Content area
Full Text
(ProQuest-CSA LLC: ... denotes formulae omitted.)
In a recent study, Ben Dor, Dynkin, Hyman, Houweling, Leeuwen, and Penninga [2007] examine the behavior of corporate bond spreads. They find that the volatilities of monthly systematic changes in spreads across sectors tend to increase linearly with the level of spreads.1 Volatility of the non-systematic component of spread change of a particular bond or issuer is proportional to its spread level as well. Furthermore, the authors show that the linear relationship between spread volatility and spread level implies that excess return volatility is roughly proportional to the product of duration and spread. This new risk measure, termed "DTS"SM (Duration Times Spread), generates better out-of-sample volatility forecasts and lower tracking error for index-replicating portfolios compared with using duration alone.
In this article we extend the analysis of credit spread behavior beyond corporate bonds and look at credit default swaps. Establishing that the conditional volatility of spread change is proportional to the level of spread makes the DTS measure of risk exposure directly applicable to portfolios of CDS. This in turn has important applications in terms of position allocation and risk management.
A priori, we would expect all the previous results to hold for credit default swaps as, in theory, changes in their spreads and those of the underlying bonds should be closely related. In practice, however, this is not always the case. Some evidence suggests that since CDS are often more liquid than the underlying bonds, their spreads incorporate new information more quickly and may exhibit higher volatility.2 In addition, in previous studies corporate bond spreads are computed relative to the Treasury curve, whereas CDS spreads represent spreads over LIBOR. Furthermore, the higher liquidity of CDS contracts as compared with corporate bonds allows us to examine whether the previous findings are still valid when spread changes are analyzed at a weekly frequency.
Another difference between this study and Ben Dor et al. [2007] is the use of quasimaximum likelihood (QML) to investigate the relation between spread volatility and spread level. This technique addresses the stochastic nature of conditional spread volatility and the fact that it is not directly observable (i.e., latent). Using QML enables us to assess the statistical validity of a pre-specified explicit functional dependence between conditional...