra dirección el trabajo de investigación “Financial Crisis, Risktaking and Early Warning Systems”. Santiago de Compostela a de de 2014 Vº Bº D. Luis Alberto Otero González D. José Antonio Redondo López For my parents Muqaddamhon Abdullaeva and Karimjon Rahimov Acknowledgement This thesis wouldn’t be possible without the help of several people, some of whom I would like to mention here. I would like to express my greatest appreciation and thanks to Luis Alberto Otero González for the noble supervision he has provided for my work. I am grateful for his continuous guidance, thoughtful feedback on numerous drafts, for encouraging my research and for allowing me to grow as a research scientist. I also want to thank José Antonio Redondo López for his co-supervision of the thesis. This work would not be possible without the help of my supervisors. Also, I would like to give special acknowledgement to the European Commission and Erasmus Mundus Programme for the opportunity to complete my doctoral study in Spain and have such a wonderful international experience here. A very cordial acknowledgement should go to my parents back home. Their unconditional encouragement and support was undoubtedly the most important contribution in keeping up high spirits throughout my PhD journey. I would like to express special appreciation for my husband Dilshod Ibragimov for his personal support and great patience at all times. Thank you for sharing these years with me. I want to thank my children for the love and warmth they shared with me during my hardships in this endeavour. I should also mention the many people who kept the PhD process remarkable. Many thanks to my sister Sevara, friends Aizhan and Selbi: their moral support was very crucial to my motivation. Last, but not least, I would like to thank the administrative staff of Westminster International University in Tashkent and my colleagues there for the help and guidance received throughout my years of study. Abstract This thesis explores three important topics which contribute to the study of bank credit risk and the global financial stability. Chapter 2 “Financial Crisis: the case of Spain” focuses on the nature of the financial crises and their common characteristics. Also, it reviews the Spanish financial crisis and features that makes it distinctive. Chapter 3 “Determinants of bank excessive risk-taking behaviour” concentrates on the main reasons behind the financial crisis and empirically analyzes bank risk-taking factors for the three types of Spanish banks: commercial, saving and cooperative banks. Chapter 4 “Early Warning Model for European banks: evidence from the recent financial crisis” analyses empirically a sample of European listed banks and tests the effectiveness of Early Warning System (EWS), based on Expected Default Frequencies and accounting ratios, in forecasting the bank defaults during the recent financial crisis. The chapters are independent of each other in terms of theoretical grounding, dataset and methodology but complement each other by investigating the recent financial crisis from three different angles. Our study suggests that the Spanish financial crisis is not an exception to the general pattern of crises. The sequences of events evident preceding the crisis have many common features with what has already been witnessed in other financial crises. Furthermore, our findings indicate that there is a strong correlation between the bank’s ownership structure and its risk-taking behaviour. The results show positive association between risk and Spanish savings banks and negative with banks with dispersed ownership. We confirm the negative influence of wholesale funding, but not in favour of deposit funding. Instead, we find adverse effect of deposit funding on bank risk. We think this may be evidence of ‘excessive’ competition in deposit markets prior to the crisis when banks raise their deposit rates too high to attract more depositors by increasing their cost of funding and decreasing their interest margins. Also results show the negative influence of non-traditional income such as commissions and fee income. We find that equity has stable risk reducing character while impaired loans have a strong harmful effect of on banks’ risk level. Last chapter findings reveal that EDF metrics combined with four CAMEL covariates and variable capturing adverse selection are able to predict the defaults of European banks up to 8 quarters before an event. When comparing the final model with that only including the EDF indicator the significance improves considerably, suggesting that added variables provide additional information and power to the model. Overall, the thesis proposes preventive mechanisms to facilitate the early detection of banks’ fragility in Europe and hence contributes to the study of financial stability and the prevention of banking crises. 5 List of Tables Table 2-1 Balance of payments in Spain (2003-2011) ................................................................................. 16 Table 2-2 Expansion of deposit institution offices in Spain ...................................................................... 19 Table 2-3 Problematic real estate exposure (2010) ..................................................................................... 21 Table 2-4 Property risk coverage (2010) .......................................................................................................... 22 Table 3-1 Criteria of the search strategy ........................................................................................................... 66 Table 3-2 Dependent variables ............................................................................................................................. 67 Table 3-3 Variables and hypotheses considered ............................................................................................ 70 Table 3-4 Sample characteristics .......................................................................................................................... 71 Table 3-5 Descriptive statistics for dependent variables before and after crisis ............................. 72 Table 3-6 Mean and Standard deviation by bank types .............................................................................. 74 Table 3-7 Descriptive statistics for independent variables before and after the crisis .................. 75 Table 3-8 Correlation matrix between dependent and independent variables ................................ 76 Table 3-9 Baseline model estimations with dependent variable Z-score ............................................ 81 Table 3-10 Parsimonious Models with Z-score & with additional independent variables ........... 84 Table 3-11 Baseline model estimations with NPL ......................................................................................... 85 Table 3-12 Parsimonious Models with NPL & with additional independent variables ................. 86 Table 3-13 Baseline model estimations with dependent variable LLR ................................................. 87 Table 3-14 Parsimonious Models with LLR & with additional independent variables .................. 88 Table 3-15 Baseline model estimations with dependent variable LLP ................................................. 90 Table 3-16 Parsimonious Models with LLP & with additional independent variables .................. 91 Table 3-17 Results of parsimonious models consolidated ......................................................................... 92 Table 4-1 Relative Classification of Credit Risk Measuring Models........................................................ 95 Table 4-2 Comparison of the selected EWS studies and their main characteristics ..................... 105 Table 4-3 Criteria of the search strategy ........................................................................................................ 108 Table 4-4 Statistics of the “default” events in the sample ........................................................................ 109 Table 4-5 Variables and expected signs .......................................................................................................... 111 Table 4-6 Summary statistics for EDF 1 .......................................................................................................... 114 Table 4-7 EDF 1: two-sample t test with unequal variances .................................................................. 114 Table 4-8 Summary statistics for EDF 5 .......................................................................................................... 115 Table 4-9 EDF 5: two-sample t test with unequal variances .................................................................. 115 Table 4-10 EDF 1: Baseline Model ..................................................................................................................... 117 Table 4-11 EDF 1: average marginal effect, marginal effect at the means & odds ratios ........... 118 Table 4-12 EDF 5: baseline model ..................................................................................................................... 118 Table 4-13 EDF 5: Average marginal effect, marginal effect at the means & odds ratios.......... 118 Table 4-14 ROAE, two-sample t-test with unequal variances ............................................................. 119 6 Table 4-15 Assets quality, two-sample t test with unequal variances ............................................. 119 Table 4-16 EDF 1 with CAMEL covariates ...................................................................................................... 120 Table 4-17 EDF 5 with CAMEL covariates ...................................................................................................... 121 Table 4-18 EDF 1 & selected CAMEL ratios, marginal effect .................................................................. 122 Table 4-19 EDF 1 with CAMEL components & adverse selection effect: assets growth ............. 123 Table 4-20 EDF 5 with CAMEL components & adverse selection effect: assets growth ............. 123 Table 4-21 EDF 1 with CAMEL components & adverse selection effect: loan growth ................. 124 Table 4-22 EDF with CAMEL components & adverse selection effect: loan growth .................... 124 Table 4-23 EDF 1 Average marginal effect and Marginal effect at the mean ................................... 125 Table 4-24 EDF 5 Average marginal effect and Marginal effect at the mean ................................... 126 Table 4-25 EDF 1Comparing predicted values vs. actual values, 15% cut-off rate .................... 127 Table 4-26 EDF 5 - Comparing predicted values vs. actual values, 20% cut-off rate ................... 128 7 List of Figures Figure 2-1 Common patterns of financial crises ............................................................................................. 13 Figure 2-2 Causes of the Spanish financial crisis ............................................................................................ 15 Figure 2-3 Evolution of the real interest rate (Spain vs. Euro zone) ...................................................... 16 Figure 2-4 Annual rate of credit growth to the non-financial private sector in the period 20002007 .................................................................................................................................................................................. 17 Figure 2-5 Private debt to non-financial sector: Spain vs. Euro area 2000-2011 (% of GDP) .... 18 Figure 2-6 Pubic Debt: Spain vs. Euro area 2000-2011 (% of GDP) ....................................................... 18 Figure 2-7 Relationship between market based financing relative to bank credits ........................ 19 Figure 2-8 Difference between the average return on investments & the average cost of funds ............................................................................................................................................................................................. 20 Figure 2-9 Structure of percentage of credit to other sectors granted by Spanish credit institutions ..................................................................................................................................................................... 21 Figure 2-10Origination model of sale in the Spanish securitization market ...................................... 24 Figure 2-11 Average time of granting mortgage loans to purchase houses (years) ........................ 25 Figure 2-12 Moody's default frequency curve (Aaa level) ......................................................................... 26 Figure 2-13 Amount of unpaid commercial bills ............................................................................................ 28 Figure 2-14 Number of late payment court proceedings ........................................................................... 28 Figure 2-15Unemployment rate by age group ................................................................................................ 29 Figure 3-1 Evolution of bank risks over 2004-2007 ..................................................................................... 71 Figure 3-2 Evolution of Z-score by bank type .................................................................................................. 72 Figure 3-3 Evolution of Impaired Loans by bank type ................................................................................. 72 Figure 3-4 Mean comparison by bank type ...................................................................................................... 73 Figure 3-5 Standard deviation comparison by bank type .......................................................................... 74 Figure 4-1 Evolution of EDF 1 ............................................................................................................................. 116 Figure 4-2 Evolution of EDF 5 ............................................................................................................................. 116 Figure 4-3 EDF 1 year - Sensitivity of prediction to probability, cut-offs 15% ............................... 127 Figure 4-4 EDF 5 years - Sensitivity of prediction to probability, cut-offs 20%............................. 128 8 Chapter 1 INTRODUCTION This thesis proposes two different models to analyse factors which played a role in massive bank failures during the recent financial crisis in Europe. The former model involves factors of banks’ excessive risk-taking during the pre-crisis period dating 2004-2007. The latter focuses on the timely prediction of bank defaults, so called Early Warning Models. These models were blamed for failing to predict the banks’ downturn in the recent financial crisis. In presenting these two models, we have analysed the following issues: first, we have reviewed common features of financial crises evident in the world and have seen how the Spanish financial crisis fits into these characteristics and how it remains unique. Secondly, we have looked at the main risk factors which have led to excessive risk-taking among Spanish commercial, saving and cooperative banks. We test the validity of risk-taking determinants, empirically evaluating them by applying the most recent data of Spanish banks. And the third, we have proposed the modified Early Warning Model through the application of conventional techniques and combination of additional variables to predict the financial distresses of European banks. A common feature of these three issues is that they are all exploring the recent financial crisis from different perspectives and add to the literature regarding bank credit risk. While all of these chapters focus on one common topic, each chapter is independent and not empirically interconnected to the others. Chapter 2 contains general revision and discussion of the issues while Chapters 3 and 4 have separate literature reviews, methodologies and results parts. The purpose of this thesis is to provide some insights into the understanding of bank defaults and to propose techniques for forewarning banks’ financial distress within the European banking system. The questions we addressed are as follows: • What is a financial crisis and how is it formed? • What are the main determinants of banks’ excessive risk-taking for the years 2004-2011 for Spanish banks? • Are conventional models able to predict the financial distress of European banks before the onset of financial crises? Even though many academic studies have analysed these issues there are still many debates which demand further investigation. The current work aims to add to the literature by consolidating major factors of risk-taking and fitting them into a new model using the most recent data of Spanish banks. It also proposes a modified Early Warning Model to predict bank distress within the European market. The thesis is organized as follows: Chapter 2 focuses on the nature of financial crises and their common characteristics. Also, it reviews the Spanish financial crisis and the features that make it distinctive. Chapter 3 concentrates on the main reason behind the financial crisis - excessive risktaking by banks - and empirically analyzes bank risk-taking factors for the three types of Spanish 9 banks: commercial, savings and cooperative banks. The choice of the Spanish banking sector is not accidental; before the crisis the Spanish banking sector was believed to be one the best and safest in Europe, but following the crisis it proved to be one of the most troubled banking sectors in the EU zone. Chapter 4 empirically analyses a sample of European listed banks and tests the effectiveness of Early Warning Systems, based on Expected Default Frequencies (EDF) and accounting ratios, in forecasting the bank defaults during the recent financial crisis. EDF is the market-based credit measure developed by Moody’s KMV which is based on Merton’s option-pricing theory’s distance to default indicator. Finally we have proposed a model with EDF and a combination of additional variables which we found to be efficient in predicting banks’ defaults for the given sample of European banks. In our analysis we applied Generalized Methods of Moments (system GMM) and model of binary choice – binomial logit model. The former method addresses the issue of endogeneity in panel data while the later model is found efficient for fitting nonlinear models with limited-dependent variable. We conclude that the Spanish financial crisis is not an exception to the general pattern of crises. The sequences of events evident preceding the crisis have many common features with what has already been witnessed in other financial crises. The main determinants of bank excessive risktaking in Spain found valid for our sample are bank ownership nature, the levels of ownership concentration, assets quality and equity structure. Our results suggest that the existence of riskinsensitive deposit insurance increases incentives for banks to exploit the deposit insurance system. As for predicting bank defaults the results reveal that Expected Default Frequency (EDF) metrics provide additional information to that of balance sheet ratios but is not a complete substitute for them. The addition of the variables representing the adverse selection effect showed that their marginal effects are insignificant in predicting the bank default. Our analyses have demonstrated the preponderance of the selected methods - system GMM and binomial logit in dynamic panel data modelling. Overall, we believe that our study provides important insights for regulators into setting up more efficient policies for controlling bank risk-taking factors and improving prevailing early warning models of bank distress. 10 Chapter 2 FINANCIAL CRISIS: THE CASE OF SPAIN 1 Introduction Concern about the stability of financial systems - and in particular the banking system - coupled with the numerous episodes experienced throughout history, has fostered research aimed at studying the causes and consequences of financial crises. The concept of financial crisis is defined by Torrero (2008) as "... an acute disorder that changes the normal functioning of markets, violently affects the valuation of assets, and may threaten the very existence of financial institutions, endangering the whole economic system. " It is a complex concept; one of the reasons for its complexity is the variety and difficulty of the various triggers that can lead to crises. There are many studies aimed at analyzing the evolution of various economic indicators that would allow us to identify a financial crisis and the consequences that flow from it to which we will refer later (Mishkin, 1991; Krugman, 1996; Allen and Gale, 2007; Reinhart & Rogoff, 2008; Rose & Spiegel, 2009; Kirmany, 2010). It should be noted that the effects of banking crises depend on the manner in which they are addressed and on the measures taken by the authorities. At this point we will refer to the spread pattern of financial crises which is known as "contagion effect or herd behaviour." This issue has been addressed by Eichengreen et al. (1996), Basu (2002) and Kaminsky et al. (2003) among others. Essentially it is considered that panic following a drastic reduction in investor optimism, also called overtrading, triggers a series of successive events and immediate herding (usually panic feeds itself and spreads) that increase the likelihood of intensification and extension to other areas. In this regard, Kindleberger (2012) refers to psychological connections: when an increase in euphoria or pessimism of investors in one country affects investors in other countries. Sometimes crises happen in conjunction with changes in the economic cycle. Economic cycles are a natural and inevitable phenomenon in market economies and to some extent can be controlled or weakened (Kindleberger, 2012). In these cycles there is certain rationality between stimuli and responses - both increases and declines are gradual. In contrast, crises do not have a natural character and cannot be explained by changes in economic fundamentals i.e. the smooth running of the economy creates expectations of future earnings that are overvalued and are not sustainable over time and somewhat irrational. In general, crises are associated with speculative phenomena that are predominantly the result of certain ideological and political options. Here, we focus on a “general euphoria”. According to Kindleberger (2012) the phenomenon of speculation often takes place in two stages: 1) Period of calm: characterized by limited and rational response of economic agents. 2) Period of euphoria: focus is on the forecasts of large capital gains characterized as irrational and unsustainable over time. An example would be when asset prices rise rapidly, much 11 faster than Gross Domestic Product (GDP) or any other measure of income. As noted by Bagehot (1873) "the first thing they look for is the high interest investment, but then it becomes a secondary. They become interested in the large profits that they can get by selling the principal”. Once past this stage what is referred to by Kindleberger (2012) as "displacement" occurs featuring some kind of external event that changes the horizon, expectations and opportunities for profit, giving rise to a "crack" and/or "panic" . Although in our case we want to study the financial crisis that began in the summer of 2007, to better understand it we should not ignore the historical perspective. In this vein, Ferguson (2008), criticizes to some extent the lack of global vision and retrospective analysis - understanding the complexities of the financial world through the analysis of its history; a large number of financial crises have occurred throughout the history from which one could draw lessons to mitigate or prevent the current and future crises. There are numerous works which analyze the triggering causes of financial crises with a view to building early-warning systems. These works include the studies of Galbraith (1991), Kaminsky and Reinhart (1999), Schwartz (2009), Reinhart and Rogoff (2008a, 2008b and 2009) and Kindleberger (2012). Galbraith (1991) reviews more known bubbles of history by placing emphasis on and analyzing shared characteristics to draw lessons for the future. Financial leverage is identified as a factor common to all. Moreover, the financial systems are not alien to the processes of liberalization. Kaminsky and Reinhart (1999) obtained evidence for the period post 1970 that banking crises, in both developed and emerging markets have occurred on average five years after the liberalization of the respective financial systems. They argue that the probability of occurrence of a banking crisis following deregulation of the system is much higher than the probability of a crisis occurring if the system is liberalized. For his part, Lordon (2009) considers that there are key fundamental aspects which lie behind every crisis: competition and innovation; the competition causes blindness of ex ante risk while innovation holds the imaginary negotiation of risks and their real accumulation. Some key works on the analysis of banking crises are those of Reinhart and Rogoff (2008a, 2008b and 2009) who analyze the historical causes and consequences of financial crises, paying attention to the great crises in developed countries post World War II. Kindleberger (2012) also analyzes the best-known economic crises that hit the financial world addressing the questions: what is a financial crisis and how it is formed? The crisis of the tulips is analyzed in the Netherlands of the 16th century as well as the crash of 1929 and the “Dot-Com bubbles”. In general, all crises follow a common pattern: Strong capital inflows and imbalances in the current account balance. At first booming economies have large inflows of foreign capital; these economies have large deficits and obtain the 12 money to pay interest to their foreign creditors for new loans. The increase in external debt, generally, is faster than the increase in GDP. Strong credit expansion. The crisis fuelled by bank credit expansion. Sometimes, rapid credit expansion is facilitated by financial deregulation. The arguments here favour more rigorous regulation and supervision of banks as the prevention of lending during periods of higher growth becomes more difficult. In addition, authors such as Minsky (1982) point out that changes in the supply of credit are cyclical, meaning credit increases when the economy is booming, investors become more optimistic and lenders reduce their risk aversion; credit decreases during downturns, investors and lenders become more cautious. Increased risk appetite. Tends to be a period in which it prevails on a widespread basis in economic agents with a greater propensity to risk that has various causes. A psychological type exists where the smooth running of the economy and profit growth leads to a stage of euphoria: contagious optimism. Others may have their origin in the financial system itself, such as increased market liberalization, financial innovations (such as securitization) and strong competition in the sector. Rapid increase of residential and commercial or other property or securities prices (Housing Bubble). The cycle of housing prices, similar to that which has occurred in the United States, Spain, Austria, Hungary, Italy, Iceland, the United Kingdom and other countries, is characterized by a steady increase leading to the purchase of securities or assets by investors to extract short term profit from increases in their prices. This form of buying grows exponentially through easy access to credit. Real assets (shares) reach a maximum before each banking crisis, usually a year earlier, and then decline substantially during the two or three years after. To a great degree, the increase in prices is explained by the increase of household and business income and increased spending. All agents become more optimistic and credit expansion continues to fuel price hikes. The change in the situation leads to an increase in defaults. To take over the pessimism of the market, in this situation, the different agents are unable even to pay the interests of debts. This puts the banks in difficulty forcing them to limit the amount of available credits which, in turn, leads to a further decline in production which increases the difficulties businesses and families have in the repayment of their loans. The aforementioned rapid increase in borrowing would allow borrowers to pay interest with the money from new loans. Bankruptcy and suspension of payments exacerbate the recession and end up affecting the solvency of financial institutions, due to securities or products in the hands of financial institutions experiencing a sharp drop in value. Banks know that other banks have accounts with these same titles or products which can lead to situations where no one trusts anyone. This results in the interbank market ceasing to function triggering a severe restriction of credit to the economy as a whole. All financial crises are accompanied by a significant reduction in employment, public debt increased as a result of the very significant drop in tax revenues and expansionary fiscal policies. 13 Figure 2-1 Common patterns of financial crises The financial crisis that began in the summer of 2007, also known as the global financial crisis, is not an exception to the common pattern discussed above. As we can see, it is possible to identify most of the factors listed above as contributors to the majority of financial crises that have occurred so far. Prior to the outbreak of the crisis, the financial market had been characterized primarily by increased globalization of financial transactions, the application of new information technology (allowing many transactions at once), the strengthening of links between banks and other financial firms and the implementation of various financial innovations. This caused an increase in the complexity and sometimes greater opacity of financial markets. In this new scenario, the possibility that one of the most important markets around the world, such as the United States or United Kingdom, could suffer a crash that would expand to other markets was not correctly assessed. Even during 2007 and part of 2008, prominent financial organizations such as the OECD, EC, IMF, Federal Reserve, the ECB and others considered the prospects for financial stability and economic prospects generally to be favourable, although they warned of some financial risks (Cabral, 2013). The financial crisis was generated in the United States following the creation of market bubbles (stock, real estate, mortgage and derivatives) and was encouraged by the following factors: a) An expansionary fiscal and monetary policy (Taylor, 2009, Allen & Carletti, 2010) which put interest rates at low levels. b) A further deregulation tending toward greater self-regulation of the financial system which led, among other things, to higher leverage (Slow, 2009, Adrian & Shin, 2009, Pozsar et al., 2009). c) An increase in the public deficit due to reduced revenues and increased expenses. Despite low interest rates economies continued investing in dollars (Borio & Disyatat, 2011). Financial Crisis Strong capital inflows & the current account imbalances Strong credit expansion & increased risk appetite Creation of the housing bubble Increase in defaults & shrinking of bank loans Reduction in employment & increase in public debt 14 There was a disproportionate increase of credit based on financial innovation mechanisms, in particular in securitization. Asset securitization means the risk of customer default is not applied to the originator resulting in abusive practices, so called "predatory lending". (Aschcraft & Schuermann, 2008). Credit growth occurred especially in the high-risk mortgage sector. Purnanandam (2009) shows how those banks that made widespread use of origination models for sale generated loans with lower credit quality. Through financial innovation financial products were created from subprime mortgages and other products. Moreover, when the housing bubble was in full swing mortgage defaults were not a problem since the growth in house prices provided a lifeline, buildings were put up for sale generating gains that allowed for debt repayment. The problem appeared when the housing market expansion ended. At this time the market value of housing started to decline making the debt value greater than the market value of homes. Borrowers then started to hand in their houses to financial institutions to cancel their debts while defaults that had been hidden thanks to the housing bubble materialized. Defaults on mortgages were transmitted to the securitized products. 2 Financial Crisis in Spain The Spanish financial crisis is not an exception to the general pattern that has been discussed in the previous section. In fact, we could say that it is a clear example of a dramatic worldwide economic change. The sequence of events described in the above section unfortunately corresponds to situations Spain has and continues to experience. Further analysis will show that the external imbalance of the Spanish economy (one of the largest in the world), the unprecedented increase in credit, the concentration of investment in real estate resulting in what is commonly known as “the housing bubble” and the use of financial innovation to raise funds or excessive risk taking have all preceded the Spanish crisis. Spain is following the foreseeable consequences of the crisis in the shape of constant business failures, restrictions to obtain credit and an increase in public debt which have been part of the everyday life of Spain in the last years. Therefore, in the years prior to the current crisis we were incubating a set of global imbalances that affected mainly the volume and direction of international capital flows (Andrew, 2009). In Spain, we can identify two areas directly affected by the crisis: the real estate business and the finance sector as a whole (Recarte, 2008). Consequently, this resulted on the one hand in the housing crisis and on the other the financial crisis (adjusting the construction sector to the current situation and the lack of liquidity in the financial system). Since both sectors are closely linked the connection and interrelationship between the two sectors is obvious. Different authors suggest several different causes of the financial crisis (Recarte 2008, Andrés 2009, Jiménez et al. (2010), Otero and Ezcurra (2012) and Maudos 2012). The most important theories and the interrelationship between them will be discussed further. 21 Figure 2-9 Structure of percentage of credit to other sectors granted by Spanish credit institutions Source: Bank of Spain (2011) Cuñat and Garicano (2009) show how saving banks which reported higher offence issues (delayed or overdue payments) were the ones concentrating more resources in real estate. In the tables below we can see the real estate exposure, exposure problems and insurance coverage held by major banks and saving banks. Table 2-3 Problematic real estate exposure (2010) Year 2010 in millions of Euros) Exposure in construction & real estate development % of lending except public sector Problematic real estate exposure Delinquency Substandard (at risk of default) Foreclosed Assets Total % of lending except public sector SANTANDER 27.334 12,2% 4.636 4.932 7.514 17.082 7,6% BBVA 16.600 8% 3.543 2.381 6.397 12.321 5,9% BFA 41.280 19,8% 7.370 7.742 9.843 24.955 11,9% LA CAIXA 26.284 14,9% 4.080 1.657 4.825 10.562 6% BANCO BASE 24.264 27% 5.222 4.565 4.207 13.994 15,5% POPULAR 17.840 18,7% 2.587 2.642 4.759 9.988 10,5% BANCO SABADELL 10.170 14,2% 4.074 3.028 3.768 10.870 15,1% BANESTO 10.354 15,1% 1.670 1.076 1.399 4.145 6,1% CATALUNYACAIXA 12.774 23,7% 1.794 1.663 5.436 8.893 16,5% NOVACAIXAGALICIA 11.150 21,8% 2.527 1.911 3.525 7.963 15,6% BANCA CÍVICA 9.187 18,8% 1.168 2.200 2.050 5.418 11,1% MARENOSTRUM 11.553 22,9% n.a. n.a. 3.854 3.854 7,6% BANKINTER 2.452 5,8% 291 99 484 874 2,1% BBK 3.574 10,4% 1.622 888 1.152 3.662 10,7% ESPAÑA+DUERO 8.067 31,3% 1.677 1.338 1.083 4.098 15,9% 22 IBERCAJA 4.636 13,9% 580 718 962 2.260 6,8% UNICAJA 2.948 12% 365 550 1.281 2.196 8,9% UNNIM 3.598 19,9% 642 599 2.345 3.586 19,8% KUTXA 1.714 11,2% 441 320 546 1.307 8,6% CAJA TRES 3.576 27% 493 1.632 609 2.734 20,6% CAJA VITAL 1.253 19% 162 211 289 662 10% Total 250.608 38,4% 44.944 40.152 66.328 151.424 9,7% Source: Bank of Spain & Author’s calculations As can be seen in the data presented in the tables the entities with greater exposure are saving banks, or financial institutions which originally were saving banks. Consequently, with an increasing delay in payments these institutions are more greatly affected by the exposure. Table 2-4 Property risk coverage (2010)3 Year 2010 (in millions of Euros) Real estate exposure Credits to construction & development with mortgage guarantee % of total real estate exposure Credit to construction & development excluding mortgage guarantee Specific provision % of problematic exposure General provisions Total % of problematic exposure (including general) SANTANDER 4.956 29,0% 768 5.724 33,5% 21.210 77,6% 6.124 BBVA 3.389 27,5% 9.765 13.154 106,8% 15.272 92,0% 1.328 BFA 9.518 38,1% 1.578 11.096 44,5% 34.985 84,8% 6.295 LA CAIXA 3.962 37,5% 1.835 5.797 54,9% 24.240 92,2% 2.044 BANCO BASE 4.591 32,8% 1.012 5.603 40,0% 21.046 86,7% 3.218 POPULAR 8.548 85,6% n.a. 8.548 85,6% 17.840 100,0% 1.552 BANCO SABADELL 3.628 33,4% 424 4.052 37,3% 9.528 93,7% 642 BANESTO 1.002 24,2% n.a. 1.002 24,2% 5.017 48,5% 436 CATALUNYA CAIXA 3.086 34,7% 181 3.267 36,7% 10.285 80,5% 2.489 NOVACAIXA GALICIA 2.823 35,5% 359 3.182 40,0% 9.241 82,9% 1.909 BANCA CÍVICA 2.080 38,4% n.a. 2.080 38,4% 8.438 91,8% 749 MARENOSTRUM n.a. n.a. n.a. n.a. n.a. 10.748 93,0% 805 BANKINTER 423 48,4% 157 580 66,4% 1.369 55,8% 1.083 BBK 1.532 41,8% 418 1.950 53,2% 3.195 89,4% 380 ESPAÑA + DUERO 1.775 43,3% 126 1.901 46,4% 6.476 80,3% 1.591 IBERCAJA 776 34,3% n.a. 776 34,3% 4.225 91,1% 412 UNICAJA 386 17,6% n.a. 386 17,6% 2.726 92,5% 222 UNNIM 658 18,3% 106 764 21,3% 3.272 90,9% 326 KUTXA 586 44,8% 19 605 46,3% 1.262 73,6% 452 CAJA TRES 601 22,0% n.a. 601 22,0% 3.234 90,4% 342 CAJA VITAL 161 24,3% 10 171 25,8% 1.163 92,8% 90 Total 54.481 36,0% 16.758 71.239 47,0% 214.772 85,7% 32.489 Source: Bank of Spain & Author’s calculations 3 n.a. – not available 23 2.6 Financial Innovation Most authors (Carbó-Valverde et al., 2012; Maddaloni & Peydró, 2010; Martín-Oliver & Saurina, 2007; Douglas & Raghuram, 2009 and Kwan, S, 1998 among others) dealing with the international financial crisis refer to securitization as origin of it. We have already mentioned that the Spanish economy fell into debt abroad and the external credit earned ended up largely financing real estate. Part of this funding was obtained through external credit which went through a process of securitization and placement in fixed income markets. The Spanish securitization market during the pre-crisis period was one of the most important in Europe. According to data published by the European Securitization Forum, the volume of Spanish Residential mortgage-backed security (RMBS) issuance represented 14.88% of the total European RMBS emissions in the year 2006, 18.49% in 2007 and 10.43% in 2008. Spain ranked second in relation to the volume of mortgage-backed securities issued in Europe in 2007, just after the United Kingdom, while in 2008 it ranked fourth, following the United Kingdom, the Netherlands and Italy. Taking into account the current outstanding asset-backed securities at the end of the third quarter of 2009, Spain is in third place with a total of 167,100 million euro (14.48% of the total), following the United Kingdom (with 458 000 million euro, representing 39.67% of the total) and the Netherlands (with 202 400 million, which account for 17.53% of the total). Diamond and Rajan (2009) argue that one of the main causes of the financial crisis was the securitization process, as it provided an excessive credit exposure related to the construction sector. Krugman (2007) summarizes the problem concluding that in the last years of growth the "big real estate cycle" banks - from 2000 to 2005 - issued a large number of the poor credit quality loans, although banks were aware of the situation and managed to get away from it with the help of securitization. Moreover, Purnanandam (2009) shows that those banks that made widespread use of origination model and sold loans of bad credit quality lessened their incentives as originators to generate mortgages making a rigorous credit risk analysis. It seems reasonable to wonder about the role played by securitization in the Spanish context. Otero and Ezcurra (2012) believe that there are certain similarities with the U.S. Subprime Crisis. Almost all research studies conducted until now on the Spanish securitization market argue that one of the fundamental differences with the U.S. model is the development of various models of securitization in Spain. Also the main characteristic of the American model is “originate to distribute” while the Spanish securization market operates under the principle of “originate to maintain”. Therefore, unlike the complete transfer of risk (thus freeing the entire use of capital which appears in the balance) as in the US, in Spain the technique of securitization has been used primarily as a funding mechanism. In this system, the bank providing credit has no incentive to undertake thorough analysis when providing too risky mortgages disregarding the borrower’s ability to pay since the real risk is kept to a greater extent on the balance sheet of the entity which administers the loan (Losada, 2006; 24 Sources, 2007; Catarineu & Perez, 2008), dissimilar to the American “originate to distribute” model, where the conflict of interest is much more evident (Purnanandam, 2009)4. The reasons and revealed evidence leading us to talk about the subprime in Spain are: • the increasing trend of operations in which the originator of the loans has been able to distribute or sell them in the market after the first loss through securitization; Figure 2-10Origination model of sale in the Spanish securitization market Source: CNMV5 • sharp rise in mortgage loans and delinquency due to late payments Mortgages have doubled in just five years (2004-2009), to account for 1,099,568 million Euros at the end of 2009, compared to 531.608 million euro in 2004. Regarding delinquency due to late payments, the latest official figures indicate a delinquency rate of credit for house purchase mortgage-backed securities of 2.42% at the end of the first quarter of 2011, compared with just 0.38% in mid-2006; • granting of increasingly riskier products with higher loan values such as mortgage lending over the appraised estimation, mortgages for second homes or loans granted to foreigners or people without stable income. Similarly, there is a general increase in the term of loans as well as the range of products that postpone the repayment of the principal loan (by paying interest only or capital repayment deferred, hybrids, etc.). For example, mortgage loans granted to foreign residents increased substantially in recent years accounting for around 7.2% of the total mortgage loans granted by the end of 2008. These loans had a delinquency rate (late payments) of 12.5% of the total at the end of the same year, while for the national residents this figure was 1.6% of the total6. 4 Purnanandam (2009) shows how the “originate to distribute” model has encouraged the granting of mortgage products of lower credit quality, as the originators did not use resources to analyze the repayment capacity of the borrowers. 5 CNMV (Comisión Nacional del Mercado de Valores) Spanish National Stock Market Commission 6The Bank of Spain’s Financial Stability Report, 2008. 25 Figure 2-11 Average time of granting mortgage loans to purchase houses (years) Source: BBVA Research Department In this regard, the works of Jimenez et al. (2010) and Otero and Ezcurra (2012) conclude that securitization encouraged poor credit quality, increased competition between banks and led to the granting of credit in more permissive conditions. The new loans were subsequently riskier and more likely to default, suggesting that eligibility rules were relaxed in order to expand credit and, consequently, securitization has proven to have a negative impact on the financial stability of Spanish credit institutions, creating conditions for bankruptcy and increasing the delay of payments. 2.7 The Credit Rating Agencies In general, in Krugman’s opinion (2007) one of the most important events was the error caused by the rating agencies assigning allegedly very high credit ratings to those securities, when they were actually the equivalent to “junk bonds” or very low level. Also, Ashcraft et al. (2009) investigate whether the potential conflict of interest in rating agencies led to what is defined as "rating inflation". The results of the study show how the rating assigned to securities were gradually increasing, even immediately after adjustment according to the level of risk and the characteristics of the loan. The potential conflict of interest among rating agencies could have impacted to their work, because the main parties involved in the process - originator, issuer and rating agency - were very interested in generating short-term commissions, without taking under consideration the risk and viability of the business in the long term. As it is outlined below, fragility in the credit rating process for mortgage securities has also been demonstrated in the Spanish market. As mortgage markets was developing, financial institutions in an increasingly more competitive environment brought about by the mobility of loans created particular mortgage products with more sophisticated characteristics and riskier profiles. But because these new products were issued in a period of economic expansion with a generalized rise in housing prices, this led to rating agencies and other market participants underestimating the risks associated. Another error found in the rating process, although not as noticeable, was the evolution or trend in housing prices. Although 26 rating agencies apply reductions in prices of real estate according to the different levels of rating assigned, the current crisis affecting the Spanish economy has shown that these reductions were not conservative enough. These shortcomings during the rating process have resulted in a large number of securitization transactions issued in the Spanish market (especially those collateralized by loans with a high Loan-to-value (LTV)7 and issued in recent years) suffering severe credit downgrades, especially the junior trenches of the capital structure. As a consequence of this, Moody's and Fitch have introduced modifications in their models. In July 2008, Moody's published a series of adjustments in their rating models for the Spanish market and in the same vein Fitch updated its matrices of default probabilities in February 2010. As can be seen in the following chart, changes are stricter as the LTV of the loan increases. Figure 2-12 Moody's default frequency curve (Aaa level) Source: Moody's 2.8 Corporate Governance Corporate governance, in particular in saving banks, was also critical to determine the level of risk taken by financial institutions. Thus, characteristics like acting on behalf of shareholders (Beltratti & Stulz, 2009) the composition of the Board of Directors (Pathan, 2009); the concentration of power and the participation of institutional investors in the Board of Directors, affected the level of risk assumed by banks. Laeven and Levine (2009) argue that banks with powerful shareholders tend to pressure managers to incur in more risky transactions. Moreover, Bai & Elyasiani (2013) conclude that banks with more incentive to incur in risk are those in which there is more likelihood of government support (systemic). In this type of bank shareholders are keener to encourage risk taking. In the Spanish case, saving banks have been the financial institutions which have incurred in higher levels of risk and endured the hardest part of the financial crisis. Cuñat and Garicano (2009) study the impact of corporate governance in saving banks by analyzing the composition and structure of the General 7 LTV is loan-to-value ratio - the ratio of a loan to the value of an asset purchased 27 Board, the training of executives and the politicization of the general council, among others. Their work concludes that banks run by a person with postgraduate education, banking experience and no previous political activities, concentrated less credit in real estate, had fewer payment delays and credit downgrades. The majority of saving banks had a politicized management team, often without any previous banking experience and acting with unprofessional criteria. In general we can say that this sector has been damaged the most by the Spanish financial crisis. 2.9 Regulation and Supervision The implementation of Basel II, the most demanding and advanced regulation covering risk, has made assume that the financial sector was prepared to deal with potential financial crises. However, its release practically coincided with one of the most severe financial crises. Further analysis led to the conclusion that regulation tends to reduce capital levels in times of economic expansion and increase them in times of crisis (Dewatripont & Freixas, 2012). Also Dewatripont and Tirole (2012) point out that Basel II failed in the design of counter-cyclical provisions. The capital buffer incorporated in Basel III can help to alleviate this problem. Another criticism has been that Basel II relinquished supervisory activity, promoting the role of the market itself as a supervisory mechanism. Still, not all countries relaxed its supervisory function to the same extent. Moreover, the work of Beltrany and Stulz (2009) makes clear that banks in countries with a stricter monitoring system also performed better during the financial crisis. In this sense, despite the prestige of the Bank of Spain as a supervisory body, much criticism has been made of its monitoring of the Spanish financial sector. So, despite knowing that banks were expanding credit beyond wise levels, the passivity of the Bank of Spain encouraged excessive risk taking. The letter sent in 2006 by the Association of Certified Credit Inspectors of the Spanish Banking (AEICA) warns about the passivity of the Bank of Spain against excessive credit growth and its concentration in real estate. The consequences have already been quoted in the characterization of financial crisis. In particular the bursting of the housing bubble caused very significant reductions in asset prices. The credit contracted rose sharply and the amount of late payment cases dramatically escalated, causing many bankruptcies and asset transfers to banks. Another consequence has been the significant increase in the unemployment rate. We can see clearly that as the crisis started to become worse, the unemployment rate increased, mostly affecting the younger generation. 28 Figure 2-13 Amount of unpaid commercial bills Source: National Statistics Institute (INE) Figure 2-14 Number of late payment court proceedings Source: National Statistics Institute (INE) 29 Figure 2-15Unemployment rate by age group Source: National Statistics Institute (INE) 3 Restructuring process Following the financial crisis many measures were taken both domestically (in Spain and EU) and internationally. Along with other changes there were proposed the new Basel III regulations, the carrying out stress tests of banks, improvement in supervision of financial institutions by EU and in national level and the creation of the Fund for Orderly Bank Restructuring (FROB). In this regard, the National Reform Plan 20128 comprises the profound changes that have been undergoing in the Spanish financial system. To solve the problems in the financial sector and mitigate their impact on the real economy, it has been adopted a fundamental reform of the financial system that focuses on deep provisioning and restructuring of balance sheets of credit institutions to increase the efficiency and competitiveness of the sector. The Bank of Spain, the national central bank and supervisors of the financial system, is coordinating this complex restructuring and recapitalization program as well as the Memorandum of Understanding (MoU). MoU has been developed jointly with the European authorities and agreed in July 2012 and primarily aimed to restore confidence in the Spanish economy, stabilize the financial sector and place them in a stronger position in future. Among the measures taken by the authorities, undertaken both internationally and nationally, there are in general two courses of actions. The one is directed to undertake preventive measures 8 While we focus on the reforms in the financial system, the reforms undertaken in Spain, as envisaged in the National Reform Programme 2012 of the Kingdom of Spain itself, have a much broader and encompass other draft lines of action such as: • Fiscal consolidation. • The modernization of public administrations and public services. • Financial system reforms • Labour market flexibility, training and education. • Growth and competitiveness. 30 mitigating the severity and consequences of future crises. A clear example of these actions is found in the decisions of the Basel Committee which can be shortly expressed as the performances focused on new capital rules far more demanding than former regulations and the reconfiguration of the financial architecture internationally. The other course are the measures to directly support financial entities in order to regain the confidence of markets, mitigate liquidity pressures and facilitate the channelling of credits to the real sector. More specifically, the measures taken by the Spanish authorities are aimed at improving confidence, credibility and strength of the financial system, supporting banks’ liquidity, promoting the consolidation and restructuring of the most fragile entities (mainly saving banks) and increasing the levels of capital and reserves. These actions also directed to cover the risks of real estate due to doubts about the valuation of real-estate assets, as well as the weight of the real estate assets held by the banking sector. The ultimate goal of these measures is to improve the outlook of the situation in Spanish financial sector. 37 financial firms from 30 countries which were at the centre of the financial crisis, find that increased risk-taking is associated with greater institutional ownership. In addition, Cheng et al., (2010) investigated whether residual compensation and risk are related to institutional ownership using data on executive compensation for financial firms from 1990-2008, and found evidence suggesting that there is heterogeneity in investor preferences, with institutional investors later wanting managers to take more risks and so having to give them incentives to do so. 2.1.2 Board of Directors: Size and Composition The board of directors is a key mechanism of bank governance. Internal and external governors need to be well coordinated to achieve the shareholders’ value maximization without unbalancing the safety and soundness of the whole banking system. In maintaining this balance a bank’s board of directors play a crucial role. The importance of boards of directors is also stated in the second pillar of Basel II, where it is accepted as an integral part of risk management (Basel Committee on Banking Supervision, 2005). However, the opacity and complexity of the banking system are major obstacles to stakeholders monitoring bank performance, diminishing the regulators’ and other stakeholders’ capacity to monitor, thus increasing the importance of banks’ board of directors in corporate governance issues. Numerous academic works have revised the relevance of banks’ board structures on risktaking (Pathan, 2009; Adams & Mehran, 2008; Andres & Vallelado, 2008 among others). For instance, Pathan (2009) examines the relationship between bank boards and bank risk-taking from an agency theory prospective. In particular he investigates the effect of strong bank board and CEO power on risk-taking behaviour. Here, ‘strong board’ is defined as board effectiveness in monitoring bank managers on behalf of shareholders. The term ‘CEO power’ is explained as the degree of a CEO’s influence on board decisions. By using a sample of 212 large US Bank Holding Companies (BHC) over the period 1997-2004 it finds that bank risk-taking is positively related to strong bank boards. These results are robust especially for small and less restrictive boards. Meanwhile, observed evidence suggests that bank risk-taking is negatively related to CEO power. Pathan (2009) proposes that the risk-averse nature of bank managers comes from a willingness to protect their non-diversifiable human capital which is mostly concentrated in their managed banks. Moreover, a negative relationship between independent directors and bank risk may imply that the former would prefer to balance between the interests of shareholders and bank stakeholders such as depositors and regulators. In general the paper suggests that bank board structure is an important determinant of bank risk taking. Meanwhile, Andres & Vallelado (2008) study the relationship between a bank’s board and its risk-taking by focusing on both the size and composition of the board since these features influence directors’ ability to monitor and advise managers. The overall findings suggest a non-monotonic relationship between board size and performance. The study covers 69 large commercial banks from 38 six developed countries over the period 1995-2005 and applies a two-step system estimator econometric model to solve unobserved heterogeneity and endogeneity problems. The authors find that the relationship between board size and its performance describes an inverted U-shape which implies a non-monotonic relationship as when boards reach an optimal size (19 directors), performance starts to diminish. They suggest that there is a trade-off between advantages of having a larger board (monitoring, advising) and disadvantages (control and coordination problems). The same inverted U-shaped relationship is observed between the proportion of outsiders and performance. In general, the findings support the idea that outside directors improve value but when the number reaches a majority of the total directors, Tobin’s Q starts to lessen. Furthermore, they argue that outside directors should hold a majority on the bank board to help minimize the conflict of interest among stakeholders through monitoring and advising in an efficient manner. And in an efficient bank board non-executive directors’ ability is complemented with the presence of executive directors and their knowledge and data concerning the bank. Overall, the authors stress the significance of having an efficient board not only for a bank’s shareholders and stakeholders but also for national economic systems to ensure the safety and health of financial intermediation. Erkens et al., 2012 analyse the importance of board composition in bank performance by focusing on three corporate governance factors - board independence, institutional ownership and the presence of large shareholders during the crisis. The study analyses why during financial crisis some financial institutions were affected more than others. The authors use data from 2007-2008 covering 296 financial firms from 30 countries that were at the centre of the financial crisis and find that firms with more independent boards and greater institutional ownership had lower stock returns during the crisis. Further analysis reveals that greater risk-taking is associated with increased institutional ownership but not with independent boards, contradicting the proposition those non-executive directors encouraged managers to take greater risks before the onset of the crisis. The poor performance of firms with independent boards is explained through the increased pressure of independent directors on managers to raise equity capital during the crisis in order to ensure capital adequacy and to lower bankruptcy risk. As equity capital rising was costly during that period, it could lead to wealth transfer from shareholders to debt holders but help them to survive the crisis. To measure the effect of large shareholders, the study uses a dummy variable with a cut-off of 10% in direct and indirect voting rights. It does not find any significant association with large shareholders and firms’ weakened stock returns. Finally, the study advocates that corporate governance had an important impact on firm performance during the crisis through its risk management and financing policies. 2.1.3 Executives and Compensation Scheme Executive compensation is often referred to as one of the key contributors to bank risk-taking behaviour. The principal question raised by many academics is whether executive compensation 39 schemes were the origin of banks’ excessive risk-taking during the recent financial crisis. According to market discipline theory, a firm’s executives are monitored and disciplined by its stakeholders such as shareholders, debt holders and by regulators so that the former act in their best interests. However, the recent financial crisis has demonstrated the ineffectiveness of the disciplinary channels through which they operate. The fundamentals of the problem may arise from the nature of bank capital structure which can directly influence the level of bank executive compensation. In line with agency theory, bank stockholders prefer that the CEO is compensated with stock options as this increases the CEO’s pay/performance sensitivity. In this manner, a higher level of stock options motivates the CEO to higher risk investments at the expense of banks’ debt holders (Dewatripont & Freixas, 2012). In measuring the ultimate influence of CEO compensation on bank risk-taking it worth using a measure of “residual compensation” introduced by Cheng et al (2010). Residual compensation is the residuals of a regression of compensation on firm size (its market capitalisation) and sub-industry level characteristics. In other words it is the compensation unexplained by firm size. For example, firms with high residual compensation include Bear Stearns, Lehman, Citicorp, Countrywide and AIG. It is also suggested that residual compensation is strongly correlated with several measures of risk-taking and with institutional ownership of the firm. A similar risk sensitivity measure of CEO compensation – “vega” - is proposed by Bai & Elyasiani (2013). “Vega” indicates the extent of change in CEO wealth relative to a one percentage point (.01) change in bank stock return volatility. The authors have analysed the relationship between insolvency risk and the executive compensation structure for large BHCs (Bank Holding Company) from 1992-2008. The study focuses on two indicators: the risk sensitivity measure of compensation – “vega”, and pay-share inequality between CEOs and other top executives. As equity based compensation became significantly more prevalent after the deregulation of markets, CEO interests became more aligned with bank stockholders’ interests as their compensation was more sensitive to banks’ stock risk. Meanwhile the increased “vega” of compensation has resulted in increased risk-taking among banks since CEOs have excessive incentives to take on risk and increase their wealth. The study proposes that this trend may lead to greater instability in the banking system at the expense of depositors, bondholders and deposit insurers. Moreover, it suggests that the relationship between managerial compensation structures and bank stability is bi-directional i.e. higher risk-sensitive compensation results in riskier investment policies by CEOs whilst riskier BHCs operate managerial compensation structures, which are more sensitive to stock return volatility. As a counterbalance to increased “vega” in managerial compensation practice the study presents the positive effect of pay-share inequality to bank stability. It is measured by the share of CEO total annual compensation in the total annual compensation of the top five executives in the same BHC and indicates pay-inequality in the top management team. The use of the this measure is based on the hypothesis that when the compensation share of a CEO increases relative to other top managers, a 40 CEO may become more risk-averse so as to not to lose his sizeable pay in the case of failure. He therefore takes less risk to avoid a greater downside loss and to lock his current position. This results in lower risk-taking and greater bank stability. The study uses natural logarithms of Z-score as a measure of bank stability. Moreover, the authors investigate whether BHCs use noninterest income activities as a channel to raise their risk-taking behaviour caused by increased “vega”. The obtained results exhibit that BHCs with higher level of “vega” have greater levels of non-traditional banking and support the hypothesis brought by the authors. It is also important to notice the relationship the authors find between the Too-Big-To-Fail effect and the level of bank’s “vega” since many large banks attempt to maximize the value of the implicit government guarantees. For this they divide BHCs into five quintile groups by the size of total assets. The empirical results confirm the effect of bank size on risk-taking behaviour through compensation. The findings show that the second largest BHCs increase “vega” more than other groups in order to encourage CEOs to take on more risk to achieve Too-Big-ToFail status and to take advantage of government guarantees. Executives are compensated for outputs such as bank performance as well as inputs such as the skills and experience they invest in the bank in so-called human capital. In fact, compensation schemes are designed to attract and retain skilful managers and it is natural to expect that higher compensation cultivates skilled corporate leaders, with significant upsides to their future earning power. However, debates over the origins of the ongoing financial turmoil are often related to the issue of a lack of human capital in banks’ management. In particular the professionalism of banks’ CEOs is repeatedly put under question. The work of Cuñat & Garicano (2010 investigates this issue of executive professionalism in a sample of Spanish savings banks known as Cajas. The authors intend to explain heterogenity in the performance of these banks from the prespective of corporate governance and human capital. The choice of Spanish Cajas is not accidental;they are characterized as an unusual segment of the Spanish financial sector which does not formally have shareholders and is heavily politicized. Moreover, shares of Cajas are not quoted on the stock market and thus major external disciplinary governance mechanisms do not work. The extended board of Cajas is formed by representatives of the local political authorities, representatives of the founders of the bank, relevant social institutions, workers and other stakeholders. Also, a substantial proportion of board members is directly appointed by local and regional governments. Considering the special features of Cajas the authors try to determine if it is beneficial to have knowledgeable chairmen through assessing their performance in portfolio allocation decisions and loan risk-taking. These decisions require relatively profound knowledge and experience while chairmen of Cajas are mainly retired politicians; their lack of required knowledge and inexperience in banking could be reflected in the CEOs key decisions. The study also analyses how corporate governance impacts on bank performance by examining the board composition and its effect on loan losses, rating changes and the composition of the loan portfolio. The authors created a synthetic index of the Chairman’s human capital for most saving banks over the last 41 9 years from 2001 onward. Their findings exhibit clear and significant patterns in governance and the human capital of the Chairman. Specifically, the human capital level of the Chairman is closely correlated with the loan portfolio of the Caja before the Crisis (in 2007) and with the loan performance of the Caja during the crisis. In particular, a Caja run by a chairman with a post-graduate education, with previous banking experience, and with no previous political appointments is expected to have significantly less real estate lending in its portfolio, or a larger share loans to individuals ratio (loan concentration), a lower rate of non-performing loans, and a lower downgrade in its bank’s rating. In general Cajas with fewer politicized board members had less exposure to real estate risks suggesting that CEO professionalization played a role in bank performance during the crisis. On the contrary, with respect to the governance of Cajas, the authors do not find a high correlation between governance and the composition of the loan book at time of the financial crisis. In practice, bank executives’ payments are also provided in the form of bonus payments to high-ability workers. The study of Bannier et al., (2013) addresses the rationale of why bonuses are paid by banks even at the expense of profit and social welfare. It presents a theoretical model of managerial talent’s remuneration where agents of differing talent (or ability) have to decide on the allocation of funds between safe and risky projects. The type of agent (ability of manager) affects bank returns received both from safe and risky investments but it is unobservable i.e. private information. In the presence of competition for talented workers, the model is characterized by hidden information on agent ability and moral hazard with respect to choice of investment (safe/risky). In a model equilibrium, only high-ability workers receive bonuses, and excessive risk-taking is deliberately accepted in order to reduce low-ability workers' information rents. Consequently, authors suggest that rising competition for workers induces banks to offer higher compensation including bonuses to be able to attract skilful workers. Moreover, it leads to higher risk-taking and higher inefficiencies both from society’s and the bank’s point of view. Therefore legal restrictions on bonuses have a positive impact on bank profits and stability as well as social welfare. Along with changes in compensation, executive turnover is also considered as an efficient internal managerial control mechanism to control top management. Schaeck et al., (2012) investigates the role of different stakeholders in disciplining bank executives through the application of two dimensions of market discipline: the ability of stakeholders to monitor and evaluate bank conditions, and their ability to influence a bank’s actions. They primarily examine the monitoring roles of stakeholders and how they influence the likelihood of an executive dismissal in small and mediumsized unlisted US banks from 1990-2007. Consequently, the study evaluates changes in post-turnover bank soundness by looking at risk-taking, losses, and profitability to examine whether executive replacements have affected a bank’s financial state. The results show that the probability of forced turnovers is robustly increasing with banks’ increasing risk-taking. But further findings reveal that debt holders or regulatory actions do not contribute to this disciplinary mechanism since in those 42 banks where debt holders have a larger stake, or in banks where the regulator is aware of the distress, the frequency of turnover does not increase with bank risk. With respect to improvements in bank performance following turnovers, the results are not supportive. The preliminary findings show weak evidence that dismissals leads to reduced losses over three years following the event, but do not reduce risk or improve profitability. Further analyses demonstrate that, on the contrary, turnovers lead to higher risk levels with greater losses, and a sustained negative impact on profitability. The study therefore put in question the effectiveness of market disciplinary mechanisms for small and medium-sized banks notwithstanding their shareholders which are also not found to have an influence on bank soundness. The experience and background of CEOs and their impact on bank risk/return efficiency is further analysed by Jonghe et al., (2012). They estimate an efficient risk/return frontier for Turkish commercial banks with the use of a stochastic frontier approach for the period 1988-2009 which includes the Turkish banking crisis of 2000-2001. The paper merges two aspects of risk/return tradeoff studies: off-balance sheet banking activities and the impact of governance mechanisms – internal and external on risk/return efficiency. Unlike other research it investigates risk and return simultaneously by relating the risk of a bank portfolio to the returns that portfolio generates. Moreover, it sets apart internal governance, which involves the attributes of the strategic decisionmaking process, from external governance which relates to market and stakeholder oversight trying to impact and control decisions. The findings suggest that a more experienced CEO in most cases increases risk/return efficiency. In relation to size, larger banks tend to be risk/return, efficient possibly due to their wider opportunities and market power. However, the political background of bank chairmen is likely to have a negative effect on risk/return. The same negative effect is observed with respect to non-interest income. Moreover, their results are in favour of CEO non-duality i.e. when the CEO is not the chair of the bank board. They find that, in general, CEO non-duality helped to achieve a higher efficiency in risk/return trade-off, especially post crisis. Overall the work sheds light on the impact of corporate governance on bank risk/return efficiency in conditions of opaque bank activities, economic, regulatory and supervisory environments. 2.2 Bank Business Models The financial crisis has highlighted a significant variability of performance and risk-taking across different banks. Academics and policymakers raise the question of if this variability could be explained through specific bank characteristics originating from their different business models. Prior to the crisis, many banks had moved away from traditional retail banking activities to “new” bank business models with complex securities, non-interest generating activities and wholesale markets funding structures. The shift towards new business models was mainly caused by financial innovations in credit markets as well as deregulation of the banking sphere. Extensive bank crashes 43 fosters debate over the characteristics of bank business models which have greater financial efficiency and stability. A cohort of academic studies has investigated the relationship between bank business models and risk-taking behaviour. The research intends to identify whether there is any association between certain business model characteristics such as bank asset structure, capital structure, income source and funding strategy and excess risk-taking, and whether bank business models can help to identify the hidden risks which could materialize in long run or cause macroeconomic shocks. Altunbas et al., (2011) using a large sample of listed banks in the EU and US, observe banks’ financial indicators before and during the crisis and investigate whether the variability across bank business models is related to their realized risk during the financial crisis. In general, the results reveal a non-linear relationship between bank business models. Specifically, a strong deposit base and income diversification are associated with lower risk, while less capital, large size, greater reliance on shortterm money market funding and rapid credit growth correlate with higher risk. In general, the study encourages bank supervisors to distinguish the impact of different business models on bank risk to explain the divergence in risk realisation during the crisis. Similarly, Köhler (2012) analyses the effect of loan growth and business models on bank risk level and revealed considerable heterogeneity in risk-taking across banks and countries. He suggests that banks with high loan growth rates are riskier. Also, he finds evidence that if banks increase their non-interest income share it positively affects stability while this effect decreases with bank size. Excessive credit growth is associated with high bank risk. Overall, the study summarizes that differences in lending activities and business models facilitate the identification of bank risk. In this section we revise the selected characteristics of bank business models such as capital, assets, income sources and funding strategy and examine how these factors affected the risk-taking of banks during the crisis. Although we present them in separate sections, in practice it is impossible to disentangle their individual impact from other macroeconomic effects; hence in many cases we present the joint impact of the factors on bank risk-taking behaviour. 2.2.1 Capital Structure The role of capital is prominent in analysing bank risk. Among other factors, insufficiency of bank capital is also extensively discussed among academics and policymakers as a contributor to the on-going financial turmoil. A strand of literature supports the hypothesis that lax regulation of banks including oversight of bank capital, restriction on bank activities and weak monitoring led banks to take excessive risks during the crisis. In fact, Basel’s recommendations on capital level served as a cornerstone of prudential regulations for banks. However, it is believed that the Basel II Accord lowered the degree of regulator and supervisor involvement and promoted financial markets as a supervisory disciplinary device. Banks are allowed to undertake internal risk assessment models in identifying their capital requirements. As a consequence, coupled with other factors it led to the 44 problem of procyclicality. Procyclicality, stemming from Basel II, is blamed for bank excessive lending, the emergence of bubbles and a financial accelerator effect during the crisis. It is also supposed that Basel II capital regulation tends to reduce capital requirements in good times and increase capital requirements in bad times (Dewatripont & Freixas, 2012). The effect of procyclicality could be particularly important in downturns when banks could face a ‘capital crunch’ that would further restrict their lending. The regulatory proposal of Basel III on countercyclical buffers is intended to solve this issue. Meanwhile, empirical literature presents contradictory results on the relationship between capital and risk-taking. Demirgüç-Kunt et al., (2010) research the role of bank capital in withstanding a shock such as the financial crisis. In particular, they investigate whether better capitalized banks had higher stock returns during the financial crisis. Also they examine which concept of capital is more relevant in stock valuation during the crisis and what items are counted as capital for regulatory purposes. The baseline model measures bank performance with a change in bank stock prices between quarters and relating it to change in its level of capital. It uses dummy variables which account for any possible omitted country-level effects such as macroeconomic shocks, systematic components, etc. and a matrix of bank-level control controls for bank-specific features (such as bank liquidity, reliance on deposits for funding, etc.) The results obtained from a large sample of international banks suggest that during the crisis banks with higher capitalization were better valued than undercapitalized banks though this trend is not observed before the crisis. Moreover, they find that big banks’ stock returns are more sensitive to the leverage ratio as a capital measure than to the risk-adjusted Basel ratio. This may be explained by a lack of reliability to Basel risk-weighted indicators by market participants at the time of the crisis. Finally, it concludes that “higher quality capital” – Tier 1 and tangible common equity are more relevant. Berger & Bouwman (2012) also examine the effect of capital on bank performance and whether it varies across financial crises and periods of economic stability. Here bank performance is measured in terms of survival and market share. Moreover, they test the joint effect of capital and size on bank performance during the crisis. The research has two baseline regressions which empirically measure the effect of capital on banks’ survival and on market share in different time periods. Potential omitted variables are covered by a broad set of control variables. The main findings of the study support in general the hypothesis that capital helps banks to survive in line with Altunbas et al., (2011), Demirgüç-Kunt et al., (2010) and others. In addition to the findings of other similar research, it reveals that for small banks capital is essential for survival at all times and for medium and large banks only during banking crises. With respect to market share and bank size, capital helps small banks to improve their market share at all times, while for medium and large banks it is helpful only during banking crises. 45 The importance of Tier 1 capital for large banks is also supported by Beltratti & Stulz (2012). By analysing the relative stock return performance of large banks across the world during the crisis they find that that large banks with more Tier 1 capital, more deposits, less exposure to US real estate, and less funding fragility performed better than banks financed with short-term funds raised in the money markets and with more exposure to US real estate. The outline of the interaction among bank capital regulation, the business cycle and the transmission mechanism is presented by Borio and Zhu (2012). In light of the evolution of the financial system, it explains the potential impact of minimum capital standards on the transmission mechanism by revising two issues: the influence of minimum capital standards on bank behaviour and their effect at the interest margin on the impact of monetary policy (Borio & Zhu, 2012). Two ways of how capital regulation effects bank behaviour are identified – through the capital minimum thresholds effect and through the capital framework effect. The former focuses on the costs associated with breaching the minimum threshold and on actions needed to prevent this. The latter looks at the broader influence of the capital framework on how a bank conducts its business. The authors suggest that with the evolution of minimum capital regulation from Basel I to Basel II the influence of prudential regulation and supervision on bank behaviour has been raised, with respect to both the threshold and framework effects. Consequently, the minimum threshold could have a greater variance over the business cycle. Obviously, risk measures tend to vary procyclically i.e. to be comparatively low during economic expansion and to be comparatively high during economic contraction. They argue that changes in the financial system and prudential regulation highlighted the importance of the risk-taking channels and that existing macroeconomic concepts and models are not sophisticated enough to capture these changes. Dewatripont and Tirole (2012) also report the negative effect of micro prudential banking regulations leading to increased bank vulnerability. They argue that Basel I/II capital regulations fail to account for macro shocks and that countercyclical capital buffers recommended by Basel III help to deal with these shocks. The authors propose a model which departs from Modigliani–Miller in that outside equity and capital requirements matter and in where they analyse banking regulation in the presence of macroeconomic shocks. They note the desirability of self-insurance mechanisms such as countercyclical capital buffers or dynamic provisioning, as well as “macro-hedges” such as CoCos and capital insurance. A number of empirical studies show that under the influence of banking competition, capital regulation may destabilize the banking sector and cause increased risk-taking. Hakenes & Schnabel (2011) present evidence on the presumed trade-off between competition and bank stability. They suggest that capital regulation may not always prevent banks’ excessive risk-taking. They develop a model where banks first solve a portfolio problem by assessing the riskiness of projects from an available portfolio, and with given limited liability and deposit insurance banks are subject to a risk- 46 shifting problem. In the second model banks solve an optimal contracting problem by extending loans to entrepreneurs who determine the risk of their projects. In this case, the entrepreneurs are subject to a risk-shifting problem. The later model is based on a paper by Boyd & Nicolo (2005) where banks compete for loans and deposits but with added portfolio problems, costly bank equity and capital regulation. The authors analyse the impact of capital requirements on the risk of loans, bank correlation and bank default rate. They suggest that stricter capital requirements weaken competition for loans and lead to higher loan rates, increasing risk-taking by entrepreneurs by raising the risk of individual loans. Moreover, strict capital requirements may induce banks to choose a more correlated portfolio by increasing the probability of default. In general, the research summarizes that the ambiguous effect of competition on banks’ risk-taking results in an ambiguous effect on capital regulation, and that capital regulation acts as a stabilizer when competition has a destabilizing effect and vice versa. Schaeck & Cihak, (2012) also examine the impact of competition on bank capital ratios. They analyse why banks maintain capital levels above regulatory requirements although it is costly and may impede banks’ ability to compete. The authors report that the observed capital ratios tended to significantly exceed minimum capital requirements in the period prior to the recent financial crisis and they seek explanations for this phenomenon. The study involves a large sample of European banks including commercial, savings and cooperative banks. Hypothetically, it is based on the theories of Allen et al., (forthcoming) which state that banks have excessive capital holdings due to market discipline arising from the banks’ assets. In other words, increased competition encourages banks to have higher capital ratios because it demonstrates their commitment to monitoring. Besides, it attracts creditworthy borrowers despite a countervailing effect of deposit insurance. They also supposed that capital ratios are higher when shareholder rights are strongly protected, and that deposit insurance lowers capital ratios. In line with Allen et al(forthcoming)., Schaeck & Cihak (2012) find robust evidence that competition motivates banks to increase capital holdings and this evidence holds true prior to the financial crisis. This result is valid mainly for commercial banks but holds true even for not profit maximizing financial institutions such as savings and cooperative banks. A 1% increase in competition raises the average bank’s capital ratio in the sample by up to 3.9%. Regarding bank size they also find that the increase in capital is greater for the average large European bank (4.2%) than for the average small bank in Europe (3.6%) supporting the view that smaller banks use different types of lending technologies than large banks, because smaller banks lend more to informationsensitive borrowers that require intensive monitoring mentioned by Berger et al., (2005). Stronger shareholder rights (or less dispersed ownership structures) are associated with higher capital ratios while agency problems and deposit insurance decreases capital ratios by reducing incentives to monitor. 53 also analyses ratings of securitized deals by identifying factors which are considered by rating agencies and the sequence of ratings provided. As stated earlier, Spain’s market is characterized by very significant loan growth in pre-crisis years, in particular during 2006 where annual loan growth was above 25% on average. Securitization activity grew dramatically, mostly together with large increases in bank credits to the private sector, from being almost insignificant in the late 1990s to financing a large portion of loan growth in the years running up to the crisis. Unlike securitization in the U.S. market, in Spain the originating bank also acts as the servicer of the loan portfolio while borrowers are not typically aware of whether their loans have been securitized or not (Martín-Oliver & Saurina, 2007). The findings of Carbó-Valverde et al. (2012) suggest that bank characteristics such as solvency, cash-flow generation and cost efficiency (on top of loan performance) affect ratings considerably. They reveal that these characteristics have a greater impact on the rating changes of savings banks compared with commercial banks. Moreover, banks located in regions with increased housing price growth in the years before the crisis also have higher impact ratings of securities issued by saving banks, suggesting their close link to regional territories. The sequence of ratings show that loan growth significantly affects loan performance with a lag of at least two years, and balance-sheet loan performance a lag of four quarters. Analyses of other factors suggest that the role of competition in stimulating loan growth has been more intense in savings banks. The general evidence from the Euro Area and the U.S. is also in line with the findings in Spain. Maddaloni & Peydró (2010) undertake a broader study of inter-relations between bank risk-taking, interest rates, securitization and bank capital supervision pre-crisis terminating in 2008:Q3. They state that low short-term rates of monetary policy –too low for too long period – led to a softening of lending standards resulting in an accumulation of risk on banks’ assets. At the same time, increased securitization activity and weak supervision of bank capital further amplified the impact of low monetary policy rates. The study supposes that low interest rates prevailing in economies for a long period of time may make riskless assets less attractive and may lead to a search-for-yield by banks (Rajan, 2005). Since the securitization of loans offered assets yielding attractive returns for investors at a time of abundant liquidity, it became widely used as an investment decision. Moreover, by securitizing their assets, bank could enhance their lending capacity, especially when the capacity constraint is binding (in times of high credit growth, partially stemming from low monetary policy rates) and grant more loans. Securitization may therefore intensify the impact of low interest rates on the softening of lending standards. 2.4 Impact of Competition on risk-taking Many academic works concentrating on factors affecting banks’ risk-taking hold the opinion that competition has a strong influence on bank risk. Since competition has traditionally been considered a source of excessive risk-taking it has been continually regulated in order to prevent bank 54 runs and instability in the banking system. These regulatory measures, as believed by many academics, gave rise to a long period of stability within the banking system. After the intense deregulation in the US and European banking sectors, banks were faced with intense competition in both national and international spheres. Consequently, there is an increasing concern that competition erodes banks’ market power and monopoly profit leading to a decrease in banks’ charter values. As there is less to lose, banks start to increase their risks which could eventually lead to bankruptcy or bank runs and contribute to overall economic distress. Salas & Saurina (2003) argue that when markets are liberalised and regulations are relaxed, competition decreases profit and banks’ charter value. In their study they have analysed 21 Spanish commercial banks within the years 1968-1998 to reveal the effects regulatory changes have on banks’ market power. The data covers 31 years, involving the principle steps to deregulation which contributed to key changes in the Spanish financial market. It analyses banks’ risk-taking behaviour in response to reduced economic profits caused by deregulation. The central question of the research is whether there is a trade-off between market efficiency and banks’ solvency in the context of financial intermediation. The study advocates that in deregulated markets such as Spain, competition may eliminate external constraints to risk-taking, as well as internal constraints assumed to be voluntary, by the banks possessing market power. The results generally confirm that the measures of liberalization have influenced bank competition resulting in reduced market power and a decrease in banks’ economic profits. Furthermore, lower economic profits caused by deregulation and increased competition fostered banks’ risk-taking as their charter values decreased and they had less to lose. Matutes & Vives (2000) have also investigated a relationship between banks’ market power and risk-taking incentives but in the presence of limited liability and the social cost of failure. The study analyses the link between imperfect competition for deposits and bank risk-taking subject to limited liability, and tries to identify whether ‘excessive’ competition for deposits exist. According to the authors, limited liability is introduced with a standard debt contract between bank and depositor. Since banks’ portfolios are not perfectly diversified, in the case of bank failure there is a social cost not borne by the bank. The paper makes the following suggestions concerning the social cost of failure, under different regimes: 1) assuming a lack of deposit insurance, where there is intense competition, banks tend to set inflated deposits rates resulting in a high social cost of failure. Under this regime banks do not internalize the cost of failure. In this case, the introduction of a proper deposit rate ceiling induces minimal risk taking; 2) flat premium deposit insurance tempts banks to engage in the highest possible risk-taking. Since investors do not have incentives to punish excessive risk-taking by the banks this induces banks to undertake maximal asset risk positions. In this case, the introduction of fair and risk based deposit insurance decreases excessive risk-taking on the deposit side because banks are fully liable for any consequences. The study summarises that high risk-taking incentives exist with flat-premium deposit insurance and could be minimized with the use of risk-based 55 insurance. It also emphasizes the role of limited liability, imperfect competition and the social cost of failure in analysing banks’ risks. The belief that more competition in banking causes a greater level of instability tends to focus on competition in deposit markets. However, far too little attention has been paid to loan markets. The study of Boyd et al. (2006) and Boyd & Nicolo (2005) argue that the conclusions of previous theoretical research are fragile since they allow competition only for deposits and not for loans, while in fact banks are simultaneously involved in both markets. The study compares two banking models, CVH (Charter Value Hypothesis) and BDN (Boyd & De Nicolo) and examines whether there is a tradeoff between bank competition and stability. CVH is based on earlier work by Allen & Gale (2000, 2004) and allows for competition in deposit markets, but not for loans and it assumes that there is no contracting problem between bank and borrower. Unlike CVH, the BDN model allows for competition in both deposit and loan markets and assumes that banks solve an optimal contracting problem with their borrowers. CVH predicts a positive relationship between competition and risk of failure whereas the opposite is predicted by BDN. Both models have an important implication in that the relationship between bank competition and profitability can easily be non-monotonic. Empirical tests conducted on 2500 cross sectional US banks’ data and a large set panel data collected from non industrialized countries find that more competition is ceteris paribus associated with a lower probability of failure. In other words, there is a positive relationship between competition and bank stability. Furthermore, the test reveals a positive link between bank competition and willingness to lend. As competition declines, banks earn more income in their loan markets through charging higher loan rates. This implies a high bankruptcy risk for borrowers due to a moral hazard problem i.e. borrowers faced with high interest costs choose higher risk-higher return projects. Consequently, the CVH model is rejected while the results are still consistent with the BDN model’s predictions. The debate continues regarding the type of relationship between competition and bank risk. The paper by Martinez-Miera & Repullo (2010) supports the Boyd and Nicolo (2006) proposition that bank competition reduces a loan’s probability of default due to reduced loan rates. This effect is referred to as the risk-shifting effect. However, they argue that increased competition may also reduce the interest payments from performing loans, which serves as a buffer to cover loan losses. This effect is referred to as the margin effect. Unlike the above mentioned models, the study suggests a U shaped relationship between competition and banks’ risk of failure where risk first decreases before starting to increase in a very competitive market. At some point, more competition leads to lower loan rates and reduces banks’ interest income from non defaulting loans used as buffer for loan losses. Consequently, in highly concentrated markets the risk-shifting effect dominates and more competition reduces bank risk, whereas in very competitive markets the margin effect dominates and the increased competition erodes banks’ franchise values and so increases risks. Accordingly, the probability of bank failure is lowest in moderate levels of competition. 56 The relationship between bank competition and risk-taking is further analysed within the Spanish national banking system by Jimenez et al., (2010). The authors support the franchise value paradigm in limiting bank risk taking. They advocate that bank managers and shareholders tend to limit and reduce their risk exposure to preserve the bank’s franchise value. The source of franchise value is assumed to be the market power of a bank and a decrease in competition among banks diminishes their appetite for risk. Conversely, an increase in competition erodes their quasi-monopoly rents and the value of the charters and may lead to greater bank risk-taking and greater financial instability. The paper examines the impact of various measures of competition in loan and deposit markets and constructs measures of market power based on Learner indexes. The Learner index is commonly used to measure the market power of a firm and indicates the degree to which it can increase its marginal price beyond its marginal cost. The dependent variable for a bank’s risk is a ratio of Non-Performing Commercial Loans (NPL) obtained from the credit register maintained by Banco de Espana. The results of loan market Learner measures indicate a negative relationship between banks’ market power and risk-taking. Similar but weaker results are shown for deposit markets, although the joint loan and deposit Learner indexes indicate a negative and very significant impact on NPL ratios. Hence, the findings based on the Spanish Banking system support the franchise value paradigm while disproving Boyd and Nicolo’s risk shifting effect i.e. the BDN model. Also, they find little evidence of a U shaped relationship between competition and risk suggested by Martinez-Miera & Repullo, (2010). Furthermore, an inverted U-shaped relationship between regional bank competition and stability is found by Liu et al. (2013). The study examines the joint impact of competition and regional economic conditions on the risk and stability of European banks from 2000 – 2008. They argue that many studies of competition and risk-taking apply national measures of competition and/or national economic activity though the majority of banks have a regional customer focus. National measure may therefore be inadequate in certain market segments like retail deposits or small business loans, in which banks operate mainly at a regional level. Consequently, the authors advocate that regional competitive and economic conditions may be more relevant in analysing the risk-taking behaviour of these kinds of banks. By analysing the relationship between regional economic conditions and competition and their subsequent impact on bank risk in European banking they confirm the prevalence of a U shaped relationship between regional competition and banks. Particularly, riskshifting effects appear to dominate in concentrated markets while margin effects appear prevalent in competitive banking markets as suggested by Martinez-Mirea and Repullo, 2009. Moreover, they advocate that regional economic conditions play a significant role in determining the stability of banks as banking risks may increase in regions with high unemployment. With regard to individual bank characteristics they find evidence that diversified banks are less stable than their smaller and more focused counterparts, whilst mutual banks seem to be more stable than their commercial banking counterparts. 57 Hakenes & Schnabel (2011) examine the joint impact of banking competition and capital regulation on bank risk-taking behaviour and find that capital regulation may destabilize the banking sector through its effect on banking competition. Specifically, evidence suggests that stricter capital requirements weaken competition for loans and lead to higher loan rates and so to higher risk-taking by entrepreneurs increasing the risk of individual loans. Moreover, strict capital requirements may induce banks to choose a more correlated portfolio by increasing the probability of default. They also find that capital regulations act as a stabilizer when competition has a destabilizing effect through the “charter value effect”. In general, the research summarises that the ambiguous effect of competition on banks’ risk-taking results in an ambiguous effect of capital regulations. Tabak et al. (2012) investigate the effects of bank competition on the risk-taking behaviours of banks in 10 Latin American countries between 2003 and 2008. In particular, they examine how size and capitalization change the relationship between competition and stability. The study applies an innovative Boone indicator method (Boone, 2008) to measure the competition by assessing the impact of efficiency on performance. Specifically, the Boone indicator within the loans market measures the intensity of the effect of the earning market share of more efficient banks. They find supporting evidence that competition influences banks’ risk-taking behaviour in a non-linear way and both high and low competition levels enhance financial stability, while they find the opposite effect for average competition. The authors suggest that the non-linearity of the effect supports both the concentrationstability (anti competition views) and the concentration-fragility (pro competition views) theories. They advocate that banks facing both high and low competition are, on average, lower level risk-takers than banks experiencing average competition. Also, the study reveals the importance of bank size and capitalization in analysing the impact of competition on bank risk. In particular, they find that the larger banks may reap greater benefits from competition since their size makes them less vulnerable. Similarly, greater capital ratio is beneficial for banks that operate in collusive markets, though capitalization only seems to have a positive impact on financial stability for larger banks. In other words, in collusive markets, banks with a larger capital ratio are more stable as shareholder capital disciplines banks under low competition. In general, the study proposes that average competition is not high enough to benefit large banks or low enough to trigger the advantages of capitalization. Another study undertaken in non-European markets analyses the effects of competition on banks’ risktaking behaviour in four South East Asian countries comprising Indonesia, Malaysia, Philippines and Vietnam from 1998-2004 (Liu et al., 2012). The results are somehow contradictory: it supports the pro competition view suggesting that competition does not increase bank risk-taking behaviour while it also asserts that concentration is inversely related to bank risk, implying a positive relationship between competition and concentration. The received evidence casts some doubt on the traditionally expected inverse link between concentration and competition. However, as is noted by the authors, Claessens and Laeven (2004) examined the drivers of competition in 50 countries 58 applying H-statistic to measure competition and found that concentration tended to be positively related to competition. Liu et al., (2012) suggest that regulatory restrictions positively influence bank risk-taking. Reductions in restrictions on banking activities, particularly on foreign bank operations, appear to lead to higher levels of competition while increases in competition reduce bank risk-taking. The results seem robust for different model specifications, estimation approaches and variable construction. 2.5 The Risk-taking Channel of Monetary Policy There are several studies which concentrate on the impact of monetary policy on bank risk. All these works intend to answer one question: how monetary policy affects risk-taking and what risktaking channels it applies. It is widely believed that banks take more risk when monetary policy is expansive. This view supports the risk-taking channel theory suggested by Borio and Zhu (2008) and identifies the transmission mechanism as the risk-taking channel. In fact the theory studies the impact of changes in monetary policy rates on either risk perceptions or risk-tolerance and so on the degree of risk in the portfolios. Extensive empirical evidence has been found suggesting that monetary policy could have a significant impact on banks’ incentives to take on risks. Besides, it has been suggested that there is a strong link between monetary policy looseness and bank risk taking (Altunbas et al., 2009). This study finds a strong connection between relatively low interest rates and bank risk supporting the idea that monetary policy can have a strong impact on banks’ balance sheet conditions and that loose monetary policy contributed to bank risk-taking. They undertake analysis of listed banks operating in the European Union (EU15) and the United States during and prior to the financial crisis using both the Taylor rule and the natural rate method. They note that changes in the financial system have contributed to strengthening this link. Moreover, the development of financial innovations and changes in the capital regulatory framework (Basel II) may further increase the risk attitude of banks. However, they admit that the mechanisms through which monetary policy may influence market participants’ risk-taking are complex and at least need to be viewed in two dimensions: through amplification of the “financial accelerator”, when monetary policy may influence the evaluation of collateral, asset prices and cash flows, thereby amplifying risk-tolerance of loan providers; through the “search for yield” process where market participants in conditions of low interest rates decide to take on riskier assets in order to increase their expected returns. The authors state that these two dimensions may be amplified if agents perceive that monetary policy will be relaxed in the case of decreasing asset prices in a financial downturn (the so-called “insurance effect” i.e. moral hazard problem). Overall, the study concludes that monetary policy is not fully neutral from a financial stability perspective Similarly, Jimenez et al. (2008) investigate the effect of monetary policy on the appetite for credit risk in banks within the Spanish banking industry. Particularly, they investigate the effect of 59 short-term interest rates on banks’ credit risk levels. Based on data from Spanish banks, the findings suggest that lower short-term interest rates motivate banks to soften their lending standards and provide more loans to borrowers with a bad or no credit history. An expansionary monetary policy is therefore most likely associated with higher credit risk. This view is widely supported by academics (Diamond and Rajan, 2006(Douglas & Raghuram, 2005); Dell'Ariccia & Marquez, 2006 and Delis & Kouretas, 2011 among others) and in general proposes that banks take more risk when monetary policy is expansive. Jimenez et al. (2008) advocate that as bank finances illiquid long-term projects with liquid demand deposits this causes a mismatch making bank reluctant to grant risky loans in times of liquidity shortages. A negative relationship between interest rates and bank risk-taking is analysed by Delis & Kouretas (2011). The study is based on data from 16 euro area countries over the period 2001-2008. The empirical findings suggest a strong negative relationship between interest rates and bank risk taking; the negative relationship is stronger for banks which have higher levels of non traditional activities i.e. with higher volumes of off-balance sheet items, while for banks with higher levels of capitalization the relationship is weaker. Therefore, the study concludes that banks’ involvement in non traditional activities and their level of capitalization is central in defining risk-taking behaviour, especially at a time of low interest rates. This paper is in line with the findings of Lepetit et al. (2008) discussed earlier in our review, stating that banks that expanded into non-interest income activities in general exhibit higher risk-taking than banks performing traditional activities. Furthermore, Achraya & Naqvi (2012) argue when there is abundant liquidity, banks’ managers may have an incentive to under-price the risk of investments; especially when managers are hedged from downside risks the risk-taking incentives amplify. This in turn induces an excessive demand for assets in the real sector and leads to asset price inflation i.e. price bubble. The authors state that when macroeconomic risk is high and investors switch from direct investment to savings in bank deposits, banks face excessive liquidity. This aggravates the risk taking moral hazard, giving rise to credit booms and asset price bubbles. Here, the situation worsens through expansionary monetary policy which results in flushing banks with even more liquidity. According to Achraya & Naqvi (2012), a central bank should adopt a contradictory monetary policy in times of excessive bank liquidity to limit banks’ risk-taking incentives. Dell'ariccia & Marquez (2006) also have found that banks’ incentives to screen depend on their cost of financing which is determined by the level of short-term interest rates. If interest rates decrease, banks’ incentives to screen borrowers also lessen. An extensive discussion of monetary transmission mechanisms are presented by Borio & Zhu, 2012. They state that more attention should be paid to the characteristics of the transmission mechanism in light of the recent evolution of the financial system. The paper examines the nexus between capital regulation and supervision, business fluctuations and traditional channels of the transmission mechanism. The authors define risk-taking channels as the transmission mechanisms 60 which provide the link between monetary policy and the perception and pricing of risk by economic agents. Based on the related literature they identify at least three ways in which such risk-taking channels may operate: through the impact of interest rates on valuations, incomes and cash flows (financial accelerator effect); through the relationship between market rates and target rates of return (search for yield effect); and through aspects of the characteristics of communication policies and the reaction function of the central bank (Borio & Zhu, 2012). In this channel the role of the regulator is very central; by increasing the degree of transparency it can decrease uncertainty resulting in a reduction in risk premium (transparency effect). By “censoring” the distribution of future outcomes, the regulator can imply that changes in rates have an asymmetric impact on behaviour, with reductions encouraging risk-taking by more than equivalent increases would curtail it (insurance effect). The study emphasizes that liquidity and risk-taking are tightly interconnected. It develops the concept by exploring the mutually reinforcing link with “liquidity” (defined as the ease with which perceptions of the value can be turned into purchasing power) and analyses its interaction with monetary policy reaction functions. The study concludes that changes in the financial system and prudential regulation elevate the importance of the risk-taking channel and that prevailing macroeconomic paradigms and related models are not well suited to fully capture it. These concepts may therefore have reduced effectiveness as prudent guides to monetary policy. 2.6 Regulatory and Institutional contexts A number of academic papers studying bank risk taking concentrate on analysing the institutional and regulatory environment of the issue. For example, Houston et al (2010) examine the links between the level of creditor’s rights, information sharing and bank risk-taking using a sample of 2400 banks in 69 countries. The study explores how these two factors affect the likelihood of a financial crisis and the overall banking system. The paper argues that the strength of creditors’ rights and the level of information sharing have an effect on the contracting environment , on creditors’ ex ante incentives as well as the recovery rates in cases of default. In particular, strong creditors’ rights make creditors more willing to grant funds since their risk is reduced with higher recovery rates. On the other hand greater protection in the event of default may induce creditors to lend to borrowers with poor credit ratings. Meanwhile, the level of information sharing among creditors may also have an important influence on bank risk-taking since it helps to reduce costly information asymmetries. The study uses a number of variables to measure information sharing and suggests that stronger creditor rights are correlated with higher bank risk-taking and therefore increase the likelihood of financial crisis. Alternatively, better information sharing among creditors reduces the risk-taking incentives of banks and significantly weakens the positive link between creditor rights and banking crises. Specifically, it serves as a post lending diciplinary or monitoring tool for borrowers by enhancing banks’ regulatory environment. In general, the study concludes that greater information 61 sharing results in greater bank profitability and lower bank risk leading to a reduced likelihood of financial crisis, and higher economic growth. 2.7 The Role of Market Discipline Lately, much emphasis has been placed on market forces by the Basel Committee on Banking and Supervision (Basel II, Accord). It suggests bank supervisors use market information to improve the assessment of banks’ financial safety and soundness. Under market discipline, the market correctly reflects individual bank risk levels as investors require a risk premium for any additional risk. This mechanism may increase banks’ cost of funding and so banks will be discouraged from taking additional risk. Therefore, market information could be used by bank supervisors as a signal and also to complement accounting data in the design of early warning systems. In the meantime, the principal question raised is whether market prices convey additional information which is not already included in accounting data or whether the benefits from employing market information outweigh the cost of using market information (Curry et al., 2002). Empirical research of US banks supports the idea that market variables improve the assessment of bank financial health when added to standard call report financial data (Curry et al., 2002; Evanoff & Wall, 2001). Additionally, it is shown that the prediction of a CAMEL (supervisory) rating downgrade to the lowest levels can be significantly improved by adding market variables to a set of accounting indicators, although the predictive power is found to be significant only for banks in great financial distress (Curry et al., 2002). Empirical findings from a sample of EU banks analysed by Gropp et al. (2002), and Distinguin et al. (2005) suggest that equity market based indicators deliver earlier signal of fragility than debt based indicators. Overall, with rapidly changing financial markets, determinants of banks’ excessive risk-taking need to adjust to a changing environment. 2.8 The Role of Credit Agencies The credibility of the Credit Rating Agencies (CRA) has been shaken in the recent financial meltdown. As financial markets became more opaque and complex, the expertise of CRAs in assessing credit-worthiness have been increasingly demanded by investors. However, CRAs have been blamed for contributing to the creation of those market conditions which led to financial turmoil (Bahena, 2010). Since credit instruments originate from a pool of many loans it is difficult to assess their creditworthiness for individual investors. Therefore, many investors used CRAs ratings which had been overoptimistic and experienced delays in reacting to changing market conditions. CRAs have also been criticized for conflicts of interest as most ratings are solicited and paid for by the firm which is rated. Many banks issuing securities have maintained a close relationship with CRAs and due to this it is unclear whether the CRAs serve the interests of public or the paying entity. Moreover, CRAs provide paid consulting to entities which want to improve their ratings which are ultimately rated by the same CRA, thus creating other conflicts of interest (Bahena, 2010). 62 Evidence regarding the dynamics of rating changes and the objectivity of credit ratings assigned during the crisis raise a number of issues on the role of credit rating agencies in the economy. For instance Carbó-Valverde et al., 2012 report a delay in rating changes of four quarters for credit derivatives in Spain. They note a considerable delay before CRAs reassess their credit views though their involvement should go beyond providing passive credit quality certification and theoretically include a more active approach over the economic cycle. Similarly, Ogut, et al., (2012) argue that the rating of a bank’s financial strength can be very misleading and investigate whether the forecast of the rating of a bank’s financial strength using the publicly available data is consistent with those of the credit rating agency. They build models to determine the significant factors that have an impact on bank financial strength ratings reported by Moody’s. Data is comprised of 26 bank ratios – both financial and proxies of qualitative data, as independent variables. Banks’ financial strength ratings by Moody’s serve as dependent variables. The period of observation covers 2003-2009 and includes only Turkish banks. The empirical findings suggest that the most important factors are efficiency, profitability (ROE), and the proportion of loans in the assets. It is observed that the rating agency assigns a higher rating to those banks that generate high net income for shareholders, use resources efficiently, and channel funds as loans to households and businesses. The authors suppose that rating agencies find it less profitable for banks to place a high proportion of their funds (mainly deposits) in government debt securities indicating that the rating of a bank is higher if its risk is shared with different groups. The results are in general consistent with those of Moody’s financial strength ratings. The study by Ashcraft et al., 2010 analyses credit ratings on subprime and Alt-A mortgagebacked-securities (MBS) deals issued between 2001 and 2007 in the US market. The study covers 3,144 MBS deals differentiated into security and loan-level data representing around 60,000 securities and 12.1m loans, and covering nearly 90% of deals issued during this period. It investigates how well initial credit ratings summarized the variation in MBS default risk across this sample of deals in period leading to crisis. In particular, it analyses the consistency of MBS ratings in two dimensions: through time, and across deals from a given vintage backed by different types of loans. Furthermore, it examines how well credit ratings order relative risks across MBS deals from within a given cohort. The empirical evidence suggests that ratings are in general informative, hence it rejects the simple story that credit rating standards deteriorate uniformly over the pre-crisis period. However, it also reveals significant time-series variation in subordination levels i.e. it finds a significant decline in risk-adjusted subordination levels between the start of 2005 and mid-2007. Moreover, the study advocates the theory that MBS ratings did not fully reflect publicly available data which reported that high-risk deals, measured by a simple ex-ante model, considerably underperform relative to their initial subordination levels. Meanwhile deals with a high share of low-documentation mortgages also perform disproportionately badly compared to other types of risky deals. The authors do not present final 69 H7: A stronger capital position (higher Tier 1 capital) achieves a better market value during crises H8: Capital enhances the banks’ probability of survival during financial crises and periods of economic stability H9: The capital of small banks helps them to survive at all times, for medium and large banks only during banking crises Altunbas et al., (2011) find that ex-post bank risk is associated with ex-ante bank size and the degree of credit expansion in the years preceding crises. During the financial crisis many large banks were often perceived as “Too Big To Fail” (TBTF effect), and thus deemed more likely to be rescued by state authorities (Huang, et al., 2011; Demirgüc-Kunt and Huizinga, 2010; Tarashev et al., 2009). Moreover, lower interest rates prior to the origination of loans gave rise to more lending to borrowers with either a bad or no credit history (Jimenez et al., 2008). Köhler (2012) finds that high rates of loan growth are associated with high bank risk. The development of securitization has also contributed to changes in the nature of the bank asset structures by allowing banks to turn traditionally illiquid claims (overwhelmingly in the form of bank loans) into marketable securities (Altunbas, Manganelli, & Marques-Ibanez, 2011). Based on empirical studies, our hypotheses in respect to bank assets are: H10: Bank size and loan ratio are positively related to risk H11: Securitization increases the overall risk of default of banks measured by Z-score and other risk-taking proxies. Another determinant of bank business models is income structure. Banks with expanded noninterest income activities exhibit greater risk-taking behaviour than banks performing traditional activities (De Young & Roland, 2001; Stiroh, 2004; Stiroh & Rumble, 2006 and Lepetit et al., 2008). The effect of non-interest income activities can be further analyzed by splitting said activities into commission and fee income and trading income. A higher commission and fee income share of noninterest income indicates increased risk and a greater risk of insolvency (Lepetit et al., 2008). Our expectations regarding bank income structure are as followings: H12: Banks with high non-interest income have higher level of risk than banks with traditional income. H13: Banks with more fee-based income exhibit higher risk The final component of bank business models is funding strategy. Huang & Ratnovski (2010) argue that with the presence of noisy public signals, fund providers’ incentives to monitor banks and impose market discipline may be distorted and lead to inefficient liquidation of a bank with higher wholesale funding. Altunbas et al., (2011) also suggest that banks with a broader deposit base were more resilient during the financial crisis. Hence we expect the following: H14: Non deposit funding shares increase bank risk H15: Short-term marketable securities increase risk 70 Table 3-3 Variables and hypotheses considered Variable Prediction Definition Source Z-Score Credit Risk Total Risk of default [Z-Score] Dependent variable Return on average assets plus the balance of capital-to-total-assets over the volatility of ROAA Bankscope Credit Risk NPL ratio % [Imparedloans] Dependent variable (non-performing loans / total gross loans) Bankscope Loan loss reserve % [Loanlossres] Dependent variable Loan loss reserve / gross loans % Bankscope Loan loss provision % [LoanLossPtoloans] Dependent variable Loan loss provision / gross loans % Bankscope Nature of ownership [Savdummy], [Comdummy] & [Coopdummy] +/- +/- Dummy variable for Saving banks, Commercial banks and Cooperative banks respectively. Bankscope Ownership concentration [BvDdummy] + - Proxy BvD Indep. Indic. With cutoff rate 25%. If widely held=1 (with the largest owner no more than 25% of share), otherwise 0. Bankscope Capital [Capitrat] + - Total Capital Ratio/Capital Adequacy Ratio: Tier 1 + Tier 2 as a percentage of risk weighted assets and off balance sheet risks Bankscope Tier 1 Capital [Tier1] + - Shareholder funds plus perpetual non cumulative preference shares as a percentage of risk weighted assets and off balance sheet risks measured under the Basle rules Bankscope Loan ratio [Netloantoas] - + Net loans/Total assets -percentage of the assets of the bank tied up in loans Bankscope Level of securitization [RMBS] - + Outstanding balance of securitized assets / total assets Annual report and Pillar III disclosures Bank size [Logtotassets] - + Natural logarithm of total assets Bankscope Non-interest income [NonIntIn] - + Non-Interest Income/ Gross Revenues % Bankscope Fee-based income [FeesCommtoOpProf] - + Net Fees & Commissions/ Operating Profit % Bankscope Trading Income [TradingfeestoOpprof] + - Net Trading fees/Operating Profit % Bankscope Non-deposit funding [Moneymarkfunding] - + Other Deposits & Short-term borrowings/Total Deposits, Money Market Instrument & Short-term Funding Bankscope Short-term Marketable Securities [otherdeptototal] - + Other Deposits and Short-term Borrowings/Total Assets Bankscope Deposits [DepostoAssets] + - Total Customer Deposits/Total Assets Bankscope Loan growth rate [Growthloans] - + Annual growth rate of Gross Loans Banksope 71 3.3 Descriptive analysis Our data includes three types of banks: commercial, savings and cooperative. Table 3-4 shows the quantity of each type of bank and their percentage weight in sample Total Assets for the year 2007. As one can see from the table, even though commercial banks comprise 27 entities they hold more than half of the total assets of the sample in the given period. Cooperative banks have the lowest weight both in terms of quantity and total assets. Table 3-4 Sample characteristics Bank ownership nature Number of banks % in Sample Total Assets Commercial banks 27 58.6% Saving banks 46 38.8% Cooperative banks 18 2.6% Total 91 100% The evolution of bank risk over the observed period (See Figure 1), represented by 4 dependent variables (Z-score, LLP, LLR &NPL), shows the evident impact of the crisis indicated primarily by a rapid increase in NPL (Net Performing Loans) starting from 2007 and a decrease in Zscore over the same period. Loan Loss Reserves (LLR) and Loan Loss Provisions (LLP) act relatively similarly and also exhibit higher levels of credit risk from mid-2007 onward with a slight decrease in 2010 when massive bank reconstructions were implemented. Figure 3-1 Evolution of bank risks over 2004-2007 We also review the evolution of Z-score and NPL for each bank type (See Figure 2 & 3). In 2004 commercial banks have the highest Z-score and lowest NPL but show a downward trend in Z-score until 2008, remaining stable thereafter. Unsurprisingly, savings banks exhibit the highest level of risk in both charts, with a dramatic increase in the level of nonperforming loans from 2007 onwards. 72 Figure 3-2 Evolution of Z-score by bank type Figure 3-3 Evolution of Impaired Loans by bank type The summary of descriptive statistics for dependent variables is subdivided into pre-crisis (2004-2007) and post-crisis (2008-2011) periods. Table 3-5presents variations of coefficientsmean and standard deviations in two periods. Before the crisis the mean Z score was higher and standard deviation lower implying lower pre-crisis insolvency risk. Consequently, credit risk variables exhibit lower mean and lower dispersion from 2004-2007 than from 2008-2011. Table 3-5 Descriptive statistics for dependent variables before and after crisis Variable Obs Mean Std. Dev. Min Max 2004-2007 Total Risk of default [Z-Score] 364 41.976 54.050 0 437.486 Loan loss reserve % [Loanlossres] 282 1.909 0.362 0.23 3.144 73 NPL9 ratio % [Imparedloans] 277 0.834 0.453 0.13 3.02 Loan loss provision % [LoanLossPtoloans] 321 0.390 0.242 -1.391 1.725 2008-2011 Total Risk of default [Z-Score] 364 30.550 68.460 -1.335 789.286 Loan loss reserve % [Loanlossres] 219 2.655 1.387 0 7.802 NPL ratio % [Imparedloans] 207 4.259 2.597 0 16.1 Loan loss provision % [LoanLossPtoloans] 254 0.804 1.299 -13.621 9.837 To reveal further differences we look at the statistics across different bank types over the same periods. Figure 3-4 Mean comparison by bank type presents results for commercial banks, savings banks and cooperative banks respectively. As it can be seen in tables, statistical indicators worsen after the financial crisis. Before the crisis commercial banks show less exposure to risk than savings and cooperative banks, exhibiting higher Z-score value and lower credit risk indicators. Conversely, credit risk variables of cooperative banks show the highest exposure to credit risk in the sample with an average of 2.09 in LLR and 0.93 in NPL. From 2008-2011, all three bank types’ indicators deteriorate with lower mean and higher standard deviations. Savings banks Z-score reduces by half while the mean NPL value increases by over 500% to 4.6. Savings banks’ maximum NPL level is 16.1, which is the highest in the sample. Mean NPL of commercial banks also increases by over 500% indicating an average of 3.9 with an increased dispersion of 2.7. Figure 3-4 Mean comparison by bank type Note: * shows values for post-crisis period (2008-2011), all values are shown in % except Z-score values which are shown in natural logarithm to facilitate comparison Descriptive statistics of independent variables for all bank types before and after the crisis are given in Table 3-7. Independent variables generally exhibit similar changes as dependent variables with decreased mean and dispersed standard deviation. The most significant changes are observed in loan growth rates which drop from an average of 22.08 to 2.4 and non-deposit funding from an average of 7.2 to 4.2. Non-deposit funding, practiced by the majority of banks before crisis, later 9 NPL – Nonperforming Loan 0.00 0.50 1.00 1.50 2.00 2.50 3.00 3.50 4.00 4.50 5.00 LLR LLP NPL Z-Score LLR* LLP* NPL* Z-Score* commercial banks saving banks cooperative banks 74 becomes unpopular because of changes in market perceptions regarding the quality of market sources. ROAA also decreases by more than half post crisis though the standard deviation is not affected. Tier 1 average value increases post crisis to 9.6 reflecting banks’ adjustments to new capital requirements. Figure 3-5 Standard deviation comparison by bank type Note: * shows values for post-crisis period (2008-2011), all values are shown in % except Z-score values which are shown in natural logarithm to facilitate comparison Table 3-6 Mean and Standard deviation by bank types Variable Mean Std. Dev. 2004-2007 Commercial Saving Cooperative Commercial Saving Cooperative Total Risk of default [Z-Score] 53.65 36.85 37.58 91.75 22.39 25.735 Loan loss reserve % [Loanlossres] 1.83 1.90 2.09 0.41 0.33 0.338 NPL ratio % [Imparedloans] 0.74 0.86 0.94 0.37 0.50 0.348 Loan loss provision % [LoanLossPtoloans] 0.36 0.41 0.37 0.22 0.25 0.237 2008-2011 Total Risk of default [Z-Score] 46.55 18.71 36.80 119.70 19.99 23.408 Loan loss reserve % [Loanlossres] 2.54 2.96 1.98 1.36 1.29 1.488 NPL ratio % [Imparedloans] 3.98 4.67 3.39 2.76 2.60 1.777 Loan loss provision % [LoanLossPtoloans] 0.71 0.96 0.65 1.84 1.16 0.374 0.00 1.00 2.00 3.00 4.00 5.00 6.00 LLR LLP NPL Z-Score LLR* LLP* NPL * Z-Score* commercial banks saving banks cooperative banks 75 Table 3-7 Descriptive statistics for independent variables before and after the crisis Variable Obs Mean Std. Dev. Min Max before after before after before after before after before after Loan ratio [Netloantoas] 90 80 72.04654 67.58026 22.28932 22.62452 2.744 0.58 99.432 96.179 Loan growth rate [Growthloans] 66 78 17.83258 3.968205 11.6679 15.41529 -20.65 -53.44 53.92 88.73 Bank size [Logtotassets] 93 83 16.42201 16.86035 1.658263 1.725138 13.59312 14.17031 20.63215 20.94763 Liquid Assets/Total Assets [Liquidtototassets] 93 83 18.48633 18.3881 24.90652 26.49303 0.004205 0.013175 98.373 99.33522 Deposits [DepostoAssets] 93 83 37.51503 38.89839 17.73386 18.67572 0 0 97.21867 96.18442 Non-deposit funding [Moneymarkfunding] 93 83 8.536654 4.225763 9.25969 6.052308 0 0 34.35599 33.45022 Short-term Marketable Securities [otherdeptototal] 90 80 13.02882 6.929541 13.92997 9.361293 0 0 49.0931 48.29235 Operating Profit ratio [Opproftototearnass] 93 83 1.132035 0.511976 0.998138 0.862901 -0.65797 -3.962 5.855005 3.77572 Operating Income ratio [OpInctoearningass] 93 83 2.569445 2.334398 1.277762 1.168178 0.014359 0.020369 6.535847 5.66448 Equity/Total Assets [EqtotAssets] 93 83 6.589366 6.362048 4.495273 4.892315 1.05 1.289 27.82 26.809 Standard deviation of ROAA [var9] 93 83 0.914796 0.444084 1.16768 0.856614 -1.186 -3.993 9.239 3.477 Operating Expenses [OperExpensestoass] 93 83 1.300393 1.300333 0.707736 0.729124 0.005385 0.001852 2.884047 3.045607 Non-interest income [NonIntIn] 93 81 32.04763 28.45185 14.10629 18.07261 -6.67 -70.95 79.38 85.49 Fee-based income [FeesCommtoOpProf] 93 83 65.48755 -562.888 72.6931 6933.576 -201.835 -62700 325.3414 4240 Trading Income [TradingfeestoOpprof] 93 83 -5.04634 -31.2814 86.86801 548.3045 -825.714 -4766.67 23.63232 975 Tier 1 Capital [Tier1] 55 51 8.938182 10.04412 3.047946 3.268181 5.62 2.02 19.6 22.4 Capital [Capitrat] 46 52 11.55457 12.14231 2.202132 2.776718 8.8 3.35 19.6 22.4 Equity / Liabilities [Eqtoliab] 93 83 7.430441 7.232578 6.09944 6.68525 1.069 1.306 38.542 36.63 Securitization [RMBS] 85 34 718707.7 1123997 1881083 3363788 0 0 1.54E+07 1.85E+07 76 Table 3-8 Correlation matrix between dependent and independent variables Zscore LLR NPL LLP Loans/TA Loans Growth LnTA Liquid Assets/TA Deposits/TA Wholesale funding Operating Profit/TA Equity/TA ROAA Non-int Income Fees&Com/OP Tier1 Equity/Liabil. RMBS Board size Recordsharh Zscore LLR -0.1378* 1 NPL -0.1743* 0.8035* 1 LLP -0.1008* 0.3751* 0.4732* 1 Loans/TA 0.1025* 0.1381* 0.0356 0.0858* 1 Loans Growth 0.0174 -0.2172* -0.4841* -0.1356* -0.0627 1 LnTA -0.2155* 0.1701* 0.2066* 0.1369* -0.2189* 0.0191 1 Liquid Assets/TA 0.3485* -0.2639* -0.1962* -0.1969* -0.8494* 0.1553* -0.0168 1 Deposits/TA -0.1838* -0.0073 -0.0435 -0.1154* 0.2518* -0.0624 -0.4795* -0.2378* 1 Wholesale funding -0.0837* -0.0527 -0.1217* 0.0274 -0.1689* 0.2139* 0.3755* 0.007 -0.3266* 1 Oper. Prof/TA -0.0047 -0.3575* -0.5267* -0.2659* 0.1301* 0.1062* -0.0027 -0.1042* 0.1122* 0.0108 1 Equity/TA -0.0479 -0.0231 -0.2056* -0.0362 0.2087* 0.0133 -0.1413* -0.2998* 0.3467* -0.0315 0.5618* 1 ROAA -0.0373 -0.2600* -0.4483* -0.1770* 0.0894* 0.1589* 0.043 -0.1190* 0.0930* 0.0908* 0.8362* 0.6615* 1 Non-int Income -0.0636 -0.0157 -0.0226 0.0619 -0.3097* 0.1304* 0.2774* 0.1773* -0.0693 0.2181* 0.1521* 0.0983* 0.1647* 1 Fees&Com/OP 0.0216 -0.1721* -0.1468* -0.1696* -0.0169 0.0118 -0.0355 0.0138 0.0285 0.031 0.031 0.051 0.0259 -0.1860* 1 Tier1 0.1799* -0.0177 -0.0285 -0.0741 -0.2569* -0.1238* -0.1102* 0.2448* 0.0316 -0.2606* 0.2920* 0.5297* 0.3148* 0.1819* 0.0408 1 Equity/Liab. -0.0454 -0.0126 -0.1857* -0.0285 0.1681* 0.011 -0.1144* -0.2673* 0.3020* -0.0192 0.5667* 0.9937* 0.6829* 0.1171* 0.0446 0.5300* 1 RMBS -0.0637 -0.0479 -0.0467 0.0194 -0.031 0.0809 0.4184* 0.0284 -0.2666* 0.1488* 0.0549 -0.1242* 0.0591 0.1998* -0.0026 -0.0922 -0.1091* 1 Board size -0.0192 -0.0243 -0.0156 0.0449 -0.1257* 0.035 0.2525* -0.0264 -0.0639 0.0922* -0.0202 0.0498 0.0127 0.1242* 0.0169 0.0373 0.0448 0.1442* 1 Record. sharh -0.029 -0.0172 0.001 0.0594 -0.1366* -0.0299 0.4967* 0.1092* -0.1359* 0.1480* 0.1371* -0.1166* 0.0537 0.1943* 0.0094 -0.039 -0.1108* 0.2880* 0.1123* 1 legend: * p<.1; ** p<.05; *** p<.01 77 4 Methodology and Results The global financial crisis of 2007-2009 has further intensified interest in understanding its possible causes. Many academics and observers blame banks’ excessive risk-taking as the core cause of the global financial crisis. This study aims to test the validity of this argument through analysis of the Spanish banking sector. Our research question is: What are the main determinants of banks’ excessive risk-taking for the years 2004-2011 for a sample of Spanish banks which practically represents the whole Spanish banking sector? In empirical analysis the baseline models are first tested and then a regressor representing a certain risk factor is added to each baseline model. In so doing, we are able to observe the individual effect of each factor in a number of parsimonious models. We build as many parsimonious models as risk determinants with each risk definition. While we revised each factor in separate models, we acknowledge that they are interrelated in affecting bank risk-taking behaviour. Their combined effect on banks’ risk-taking is, therefore, hard to predict. We establish the following objectives in answering our research question: 1) to measure the influence of a range of risk determining factors on banks’ insolvency risk – Z score; 2) to assess the effect of the same risk factors on alternative risk measures – banks’ credit risk measures represented by Impaired Loans, Loan Loss Provision and Loan Loss Reserves; 3) to draw inference as to the nature of the factors which caused excessive risk-taking for Spanish banks from 2004-2011. For this we apply dynamic panel data modelling and the system Generalized Methods of Moments (GMM) method of estimation. By applying dynamic modelling we not only take into account temporal autocorrelation in the residuals, but we are also able to reduce the amount of potential spurious regression, which may lead to incorrect inferences and inconsistent estimation in static models. Besides, the coefficient of the lagged dependent variable itself is of interest to us. DPD models contain one or more lagged dependent variables, allowing for the modelling of a partial adjustment mechanism. According to Wooldbridge (2012?), models with lagged dependent variables are hard to estimate when heterogeneity and other sources of endogeneity are present. The problem of endogeneity usually appears when explanatory variables are not fully exogenous. When one applies Fixed Effect and Random Effect methods to dynamic models in the presence of unobserved heterogeneity and endogeneity, estimators are inconsistent and biased. A serious difficulty arises when using the one-way fixed effects model in the context of a DPD model when the number of years is small while the number of individual units is large -“small T, large N” data (Baum, 2012?). This happens because of a demeaning process which subtracts the individual’s mean value of y and each X from the respective variable creating a correlation between regressor and error. The resulting correlation creates a bias in estimating the coefficient of the lagged dependent variable which is not 78 mitigated by increasing N. Similar problems affect the one-way random effects model where the lagged dependent variable cannot be independent of the composite error process. Since each bank has its own culture and its own way of managing risk, and considering the possibility of an endogenous relationship between variables, we have opted for a methodology based on dynamic panel data, making estimates using the system generalized method of moments (GMM). System GMM is designed for dynamic models and is well suited to tackle the endogeneity problem. By applying Generalized Method of Moments (GMM), we believe we can construct more efficient estimates of the dynamic panel data model. The difference and system GMM estimators developed by Holtz-Eakin et al., (1988); Arellano & Bond (1991); Arellano & Bover (1995) and Blundell & Bond (1998) are designed for situations with “small T, large N” panels such as ours. They deal well with independent variables that are not strictly exogenous i.e. correlated with past and possibly current realizations of the error, with fixed effects, heteroskedasticity and autocorrelation within individuals (Roodman, 2009). In difference GMM all regressors are usually transformed by differencing (also referred to as Arellano–Bond estimation). System GMM is an extension of difference GMM (also referred as the Arellano–Bover/Blundell–Bond estimator) which augments Arellano–Bond by building a system of two equations -the original equation and the transformed equation - and making an additional assumption that first differences of instrument variables are uncorrelated with the fixed effects. System GMM was invented to tackle the weak instrument problem and allows for the introduction of more instruments and the improvement of the models’ efficiency. Our model is as follows: 𝑌𝑖𝑡 =𝛼𝑖𝑡 +𝛽1𝑌𝑖𝑡−1 +𝛽 [𝑋]𝑖𝑡 +𝛾[𝐶]it +∑𝑌𝑒𝑎𝑟𝑡 8 t=1 + 𝜀𝑖𝑡 [1] Where 𝑌𝑖𝑡 is a dependent variable representing alternative risk measures of a particular entity i in period t and 𝑌𝑖𝑡−1 is its one period lag. [X]it is a set of independent variables and [𝐶]ita set of control variables which we have already presented above. ε𝑖𝑡 represents the error term, whereas α, β and γ denote the parameters to be estimated. We have built separate baseline models for each dependent variable based on theories and empirical literature. For Z-score (measure of bank insolvency risk) our baseline model is: Equation 1 Baseline model for Insolvency risk - Z-score [𝑙𝑜𝑔𝑍𝑠𝑐𝑜𝑟𝑒]𝑖𝑡 =𝛼𝑖𝑡 +𝛽1[𝑙𝑜𝑔𝑍𝑠𝑐𝑜𝑟𝑒]𝑖𝑡−1 +𝛽2 [logGrowthloans]𝑖𝑡 +𝛽3 [logEqtoliab]𝑖𝑡 + +𝛾1[Logtotassets]it +𝛾2[Netloantoas]it+ + 𝛾3[Netloantoas]it−1 +∑𝑌𝑒𝑎𝑟𝑡 8 t=1 + 𝜀𝑖𝑡 [2] Where: 𝑙𝑜𝑔𝑍𝑠𝑐𝑜𝑟𝑒𝑖𝑡 - log of Z-score of bank i in period t and 𝑙𝑜𝑔𝑍𝑠𝑐𝑜𝑟𝑒𝑖𝑡−1 is its one period lag 𝑙𝑜𝑔𝐺𝑟𝑜𝑤𝑡ℎ𝑙𝑜𝑎𝑛𝑠 - Logarithm of annual loan growth rate 85 Table 3-11 Baseline model estimations with NPL Variable PoolOLS FE RE SysGMM L1.logNPL 0.8111*** 0.5088*** 0.8050*** 0.8959*** Growthloans2 0.0001** 0.0000 0.0001** 0.0001** logEqtoliab -0.1248*** -0.4545*** -0.1299*** -0.3545*** Logtotassets 0.0309*** 0.4880** 0.0318*** 0.0339 L1.logNetloantoas 0.1184*** 0.383 0.1215*** 0.1941* _cons -0.2605 -7.9031** -0.2689 -0.9274 R2_within 0.9325 0.917 corr(x_i,mu_i) -0.5706 sigma_u 0.6650594 0.0333496 sigma_e 0.2561225 0.2561225 rho 0.8708439 0.0166719 F 401.33 377.96 616.08 Wald chi2(12) 4321 diff AR(2) 0.274 Hansen test 0.685 No. of instruments 70 No. of groups 83 No. of observations 395 NOTE: Table reports the panel data estimates for Pooled OLS, Fixed Effect, Random Effect and the system Generalized Method of Moments where the dependent variable is the Log of NPL [logNPL] and GMM style lag limits (3 3) and estimates are robust. Year dummies are included. Hansen is a test for overidentifying restrictions, asymptotically distributed. Legend: * p<.1; ** p<.05; *** p<.01 The effect of equity ratio is consistent with earlier discussions - negative and strongly significant in all models. In system GMM estimation the size of the bank is no longer significant whilst one period lag of the total landing ratio is significant and has a positive effect. The diagnostic tests of GMM estimation show that it is a well-fitting model with statistically insignificant test statistics for both second order autocorrelation and Hansen J-statistics of overidentifying restrictions. To summarize, the results of alternative risk definitions (Z-score and NPL) in general are consistent, suggesting that the estimates do not depend on a specific definition of bank risk. Next, we test additional explanatory variables by adding them to the baseline model one by one. Table 3-12 presents the results of our estimations. Unlike the Z-score model, our NPL model exhibits more sensitivity in response to increased risk factors. Ownership Nature & Concentration In Model 1 we found that commercial banks exhibit less risk than two other types (savings and cooperative banks) which is in line with our Z-score model and Iannotta et al., (2006) who suggest that private banks are more efficient than public banks (savings banks). 86 Moreover, Model 2 exhibits its sensitivity to the ownership concentration proxy [Recordshar] which used for widely held banks and in line with Z score model results. These findings may provide evidence that banks with concentrated ownership exhibit higher risk than widely held banks. Consistent with agency theory, managers of banks with dispersed ownership exhibit lower risk than is optimal for shareholders (Iannotta et al., 2006). Moreover, Laeven & Levine (2009) argue that banks with concentrated ownership seek to compensate for utility loss from capital regulations and stringent activity restrictions by increasing the bank risk. Table 3-12 Parsimonious Models with NPL & with additional independent variables Variable Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 L1.logNPL 0.8565*** 0.8732*** 0.9032*** 0.4389** 0.6830*** 0.7433*** 0.8613*** Growthloans2 0.0001* 0.0001** .0001*** -0.0009* 0.0001*** 0.0001*** 0.0001** logEqtoliab -0.3536** -0.3884*** -0.3523*** -0.451 -0.6372** Logtotassets 0.0457 0.0645 0.0250 -0.1424 0.0438 0.0639 0.0616 L1.logNetloantoas 0.2085* 0.2043 0.2164** -0.2318 0.5607*** 0.2613*** 0.1557* Comdummy -0.1338* Recordshar -0.0013** L1.logNonintIn 0.2425* L1.logRMBS 0.1547** Tier1 0.0077 Capitrat -0.0547*** EqtotAssets -0.0504** _cons -1.1621 -1.3872 -1.6966** 2.6263 -2.2041* -1.8649** -1.5755* diff AR(2) 0.254 0.231 0.456 0.234 0.256 0.351 0.339 Hansen test 0.66 0.67 0.761 0.978 0.283 0.739 0.737 No. of instruments 70 71 78 55 42 68 70 No. of groups 83 83 83 42 69 72 83 No. of observations 395 395 394 130 332 337 395 NOTE: Table reports the panel data estimates for the system Generalized Method of Moments where the dependent variable is the Log of NPL [logNPL] and GMM style lag limits (3 3) and all estimates are robust. Year dummies are included. Hansen is a test for overidentifying restrictions, asymptotically distributed. Model1 to Model 7 are parsimonious estimates with addition of one independent variable to baseline model. Legend: * p<.1; ** p<.05; *** p<.01 In Model 3, the past percentage change of non-interest income has a positive, significant coefficient on bank risk confirming the findings of Jonghe et al., (2012), De Young & Roland, (2001), Stiroh, (2004), Stiroh & Rumble, (2006) and Lepetit et al., (2008). The positive sign for securitization proxy [logRMBS] is significant and suggests that banks may use securitization to acquire more risky assets eventually resulting in a rise in bank risk (Kero, 2010 and Carbó-Valverde et al., 2012 ). Our findings are congruous with those of Otero González et al. (2012), Kero, (2010) and Carbó-Valverde et al., (2012) and contradict those of Altunbas et al (2011) and Martín-Oliver & Saurina, (2007) who do not find evidence that banks exploit securitization to undertake riskier strategies. 87 With regard to capital ratios, Tier1 capital is insignificant whilst capital ratio and equity to assets ratio are significant and have a negative influence on the level of credit risk. These findings are supported by Berger & Bouwman (2012) Altunbas et al., (2011), Demirgüç-Kunt et al., (2010), Garcia Marco et al. and Demirgüç-Kunt et al., (2010). They state that during the crisis more focus is given to components of capital that is able to absorb losses and Tier1 ratio may not be viewed as informative in capturing the true risk in bank portfolios at this time. 4.1.2 Loan Loss Reserves Loan Loss Reserves (LLR) reflects banks’ estimated losses on loans due to defaults and nonpayment. It indicates a bank's sense of how stable its lending base is and its approach in estimating its reserves. On the one hand an increase in LLR may indicate an increase in the level of credit risk. On the other hand banks may vary when it comes to deciding how much of a loan to write off and when increased LLR may constitute evidence of prudent bank behaviour i.e. a conservative approach. Moreover, increased LLR is not always the result of bad lending decisions or risky lending decisions because macroeconomic changes may also increase banks’ expected loan losses by hitting responsible borrowers. In summary, a higher provision could mean higher expected losses and more risk exposure but at the same time indicate prudent behaviour of managers, while lower LLR could be an indication of risky behaviour if there are not enough provisions set relative to the risk they bear. The regressions with LLR are presented in Table 3-13. The baseline models demonstrate consistency of signs of explanatory variables across different regression methods, and system GMM estimates of lagged dependent variable lies within the “credible range” referred by Roodman (2009). As in previous models capital has a negative and size has a positive effect on risk though the latter does not demonstrate consistency in significant levels. Table 3-13 Baseline model estimations with dependent variable LLR Variable PoolOLS FE RE SysGMM L1.logLLR 0.9707*** 0.5311*** 0.9373*** 0.9523*** Growthloans -0.0033*** -0.0031*** -0.0035*** -0.0046*** logEqtoliab -0.1161*** -0.4755*** -0.1633*** -0.3257*** Logtotassets 0.0291** 0.1833 0.0394** 0.0619 _cons -0.1737 -1.6401 -0.1941 -0.2071 R2_within 0.4403 0.3766 corr(x_i,mu_i) -0.1002 sigma_u 0.4858954 0.1226651 sigma_e 0.28408 0.28408 rho 0.7452572 0.1571489 F 125.98 24.62 163.59 Wald chi2(12) 1009.4 diff AR(2) 0.425 Hansen test 0.334 No. of instruments 82 88 No. of groups 85 No. of observations 408 408 408 408 NOTE: Table reports the panel data estimates for Pooled OLS, Fixed Effect, Random Effect and the system Generalized Method of Moments where the dependent variable is the Log of LLR [logLLR] and GMM style lag limits (2 3) and estimates are robust. Year dummies are included . Hansen is a test for overidentifying restrictions, asymptotically distributed. Legend: * p<.1; ** p<.05; *** p<.01 The most interesting findings are the negative effect of loan growth on banks’ loan loss reserves. We originally expect a positive association with bank risk as rapid loan growth is generally blamed for an increase in bank risk by many academics (Altunbas et al., 2011; Köhler, 2012; of MartínOliver & Saurina, 2007 & Jimenez et al. 2008). Nevertheless, the results contradict the logic of prudent bank behaviour which implies reserving more loan provisions during periods of rapid credit growth. Since the provision of loan reserves involves a high degree of subjective judgement it could be used by bank managers as a tool to present a bank’s earnings in a better than realistic light. Despite the fact that loan loss reserves must be estimated looking forward it seems that this does not happen in practice; this could be evidence for the validity of income smoothing practices among Spanish banks. Laeven & Majnoni (2003), by analysing large commercial banks around the world, find that banks appear to have increased provisions during periods of positive profits whilst being less prudent during periods of rapid credit growth. Table 3-14 Parsimonious Models with LLR & with additional independent variables Variable Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8 L1.logLLR 0.9335*** 0.9221*** 0.9597*** 0.9738*** 0.9057*** 0.9386*** 0.8694*** 0.9111*** Growthloans -0.0047*** -0.0047*** -0.0045*** -0.0048*** -0.0042*** -0.0051*** -0.0041*** -0.0042*** logEqtoliab -0.3094*** -0.3386*** -0.3092*** -0.3189*** -0.4778*** -0.3493*** Logtotassets 0.0582 0.1156** 0.0511* 0.0544 0.0719** 0.0614* 0.0082 0.0024 Savdummy 0.1489** Recordshar -0.0022** FeesCommtoOP -0.0000* TradingfeestoO -0.0001* Liquidtotoasset -0.0064** logOperatexpe 0.1917** Tier1 -0.0303*** Capital ratio -0.0339*** _cons -0.2585 -0.9888 -0.0706 -0.1089 0.0422 -0.2036 0.2866 0.51 diff AR(2) 0.475 0.381 0.76 0.897 0.801 0.302 0.274 0.191 Hansen test 0.386 0.304 0.326 0.832 0.359 0.749 0.587 0.533 No. of instruments 88 83 80 80 80 99 68 66 No. of groups 85 85 85 85 85 85 69 72 No. of observations 408 408 408 408 408 408 340 345 NOTE: Table reports the panel data estimates for the system Generalized Method of Moments where the dependent variable is the Log of LLR [logLLR] and GMM style lag limits (2 3) and all estimates are robust. Year dummies are included. Hansen is a test for overidentifying restrictions, asymptotically distributed. Model1 to Model 7 are parsimonious estimates with addition of one independent variable to baseline model. Legend: * p<.1; ** p<.05; *** p<.01 89 As a result, banks on average create insufficient provisions in good times and are then forced to increase them during cyclical downturns, magnifying losses and the size of negative capital shocks. The risk factor analysis presented in Table 3-14 reveals the significant effect of the following variables: the savings banks dummy shows a positive association with LLR suggesting a higher credit risk in this type of bank or, at least, a higher level of funds in expectation of higher future losses or a conservative behaviour applied by this type of banks. The recorded shareholders variable which is the proxy for widely held banks has a negative influence on bank risk, consistent with our results for the NPL model; fee and commission income along with Trading income exhibit a weak negative influence on LLR which may prove evidence of the positive effect of income diversification also found by Altunbas et al.,(2011) and Köhler (2012), although Köhler (2012) finds that the positive effect of income diversification decreases with bank size. This could explain the small magnitude of coefficient for our sample as it comprises principally of large banks; Liquid assets have a negative association with LLR as we expected and operating expenses are positively associated with credit risk. This could be explained through the fact that banks encountered high operating expenses in periods of rapid credit growth and increased involvement in non-interest activities. As it is argued by De Young & Roland (2001), non-interest income might augment the bank’s fixed costs, such as the cost of hiring additional staff, which could increase the operational leverage of banks along with its risk-taking; capital negatively affects bank risk level and is highly significant and consistent across all models and across different capital ratios (Tier1, Capital ratio, and Equity to Liability ratio). 4.1.3 Loan Loss Provisions Our last dependent variable representing bank credit risk is Loan Loss Provision (LLP). LLP is a key accounting indicator which directly influences the volatility and cyclicality of bank earnings. In banks’ financial reports it reflects the risk of loan portfolios. Table 3-15 presents the results of the baseline regressions. Unlike other models, here the lagged dependent variable exhibits lesser dependence on its one period lag. The log of loan growth rate is insignificant [logGrowthloans] while its squared value [Growthloans2] is highly significant and positive across all regression methods. Total loan ratio and its lagged value are added to control the bank’s involvement in loans activity. They both show their influence on LLP, with the same period loan ratio having a negative effect and one period lagged value having a positive effect. Size (logtotassets) was significant and the sign indicates a positive relationship to the level of bank credit risk. Again, our system GMM lagged dependent variable lies within the “credible range” of pooled OLS and FE coefficients. The diagnostic tests of GMM estimation show that it is a well-fitting model with statistically insignificant test statistics for both second order autocorrelation and Hansen J-statistics of overidentifying restrictions. 90 The addition of risk determining factors into the baseline model reveals that LLP has positive associations with savings banks within other risk definitions. However, the result of widely held banks is insignificant indicating a lack of association between Loan Loss Provisions and the bank ownership concentration. Table 3-15 Baseline model estimations with dependent variable LLP Variable PoolOLS FE RE SysGMM L1.logLLP 0.6644*** 0.1898*** 0.6644*** 0.5636*** logGrowthloans -0.0373 0.0358 -0.0373 -0.0638 Growthloans2 0.0002*** 0.0002*** 0.0002*** 0.0002** logNetloans -0.7566* -2.0896*** -0.7566* -1.8530* L1.Netloans 0.7550* 1.1071* 0.7550* 1.8561* Logtotassets 0.0532*** -0.7158 0.0532*** 0.0990** _cons -1.1784*** 14.4712 -1.0297** -1.9147* R2_within 0.436 corr(x_i,mu_i) -0.9107 sigma_u 1.3168 0 sigma_e 0.4274 0.4274 Rho 0.9047 0 F 30.9515 16.5588 22.2701 Wald chi2(12) 371.4175 diff AR(2) 0.467 Hansen test 0.831 No. of instruments 78 No. of groups 88 No. of observations 357 357 357 357 NOTE: Table reports the panel data estimates for Pooled OLS, Fixed Effect, Random Effect and the system Generalized Method of Moments where the dependent variable is the Log of LLP [logLLP] and GMM style lag limits (2 2) and estimates are robust. Year dummies are included . Hansen is a test for overidentifying restrictions, asymptotically distributed. Legend: * p<.1; ** p<.05; *** p<.01 Similar to Loan Loss Reserves, Loan Loss Provisions exhibit a positive association with Operating Expenses and the log of Operating Expenses. As in LLR, this is explained through high operating expenses in periods of rapid credit growth as can be seen before the crisis. Therefore operating expenses increased relative to the extent of loan activities and size of banks. Remarkably, we find a positive association between LLP and deposit funding as observed in the Z –score model. This association is quite odd since we hypothesized that deposit funding is safer than its alternative wholesale funding and would be expected to reduce bank risk. We suppose that this result is not coincidental and could be evidence of ‘excessive’ competition in deposit markets prior to the crisis as mentioned by Matutes, C & Vives, X (2000) and Craig & Dinger, (2013). They argue that in response to intense deposit competition banks raise their deposit rates too high. By doing so, they attract more depositors by increasing their cost of funding while decreasing their net interest margin. Moreover, banks may use deposits for risky investments since they do not internalize the cost of 91 failure thanks to the existence of deposit insurance. This also lessens the monitoring incentives of depositors and other stakeholders. As a result banks may have higher levels of bad debts leading to increased loan loss provisions though banks use mainly deposit funding. Positive and significant effects of fee and commission income were observed as in our LLR model. This we again relate to the positive effect of income diversification found by Altunbas et al., (2011). As in the case of LLR, the significance of operating expenses may be associated with banks’ increased operational leverage referred by De Young & Roland (2001). Table 3-16 Parsimonious Models with LLP & with additional independent variables Variable Model 1 Model 2 Model 3 Model 4 Model 5 L1.logLLP 0.4865*** 0.5004*** 0.5394*** 0.5391*** 0.4398*** logGrowthloans -0.0159 -0.0005 -0.0144 -0.0156 0.011 Growthloans2 0.0002** 0.0002 0.0002 0.0003** 0.0002** logNetloans -1.8688* -2.0487** -1.8149* -2.3327** -2.7724** L1.Netloans 1.8850* 2.0068* 1.8654* 2.3886** 2.7809** Logtotasse~_ 0.054 0.1457*** 0.0521 0.1204** 0.1140** Savdummy 0.3441** DepostoAss~_ 0.0107** FeesCommto~_ -0.0000** OperExpens~_ 0.2687* logExpens 0.3897*** _cons -1.6835 -3.3125*** -1.5046 -3.1281*** -2.7348*** diff AR(2) 0.515 0.698 0.595 0.547 0.629 Hansen test 0.877 0.822 0.922 0.738 0.697 No. of instruments 84 88 88 86 86 No. of groups 88 88 88 88 88 No. of observations 357 357 357 357 357 NOTE: Table reports the panel data estimates for the system Generalized Method of Moments where the dependent variable is the Log of LLR [logLLR] and GMM style lag limits (2 2) and all estimates are robust. Year dummies are included. Hansen is a test for overidentifying restrictions, asymptotically distributed. Model1 to Model 5 are parsimonious estimates with addition of one independent variable to baseline model. Legend: * p<.1; ** p<.05; *** p<.01 92 Table 3-17 Results of parsimonious models consolidated Risk Factors Z-score Impaired Loans Loan Loss Reserves Loan Loss Provisions Ownership nature Risk increase with saving banks Risk decrease with commercial banks Risk increase with saving banks Risk increase with saving banks Ownership concentration Risk decrease with widely held banks Risk decrease with widely held banks Risk decrease with widely held banks - Bank size Risk increase with bigger size banks - Risk increase with bigger size banks Risk increase with bigger size banks Securitization - Risk increase with higher level of securitization* - - Liquid Assets - - Risk decrease with higher level of Liquid Assets - Loan growth Risk increase with higher loan growth Risk increase with higher loan growth Risk decrease with higher loan growth Risk increase with higher loan growth * Deposit funding Risk increase with higher deposit ratio - - Risk increase with higher deposit ratio Wholesale funding Risk increase with higher proportion of wholesale funding - - - Non-Interest income - Risk increase with higher proportion of non-interest income - - Fee-based income Risk increase with higher proportion of fee-based income - Risk decrease with higher proportion of fee-based income Risk decrease with higher proportion of fee-based income Trading income - - Risk decrease with higher proportion of trading income than income from commissions - Tier I ratio - - Risk decrease with higher level of Tier I ratio - Leverage ratio Risk decrease with higher leverage ratio Risk decrease with higher leverage ratio Risk decrease with higher leverage ratio - Capital ratio - Risk decrease with higher capital ratio Risk decrease with higher capital ratio - Equity to Assets ratio Risk decrease with higher Equity-toAssets ratio Risk decrease with higher Equity-toAssets ratio - - Operating expenses - - Risk increase with higher level of operating expenses Risk increase with higher level of operating expenses NOTE: Table reports the associations of factors with alternative risk based on baseline and parsimonious models presented in tables and do not necessarily show the sign of the factor but its effect on risk (Z-score, NPL, LLR and LLR). ‘–’ indicates there is no significant coefficient for this factor. * signifies non linear relation 93 Chapter 4 EARLY WARNING MODEL FOR EUROPEAN BANKS: EVIDENCE FROM THE RECENT FINANCIAL CRISIS 1 Introduction The on-going financial turmoil evident in European financial markets and the rest of the world re-emphasizes the economic importance of banks’ stability. Since the banking system is the main mechanism of financial stability, monitoring the appetite for risk within banks has become a central issue for regulators. Many economical models have been developed to forewarn regulators about the possible vulnerabilities of banks at both the systemic and bank-specific levels. These models were broadly classified as Early Warning Systems of bank fragility. However, the most recent banking turmoil has demonstrated that the majority of existing EWS models cannot fully capture the growing complexities of current financial markets. For this reason existing approaches should be reassessed critically and, where required, extended or modified to incorporate the latest changes in financial markets and risk factors. This chapter critically reviews existing approaches and models of EWS by placing more emphasis on market indicators of bank fragility. The paper undertakes a proper study of European banks and proposes possible directions for the extension or modification of existing EWS models. EWS models are widely considered as important complementary tools for bank regulators as off-site detection means. Recent empirical studies have advocated that bank supervisors should devote more attention to equity securities issued by banks. However, there has been proportionally less discussion regarding the role of equity market indicators in predicting bank distress, especially within the context of European markets. The majority of research is performed primarily in the US market and focuses on bond market indicators - subordinated debt spreads. But bond spreads are difficult to calculate in the European context due to market illiquidity and problems in constructing appropriate risk free benchmarks, especially for smaller European countries (Gropp, Vesala, & Vulpes, 2002). In contrast, market data of bank equities are available in a high frequency and the European market is itself quite efficient in processing available information. Nevertheless, relatively few empirical works focus on the predictive ability of equity market indicators for European banks. For this reason, in our literature review we concentrate mainly on bank vulnerability indicators based on equity prices and on empirical research undertaken in the scope of EU market. Our review is noteworthy as these days more European banks are downgraded by ratings agencies leading to a greater availability of data for the empirical implementation of our research. Besides, it may positively contribute to the statistical power of our estimates. 94 Our literature review has been organized in the following way: first, we give a brief theoretical overview of the main approaches of credit risk measuring models; next we examine in more detail an option theoretical structural approach in the opposition to existing alternative viewpoints and also focus on existing conflicts and debates within the approach; we then discuss the latest empirical studies on bank fragility undertaken within the European market and other international markets; finally we present our general conclusions and propose approaches to conducting our research. 2 Literature review 2.1 Brief theoretical background of default risk modelling techniques While we focus on EWS models of bank fragility we would first like to mention the concepts of Early Warning Systems in general and their assumptions. EWS are defined as functional, data-driven approaches which concentrate on variables related to past crises in order to forewarn policy makers of possibilities of future crises (Gramlich et al., 2010). EWS are based on economic theories of financial crises and make two fundamental assumptions: 1) there is causality between crisis and crisis-driving factors and 2) crisis-driving factors can be identified ex ante. Since financial markets and risk factors change rapidly, neither type of EWS is considered static and are subject to continuous reassessment and upgrading. EWS may be tailored to user objectives, model complexity and data availability. However, in light of the latest crisis there is increasing concern that existing EWS models are not sophisticated enough to capture the rapid change and growing complexity of the banking industry and financial markets. King, Nuxoll and Yeager (2005) argue that prevailing EWS models can be adjusted to those changes through the adoption of forward-looking approaches and/or by addressing various types of bank risk individually. Much emphasis has recently been placed on market forces by the Basel Committee on Banking and Supervision (Basel II). It suggests bank supervisors use market information to improve the assessment of banks’ financial safety and soundness. Under market discipline, the market correctly reflects individual bank risk levels as investors require a risk premium for any additional risk. This mechanism may increase banks’ funding costs meaning banks will be discouraged from taking additional risk. Therefore, market information could be used by banks’ supervisors as a signal and also to complement accounting data in the design of EWS. In the meantime, the principal question raised is whether market prices convey additional information which is not already included in accounting data or whether the benefits from employing market information outweigh the cost of its use (Curry et al., 2002). Empirical research of US banks supports the idea that market variables improve the assessment of banks’ financial health when added to standard call report financial data (Curry et al., 2002; Evanoff & Wall, 2001). Additionally, it is shown that the veracity of predicting a CAMEL (supervisory) rating downgrade to the lowest levels can be significantly improved by adding market variables to a set of accounting indicators, though the predictive power is found to be 101 Grop et al. (2002) argue that the negative distance to default measure is both complete and unbiased and may prove a useful leading indicator of bank fragility. It was compared to the standard bond spreads indicator; overall these two indicators were found to provide complementary information by reducing Type I errors. By estimating logit and proportional hazard models it was found that both indicators could predict bank distress up to 18 months prior to an event even taking into account the safety net effect. However, the predictive power of the indicators is different: negative DD shows poor predictive power close to the default while spread signals come only close to the event. Based on their findings, spreads signals predict defaults of only smaller banks with uninsured securities. Using a synthetic measure of the financial situation of banks based on accounting data it can be demonstrated that DD provides supplementary information relative to accounting information. Another test, developed for European banks through a specifically designed logit econometric model (Distinguin, Rous, & Tarazi, 2006), investigates how well stock market prices contribute to the improvement of predicting a bank’s distress. It has been Since European markets generally suffer from insufficient liquidity in bond markets, the work only considers a great variety of equity market indicators. It focuses only on the prediction of any downgrading of banks credit ratings (by Fitch, Standard & Poors, and Moody’s) to test the actual information content of stock prices, not on bank failure/severe financial distress. As accounting indicators and equity market data are not available at the same frequencies, the study departs from each date at which accounting data information is available (31 December) and then considers events which take place in four subsequent quarters following this date. The sample comprises 64 European banks listed on the stock market and which have at least one rating from three major ratings agencies through the period 1995-2002. Accounting indicators are ratios which are commonly used to assess a bank’s financial state and are presented in their time changes. Market indicators are constructed from daily equity prices to capture: 1) the effect of shocks or presence of abnormal returns; 2) risk changes and 3) changes in the probability of failure. The research findings confirm that the market information can act as a substitute to accounting information and conveys additional information regarding the probability of bank’s being downgraded. The accuracy of the predictive power depends on the extent to which bank liabilities are traded in the market. For those banks which rely heavily on insured and non-market priced deposits, larger subordinate debt issues do not improve prediction and the market seems to be unable to convey useful information. Size and opacity effects, at odds to the study of Gropp at al. (2002), show that they may undermine the ability of stock prices to transmit useful information on banks’ future financial health. For instance, a higher degree of opacity tends to weaken the existing link between equity market indicators and the probability of future downgrading. Earlier we debated that recent regulatory changes, financial and technological innovations have changed the existing banking environment and the causes of financial distress both in US and European markets. To capture these changes, the introduction of more risk focused and growth 102 focused indicators has been suggested so that to adapt risk measuring models to the new banking environment (King et al., 2005). This implication is empirically tested using a sample of 82 EU banks observed from 1991-2005 by Brossard et al (2007). The study constructs DD indicators to test the predictive power of bank failure, similar to the test implemented by Gropp et al (2005), while introducing a variable detecting the adverse selection effect of rapid growth strategies in their model. This indicator accounts for problems arising from banks undertaking aggressive growth strategies where they might employ lower standards in the selection and monitoring of their new assets. This is considered in the FDIC Growth Monitoring System, but its significance is still not well evidenced in EU markets. Their empirical findings confirm the robustness of DD as an early indicator of banks’ failure supporting the study of Gropp et al (2002) though a more restrictive definition of the “failure” is used. To define credit events the Individual Ratings from Fitch/ICBA are used. They also include a Support Rating from the same agency to explain the extent of the safety net a bank might benefit from in case of financial difficulties controlling “To-Big-To-Fail” effect. DD remains significant when it is joined with CAMEL accounting indicators and after the introduction of the “Too-Big-Too-Fail” effect. Another important finding is that the new indicator of the adverse selection effect improves the predictive power of the model. As we mentioned above, Merton’s model is subject to various modifications and additions. Since it uses market signals as a primary input, the model may suffer from potential shortcomings associated with the accuracy of market indicators. Well-known factors which influence data accuracy are opacity, option value effect, and moral hazard due to the safety net. These factors have been extensively discussed in academic literature. The work Auvray and Brossard, forthcoming focuses on another factor which may have a negative effect on the reliability of share prices in predicting bank distress. This factor is referred to as the level of ownership dispersion suggesting that too much ownership dispersion may impair the information content of share prices due to weaker monitoring. The study tests a sample of 76 European banks applying Merton-KMV DD and investigates whether dispersed ownership leads to weaker monitoring from shareholders and consequently a poor power of predictability of the DD indicator. The study also examines the quality of information gathered by banks’ shareholders and how well it is incorporated into banks’ share prices. The sample only contains data from the biggest and most actively traded European banks which have individual ratings from Fitch/IBCA agency between the years 1997-2005. First EWM are built using five accounting variables (CAMEL) and the DD indicator. A dummy variable is introduced to take into account the possible impact of the degree of public support. In the second step, DD is multiplied with another dummy variable which captures ownership type (dispersed or not dispersed). The test confirms that ownership dispersion of a bank’s shareholders clearly reduces the effectiveness of distance-to-default as a predictor of bank distress and bank recovery. In contrast, when ownership is concentrated it raises the predictive power of the indicator. 103 Empirical findings in emerging markets also are favourable towards equity market signals in predicting banks’ financial distress. Chan-Lau, Jobert, & Kong (2004) in their study of emerging markets’banking vulnerabilities follow Merton’s option-based structural model of credit risk (Merton, 1974) by deriving normalized distance-to-default as a risk neutral indicator of bank vulnerability. The sample period covers July 1997 to July 2003 comprising 38 banks from 14 different emerging countries. Their findings show that indicators can forewarn bank distress, defined as ratings downgrades to CCC or below, up to nine months in advance within the sample. Correspondingly, out of the sample results prove that the indicator could have signalled bank failures in Argentina by the end of 2001. The authors suggest using these indicators in real time as a policy maker’s toolkit to forecast bank crises. However, the distance to default indicator is considered to have an inherent weakness which stems from the fact that it is only a “risk neutral” measure, thus making it difficult to apply it as a “real world” objective measure of financial distress. The concept of using distance to default as a pre-default regulatory action was further developed through a number of pieces of research and case studies. Chan-Lau and Sy (2006) offer an alternative risk measure named Distance-to-Capital (DC) to serve as the regulatory purpose of bank supervisors in intervening well ahead of a bank’s default, since it involves substantial welfare costs. They argue that the original definition of DD may understate the likelihood that a bank may be required to undertake corrective actions by regulators and thus the DD may be “a bridge to far” for regulatory purposes. In contrast DC incorporates triggers embedded in the prompt-corrective-action (PCA) frameworks providing better signals as to when a bank would be required to take corrective actions or require regulatory intervention. The study suggests DC as a tool for policy makers to monitor the stability of the financial system as a whole, but it emphasizes its limitations in its application such as the existence of numerous capital thresholds and problems with the aggregation of individual bank data. Another test of the predictive power of DD for eight failed Japanese banks is proposed by Harada et al (2010). DD is calculated by a structural model of credit risk assessment based on Merton’s (1974) option pricing theory. Banks are placed into two groups based on their asset size: 3 large banks and 5 smaller regional banks. DD becomes smaller in predicting banks’ failure in many cases. They find that DD is generally a reliable measure, but a lack of transparency in financial statements and disclosed information deteriorates its predictive power. The DD spreads, defined as DD of failed banks minus DD of sound banks is also found to be a helpful indicator in predicting bank failures. Given that here we discuss bank fragility indicators based on publicly observable information, most surveys refer to a central question: to what extent market signals of bank fragility are reliable. This question is addressed by a study of the South East Asian crisis countries over the years 19961998 with collective data from 246 financial institutions (Bongini, Laeven, & Majnoni, 2002). They explore the performance of three publicly available indicators of bank fragility namely accounting 104 data, stock market prices and credit ratings by means of three different tests. First, they investigate the degree of market discipline imposed by credit ratings and stock prices before and during the onset of the financial crisis. The results confirm that neither rated nor listed banks were subject to a significant degree of market discipline. The next test is referred to as a horse race since it investigates which of the three indicators of bank fragility has more power in predicting actual bank distress. A balance sheet indicator is constructed using CAMEL ratios, by transferring them into dummy variables whenever their value is worse than that of 75% of all the sampled banks and zero otherwise. Market signals are built by using deposit insurance premiums via the implementation of Merton’s model (1977) suggested by Ronn and Verma (1986). Two key assumptions of the model are (1) that the bank’s asset values follow geometric Brownian motion and (2) that all bank debt is insured. East Asian countries governments are found to implicitly fully guarantee depositors funds. Credit ratings are taken from Moody’s ratings. The results of the second test suggest that after controlling for country specific and size factors none of the three indicators exhibits a significant amount of information with regard to discriminating distressed banks from non-distressed ones. Among these three indicators implicit deposit insurance premiums demonstrate a relatively higher power, followed by the balance sheet indicator. In general, the investigation revealed that all three indicators of bank fragility did not demonstrate common behavioural characteristics during the onset of the East Asian Crisis and did not provide considerable predictive power in the forecast of bank failure. From a dynamic perspective, stock market based indicators proved to react faster than the other two. The work concludes that in less developed financial systems it is important to use simultaneously a plurality of indicators to assess bank fragility. Though the revised empirical research has applied different approaches and methods of EWS, we can draw the following conclusions: • Distance to default indicator may prove useful for banks monitoring purposes though it suffers from its restrictive assumptions and other limitations • Majority of studies advocate paying more attention to the equity market and to information embedded in the market prices of bank’s securities • Many studies present evidence in favour of using market price based measures as early indicators of bank fragility • Stock market based information responds more quickly to changing financial conditions • The accuracy of market based indicators can deteriorate as a result of a number of factors • Further extensions of the bank fragility indicator should capture the increasing complexity and transmission of changes in financial markets. 105 In the following table we have made a comparison of the main characteristics of some of the revised EWS studies. Table 4-2 Comparison of the selected EWS studies and their main characteristics Author(s) Data analyzed Period of study Type of bank’s fragility measure Predictive power, time span Limitations Gropp, Vesala, & Vulpes, 2002 European banks 19912001 Equity-based distance to default 6-8 months, up to 24 months Fails to predict close to event Subordinate bond spread 3-6 months Predicts only in short run Chan-Lau & Sy, 2006 Japanese banks 20012003 Distance-to-Capital - Cannot be applied system wide due to existence of numerous capital thresholds Chan-Lau, Jobert, & Kong, 2004 Emerging market banks 19972003 Normalized DD (derived from Merton’s formula) Up to 9 months 1) Risk neutral measure 2) It assumes that default barrier remains constant during the period Brossard, Ducrozet & Roche, 2006 European banks 19912005 DD with CAMEL (accounting ratios) and adverse selection effect variable Up to 24 months Could be further improved with bank sensitivity to contagion effects and systematic risks. Distinguin, Rous, & Tarazi, 2005 European banks 19952002 Accounting indicators: CAEL ratios and market indicators (11 indicators including Z-score and DD) Up to 4 quarters before downgrade Depends on extent to which bank liabilities are market traded Auvray & Brossard, (2012) European banks 19972005 EWS using five accounting variables (CAMEL) and DD (Merton’s KMV model)indicator 24 quarters If bank’s ownership is dispersed this diminishes predictive power of DD 3 Other indicators of bank’s fragility Some academic papers do not focus on a set of indicators as a part of EWS models but rather concentrate on specific factors such as capital structure, funding structure, business models, etc. and investigate their power in predicting bank distress. Similarly, a number of studies concentrate on analyzing how well certain indicators performed in reflecting financial soundness/ fragility of banks during the recent financial crisis. In this section we review a few of these studies and discuss the validity of the proposed factors. Demirgüç-Kunt et al., (2010) research the role of bank capital in withstanding shocks such as the financial crisis. In particular, they investigate whether better capitalized banks have higher stock returns during financial crises. Also, they discuss which concept of capital is more relevant in stock valuation during crises and what items are counted as capital for regulatory purposes. The baseline model measures bank performance through changes in a bank’s stock prices between quarters relating it to changes in its level of capital. It uses dummy variables which account for any possible omitted country-level effects such as macroeconomic shocks, systematic components, etc. and a matrix 106 of bank-level control controls for bank-specific features (such as bank liquidity, reliance on deposits for funding, etc.) The results obtained from a large sample of international banks suggest that during the crisis banks with higher capitalization were better valued than undercapitalized banks, though this trend is not observed before the crisis. Moreover, they find that big banks’ stock returns are more sensitive to the leverage ratio as a capital measure than to the risk-adjusted Basel ratios. This may be explained by a lack of reliability of later indicators by market participants at the time of the crisis. Finally, it concludes that “higher quality capital” – Tier 1 and tangible common equity are more relevant. The role of bank capital is analysed by Berger & Bouwman (2012). The study examines the effect of capital on bank performance and whether it varies across financial crises and periods of financial stability. Here bank performance is measured in terms of survival and market share. The research has two baseline regressions which empirically measure the effect of capital on bank survival and on market share over different time periods. Potential omitted variables are covered by a broad set of control variables. The main findings of the study support the general hypothesis that capital helps banks to survive. It demonstrates that for small banks capital is essential for survival at all times while for medium and large banks it si essential only during banking crises. Capital helps small banks to improve their market share at all times, while for medium and large banks it is helpful only during banking crises Similar analysis has been performed by Altunbas et al., (2011) for a large sample of listed banks in the EU and US. They observe complex financial indicators before and during the crisis and investigate whether the variability across bank business models is related to their realized risk during the financial crisis. Realized bank risk is measured by several indicators such as the likelihood of bank rescue, systematic risk and intensity of recourse to central bank liquidity. Probit and linear regressions are applied to three measures of risk and a group of independent variables. To measure bank distress during a crisis the study employs regression quantile techniques. The results reveal that higher levels of capital decrease bank risk, though this is argued to be a non-linear relationship. They also find that ex-post bank risk is associated with ex-ante bank size and the degree of credit expansion in the years preceding a crisis. Moreover, they argue that banks with more deposit base funding are less risky than banks with a higher market funding. In general, the study encourages bank supervisors to distinguish the impact of different business models on bank risk to explain the divergence in risk realization during crises. The idea of the importance of funding stragegy in defining the level of bank riskiness is now quite popular among academics. The implications of a bank's funding strategy for bank risk and return is investigated by Demirgüç-Kunt & Huizinga (2009) among others. The study examines how bank activity and short-term funding strategies affect risk and return trade-off. The period covered is 19952007 and comprises international banks with stock exchange listings. The study first intends to 107 explain variations in income and funding share through a range of bank level, bank environment variables. Next, the relationship between fee income and non-deposit funding on bank risk and return is tested. For the purpose of robustness two alternative measures of bank risk and return are tested. The possibility of endogeneity in bank risk and return is also revised. Research findings support the idea that a higher non-interest income or non-deposit funding level contributes to higher bank risk, though the impact of both variables on bank return is difficult to explain due to endogeneity concerns. The study concludes that, overall, traditional banks – with heavy reliance on interest income and deposit funding - are safer. Alternatively, Huang & Ratnovski (2010) also research the effect of bank funding strategies and specifically wholesale funding impact. They present two alternative models: “bright side” and “dark side” of wholesale bank funding. As a benchmark of “bright side” wholesale funding the CK (Calomiris & Kahn, 1991) model is taken. It is then contrasted with an alternative “dark side” model with the introduction of costless and noisy signals of bank project quality. The results reveal that wholesale funding is beneficial when providers are informed. But with the presence of noisy public signals the incentives of fund providers to monitor banks and impose market discipline may be distorted and lead to inefficient liquidation of a bank. The negative effects of wholesale funding relate to banks with extended exposures to standardized and tradable arm’s length assets, with readily available public information and when wholesale funds are senior claimants. 4 Data and Empirical Methodology Despite the fact that the sphere of credit risk measurement has been well explored, the fundamental techniques still have the same objective - that is to identify factors, using a common set of variables and/or financial ratios that differ in a systematic way between failed and non-failed banks. We measure European banks’ default predictability using Moody’s Analytics’ Expected Default Frequency (EDF). EDF is the market-based credit measure developed by Moody’s KMV which is based on Merton’s option-pricing theory’s distance to default indicator. It calculates distance to default of an entity through mapping it to MKMV’s empirical default database and provides with the regularly updated probabilities that a company defaults within a given time horizon, where default means the failure to make scheduled debt payments. EDF has two distinctive characteristics: it is grounded on corporate theory unlike reduced models and incorporates market information unlike structural credit risk models. Moreover, it incorporates more realistic assumptions which better adjusts it to reflect real-world default dynamics. EDF metrics range from 1 basis point to 35% and provides with absolute default probability estimates. It is argued by number of empirical works (Bohn, 2000 & Agrawal, Arora, & Bohn, 2004) as being efficient measure of firms’ defaults which generally outperforms the basic Merton structural model and Hull and White’s reduced form models. 108 Our methodology is based on using 1 year and 5 years EDFs of European banks and test whether they discriminate distressed banks from non-distressed ones. We take Moody’s credit ratings as a “default” variable. Whenever the bank’s rating falls to D+ or below it is accepted as “default” event. We add other variables suggested by previous studies such as adverse selection effect variable and accounting variables to test if the efficiency of our baseline model increases with their addition. Our final model comprises of the combination of the variables which is found efficient in predicting banks’ defaults for the given sample of European banks. Our main research question is: Are conventional models able to predict the financial distress of banks before the onset of financial crises? By answering this question we also examine if EDF is a good measure for distinguishing fragile banks from not fragile ones? If so, what parameters can be improved and how can the model be improved so as to make a robust estimator of EU bank defaults? Our sample comprises of 93 European10 banks which are selected by applying the following selection criteria (Table 4-3): Table 4-3 Criteria of the search strategy World Region/Country European Union of 15 Accounting standards International Accounting Standards, International Financial Reporting Standards (IFRS) Specialisation Commercial banks, Savings banks, Cooperative banks, Real estate &mortgage banks, Investment banks, Bank holdings & Holding companies, Private banking / Asset management companies Listed Banks Listed Banks Moody’s EDF data Expected Default Frequency for 1 year and 5 years Time Period 2005 1st quarter – 2011 4th quarter We start our sample selection process from a list of European banks which have the Expected Default Frequency (EDF) data for the selected time period. The Accounting and Moody’s rating data is taken from the BankScope International Bank Database provided by Fitch/Bureau Van Dijk. Since downgrades are announced in an irregular manner and do not occur at equally spaced intervals, we decided to use quarterly data to better match the announcement times. 4.1 Dependent Variable Our dependent variable should indicate the occurrence of bank default. However, among European banks formal defaults are very rare. Prior to actual default there is no way to explicitly discriminate between banks that may default and those that may not. We can, therefore, only use proxies which are based on the probabilistic likelihood of default. In the majority of previous studies academics use credit agencies’ ratings as a proxy. The credit ratings discriminate between sound banks and fragile ones by assigning them a particular score indicating the financial state of an entity. 10 List of analysed banks are provided in Appendix C 109 In our research we use Moody’s Bank Financial Strength ratings as a default indicator. The rating reflects a bank’s intrinsic financial strength and ranges from A to E including ‘+’ and ‘-’ qualifiers, where A is assigned to the strongest banks. In line with Gropp et al. (2002) and Brossard et al. (2006) we consider a bank’s downgrade to D+ or below as a proxy for being “defaulted” as it reflects a substantial weakening of a bank’s financial strength. Banks with a D rating display modest intrinsic financial strength and may require external support. BFSR D+ ratings are mapped to the Baseline Credit Assessment (BCA) rating as baa3-ba1. BCA ratings reflect opinions of issuers’ standalone intrinsic strength, notwithstanding any extraordinary support from an affiliate or a government. Banks which have a rating lower than baa3 are characterised as having speculative intrinsic, or standalone, financial strength and are considered as having substantial credit risks. Table 4-4 shows the total number of observations for each group in our sample. Defaulted banks account for 90 observations in the sample. These are banks which were downgraded by Moody’s agency to D+ or lower levels (D+, D, D- & E) within the observed quarters. Not rated banks signify banks which did not receive Moody’s ratings for the observed period. Table 4-4 Statistics of the “default” events in the sample Freq. Percent Cum. Not defaulted Banks 669 8.31% 8.31% Defaulted Banks 90 1.12% 9.42% Not rated Banks 7296 90.58% 100% Total 8055 100% A general statistics table of the sample shows that about 12% of all rated banks have experienced a downgrade event, with a frequency of 90 out of 759 rated banks. It comprises only 1.12% of the overall sample. Only 9.4% of the banks in the sample received Moody’s ratings in the observed period (from 2005 1st quarter to 2011 4th quarter). As we mentioned above, formal bank bankruptcy is a rare event for European banks. We are aware that rare events could bring bias to model estimations, though with the given sample size and the proportion of positive outcomes we believe there is no bias in maximum likelihood estimation. 4.2 Independent Variables and hypothesis Our main independent variable is Expected Default Frequency – EDF. Essentially, EDF is based on Merton’s option-based structural model of credit risk. The first structural model presented by Merton (1974) applies option pricing theory developed by Black and Scholes (1973) to model a firm’s debt. The classical Merton model implies a relationship between the unobservable variable (Asset) and observable variables (Equity and Debt). It supposes that a firm’s equity value is equivalent to a European call option on the firm’s asset with the strike price equal to the debt’s face value. The model measures the creditworthiness of an entity through the “distance to default” indicator. Distance to 110 default denotes the number of standard deviations of assets’ volatility to an entity’s default point. The higher the distance to default, the lower is the default risk. Vasicek and Kealhofer have extended the Black-Scholes-Merton framework to produce a model of default probability known as the VasicekKealhofer model (VK model). KMV (Moody’s KMV) is a commercial version of the VK model which provides commercially available EDF measures for firms and financial institutions. EDF, unlike the distance to default measure, represents the probability of default i.e. the likelihood of a bank being insolvent within a specified period of time. Here, default is an event when a company is not able to meet its debt obligations or when the market value of the firm's assets is less than the book value of the firm's liabilities by the time the debt matures (Crosbie & Bohn, 2003). To calculate the probability of default one needs to determine the distance to default value first. Distance to default measures the difference between the asset value of a firm and the face value of its debt, scaled by the standard deviation of the firm’s asset value:             − −−−− =tT LtTA DD t σ σ µ φ ln))( 2 ()ln( 2 [2] Consequently, the probability of default or EDF can be defined as the cumulative normal distribution of the distance to default: [ ] DDPD −= φ [3] As can be seen from the equation [2], the probability of default (PD)/EDF are inversely proportional to the distance to default measure where the higher PD/EDF values indicate lower distance to default. Despite being well grounded in economic theory, EDF is a forward looking measure based on market information. It uses the asset value of entities based on information from equity prices, and these prices reflect current and future estimates of market participants. EDF is measured in regular base and provides a continuous credit monitoring process that is difficult and expensive to duplicate. Its values are estimated based on Moody’s historical database identifying the proportion of entities with particular distance to default who actually defaulted within a particular time period. It is believed by many practitioners that using such a database improves default predictions enormously. Considering all the above, we believe that EDF is a reliable measure of banks’ credit quality and thus an appropriate measure for our analysis. We use both the bank ratings and EDF of Moody’s Analytical Services in our estimates; it is logical to question whether these two measures replicate each other. EDF does not correspond fairly to default probabilities mapped to agencies’ ratings. Moody’s explanation is that these discrepancies reflect the difference in nature of the measures. Ratings represent long-term views of credit risk and 117 It is worth mentioning, except dependent variable, when there are missing values in quarterly data, in line with study of Brossard et al. (2006) we duplicate the previous quarter value until end of the calendar year. In doing so we are able to keep more observations while not stretching out financial information of one year to another. To avoid supplementary autocorrelation which could arise with duplication of data we use robust standard errors adjusted for clustering between banks. Consistent with our a priori expectations all lagged EDF signs are positive in all estimations suggesting positive correlation between bank defaults and 1 year EDF. The results are significant at 99% level in all 8 lags. The highest Pseudo R is observed in lags 4, 5 & 6 lags. Table 4-10 EDF 1: Baseline Model Pooled logit 1 lag Pooled logit 2 lags Pooled logit 3 lags Pooled logit 4 lags Pooled logit 5 lags Pooled logit 6 lags Pooled logit 7 lags Pooled logit 8 lags edf1year L1. 0.635*** L2. 0.963*** L3. 1.012*** L4. 1.138*** L5. 1.186*** L6. 1.235*** L7. 1.205*** L8. 1.068*** _cons -2.968*** -2.773*** -2.645*** -2.560*** -2.456*** -2.355*** -2.285*** -2.209*** N 495 556 587 620 653 686 683 680 𝑃𝑠𝑒𝑢𝑑𝑜 𝑅2 0.048 0.087 0.087 0.098 0.096 0.095 0.081 0.065 𝑊𝑎𝑙𝑑 𝜒2 14.064 12.051 11.441 11.949 11.607 11.282 8.322 7.849 Log likelihood -115.381 -153.497 -174.74 -195.349 -217.668 -241.388 -244.678 -248.413 edf1year is EDF for one year; L1.- L8. are EDF values lagged for 1-8 quarters; _cons – constant; legend: * p<.1; ** p<.05; *** p<.01 Table 4-11 presents average marginal effect (AME), marginal effect at means (ME) and odds ratios of the baseline model with 4 quarter lags. Both marginal effect methods show almost the same magnitude of EDF variable - about 10%. AME represents the average of each bank’s marginal effect while ME is computed at the average bank. The ME implies that one unit change in EDF from the average 0.303 is associated with 10.1% increase of likeliness of default. Odds ratios represent odds of being defaulted when EDF increases by one unit. In other words, for one unit increase in edf1year value, the expected change in log odds is 3.121 i.e. more than 3 times more likely to be defaulted. 118 Table 4-11 EDF 1: average marginal effect, marginal effect at the means & odds ratios AME ME (Mean) Odds ratios edf1year L4. 0.100*** 0.101*** 3.121*** (0.303) Cons 0.077323 Note: AME – Average Marginal Effect, MEMarginal Effect at means, Mean values of ME are given in brackets; legend: * p<.1; ** p<.05; *** p<.01 Table 4-12 presents the results of baseline regression with EDF 5 years. The coefficient of the variable is positive and highly significant in all lags. Comparing test statistics to EDF with 1 year horizon, the 5 year EDF exhibit slightly higher pseudo R squared results in respective models (4, 5 and 6 lags). Table 4-12 EDF 5: baseline model Pooled logit 1 lag Pooled logit 2 lags Pooled logit 3 lags Pooled logit 4 lags Pooled logit 5 lags Pooled logit 6 lags Pooled logit 7 lags Pooled logit 8 lags edf5years L1. 0.499*** L2. 0.581*** L3. 0.609*** L4. 0.663*** L5. 0.682*** L6. 0.703*** L7. 0.686*** L8. 0.613*** _cons -3.016*** -2.922*** -2.793*** -2.698*** -2.586*** -2.480*** -2.403*** -2.312*** N 525 556 587 620 653 686 683 680 Pseudo R2 0.084 0.101 0.102 0.108 0.105 0.102 0.09 0.074 Wald χ2 18.103 15.647 14.512 14.015 13.549 13.102 11.257 10.993 Log likelihood -131.866 -151.076 -171.998 -193.125 -215.615 -239.408 -242.233 -246.031 edf1year is EDF for one year; L1.- L8. are EDF values lagged for 1-8 quarters; _cons – constant; legend: * p<.1; ** p<.05; *** p<.01 Table 4-13presents average marginal effect (AME), marginal effect at means (ME) and odds ratios of the baseline model with 6 quarter lags. We take deeper lag than in edf1year case because we want to test if edf5years may predict bank defaults for more than one year in advance. The marginal effect of EDF 5 years is about 7%. Odds ratio reports that the likeliness of being downgraded with one unit increase in EDF value is more than two times. Table 4-13 EDF 5: Average marginal effect, marginal effect at the means & odds ratios AME ME (Mean) Odds ratios edf5years L6. 0.070*** 0.073 2.021*** (0.648) Cons 0.084 Note: AME – Average Marginal Effect, MEMarginal Effect at means, Mean values of ME are given in brackets; legend: * p<.1; ** p<.05; *** p<.01 119 5.1 Baseline model and CAMEL covariates We follow the existing literature and add CAMEL covariates to our baseline model. Before showing the regression results we review each CAMEL ratio with two-sample t-tests on the equality of means with unequal variances. The tests show whether the value of ratios for two groups of banks are different from those observed before the onset of the crisis, signalling the deterioration of the bank’s financial state. Table 4-14 ROAE, two-sample t-test with unequal variances ROAE Status N Mean Difference T 1-quarter lag 0 323 9.358 7.567 3.893*** 1 51 1.792 2-quarter lag 0 321 9.490 7.483 4.320*** 1 60 2.008 4-quarter lag 0 327 9.674 6.381 4.378*** 1 76 3.293 6-quarter lag 0 282 10.319 7.294 4.676*** 1 73 3.024 8-quarter lag 0 244 10.694 6.790 3.894*** 1 66 3.904 Note: Status 0 is Not downgraded banks, Status 1is Downgraded banks; Difference is mean (Status=0) – mean (Status=1); t is t-statistics for testing the hypothesis that difference is not equal to 0; legend: * p<.1; ** p<.05; *** p<.01. Table 4-15 Assets quality, two-sample t test with unequal variances Impaired loans/ Gross Loans Status N Mean Difference T 1-quarter lag 0 200 3.074 -2.222 -4.445*** 1 41 5.296 2-quarter lag 0 200 3.025 -2.144 -4.289*** 1 46 5.169 4-quarter lag 0 205 2.907 -1.981 -4.552*** 1 57 4.888 6-quarter lag 0 169 2.666 -1.641 -3.457*** 1 50 4.307 8-quarter lag 0 141 2.480 -0.948 -2.046** 1 42 3.428 Note: Status 0 is Not downgraded banks, Status 1is Downgraded banks; Difference is mean (Status=0) – mean (Status=1); t is t-statistics for testing the hypothesis that difference is not equal to 0; legend: * p<.1; ** p<.05; *** p<.01. As can be observed from the table the differences between non-defaulted and defaulted banks are positive in all lagged periods suggesting that non-defaulted banks have higher equity ratios than defaulted banks. The 1, 6 and 8 quarter lags exhibit 99% significance, unlike CAMEL’s capital ratio assets quality indicator which shows no significance at any lagged period with a positive difference up 120 to 1 year and a negative difference afterwards. The t-tests of the rest CAMEL’s covariates are provided in Appendix C. EDF is market based indicator and we expect that it reflects past balance sheet information as well as future expectations about the bank’s financial state. Nevertheless we test the predictive power of the model with addition of all CAMEL covariates lagged up to 4 quarters. Table 4-16 EDF 1 with CAMEL covariates Variable Pooled logit CAMEL Pooled logit 4 lags Pooled logit 5 lags Pooled logit 6 lags edf1year L4. 1.070*** L5. 1.058*** L6. 1.162*** equitytototassets L4. -0.236*** 0.012 -0.004 0.01 imploanstogrossloans L4. 0.347*** 0.256*** 0.275*** 0.276*** costtoincome L4. 0.038*** 0.042** 0.045*** 0.043** roae L4. -0.081*** -0.095** -0.095*** -0.094*** liqassettototdepbor L4. -0.073*** -0.068*** -0.070*** -0.072*** _cons -1.645 -3.579** -3.573** -3.401** N 262 204 229 252 Pseudo R2 0.279 0.382 0.372 0.362 Wald χ2 61.554 42.473 52.747 59.046 Log likelihood -98.969 -63.238 -74.652 -85.942 edf1year is EDF for one year; L1.- L6. are variable values lagged for 1-6 quarters; _cons – constant; legend: * p<.1; ** p<.05; *** p<.01 Table 4-16 shows the results of regressions with both CAMEL ratios only and a combined model with EDF1 and CAMEL. When we regress CAMEL covariates without EDF, all variables’ coefficients become significant at the 99% level. Moreover, they are also in line with our hypothesis. The results show that bank defaults are negatively associated with capital ratio, earning and liquidity ratios. Defaults are positively associated with impaired loans ratio and management efficiency measure. When we combine edf1 with 4, 5 and 6 lags and CAMEL components, edf1 demonstrates high level of significance in all three specifications. The test statistics of the combined models are also improves in comparison with their earlier specifications. Pseudo R increases to 38.2%. All CAMEL components except capital ratio remain significant. Brossard et al., 2006 also combine distance-todefault measure with CAMEL covariates. They explain lower insignificance of capital ratios with the relative homogeneity of European banks’ capital indicators. They argue that European banks maintain their capital ratios in accordance with the Basel II regulatory framework and thus do not vary to a great extent similar to US banks’ capital ratios. This fact may reduce the signalling power of equity 121 ratio in our model too. Our t-test on the equality of means also reveals that there is no significant difference between defaulted and non-defaulted banks’ equity indicators. We combine CAMEL covariates with EDF 5 years (see regression results in Appendix C). In general the results are similar to what we received for 1 year EDF. All CAMEL ratios are significant except equity ratio. The EDF 5 variable remains significant at 99% in all 3 models. In general the test results for both 1 year EDF and EDF 5 years suggest that they convey additional information to that already provided by CAMEL ratios, and that such a combination of variables improves the predictive power of the models. Table 4-17 EDF 5 with CAMEL covariates Variable Pooled logit 6 lags Pooled logit 7 lags Pooled logit 8 lags edf5years L5. L6. 0.865*** L7. 0.790*** L8. 0.757*** equitytototassets L4. 0.077 0.054 0.052 imploanstogrossloans L4. 0.258*** 0.258*** 0.267*** costtoincome L4. 0.047*** 0.049*** 0.052*** roae L4. -0.088** -0.090** -0.098*** liqassettototdepbor L4. -0.074*** -0.077*** -0.083*** _cons -4.333*** -4.069*** -4.090*** N 252 252 252 Pseudo R2 0.38 0.37 0.378 𝑊𝑎𝑙𝑑 𝜒2 59.917 60.539 58.893 Log likelihood -83.541 -84.91 -83.864 Edf5 year is EDF for five years; L1.- L8. are variable values lagged for 1-8 quarters _cons – constant; legend: * p<.1; ** p<.05; *** p<.01 We remove equity ratio from our models and run the regression with EDF and 4 CAMEL components. The marginal results are provided in Table 4-18. Average magnitude of EDF 1 year ranges from about 10% to 15%. Average marginal effect of impaired loans is between 2.4%-3.6%. Conditional means which are computed at the average bank show that one unit change in NPL ratio from the average 3.1%-3.3% is associated with 2.8%-3.6% increase of likeliness of default. Marginal effect of the combined mod with EDF 5 years and selected CAMEL variables are provided in Appendix C. Similar to EDF 1 year edf5years is significant at 99% level in all three models. Its average margin ranges from 7.5% to 9.7% which is slightly lower than one of edf1year. Magnitudes of CAMEL covariates are similar to that of EDF 1 year. 122 Table 4-18 EDF 1 & selected CAMEL ratios, marginal effect 4 lags 5 lags 6 lags edf1year AME ME (Mean) AME ME (Mean) AME ME (Mean) L4. 0.099*** 0.116** (0.476) L5. 0.106*** 0.126** (0.456) L6. 0.122*** 0.149*** (0.439) imploanstogrossloans L4. 0.024*** 0.028*** 0.027*** 0.032*** 0.029*** 0.036*** (3.051) (3.192) (3.304) costtoincome L4. 0.004** 0.005** 0.004*** 0.005** 0.004** 0.005** (63.188) (62.943) (62.799) roae L4. -0.009*** -0.010*** -0.009*** -0.011*** -0.010*** -0.012*** (8.154) (8.118) (8.083) liqassettototdepbor L4. -0.006*** -0.007*** -0.007*** -0.008*** -0.008*** -0.009*** (23.797) (23.695) (23.737) Note: AME – Average Marginal Effect, MEMarginal Effect at means, Mean values of ME are given in brackets; legend: * p<.1; ** p<.05; *** p<.01 We also test the validity of adverse selection effect for our models, similarly to Brossard et al (2006) and other studies. This test demonstrates whether aggressive growth undertaken at an earlier time affects a bank’s subsequent default. We examine two alternative measures of adverse selection effect: moving averages of annual growth - total assets and gross loans - with different time lags. To estimate we add an adverse selection variable to models with the selected CAMEL components, producing two different results for each definition of adverse selection effect. The addition of the assets growth variable with no lag (moving average of assets growth in the same year when the default happens) does not give significant results. After some iteration of the process we find that 4 quarter lagged moving average of assets growth has a more apparent effect. The results for 1 year EDF with past assets growth are given in Table 4-19. In line with other studies, our results suggest that past assets growth is positively associated with the probability of bank default and is significant at 95%. It suggests the perilous consequences of banks’ rapid/aggressive growth strategies that may trigger asset quality deterioration and lead to severe downgrades of bank ratings. 123 Table 4-19 EDF 1 with CAMEL components & adverse selection effect: assets growth Pooled logit 4 lags Pooled logit 5 lags Pooled logit 6 lags edf1year L4. 1.256** L5. 1.257*** L6. 1.466*** ma4growthtotas L4. 0.119** 0.096** 0.088** imploanstototalassets L4. 0.241*** 0.270*** 0.283*** costtoincome L4. 0.035* 0.040** 0.038** roae L4. -0.104** -0.100*** -0.098*** liqassettototdepbor L4. -0.067*** -0.068*** -0.071*** _cons -0.067*** -0.068*** -0.071*** N 198 223 246 Pseudo R2 0.41 0.394 0.383 Wald χ2 35.254 55.368 61.267 Log likelihood -59.61 -71.132 -82.128 edf1year is EDF for one year; L1.- L6. are variable values lagged for 1-6 quarters; _cons – constant; legend: * p<.1; ** p<.05; *** p<.01 The results for EDF 5 years are provided in Table 4-20. Similar to 1 year EDF, the signs of past assets growth are positive and significant in all three specifications. The addition of the new variable does not weaken the significance of EDF but brings more predictive power to the models. Table 4-20 EDF 5 with CAMEL components & adverse selection effect: assets growth Pooled logit 4 lags Pooled logit 5 lags Pooled logit 6 lags edf5years L6. 1.022*** L7. 0.902*** L8. 0.728*** ma4growthtotas L4. 0.093** 0.081** 0.052 imploanstototalassets L4. 0.279*** 0.272*** 0.279*** costtoincome L4. 0.043** 0.045** 0.050*** roae L4. -0.092** -0.093*** -0.099*** liqassettototdepbor L4. -0.078*** -0.080*** -0.085*** _cons -4.116*** -3.913*** -3.780*** N 246 246 246 Pseudo R2 0.403 0.387 0.381 Wald χ2 62.279 61.22 61.235 Log likelihood -79.483 -81.643 -82.472 Edf5years is EDF for five years; L1.- L8. are variable values lagged for 1-8 quarters; _cons – constant; legend: * p<.1; ** p<.05; *** p<.01 The test statistics are also improved: pseudo R increases to 38.3%-41% and the probabilities of 𝑊𝑎𝑙𝑑 𝜒2, which tests if all the coefficients are different than zero, are less than 0.05. In general, the 124 addition of 4 quarter lagged average assets growth supplement to the models’ overall test statistics and significance of the variables. Surprisingly, our alternative measure for adverse selection effect - the past growth of gross loans exhibits negative relation with the bank defaults. The sign does not change with a variation of lags of the added variable for both EDF 1 year and 5 years. The following tables show the results of the combined models with loan growth. Table 4-21 EDF 1 with CAMEL components & adverse selection effect: loan growth Pooled logit 4 lags Pooled logit 5 lags Pooled logit 6 lags edf1year L4. 0.830** L5. 0.977** L6. 1.345*** ma4growthgrossloans -0.060*** -0.063** -0.070*** imploanstototalassets L4. 0.254** 0.231** 0.223* costtoincome L4. 0.063*** 0.065*** 0.069*** roae L4. -0.079** -0.081** -0.085** liqassettototdepbor L4. -0.115*** -0.114*** -0.124*** _cons -4.291*** -4.398*** -4.517*** N 159 159 159 Pseudo 𝑅2 0.467 0.477 0.498 Wald 𝜒2 36.931 36.892 32.084 Log likelihood -38.578 -37.912 -36.335 edf1year is EDF for one year; L1.- L6. are variable values lagged for 1-6 quarters; _cons – constant; legend: * p<.1; ** p<.05; *** p<.01 As it is seen from results, the test statistics of the all models improved with the addition of loan growth variable. The EDF is significant in 95% and 99% confidence levels in 1 year and 5 years models respectively. Variable accounting for loan growth has negative sign and significant. Pseudo R squared is raised up to 50% and 51% approximately for one year and five-year EDF respectively. Table 4-22 EDF with CAMEL components & adverse selection effect: loan growth Pooled logit 6 lags Pooled logit 7 lags Pooled logit 8 lags edf5years L6. 0.855*** L7. 0.955*** L8. 1.717*** ma4growthgrossloans -0.069** -0.075*** -0.069*** imploanstototalassets L4. 0.228* 0.233* 0.236* costtoincome L4. 0.072*** 0.083*** 0.085*** roae L4. -0.076* -0.078* -0.069 liqassettototdepbor L4. -0.128*** -0.144*** -0.140*** 125 _cons -4.960*** -5.203*** -5.533*** N 159 159 159 Pseudo 𝑅2 0.503 0.506 0.495 Wald 𝜒2 32.57 34.165 30.699 Log likelihood -35.999 -35.751 -36.604 Edf5years is EDF for five years; L1.- L8. are variable values lagged for 1-8 quarters; _cons – constant; legend: * p<.1; ** p<.05; *** p<.01 Table 4-23 and Table 4-24 report the marginal effect of models with adverse selection variables: with asset growth and with growth of loans. The average marginal effect of 1 year EDF in models with asset growth is 11.3%, meaning a one-unit increase in EDF on average leads to an increase in probability of default of 11.3%. The alternative measure - marginal effect at means for average bank - reaches 13.7%. The impaired loans ratios have magnitudes of 2.2% and 2.6% with a significance of 99% in both methods. There is a positive association of past assets growth with the likelihood of bank default with marginal effects of 1.1% and 1.3% in the two alternative methods. Marginal effects of EDF in the loan growth model are lower than in the assets growth model, 5.9% and 5.6% respectively in the two alternative methods. Past loan growth has a negative sign suggesting favourable influence on banks’ ratings although with a marginal effect of less than 1%. Table 4-23 EDF 1 Average marginal effect and Marginal effect at the mean Assets growth Loan growth AME ME (Mean) AME ME (Mean) edf1year L4. 0.113** 0.137** 0.059** 0.056* (0.490) (0.460) ma4growthtotas L4. 0.011** 0.013** (2.592) ma4growthgrossloans -0.004*** -0.004** (-1.960) imploanstogrossloans L4. 0.022*** 0.026*** 0.018** 0.017** (3.070) (2.761) costtoincome L4. 0.003* 0.004 0.005*** 0.004*** (63.317) (63.993) roae L4. -0.009*** -0.011*** -0.006** -0.005** (7.954) (8.107) liqassettototdepbor L4. -0.006*** -0.007*** -0.008*** -0.008*** (23.627) (24.750) Note: AME – Average Marginal Effect, MEMarginal Effect, (Mean)-mean values of ME; legend: * p<.1; ** p<.05; *** p<.01 126 Table 4-24 EDF 5 Average marginal effect and Marginal effect at the mean Assets growth Loan growth AME ME (Mean) AME ME (Mean) 0.092*** 0.117*** 0.064*** 0.062** edf1year L4. (0.894) (0.741) ma4growthtotas L4. 0.008** 0.011** (3.136) ma4growthgrossloans -0.005*** -0.005** (-1.960) imploanstogrossloans L4. 0.028*** 0.035*** 0.016* 0.015** (3.326) (2.761) costtoincome L4. 0.005** 0.006** 0.006*** 0.005** (62.893) (63.993) roae L4. -0.009*** -0.012*** -0.005** -0.005* (7.920) (8.107) liqassettototdepbor L4. -0.008*** -0.010*** -0.010*** -0.009*** (23.599) (24.750) Note: AME – Average Marginal Effect, MEMarginal Effect, (Mean)-mean values of ME; legend: * p<.1; ** p<.05; *** p<.01 The average marginal effect of 5 year EDF in the assets growth model suggests that a one-unit increase in EDF on average leads to an increase in probability of default of 9.2% and 11.7% respectively in the two alternative methods. Past assets growth has a positive association with magnitudes of 0.8% and 1.1% with a significance of 95%. The 4 quarters lagged NPL ratio demonstrates marginal effects of 2.8% and 3.5% in the respective methods. In contrast, the model with loan growth exhibits a negative correlation with defaults but with a relatively lower marginal effect – 0.5%. In the second model we observe lesser impacts of the 5 year EDF variable which decrease to about 6.4% and 6.2% respectively. 5.2 Prediction Using our two alternative models – selected CAMEL covariates adverse selection variables - we predict the probability of default for two EDF categories. The predicted probability indicates the likelihood that the bank defaults. We compare our predictions with the actual results of our final models in Table 4-25 and Table 4-26 below. The tables present the sensitivity to Type I and Type II errors. A Type I error is when the model fails to identify the defaulted banks. A Type II error is when the model falsely identifies sound banks as defaulted. The cut-off value specifies whether an observation has a predictive positive outcome or not. A higher cut-off results in fewer banks determined to have a positive outcome i.e. being defaulted, hence Type I errors increase. A lower cutoff conversely leads to an increase in Type II errors by identifying sound banks as failed ones. Poghosyan & Cihak (2009) suggest putting a larger weight on Type I errors in analysing bank defaults 133 Our tests reveal that all CAMEL components except capital ratio exhibit significance in predicting bank defaults. As argued by Brossard et al., 2006 equity ratios may weaken their signalling power because of the relative homogeneity of European banks’ capital assets. They argue that European banks maintain their capital ratios in accordance with the Basel II regulatory framework and thus they do not vary much like in US banks’ capital ratios. Since our data comprises that of European banks, equity ratio may not be valid as one of the default predictors. As for adverse selection effect, both definitions contribute to the predictive power of the model, although their marginal effects are not very high in all models’ specifications. Past asset growth exhibits a positive association with defaults, while loan growth persistently displays a negative sign. At first sight these are contradictory results, but we suppose that this inconsistency is brought about by the following issues: • The measure of bank assets is broader in definition than gross loans and includes bank loans as well as other items. As was evident in the recent financial crisis, many loans were securitized and transferred by use of Mortgage-Backed Securities (MBS). By doing so, a bank may not exhibit excessive growth of loans but may still have risky MBS and other derivatives in its assets which will have a negative effect on its solvency • The timing of loan loss may also influence regression results. Usually the peak of delinquency on loans comes between two to four years after their origination. Since in our model we do not include time intervals for more than two years the negative effect of rapid loan growth may not yet be apparent • During the financial crisis credit rating agencies have been criticized for delaying banks’ downgrades. This was especially relevant to big banks. Since we use downgrades as a proxy for default and our sample comprises of mainly big banks this could bring some bias to our model. • Large banks may also have implicit government guarantees and thus deemed to be guaranteed rescue by state authorities. This effect is referred as implicit safety nets in the academic literature. Banks may therefore not receive significant downgrades from rating agencies. To summarize, in Early Warning Models it is beneficial to use both market based indicators and balance sheet indicators as they mutually complement each other. It also implies that EDF provides additional information to that of balance sheet ratios but is not a complete substitute for them. There are some issues which need to be taken into account in generalizing the thesis findings: • We estimate our models in Chapter 4 based on the data of EU banks which is perhaps not representative of the entire universe of European banks. As such, our conclusions apply mainly to the EU-15 banks with specified specializations and with market traded securities. Also we use the downgrade of banks’ ratings as a proxy for bank default, thus it may bear some bias to the estimated models. 134 • Since the financial environment of banks is changing rapidly and more financial innovations are introduced, the statistical models of predicting bank failures and explaining bank risktaking behaviour need to be re-estimated periodically in order to adjust them to fit new conditions prevalent in the economy. • It is important to stress the importance of general macroeconomic states which have a huge impact on the bank-specific risk factors we have analysed. Though our aim in this study is not to estimate their influence we acknowledge that we cannot ignore the prevailing macroenvironment when looking to understand the effect of bank-specific factors. The impact of the general macroeconomic state of the country, competition in the banking industry, regulatory and institutional factors, macroeconomic shocks etc. inevitably play a role in bank risk-taking and in predicting bank distress. However, we would like to leave this issue for future studies. Our main intention in this thesis is to reveal risk-taking factors contributing to an excess of bank risk and to show how these risks can be foreseen or predicted in order to prevent the future realization of financial crises. In so doing we believe we can put our own contributions toward understanding the causes of the Spanish financial crisis and suggesting possible improvements to early warning models for European banks. The work could provide important insights for regulators into setting up more efficient policies for controlling bank risk-taking factors and improving prevailing early warning models of bank distress. One possible direction for further research in modelling bank credit risk may be the inclusion of macroeconomic variables accounting for macroeconomic conditions in the observed market, as well as the addition of variables for realized and/or expected bank bail-outs. Another interesting topic for further research can be the use of a dataset with actual bank defaults, not its proxies, as it can better discriminate between sound and fragile banks and improve the predictive quality of the model. Our empirical work can be expanded by using the research framework employed in Chapter 3 and Chapter 4 to study smaller size banks and/or the broad nature of bank ownership. 135 REFERENCES Achraya, V., & Naqvi, H. (2012). The Seeds of a Crisis: A Theory of Bank Liquidity and Risk-Taking Over the Business Cycle. CEPR Discussion Paper N 8851. Adams, R., & Mehran, H. (2008). Corporate Performance, Board Structure and its Determinants in the Banking Industry. Working Paper, Federal Reserve Bank of New York. Adrian, T., & Shin, H. S. (2009). The shadow banking system: implications for financial regulation (No. 382). Staff Report, Federal Reserve Bank of New York. Adrian, T., & Shin, H. S. (2010). Liquidity and leverage. Journal of Financial Intermediation, 19(3), 418-437. Agrawal, D., Arora, N., & Bohn, J. (2004). Pasimony in Practice: An EDF-based Model of Credit Spreads. White Paper, Moody's KMV . Allen, F., & Carletti, E. (2010). An overview of the crisis: Causes, consequences, and solutions. International Review of Finance, 10(1), 1-26. Allen, F., & Gale, D. (2000). Comparing Financial Systems. Cambridge, Massachusetts: MIT Press. Allen, F., & Gale, D. (2004). Competition and financial stability. Journal of Money, Credit and Banking, 453480. Allen, F., & Gale, D. (2007). Understanding financial crises Oxford University Press, USA. Allen, F., Carletti, E., & Marquez, R. (forthcoming). Credit market competition and capital regulation. Review of Financial Studies. Allen, L. (2002). Credit Risk Modeling of Middle Markets. Conference on Credit Risk Modeling and Decisioning. Pennsylvania: the Wharton Financial Institutions Centre, University of Pennsylvania. Altman, E. (1968). Financial Ratios, Discriminant Analysis and the Prediction of Corporate Bankruptcy. Journal of Finance , 589-609. Altunbas, A., Gambacorta, L., & Marques-Ibanez, D. (March 2010). An Empirical Assessment of The Risktaking Channel. European Central Bank Working Paper Series, 1166. Altunbas, Y., Manganelli, S., & Marques-Ibanez, D. (2011). Bank Risk During Financial Crisis. Do Business Model Matter? Working Paper Series N 1394. Anderson, R., & Fraser, D. (2000). Corporate control, bank risk taking, and the health of the banking industry. Journal of Banking & Finance , 1383-1398. Andrés, J. (2009). España & los desequilibrios globales. In A. Cabrales, J.J. Dolado, F. Felgueroso & P. Vázquez (coords.) (Ed.), La crisis de la economía española: Lecciones & propuestas () Madrid: Fedea. 136 Andres, P., & Vallelado, E. (2008). Corporate Governance in Banking: The Role of Board of Directors. Journal of Banking and Finance 32 , 2570-2580. Arellano, M., & Bover, O. (1995). Another look at the instrumental variable estimation of error-components models. Journal of econometrics 68.1 , 29-51. Aríztegui Yáñez, J. (2011). El proceso de reforma del sistema financiero español. 40 Años De La Unacc: El Nuevo Mapa Del Sistema Financiero, pp. 15-32. Arrora, N., Bohn, J., & Zhu, F. (2005). Reduced Form vs. Structural Models of Credit Risk: A Case Study of Three Models. White paper, Moody's KMV Company . Ashcraft, A. B., Goldsmith-Pinkham, P., & Vickery, J. I. (2010). MBS ratings and the mortgage credit boom. FRB of New York Staff Report, (449) Ashcraft, A., & Schuermann, T. (2008). Understanding the securitization of subprime mortgage credit. Federal Reserve Bank of New York Staff Reports, (318) Ashcraft, A., Goldsmith-Pinkham, P., & Vickery, J. (2010). MBS Ratings and the Mortgage Credit Boom. New York: Federal Reserve Bank of New York Staff Reports. Auvray, T., & Brossard, O. (2012). Too Dispersed to Monitor? Ownership Dispersion, Monitoring and the Prediction of Bank Distress. Journal of Money, Credit and Banking, 685-714. Bagehot W. (1873), Lombard Street: A Description of the Money Market, Johnston and Hartley Withers editors. Bahena, A. J. (2010, March). What Role Did Credit Rating Agencies Play in the Credit Crisis? The University of Iowa, Center fro International Finance and Development, Iowa, USA. Bai, G., & Elyasiani, E. (2013). Bank stability and managerial compensation. Journal of Banking & Finance, 799-813. Bai, G., & Elyasiani, E. (2013). Bank stability and managerial compensation. Journal of Banking & Finance, 37(3), 799-813. Banco de España. (2010). La reestructuración de las cajas de ahorro en España. Situación a 29 de junio de 2010. Nota informativa. Retrieved 06/01, 2013, from www.bde.es Banco de España. (2011). El banco de españa presenta el balance del proceso de recapitalización del sistema financiero previsto en el RD-l 2/2011. Retrieved 04/10, 2013, from http://www.bde.es/f/webbde/GAP/prensa/ficheros/es/presbe2011_37.pdf Banco de España. (2011). Presentación del gobernador "reestructuración del sector financiero. Resultados del proceso de recapitalización". Retrieved 04/10, 2013, from http://www.bde.es/f/webbde/GAP/prensa/ficheros/es/mfo300911.pdf Banco de España. (2012). Proceso de recapitalización & reestructuración bancaria. 137 Bannier, C., Feess, E., & Packham, N. (2013). Competition, Bonuses, and Risk-Taking in the Banking Industry. Review of Finance , 653-690. Basu, R. (2002). Financial contagion and investor" learning": An empirical investigation International Monetary Fund. Beltratti, A., & Stulz, R. (2009). Why did some banks perform better during the credit crisis? A cross-country study of the impact of governance and regulation. ECGI Working Paper Series in Finance . Beltratti, A., & Stulz, R. (2012). The Credit Crisis Around the Globe: Why Did Some Banks Perform Better? Journal of Financial Economics 105 , 1-17. Berger, A., & Bouwman, C. (2012). How does capital affect bank performance? Working paper. Bernanke, B., Gertler, M., & Gilchrist, S. (1996). The financial accelerator and the flight to quality. The Review of Economics and Statistics, MIT Press, 78(1), 1-15. BIS. (2005). International Convergence of Capital Measurement and Capital Standards: A Revised Framework. Basel: Basel Committee on Banking Supervision. Blundell, R., & Bond, S. (1998). Initial conditions and moment restrictions in dynamic panel data models. Journal of econometrics, 87(1), 115-143. Bohn, J. (2000). An Emprirical Assessment of a Simple Contingent-Claims Model for the Valuation of Risky Debt. Journal of Risk Finance , 55-77. Bongini, P., Laeven, L., & Majnoni, G. (2002). How Good is the Market at Assessing Bank Fragility? A Horse Race Between Different Indicators. Journal of Banking and Finance, 1011–1028. Borio, C., & Disyatat, P. (2011). Global imbalances and the financial crisis: Link or no link? (No. 346). Bank for International Settlements. Borio, C., & Zhu, H. (2012). Capital Regulation, Risk-taking and Monetary Policy: A Missing Link in the Transmission Mechanism? Journal of Financial Stability, 236-251. Boyd, J. H., & De Nicolo, G. (2005). The theory of bank risk taking and competition revisited. The Journal of Finance, 60(3), 1329-1343. Boyd, J. H., De Nicolò, G., & Jalal, A. M. (2009). Bank competition, risk and asset allocations (No. 20092143). International Monetary Fund. Boyd, J. H., Nicolo, G., & Jalal, A. M. (2006). Bank Risk-Taking and Competition Revisited: New Theory and New Evidence. IMF Working Paper WP/06/97. Brossard, O., Ducrozet, F., & Roche, A. (2007). An early warning model for EU banks with detection of the adverse selection effect. Cahiers du GRES 08 . 138 Bushman, R., & Williams, C. (2012). Accounting discretion, loanlossprovisioning, and discipline. Journal of Accounting and Economics , 1–18. Cabral, R. (2013). A perspective on the symptoms and causes of the financial crisis. Journal of Banking & Finance, 37(1), 103-117. Calvo Bernardino, A., & Paúl Gutiérrez, J. (2012). El crédito al sector privado en españa: Del crecimiento desbordado a la fuerte restricción. REVESCO.Revista De Estudios Cooperativos, 109, 13-44. Carbó, S. (2011). Lecciones del proceso de reestructuración bancaria en españa. Mecanismos De Prevención & Gestión De Futuras Crisis Bancarias, Papeles De La Fundación, 42, 35-47. Carbó-Valverde, S., Marques-Ibanez, D., & Rodríguez-Fernández, F. (2012). Securitization, risk-transferring and financial instability:The case of Spain. Journal of International Money and Finance , 80-101. Catarineu, E., & Pérez, D. (2008). La titulización de activos por parte de las entidades de crédito: El modelo español en el contexto internacional & su tratamiento desde el punto de vista de la regulación prudencial. Estabilidad Financiera, 14, 88-121. Chan-Lau, J., & Sy, A. (2006). Distance-to-Default in Banking: A Bridge Too Far? IMF Working Paper. Chan-Lau, J., Jobert, A., & Kong, J. (2004). An Option-Based Approach to Bank Vulnerabilities in Emerging Markets. IMF Working Paper. Cheng, I., Hong, H., & Scheinkman, J. (2010). Yesterday's Heroes: Compensation and Creative Risk-taking. NBER working paper 16176 . Consejo de Ministros. (26 de abril de 2013). Balance & actualización de la estrategia española de política económica. Madrid. Craig, B. R., & Dinger, V. (2013). Deposit market competition, wholesale funding, and bank risk. Journal of Banking & Finance , 3605–3622. Crespi, R., Garcia-Cestona, M., & Salas, V. (2004). Governance Mechanisms in Spanish Banks. Does Ownership Matter? . Journal of Banking and Finance , 2311-2330. Cuñat, V., & Garicano, L. (2009). Did Good Cajas Extend Bad Loans? The Role of Governance and Human Capital in Cajas’ Portfolio Decisions. FEDEA Monograph," The Crisis of the Spanish Economy". Curry, T., Elmer, P., & Fissel, G. (2002). Regulator Use of Market Data to Improve the Identification of Bank Financial Health. FDIC Working Paper . Curry, T., Elmer, P., & Fissel, G. (2003). Using Market Information to Help Identify Distressed Institutions: Regulatory Perspective. FDIC Banking Review , 15, n3. Davies, H., & Green, D. (2009). Regulación financiera mundial: Guía básica Paidós. 139 De La Vega, J. 1688. Confusion De Confusiones: Portions Descriptive of the Amsterdam Stock Exchange.(Translation by H.Kellenbenz, Harvard University, 1957), Delis, M., & Kouretas, G. (2011). Interest Rates and Bank Risk-Taking. Journal of Banking and Finance 35 , 840-855. Dell'ariccia, G., & Marquez, R. (2006). Lending Booms and Lending Standards. Journal of Finance , 25112546. Demirgüç-Kunt, A., & Huizinga, H. (January, 2009). Bank Activity and Funding Strategies: the Impact on Risk and Return . European Banking Center Discussion Paper . Demirgüç-Kunt, A., Detragiache, E., & Merrouche, O. (2010). Bank Capital: Lessons from Financial Crisis. IMF Working Paper , 286. Dewatripont, M., & Freixas, X. (2012). Bank resolution: lessons from the crisis. The Crisis Aftermath: New Regulatory Paradigms, 105. Dewatripont, M., & Freixas, X. (2012). The Crisis Aftermath: New Regulatory Paradigm. Centre for Economic Policy Research Working Paper . Dewatripont, M., & Tirole, J. (2012). Macroeconomic Shocks and Banking Regulation. Journal of Money Credit and Banking , 237-254. Dewatripont, M., Rochet, J., & Tirole, J. (2010). Balancing the banks: Global lessons de the financial crisis Princeton University Press. Diamond, D. W., & Rajan, R. (2009). The credit crisis: Conjectures about causes and remedies (No. w14739). National Bureau of Economic Research. Diamond, D., & Raghuram, R. (2009). Illiquidity and Interest Rate Policy. NBER Working Paper , 15197. Distinguin, I., Hasan, I., & Tarazi, A. (2013). Predicting rating changes for banks: how accurate are accounting and stock market indicators? Annals of Finance 9(3) , 471-500. Distinguin, I., Rous, P., & Tarazi, A. (2006). Market discipline and the use of stock market data to predict bank financial distress. Journal of Financial Services Research 30(2) , 151-176. Domínguez, D. G., & Cueto, M. J. V. (Sept. 2006). Análisis de las crisis de los sistemas bancarios en la década de los noventa. Revista Electronica De Comunicaciones & Trabajos De ASEPUMA. Douglas, D., & Raghuram, R. (2005). Liquidity Shortages and Banking Crises. Journal of Finance , 615-647. Eichengreen, B. J., Rose, A., & Wyplosz, C. (1996). Contagious currency crises National Bureau of Economic Research. Elizalde, A. (2005). Credit Risk Models II: structural models . 140 Elmer, P., & Fissel, G. (2001). Forecasting Bank Failures from Momentum Patterns in Stock Returns. Washington: Federal Deposit Insurance Corporation. Erkens, D., Hung, M., & Matos, P. (2012). Corporate governance in the 2007–2008 financial crisis: Evidence from financial institutions worldwide. Journal of Corporate Finance , 389-411. European Commission. (July 2012). Macroeconomic imbalances - Spain. European Economy Occasional Papers 103 . Evanoff, D., & Wall, L. (2001). Sub-debt Yield Spreads as Bank Risk Measures. Working Papers, Federal Reserve Bank of Chicago . Ferguson, N. (2008). The ascent of money: A financial history of the world Penguin. Fernández Ordóñez, M. (2011). In Banco de España (Ed.), La reestructuración del sector bancario español & el real decreto-ley para el reforzamiento del sistema financiero Fondo Monetario Internacional. (June 2012). Spain: Financial stability assessment. IMF Country Report No. 12/137. Fridson, M. S. (1841). Extraordinary popular delusions and the madness of crowds Charles Mackay. Friedman, M., & Schwartz, A. J. (2008). A monetary history of the United States, 1867-1960 Princeton University Press. Fuentes Egusquiza, I. (2007). La titulización en España: Principales características. Boletín Económico.Banco De España, (12), 89-98. Galbraith, J. K. (1991). Breve historia de la euforia financiera Ariel. Garcia Marco, T., & Roblez Fernandez, D. (2008). Risk Taking Behaviour and Ownership in the Banking Industry: The Spanish Evidence. Journal of Economics and Business , 332-354. Gramlich, D., Miller, G., Oet, M., & Ong, S. (2010). Early Warning Systems for Systematic Banking Risk: Critical Review and Modeling Implications. Banks and Bank System , 5 (2). Gropp, R., & Köller, M. (2010). Bank Owners or Bank Managers: Who is Keen on Risk? Evidence from Financial Crisis. Centre for European Economic Research. Gropp, R., Vesala, J., & Vulpes, G. (2002). Equity and Bond Market Signal as Leading Indicators of Bank Fragility. Frankfurt: European Central Bank. Haensel, D., & Krahnen, J. (2007, ). Does credit securitization reduce bank risk? Evidence de the European CDO market. Message posted to http://papers.ssrn.com/soL3/papers.cfm?abstract_id=967430 Hakenes, H., & Schnabel, I. (2011). Capital regulation, bank competition, and financial stability. Economics Letters , 256-258. 141 Harada, K., Ito, T., & Takahashi, S. (2010). Is the Distance to Default a Good Measure in Predicting Bank Failures? Case Studies. NBER Working Paper, 16182. Houston, J. F., Lin, C., Lin, P., & Ma, Y. (2010). Creditors Rights, Information Sharing, and Bank Risk Taking. Journal of Financial Economics 96 , 485-512. Huang, R., & Ratnovski, L. (July, 2010). The dark side of bank wholesale funding. ECB Working Paper Series , N 1223. Hull, J., & White, A. (2000). Valuing Credit Default Swaps 1: No Counterparty Default Risk. Journal of Derivatives , 29-40. Hyman, M. (1982). The financial instability hypothesis: Capitalist processes and the behaviour of the economy. Financial Crises: Theory, History, and Policy, Cambridge University Press, Cambridge-London, Iannotta, G., Nocera, G., & Sironi, A. (2007). Ownership structure, risk and performance in the European banking industry. Journal of Banking & Finance , 2127-2149. International Monetary Fund. (2009). Global Financial Stability Report, Responding to the Financial Crisis and Measuring Sytematic Risks. Washington: IMF. Jarrow, R., & Potter, P. (2004). Structural versus Reduced Form Models: A New Information Based Prospective. Working Paper, Cornell University . Jiménez, G., Mian, A., Peydro, J., & Saurina, J. (2010). Estimating the aggregate impact of credit supply channel: Evidence de securitization in Spain. Mimeograph, Banco De España, Jimenez, G., Ongena, S., Peydro, J., & Saurina, J. (2008). Hazardous times for monetary policy: What Do Twenty-Three Million Bank Loans Say About the Effect of Monetary Policy on Credit Risk? CERP Discussion Paper. Jimenez, G., Lopez, J., & Saurina, J. (2010). How Does Competition Impact Bank Risk-Taking? Madrid: Banco de España Working Paper No. 1005 . Jones, P., Mason, S., & Rosenfeld, E. (1984). Contingent Claims Analysis of Corporate Capital Structures: an Empirical Investigation. Journal of Finance , 611-625. Jonghe, O., Disli, M., & Schoors, K. (2012). Corporate Governance, Opaque Bank Activities, and Risk/Return Efficiency: Preand Post-Crisis Evidence from Turkey. Journal of Financial Services , 51-80. Kaminsky, G. L., & Reinhart, C. M. (1999). The twin crises: the causes of banking and balance-of-payments problems. American economic review, 473-500. Kaminsky, G. L., Reinhart, C. M., & Vegh, C. A. (2003). The unholy trinity of financial contagion. The Journal of Economic Perspectives, 17(4), 51-74. Kehoe, T. J., & Prescott, E. C. (2007). Great depressions of the twentieth century Research Department, Federal Reserve Bank of Minneapolis. 142 Kero, A. (2013). Banks Risk Taking, Financial Innovation and Macroeconomic Risk. The Quarterly Review of Economics and Finance 53(2) , 112-124. Kindleberger, C. P., 1910-2003. Manías, pánicos & cracs: Historia de las crisis financieras / Charles P. Kindleberger, Robert Z. Aliber (Barcelona : Ariel, 2012 ed.) King, T., Nuxoll, D., & Yeager, T. (2005). Are the Causes of Financial Distress Changing. Center for Financial Research Working Paper , 2005-03. Köhler, M. (2012). Which banks are more risky? The impact of loan growth and business model on bank risk-taking. Deutsche Bundesbank Discussion Paper , No 33. Konishi, M., & Yasuda, Y. (2004). Factors Affecting Bank Risk Taking: Evidence from Japan. Journal of Banking and Finance 28 , 215-232. Krugman, P. (1996). Are currency crises self-fulfilling? NBER macroeconomics annual 1996, volume 11 (pp. 345-407) MIT Press. Kwan, S. (1998). Securities activities by commercial banking firms' section 20 subsidiaries : risk, return, and diversification benefits. Economic Research . La Porta, R., De Silanes, F., Shleifer, A., & Vishny, R. W. (1998). Law and Finance. Journal of Palitical Economy , 1113-1155. La Porta, R., Lopez-de-Silanes, F., & Shleifer, A. (1999). Corporate Ownership Around the World. Journal of Finance , 471-517. Laeven, L., & Levine, R. (2009). Bank governance, regulation and risk taking. Journal of Financial Economics, 93(2), 259-275. Laeven, L., & Majnoni, G. (2003). Loan loss provisioning and economic slowdowns: too much, too late? Journal of Financial Intermediation , 178–197. Lauck, W. J. (1907). The causes of the panic of 1893 Houghton, Mifflin. Lepetit, L., E, N., Rous, P., & Tarazi, A. (2008). Bank Income Structure and Risk: An Impirical Analysis of European Banks . Journal of Banking and Finance , 1452-1467. Liu, H., Molyneux, P., & Nguyen, L. (2012). Competition and risk in South East Asian Commercial Banking. Applied Economics , 3627-3644. Lordon, F. El porqué de las crisis financieras & cómo evitarlas (Madrid : Los libros de la catarata, [2009] ed.) Losada, R. (2006). Estructuras de titulización: Características e implicaciones para el sistema financiero. Monografias CNMV(14)