Estimating Loss Given Default from CDS under Weak Identification[Estimation and Inference with Weak, Semi-Strong, and Strong Identification]

Existing reduced-form default intensity models that jointly estimate probability of default (PD) and loss given default (LGD) from credit default swaps (CDSs) produce dissimilar results, and there is little guidance on which time series specification to choose. This article develops a model of CDS term structure without parametric time series restrictions for PD and uses weak-identification robust methods to investigate whether separate identification of PD and LGD is still possible. Consistent with intuition about the identification strategy, the model is not globally identified. However, in my empirical application, LGD is precisely estimated for half of the firm-months under study, with resulting values much lower than conventional values. This implies that the risk-neutral PD and the risk premia on PD are underestimated when LGD is set to conventional values.