In the lead up to the Global Financial Crisis, the development of financial markets largely brought benefits to the community, but also brought with it significant risks. Risk management in financial markets has consequently become a growing concern for most financial institutions and governments.
A sound risk management system is important to both financial institutions and governments. This thesis consists of three essays that contribute to the literature on risk management, specifically, in the area of corporate credit risk prediction. The first essay relaxes the constant volatility assumption in Merton’s structural model to develop a stochastic volatility (SV) structural credit risk model. The SV structural model is evaluated by comparing with the Merton model in terms of their credit spread predictions through a Monte Carlo study and an analysis based on empirical data of the Down Jones firms. The simulation study verifies the better performance of SV model than Merton model when the asset returns actually have a stochastic volatility. The empirical analysis ascertains the importance of recognizing the stochastic property of the asset return’s volatility in the credit risk prediction, by showing that the SV structural model predicts the actual Credit Default Swap (CDS) spread better than the Merton model.
The second essay, motivated by the importance of the jump process in stock returns, examines the impact of allowing for jumps in the Stochastic Volatility (SV) structural model on corporate credit risk prediction. Bates (1996) model is employed as an example of the Stochastic Volatility and Jumps (SVJ) structural model to describe the evolution of the asset returns. The empirical analysis ascertains the importance of recognizing the jumps in the SV structural model by showing that on average the SVJ model raises the credit risk spread prediction from the Merton structural model and SV model by 6.5 basis points and 2.5 basis points in the Dow Jones firms and 8 basis points and 3 basis points in 200 CRSP firms respectively. This helps explain up to 8% and 10% of the time-variation in actual credit spreads.
Despite both the SV and SVJ model significantly improving CDS spread prediction, the empirical evidence suggests that the superior performance is not constant throughout the sample period. Model uncertainty and instability seriously impair the prediction ability of these models, especially when the model is misspecified. In essay three, we consider three alternative structural models including the Merton model, a structural model with stochastic volatility (SV) and a structural model with stochastic volatility and jumps (SVJ). To mitigate this prediction problem we propose a bias-corrected global minimum variance (GMV) combined forecast procedure. We illustrate the proposed method using both a simulation study and an empirical analysis of the Dow Jones firms in order to further improve CDS spread prediction. Both simulation and empirical results show that the optimal combination significantly improves CDS spread prediction.