The evaluation of the likelihood of defaults of corporations with hidden factors
This thesis aims to evaluate the likelihood of corporations' defaults based on a dataset of U.S. public non-financial firms over the period January 1980-June 2019 by incorporating both observable firm-specific/macroeconomic factors and latent factors using the Bayesian approach coupled with Particle filters. We consider a reduced-form model and adopt a Particle Markov Chain Monte Carlo method combined with the Expectation- Maximization algorithm to scrutinize this relationship. To draw samples for the hidden factors, we adopt a Particle Independent Metropolis-Hastings algorithm. The key to our result is the realization that the Particle Markov Chain Monte Carlo method can work well in models with latent factors. It is feasible to draw samples through the Particle Independent Metropolis-Hastings algorithm in a nonlinear non-Gaussian state space model. Our empirical results provide strong evidence that the variation of the default rates of U.S. industrial firms can be significantly explained by both observable and hidden factors.