Detecting abnormal changes in credit default swap spreads using matching-portfolio models

We evaluate the size and power of different statistical tests and adjustment methods for matching-portfolio models to detect abnormal changes in credit default swap (CDS) spreads. The sign-test generally dominates the signed-rank test in terms of size, and dominates both the t-test and the signed-rank test in terms of power. Traditional adjustment methods often lead to a misspecified sign-test. We propose a new and parsimonious method (the spread-matched method), which leads to a well-specified and more powerful sign-test. The superiority of the spread-matched method is particularly evident for observations characterized by extreme levels of CDS spread. Analyses of CDS samples differing by contract maturity, data source, and time period confirm these results. We perform an event study on rating downgrades to illustrate how the choice of tests and adjustment methods can affect inference.