A Mixed Ordered Probit Analysis of Corporate Credit Ratings

Abstract

Using a U.S. entity credit ratings dataset, I examine the performance of Mixed (Random Parameters) Ordered Probit (MOP) model with more conventional Standard (Fixed Parameters) Ordered Probit (SOP) model in analysis and explanation of corporate credit ratings. Consistent with the discrete choice literature, I find that a MOP model is better able to extract, to a fuller extent, the underlying behavioral information in the model covariates than SOP. I show that the likelihood ratio test rejects the SOP model in favor of the MOP model. The understanding of behavioral responsiveness of variations in levels of model covariates on probability outcome (i.e. levels of credit ratings) is significant for economic decisions.