Introduction. Edward I. Altman; Andrea Resti, Andrea Sironi. NYU Salomon Center and NYU Stern School of Business; Bocconi University

Transcription

1 Introduction Edward I. Altman; Andrea Resti, Andrea Sironi NYU Salomon Center and NYU Stern School of Business; Bocconi University The subject of credit risk management has recently emerged as perhaps the principal and most challenging area of risk management in financial markets. This prominence has been motivated by a number of factors in the last decade, including the introduction and debate on the new Basel II Capital Adequacy Accord, the record number of bankruptcies and defaults in the United States and elsewhere, and the introduction and phenomenal growth of the use of credit derivatives. It is now generally agreed that the three main variables affecting the credit risk of a financial asset are: (1) the exposure at default; (2) the probability of default; and (3) the loss given default. While significant attention has been devoted by the credit risk literature over many years to the estimation of the probability of default of a counterparty, whether it be a corporation, bank, sovereign, individual or structured product, much less attention, until recently, has been dedicated to the estimation of the recovery rate and the resulting loss given default. This volume seeks to assemble many of the most important works in the area of loss given default estimation and analysis in order to better understand this critical variable in credit risk management and measurement. The importance of loss given default has been punctuated by Basel II s continued emphasis on both expected and unexpected loss of credit assets and its recent focus on a better understanding of recovery rates in economic downturns. For many years, credit risk models traditionally assumed the recovery rate on a counterparty s xxi

2 RECOVERY RISK: THE NEXT CHALLENGE IN CREDIT RISK MANAGEMENT asset to be dependent primarily on individual features (eg, seniority or collateral) that do not respond to systematic factors. Lately, researchers have been questioning the independence of recovery and default incidence and are thinking of more complex relationships in the estimation of default recoveries. Financial markets have given so much prominence to the loss given default variable such that some major credit rating agencies (like S&P and Fitch) have introduced recovery ratings and recovery scales which seek to quantify the expected ultimate recovery on defaulted loans and other speculative-grade exposures; on the other hand, institutions like Moody s KMV have invested considerable resources on developing ad hoc loss estimation models. We also expect that the growth in the credit derivative market and the desirability of adding extra credit protection for structured, asset-backed products will motivate additional interest by the credit insurance industry in the loss given default phenomenon. Indeed, we already observe that some derivative-insurance products carry with them certain protective features dealing with recovery rates. For all of the above reasons, we have compiled an impressive group of papers on this critical subject. These materials are grouped into four parts, which constitute the main streams of research of this book. xxii

3 Part I Defining and Measuring Recovery Risk Part I deals with the definition and measurement of the loss given default. In fact, while there is a wide consensus on how other components of credit risk (above all, default probabilities) should be identified and characterised, the criteria to effectively delineate recovery risk and its components have still to be fully agreed upon by risk managers, scholars and regulators. This part of the book includes five chapters ranging from an analysis of the actual meaning of LGD, and the factors driving its variability, to a survey of the role of this risk component in the new Basel II architecture, to the different approaches (quantitative versus qualitative) adopted to model its behaviour. Chapter 1, written by Til Schuermann of the Federal Reserve Bank of New York, looks at some stylised facts on recoveries and losses, and outlines the different perspectives that can be adopted to quantify LGD based on market data and/or workout experience. Special care is given to the time-profile of the process leading a firm into default and restructuring, and to the main factors affecting recovery rates (the capital structure of the borrower, seniority and collateral, industry, and business cycle); the peculiar, bimodal form of the probability distribution observed for LGDs is also briefly scrutinised. Chapter 2, by Andrea Resti and

4 RECOVERY RISK: THE NEXT CHALLENGE IN CREDIT RISK MANAGEMENT Andrea Sironi of Bocconi University, looks at how recovery risk is dealt with by the new Basel II regulation: recoveries are found to play a key role not only in the advanced IRB approach (where banks are responsible for the development and implementation of their own LGD measurement systems) but also in the foundation approach to internal ratings, and even in the simpler, standardised approach. Moreover, the main requirements faced by institutions wishing to adopt an internal system for LGD estimation are outlined and critically analysed. Chapter 3, by Edward Altman (Stern School of Business, NYU), Resti and Sironi, reviews the theoretical and empirical literature concerning recovery rates, their determinants, variability, and the link with default risk. After summarising the implications for LGD of both structural and reducedform credit-risk models, this survey highlights the treatment of recovery risk in last-generation credit portfolio models and the results from several recent empirical studies; a special section is also devoted to the correlation between loss rates and default probability, a topic further discussed in Part III of the book. In Chapter 4, Greg Gupton of Moody s KMV and Defaultrisk.com presents a quantitative, statistical approach to LGD measurement: the updated release of the LossCalc model, a cutting-edge solution for measuring and forecasting expected recovery rates, is presented with all of its main features and implications; a final section is dedicated to some new schemes for the validation of LGD-forecasting techniques. Finally, Chapter 5, by William Chew and Steven Kerr of Standard & Poor s, presents a totally different approach to recovery estimation: a rating system consistent with the procedures traditionally followed by major rating agencies, aimed at analysing and decomposing ultimate recovery risk by means of qualitative judgements, quantitative estimates and human expertise. The empirical performance of this (relatively new) system in the first year of implementation is also presented in the final part of the chapter. 2

5 Part II Measuring LGD on Specific Portfolios Part II addresses the issue of measuring and understanding LGDs on some specific asset portfolios. Indeed, while the main features of the general framework presented in Part I still hold true, further challenges lie ahead for risk managers wishing to assess the degree of recovery risk surrounding specific classes of exposures such as bank loans, leases and MBO-related loans. More precisely, Chapter 6, by Jean Dermine (INSEAD) and Cristina Neto de Carvalho (Universidade Catolica Portuguesa), presents a mortality-based methodology to estimate LGDs and fair provisioning on bank loans, which is empirically tested on a dataset of Portuguese bank loans to small and medium-sized enterprises. Using a multivariate statistical analysis of the determinants affecting these LGDs, bank loans are found to behave in a significantly different manner compared to corporate bonds. In Chapter 7, Pierpaolo Grippa and Simonetta Iannotti (both of the Bank of Italy) and Fabrizio Leandri (Monte dei Paschi di Siena) present the main results of an empirical survey on the loan recovery activity of Italian banks, offering an unprecedented insight into recovery rates, timing and costs of workout procedures. Collateral, personal guarantees, the size of the loan, the use of private agreements with debtors to circumvent

6 RECOVERY RISK: THE NEXT CHALLENGE IN CREDIT RISK MANAGEMENT lengthy court procedures, geographic area and industry are all found to be relevant in explaining the variability in recovery levels, which, nevertheless, remains mostly idiosyncratic. Chapter 8, by Giacomo De Laurentis (Bocconi University) and Marco Riani (Parma University), focuses on leasing portfolios, a category of exposures characterised by the fact that the lender retains legal ownership of the leased asset and therefore can quickly gain control of it in case of default. Based on a dataset of more than 1,000 contracts, an econometric analysis is performed to assess the significance of some 20 variables, which may be potentially related to recovery levels. Results indicate that the bulk of recoveries come from asset sales; however, many non-asset related factors prove significant, such as the type of business, leasing period, bank guarantees and repurchasing agreements. Finally, Chapter 9, written by David Citron (Cass Business School, London) and Mike Wright (Nottigham University Business School), looks at how bankers monitor highly-leveraged MBO transactions by investigating the rates of repayment achieved on these types of exposures. The empirical analysis is based on data from the UK, a market generally viewed as being relatively creditor-oriented, in contrast to the US insolvency regime, which has a debtor focus. Factors associated with repayment levels to secured creditors are investigated. 100

7 Part III The PD/LGD Correlation Part III addresses the PD/LGD correlation, a topic that has become increasingly popular in recent years, both in the Basel II debate and in the analytical schemes developed by scholars and risk managers alike. The theoretical underpinnings, empirical evidence and practical implications of the link between default and recovery risk are investigated by means of a set of contributions described below. Chapter 10, by Jon Frye (Federal Reserve Bank of Chicago) revises and enhances the results developed by him in a series of seminal works on the subject. The classic assumption of LGD/PD independence is here challenged, based on a double intuition. First, as recoveries stem from the assets of the defaulting party, and those assets are sensitive to systematic risk, recovery should be lower in a downturn; second, as LGDs are mostly capped at 100%, such a sensitivity must be stronger for debt instruments, which usually show lower LGDs (such as guaranteed senior secured loans), since there is more scope for an increase. The chapter presents evidence that supports both intuitions. Chapter 11, by Samu Peura (Sampo plc) and Esa Jokivuolle (Bank of Finland) further elaborates on these intuitions by introducing a model for bank loans in which collateral value is correlated with the PD. To model risky debt, the

8 RECOVERY RISK: THE NEXT CHALLENGE IN CREDIT RISK MANAGEMENT 184 authors use a Merton-like option-pricing framework, where default is triggered as the borrowing firm s total assets strike some default threshold. Based on such a model, an expression for expected LGDs is provided, defined as a function of the collateral value as well as the default probability. However, the response of the LGD to an increase in the borrower s PD is found to be negative, a result which differs from mainstream PD/LGD models. Chapter 12, by Altman, Brady, Resti and Sironi, builds on Frye s intuition by empirically investigating the PD/LGD link based on high yield junk bond market data. Their univariate and multivariate models suggest that, as aggregate default rates tend to increase, expected recoveries (measured through post-default prices) become lower; besides the common dependence of defaults and recoveries on some macroeconomic background factors, demand and supply equilibria are found to play a pivotal role in strengthening this negative correlation. In Chapter 13, Klaus Düllmann (Deutsche Bundesbank) and Monika Trapp (University of Mannheim) offer two contributions to the study of the PD/LGD relationship. First, they produce an empirical analysis of a sample of default and recovery rates taken from the S&P Credit Pro database, based on an extended single risk-factor model that leverages off Frye s framework; a special emphasis is given to a comparison of recovery rates based on post-default prices and values at emergence from bankruptcy. Second, they study the impact on economic capital of the systematic risk embedded in recovery rates. They show that, if systematic risk can inflict losses either through a higher default frequency or through a lower recovery rate, none of these effects can be overlooked when computing unexpected losses. This subject is further developed by Altman, Resti and Sironi in Chapter 14, where two different simulation exercises are performed. First, the impact of a positive correlation between PD and LGD on portfolio risk measures is assessed by means of an extended version of Credit Suisse Financial Products Creditrisk model; the results show that, besides increasing the volatility of future losses and expanding the economic capital required to cope with extreme scenarios, such a correlation also impacts on expected losses. Second, the consequences of the PD/LGD link for the issue of procyclicality are investigated. A numerical exercise, where historical rating migrations experienced in the US from 1985 to 2005

9 THE PD/LGD CORRELATION are used to simulate capital requirements under an internal ratingsbased capital regime, shows that cyclical swings can be exacerbated when recoveries and defaults respond to the same drivers. Part III ends with Chapter 15, by Ali Chabaane (ACA Consulting and BNP Paribas), Jean-Paul Laurent (University of Lyon and BNP Paribas) and Julien Salomon (BNP Paribas). Their contribution is based on a generalisation of the credit value-at-risk (VAR) model underlying the Basel II accord; next to the uni-factorial, Merton-like scheme driving defaults, a separate stochastic process is introduced to account for the uncertainty surrounding LGDs. The two random variables can be independent, but in the most general form of the model they can also show a common dependence, increasing risk. The standard formulae for VAR and expected shortfall originating from the Basel II framework are then contrasted to those derived from this broader, more general model. 185

10 Part IV Advanced Methodologies Part IV deals with a set of advanced methodologies recently proposed by several authors to refine our understanding of recovery risk and/or to circumvent some practical problems surrounding LGD estimation. These range from specific issues, such as the choice of a risk-adjusted discount rate to be used in the computation of workout LGDs or the measurement of downturn LGDs at a portfolio level, to more wide-ranging topics such as the estimation of the entire probability density function of recovery rates (as opposed to the estimation of just one expected value). In Chapter 16, Iain Maclachlan, of the Australia and New Zealand Banking Group, first reviews the different approaches that have been used by previous researchers, or suggested by regulators, regarding the spread used to discount nominal recoveries; he then proposes his own original solution. Such a solution is aimed at making the discount rate used in LGD computation fully consistent with the amount of systematic risk, which, according to the Basel Committee, is present in the different exposure portfolios regulated in the Basel II accord. An empirical application of some 200 loans to small/medium sized enterprises shows how this approach can be made operational. Chapter 17, by Marie-Paule Laurent and

11 RECOVERY RISK: THE NEXT CHALLENGE IN CREDIT RISK MANAGEMENT Mathias Schmit from Université Libre de Bruxelles, presents an innovative approach to the estimation of downturn LGDs on diversified portfolios. To verify to what extent the variability in recovery rates can be ascribed to systematic factors, the authors resort to historical simulations based on data from a wide dataset of more than 6,000 defaulted leasing contracts. Although each simulated portfolio only includes exposures that defaulted in the same year (to avoid averaging-out cyclical effects by combining recoveries measured at different points in time), the results suggest that, in the case of leasing, recovery risk is mostly idiosyncratic and can be eliminated through adequate diversification. Expected LGDs, therefore, do not dramatically differ from the extreme percentiles of the portfolio losses distribution, hence the downturn effect on recovery rates does not appear to be dramatic. The last two chapters of the book focus on methodologies to estimate recovery rate densities, that is, the probability distributions surrounding expected recovery rates. Chapter 18, written by Matthias Hagmann and Olivier Scaillet (from HEC Lausanne and FAME and HEC Genève and FAME, respectively) together with Olivier Renault (Citigroup), estimates the recovery risk (RR) density function using three different techniques: a non-parametric algorithm based on a beta-kernel density estimator, the standard market practice to parametrically model the RRs by means of a beta distribution, and a semi-parametric method based on a transformation approach. The authors find that bond seniority and obligor industry are important determinants of recovery rate distributions. Moreover, recovery rates are far from being beta distributed: this means that the parametric approach followed by most practitioners can lead to a substantial underestimation of credit VAR. Finally, in Chapter 19 Craig Friedman and Sven Sandow of Standard & Poor s and NYU respectively, focus on ultimate RRs measured at the time of emergence from bankruptcy, and estimate their conditional probability distribution as a function of collateral, amount of debt below-class, debt above-class and economy-wide default rates. The approach followed in this chapter is rather unique, as it tackles the problem by maximising a creditor s utility function, based on the RR probability distribution, conditional on information that ought to influence it, such as collateral quality and debt seniority. 284