違約強度模型之計量分析與應用

Abstract

Default/Bankruptcy prediction has a long history in finance. Intensity (instantaneous default probability) model has gain more and more attentions in recent default/bankruptcy studies. In this work, we discussion two aspects in intensity modeling: model specification test and regime-switching generalized intensity model. Model specification test is an important issue for empirical researchers to judge the goodness of fit of the model to the data. In intensity models, Das et. al. (2007) and Lando and Nielsen (2010) needs to specified a bin size that may affect the test results. I will introduce a new specification test based on the martingale properties of aggregate default intensities. Simulation results show that martingale-based test has better power than time-transformed specification test without considering the estimation effects. The second part of this thesis, I will introduce a default prediction model with regime-switching effects in the intensity function, such that the default intensity is affected by both observable risk factors and unobservable regime indicators. In particular, the level of the intensity function and the risk exposures to observable risk factors in the intensity function are specified as state dependent. We provide an estimation algorithm when the state variable is Markovian and illustrate the proposed model using the default data of US-listed companies during 1990-2009. Our test indicates that the regime switching effect in the intensity function is significant. Regime-switching intensity model outperforms classical intensity model in the in-sample fit and out-of-sample default predictions.