Bond Default Correlations are Not StableThe March 2009 issue of Wired has an article by Felix Salmon that lays out in clear terms another of the flawed tools many financial firms have been using to evaluate some of the risks in their portfolios. (See this earlier post for reference to the shortcomings of the Value at Risk tool, also dealt with exhaustively in this January article by Joe Nocera.)
The tool in question is called a "Gaussian copula function," not something a layman needs to attempt to understand in all its technical glory. The tool was developed by David X. Li, a quantitative analyst and actuary. Li's formula generated a measure of the degree to which defaults on the bonds in a pool were correlated. The higher the default correlation (= the likelihood that all companies represented in the pool will default on their debt at the same time), the higher the risk and, therefore, the higher the return an investor will demand.
Li intended his formula to be used for setting prices for newly created assets comprised of pooled bonds, not really for predicting future values of the various tranches in such pools. Future values will reflect hard to predict changes in the default correlations. Such changes occur when default probabilities for individual bonds or tranches do not move in sync.1
For an overview of Salmon's story, you can read an interview broadcast today by American Public Media.
1 See this Wall Street Journal article by Mark Whitehouse for a cautionary discussion of Li's formula, and how financial firms have used it, that antedates Salmon's article by 3½ years.