Public and private debt markets : an empirical analysis of credit risk

Abstract

Abstract : Credit risk models are generally separated into two families: structural models and reduced-form models. The first structural model was introduced by Merton (1974) and relies on public financial information. On the other hand, reduced-form models use mainly the credit spread as an input and benefit from easy implementation. The following research aims at deepening our understanding of public and private debt credit risk. In our study, public debt (bonds) is modeled with a structural model while private debt (loans) is modeled with a simple reduced-form model. The first objective of our study is to compare credit risk between both markets through default probability estimations. Public and private debt markets are known to be only partially integrated which led to studies on the information discrepancies for market participants. Mostly studied through credit risk ratings and derivative markets, the general consensus is that private debt markets benefit from non-public information. The implications of possessing additional non-public information for lenders are manifold. For instance, the information asymmetry could benefit lenders active in both markets and disadvantage lenders restricted to public debt investments. Our second objective is to determine how both credit models, structural and reduced-form, adjust to new information by analyzing the trends over time for both models. Identifying such trends can help in understanding the inherent credit risk perceived by both markets. More importantly, we can deduce which market integrates new information faster, as well as what justifies the difference in terms of information assimilation. The third objective is to identify which variables significantly influence the probability of default (PD) from each model. Parameters estimations is essential in structural models. Knowing which variables are responsible for the PD can help point towards future research in parameter estimation techniques. Estimates that are more accurate will lead to increasingly more precise probability of default outputs for not only the original Merton (1974) model but all structural models.