Estimating Asset Correlation from CDS Spreads: An Analysis of the Hedge Effectiveness of FTD Baskets

Abstract

Modeling default dependence for measuring and managing portfolio credit risk is one of the most challenging problems in modern finance. The standard industry model is a multi-variate Gaussian latent-variable model, where the latent variables are associated with log asset value processes. These processes are most commonly modeled by a firm-specific and a single market factor with a constant factor loading determining the strength of correlation between the asset values. Estimating asset correlations is not straightforward because asset values are per se not observable. Standard practice suggests estimating asset values from stock prices using the Black-Scholes-Merton framework and computing the correlation parameter from the obtained asset values. However, it is well known that this model is based on several simplifying assumptions and, moreover, relies on infrequently updated accounting data. In this paper we apply a latent variable model to estimate the log asset values from observable CDS spreads. In a market without reliable prices, such as for first-to-default (FTD) baskets, we compare our model with the standard industry model by analyzing the hedge effectiveness on FTD baskets for three different regimes: the GM/Ford crisis, the sub-prime crisis and the period in between, when market conditions were more normal. We find both models to perform similarly but in periods of severe market turbulences such as the ongoing sub-prime crisis the spread model is more favorable, since it results in hedging error distributions with substantially lower dispersion.