Resumen de Topics on recent developments in banking regulation and supervision
Chapter 1 Motivated by the recent financial turmoil and as a preface to the description of Basel II presented in next chapter, we comment a series of connections between the reasons that motivated the development of Basel II and some of the main concerns that the current global financial crisis has raised on banking regulation and supervision. Whereas some of the concerns are recurrent and have been partially addressed by the Basel II framework, the current crisis has revealed the need of further improvements and has given light on some novel challenges as well.
Chapter 2 In this Chapter we briefly describe the Basel II regulatory framework with special emphasis in the IRB Approach for credit risk as it is the approach in which we focus most of the analysis in the following chapters. We analyze the theoretical framework in which the IRB is based on, and provide a description of the different risk measures applied. The core elements of the IRB, the risk weight equations, are analyzed in order to review the key assumptions and requirements introduced and their relation with the internal risk measures. The relation between the assumptions of the framework and the requisites on the estimation of probabilities of default is remarked.
Chapter 3 Despite the many advances introduced and a baseline framework already released, Basel II is still on the planning phase in many countries around the world, particularly in emerging economies, where fewer resources are available and the initial conditions of the regulatory systems are quite diverse. In this context the adoption of more sophisticated approaches, as the IRB, can create some significant effects derived from distortionary actions introduced by the regulated institutions. In environments were banking supervision is weak, the quality of reporting information is low and banks are heterogeneous in what respects to their risk management skills, these negative effects may dominate the positive gains from attempting to link bank capital closer to risk. In the previous Chapter, we study the possible effects on the banks' behavior of adopting a regulatory framework (e.g. Basel II) that necessarily increases the quality of reporting information provided by financial institutions. We particularly focus on a hypothetical implementation period were banks must prove to the banking supervisor their technical skills to provide accurate estimations of the risk inputs required by the IRB Approach. We assume that during the implementation period supervisors are less informed about the banks' real ability and the consequent performance in the estimation of the set of risk inputs required by the IRB approach.
Due to the extra complexities in the regulation and supervision of financial institutions under the Basel II environment, the quality of information required to be eligible for the IRB Approach has been increased. The reliability on the credit risk inputs supplied by the regulated institutions is a critical element of the IRB that explains why more precise information about the design of the rating systems and of the mechanisms to estimate the risk factors is required. The proper estimation of risk inputs will depend in great extent on the banks' internal risk management operations. One point that we highlight in this chapter is that banks with different skills levels and expertise in their risk management operations can present different behaviors towards the implementation of the framework. These different internal capacities in risk management will also affect the banks' ability to provide accurate estimations of the risk inputs required by the IRB Approach. We argue that this ability will be hardly observable by supervisors, particularly during the early years of implementation. If the internal estimations are unreliable or inaccurate, the IRB Approach will clearly fail in its attempt to provide a more risk-sensitive regulatory capital, one of the main objectives of the new framework.
For the purpose of this chapter we use a formal economic model to study the effects of Basel II on banks' behavior. In particular, we adapt and interpret a contract theoretical framework developed for the analysis of corporate governance reforms (Hermalin and Weisbach, 2006 and Hermalin, 2005) to study a banking regulatory reform such as Basel II. One of the main characteristics of the model is that the participants' responses to a regulatory reform are endogenously determined in the model. More specifically in the context of the implementation of Basel II, we found four suitable features in this theoretical framework that are useful for the analysis: 1) the limited observability in the short run of the bank's true ability to provide accurate estimations of the Basel II risk inputs, 2) the possibility that banks can exert some actions in order to distort the information that is provided to the supervisor, 3) the existence of random effects that can influence their risk estimations, and 4) the fact that the supervisors can get some signals about the bank's real performance along its early activity. Another characteristic of the model is the existence of a penalization scheme in an intermediate period where candidature to the IRB Approach can be canceled by the supervisor depending on the banks performance. We interpret this decision as a penalization that causes a reputation loss coming from the unsuccessful adoption of an international regulatory standard. The analysis proposed in this chapter constitutes a first step to the use of a career-concerns-type framework to a different issue other than the typical executive compensation by extending the interpretations and views of the model to the analysis of a banking regulatory reform.
One of the main policy recommendations in this section is the suitability in particular cases (e.g. a banking system with heterogeneous banks and an initial low level of quality of information), of a more "centralized" version of the framework during a "transitional period". During this transition period, the supervisor should fix a set of benchmarks for the estimation of risk inputs in order to guarantee a minimum level of soundness in the system. In addition, both the supervisor and banks would learn about their authentic performance under the new framework. Finally, we provide empirical evidence that supports the non-lineal relation between the quality of reporting information required to banks and their actions distorting the reported information. We found that a U-shaped relation between performance and quality of information is supported by the data on banking systems for a group of 45 countries.
Chapter 4 As it was commented before, the process of estimating the internal risk measures is a crucial part of the IRB Approach. In particular, under the IRB Approach banks should accurately discriminate among different grades of obligors in their credit portfolios taking into account not only the obligor-specific characteristics but also their sector and macro-economic environment. The final objective is to assign a rating grade to each borrower based on a set of estimated probabilities of default. With this purpose in mind, the industry has been gradually moving to more advanced techniques in financial distress prediction, with survival models playing a prominent role in the last few years. In relation with the internal estimation of the corresponding probabilities of default for the credit portfolio, we consider the use of a discrete-time proportional hazard model to the estimation of financial distress in a large sample of Spanish firms during 1994 and 2005. One fundamental advantage of hazard models is that they use not only static but also dynamic characteristics of the subjects at risk, which makes them a very valuable methodology in the design of a through-the-cycle (TTC) rating system. Previously to the estimation of the probabilities of financial distress, we summarized the minimum criteria established by Basel II for the construction and validation of the credit ratings into four areas (general guidelines, estimation of PDs, credit rating construction, and validation and tests). The methodologies we applied in this chapter will be related to these four areas.
The use of hazard models has recently become one of the most popular techniques in financial distress prediction. Several authors have highlighted the superiority of this type of models compared to discriminate analysis and Logit/Probit models. Duration analysis has its origins in what is typically called survival analysis, were the duration of interest is the survival time of a subject. The concept of conditional probabilities is a central point in this type of methodology. Duration analysis will be useful to answer the following type of questions: What is the probability that a firm goes bankrupt next year given that it has already survived in the market for x number of years? This probability will be the result of computing a sequence of events (i.e.: the probability of going bankrupt the first year, then the probability of going bankrupt the second year given the fact that firm has survived the first year, and so on). Given the common practices in the field, the previous theoretical and empirical literature on the subject, and the particular characteristics of the data, we decided to use a proportional hazard (PH) model in discrete time.
In what respects to the data sample we use the period between 1994 and 2005 which has been characterized by a notable expansion of the Spanish economy, characterized by positive GDP growth rates and declining interest rates. The favorable macroeconomic environment has encouraged the emergence of many new firms, but a large number of them have not been able to grow successfully. Under this context, the age of the firm (or duration) will play an important role in the prediction of financial distress and consequently in the design of a risk rating system based upon the estimated probabilities of financial distress. We show that the duration-dependence effect will be significant even after including system-level variables, such as macro-economic factors (that cause temporal dependence in the data) and after considering unobserved individual heterogeneity in the model. The hazard model provides a sound and relatively simple way to incorporate duration-dependence effects jointly with time-varying firm-specific variables, system-level variables (e.g.: macroeconomic variables), and other controls usually present in the literature into the econometric estimation of default probabilities.
Giving the lack of a comprehensive theoretical founded model guiding the selection of the financial ratios to be used as explanatory variables, we have selected a set of standards financial ratios for the baseline model which try to capture the four dimensions usually present in the literature on financial distress prediction: profitability, productivity/activity, liquidity, and solvency/leverage. In addition to financial indicators, we introduce two systemic factors are introduced (the real GDP growth rate and the short term deposit interest rate), together with controls by sector activity and size.
The relevance of including age-dependence effects into the analysis has been suggested in the recent Resource-based view literature (see for example Thornhill and Amit, 2003). This stream of literature identifies systematic differences in the determinants of firm failure between: (1) firms that fail early in their life and (2) those that fail after having successfully negotiated the early liabilities of newness and adolescence. The "liability of the adolescence" or "honeymoon effect" feature has been also reported in the Spanish case by Lopez-Garcia and Puente (2006), but they do not explore explicitly the possible effect of fluctuations in the macroeconomic environment. We think that the explicit inclusion of age or duration-dependence effects together with macroeconomic variables is particularly relevant for understanding the origin of the availability of resources during a long lasting expansionary period. Using a non-parametric and a parametric specification we find an inverted-U shape curve in the relation between firms' exit and its age. This type of duration-dependence has also been found in other countries such as the United States, United Kingdom, Italy, and Germany.
Given that the recovery and expansion of the Spanish financial system after the 1992 crisis has presented some delay respect the real economic expansion (summarized in the GDP growth), we use the Spanish average deposit interest rate as a common risk factor that not only would affect the firms' access to the financial system but also their assets values and financial costs, irrespective of other financial features. In addition to this last feature we have also observed that the interest rate has suffered more pronounced fluctuations during the time window we are considering. We find the short-term deposit interest rate as a proxy of the general interest rate level of the economy to be more representative of the common macroeconomic effects present in the dataset. These improvements in the overall financial conditions could explain part of the availability of resources by new firms (many of which are not able to survive more than 3 o 4 years).
Chapter 5 Basel II differentiates between two stages for the validation of estimated probabilities of default: (1) validation of the discriminatory power of a rating system and (2) validation of the accuracy of the PD quantification (calibration). In the present Chapter, based on the PFD estimates from Chapter 4, we study the discriminatory power of the model's prediction applying a common measure of discriminatory power, the Accuracy Ratio (AR). The proposed financial distress prediction model has shown an adequate discriminatory power that is reflected in relatively higher AR values. Even when differences in the samples (e.g. different credit portfolios) may lead to different assessments of discriminatory power of alternative models we find the corresponding discriminatory measures for the hazard model to be highly satisfactory if compared with other similar works on the field.
Given the difficulties of contrasting power measures between different samples we also evaluate the power of the PH model using two competing binary outcome panel data models. The proposed panel data models will also be useful to check the robustness of the duration-dependence effects and the overall specification of the hazard model. At the end of this section we construct a rating system using cluster analysis and we explore its calibration level. First, we perform a binomial test in order to detect large deviations from the pooled PFDs in each grade; second, we study the stability of the rating applying a Hoshmer-Lemeshow (HL) test to the whole rating system.
The results from the comparison with respect to two alternative specifications for a Random Effects (RE) binary outcome model for panel data shows that these later models have been less able to detect the effect of common risk factors than the hazard model. As it was commented before, the consideration of these effects is a core part of the construction of TTC rating systems. In addition, the duration-dependence terms (both in the parametric and in the non-parametric forms) have remained significant and with an important contribution to the log-likelihood function even after including unobserved heterogeneity. We interpret this result as a robustness check to the relevance of including duration-dependence into the estimation model of financial distress. Finally, the discriminatory power of the hazard model (measures by the ARs) has over-performed the two competing panel data models.
Given the satisfactory performance of the forecasted PFDs using the proposed hazard model, we next proceed to the mapping of the estimated PFDs into risk buckets or categories in order to get a rating system for corporate firms in Spain. We perform a k-mean cluster analysis on the predicted PFDs and the cumulative frequency of the firms to build the rating system using the whole data span of year-firms. After rescaling the rating obtained we undergo two tests, a Binomial test and HL test, with satisfactory results in both cases. Finally, we highlight some difficulties that arise in the construction of TTC rating systems based on a period of long and persistent expansion. Estimations and ratings based on this period should be probably extra-adjusted in order to account for a possible deterioration in the macroeconomic environment.
Chapter 6 In this last Chapter we present two future lines of research. One line of research is aimed to the study of systemic effects and contagion in Spanish firms. The possible existence of contagion is very relevant for countries planning to apply the IRB Approach, mainly because the theoretical framework were it has been built is based on the "doubly stochastic assumption" or "conditional independence" of the default events. A second line of research proposed is the study of the interaction in the risk taking behavior by banks and firms during a long expansionary period with wide and flexible access to financial credit. The combination of both elements can encourage the arising of a full cohort of new firms, which have not been able to growth efficiently and consequently have gone bankruptcy or have quit the market in the following few years.
Summary of the main conclusions The main conclusions and outcomes of this thesis are summarized in the following points:
¿ The concurrence between a series of issues raised by the current financial crisis and those factors that have motivated the development of Basel II.
¿ Given the potential large negative effects due to distortionary actions by the regulated institutions, the suitability in particular cases (especially in the context of emerging countries) of a more "centralized" version of the IRB Approach during the "transitional period".
¿ The availability of empirical evidence supporting the non-lineal relation between of information quality required to banks and their actions distorting the reported information.
¿ The effective use of a hazard model to the construction of a TTC rating system.
¿ The importance of including age-dependence in the prediction of financial distress in addition to idiosyncratic and system-level variables.
¿ The robustness of the U-shaped hazard function to a parametric and a non-parametric specification. This result is in line with empirical evidence reported in others countries and connected to the Resource-based view literature.
¿ The significant role of interest rates levels in the prediction of financial distress, particularly during a highly expansionary period. The potential relation of interest rates, the relaxed of financial restrictions (and the consequent large availability of financial resources), and the risk taking behavior of firms.
¿ The robustness of the age-dependence effects to alternative prediction models that include unobserved heterogeneity.
¿ The reported over-performance of the hazard models with respect to other comparable competing models.
¿ The satisfactory validation of the estimated financial distress probabilities and the construction and calibration of a risk rating system based on the estimated probabilities.
