The Dynamic Splitting Method with an application to portfolio credit risk

TL;DR

This paper introduces a modified dynamic splitting algorithm tailored for accurate estimation of rare large losses in credit portfolios, improving upon existing importance sampling methods.

Contribution

It proposes a novel dynamic splitting method that leverages quasi-monotonic properties for unbiased rare-event probability estimation in credit risk models.

Findings

The new algorithm outperforms traditional importance sampling in numerical experiments.

It provides unbiased estimates of large loss probabilities in complex credit portfolios.

The method is adaptable to various credit risk models including factor and copula models.

Abstract

We consider the problem of accurately measuring the credit risk of a portfolio consisting of loss exposures such as loans, bonds and other financial assets. We are particularly interested in the probability of large portfolio losses. We describe the popular models in the credit risk framework including factor models and copula models. To this end, we revisit the most efficient probability estimation algorithms within current copula credit risk literature, namely importance sampling. We illustrate the workings and developments of these algorithms for large portfolio loss probability estimation and quantile estimation. We then propose a modification to the dynamic splitting method which allows application to the credit risk models described. Our proposed algorithm for the unbiased estimation of rare-event probabilities, exploits the quasi-monotonic property of functions to embed a static simulation problem within a time-dependent Markov process. A study of our proposed algorithm is then conducted through numerical experiments with its performance benchmarked against current popular importance sampling algorithms.