Credit Spreads, Factors and Noise

Abstract

With a large dataset of credit default swap spreads (CDSs), this paper shows that previous studies might have spuriously generated a strong latent factor, which appeared to drive the co-movement in the data. The existence of such a strong latent factor is one of the main arguments for the misspecification of structural models of credit risk. However, a careful signal-to-noise analysis of the latent factor and some insights from the random matrix theory reveal that structural models need not be so badly misspecified. Moreover, changes in spreads of CDSs are very noisy and difficult to explain in linear regressions even with very good regressors. Therefore, linear regressions and latent factors in the residuals may give untrustworthy evidence against structural models of credit risk.