PDF Ebook The Determinants of Default Correlations
Corporate defaults exhibit two key characteristics that have profound implications for default risk management. First, default risk is correlated through time. Bankruptcies are normally the end of a process that begins with adverse economic shock and end with financial distress. Although some bankruptcies are unexpected and, therefore, are point events, like Enron and Worldcom, investors become aware of the company’s difficulties some years prior to the bankruptcy event. Second, financial wealth of companies in the same industry, or within the same economic area, is a function of managers’ skills and common factors that introduce correlations.
Companies’ default risk is linked through sector-specific and/or macroeconomic factors. Whilst a great deal of effort has been made by practitioners to measure and explain companies’ default correlations, academics have only recently began to devote attention to this issue. The existing literature on default correlations can be divided into two approaches: the structural approach that models default correlations through companies’ assets values; and the reduced-form approach that models default correlations through default intensities. While financial institutions, namely banks, are aware of these relationships, their ability to model such correlations is still not fully developed. Basle Committee on Banking and Supervision (BCBS 1999, p. 31) states “… the factors affecting the credit worthiness of obligors sometimes behave in a related manner…” which “… requires consideration of the dependencies between the factors determining credit related losses”. Whilst there are many different models and approaches to compute default probabilities, there is no consensus on the importance of different factors that drive default correlations. BCBS (1999) report points out that whilst practitioners have been managing and studying this dependence, there is a lack of theoretical and empirical work on this issue that tests the robustness of the frameworks.
In this paper, we concentrate our empirical investigation on the determinants of default correlation. Our analysis comprises three stages: First, we apply a set of structural models, Merton (M, 1974), Longstaff and Schwartz (LS, 1995) and Ericsson and Reneby (ER, 1998), to compute companies’ expected default probabilities (EDPs). Second, based on cross-sectional tests we analyse the effect of volatility and idiosyncratic risk on EDPs. Given that unexpected events or fraudulent defaults lead to market-wide jumps in credit spreads, which reduce the ability to diversify this risk, it is important to examine the relationship between company’s idiosyncratic risk and bankruptcy. Third, using a factor-analytical technique, we extract common or latent factors that explain default correlations. This analysis enables us to assess the extent to which default correlation can be ascribed to the latent factors and to the systematic variables from capital and bond markets.
The most popular credit risk frameworks used and sold by financial institutions are the KMV1 (building on Merton (1974) model) and CreditMetric. According to the Merton model, dependence between companies’ defaults is driven by dependence between assets and threshold values. In the actuarial CreditRisk+2 framework, default correlations are driven by common factors. For each pair of obligors, the asset value is assumed to follow a joint normal distribution. The efficacy of diversification within a portfolio of claims requires accurate estimates of correlations in credit events for all pairs of obligors. For example, Collateralised Bond Obligations (CBO) and valuation of credit derivatives examined by Hull, Predescu and White (2005) require estimates of the joint probability of default over different time periods and for all obligors. Default correlations can lead to a dramatic change in the tails of a portfolio’s probability density function of credit losses (PDCL) and, consequently, in the economic capital required to cover unexpected losses. The common assumption of independence between events produces the right tail of the theoretical PDCL to be thinner than the one observed in practice, which implies that observed unexpected losses are higher than the ones estimated. BCBS (1999) points out that PDCL of portfolios are skewed toward large losses and are more difficult to model. The PDCL that result from the combination of single credit exposures depends on the assumptions made about credit correlations.
Posted in :
