关联结构（copula）在信用风险评量之使用.docVIP

The Application of Copula in Credit Risk Management JCIC Risk Research Team Lai Bo-Chih 1. Introduction With the final version of Basel II released at the end of June this year, the professional techniques for creating and managing credit risk models has become a vital issue for financial institutions. The approaches to risk measurement in the past focused on measuring the risk of individual obligor and then summing them up. In recent years, more attention is paid to the assessment of portfolio risk. One critical yet thorny question faced by banking supervisory authorities and risk managers in gauging portfolio risk is: how to determine and estimate the joint change of the credit rating and probability of default of counterparties to various credit assets in the portfolio (e.g. bonds, loans, credit derivatives, etc.). The credit risk models CreditManager from RiskMetrics and PortfolioManager from KMV all assume that the joint change of credit rating and probability of default observe multinormal distribution. But empirical studies show that few data in finance and insurance completely follow the rules of multinormal distribution (Embrechts et al.1999). Also given that macroeconomic cycle would bring about the time series behavior of transition matrix (Coleman 2002, Bangia et al 2000), the hypothesis of multinormal distribution tends to underestimate portfolio credit risk by underestimating the probability of a catastrophic event (e.g. financial crisis) or simultaneous decline in equity prices or simultaneous default of several counterparties during global economic slump (e.g. in the early 2000s). In the past, it was a highly complex task in both theoretical deduction and computation to fit to a multivariate joint probability distribution. In particular when the number of assets in the portfolio is huge, it is almost unlikely to accurately estimate the joint probability distribution. The common approach to this challenge was to assume that the return on asset observe