Credit Risk Portfolio's Modelling And The Research On Calculation Of Marginal VaR

Recent years, credit risk portfolio models have achieved great advancement.Since the public release of J.P. Morgan's CreditMetrics and Credit Suisse'sCreditRisk+ have become influential benchmarks for internal credit risk models.This paper introduces the two models, concludes their assumptions, input data, andcontents, and builds a generalized framework for the two credit risk portfoliomodels.After introduction of recent years' achievements in credit risk measurement, weare engaged on solving the computation of components' risk contribution, and thenoptimizing the portfolio. In comparison with risk measurement, the computation ofrisk contribution is a novel field. We build the mean-VaR model to optimize thecredit asset portfolio like the mean-variance model of Markowitz, the key of solvingis to compute the marginal VaR.The differentiation of VaR is the precondition of computing marginal VaR, andinvolves a series conceptions and properties like the conception of homogeneity, etc.There are two means of computing marginal VaR, namely conditionalexpectation approach given loss and saddle-point approximation approach. PanjerRecursion is used to compute loss probability in conditional expectation approach,but Hermann Haaf points out its drawbacks. In contrast with conditional expectationapproach, saddle-point approximation avoids the use of Panjer Recursion. Anillustrative example shows that the result of two methods is near. In addition, wegive an illustrative example of mean-VaR model to optimize a portfolio. Because itis difficult to express VaR in analytic expression, we solve the model using geneticalgorithm with constraints. Result shows that optimized portfolio reduces risk oncondition of the same return.