2010년 8회 세미나 개최를 알려드리고자 합니다.
일시: 2010년 10월 27일 수요일 오후 5시
장소: 정경관 507호
연사: Saebom Jeon
(Institute of Statistics, Korea University)
"A Hierarchical Bayesian Method with Dynamic Latent Variable for Credit Rating"
<Abstract>
Evaluating the distribution of credit rating rather than simply estimating it is a lot more informative to measure the credit risk of obligors. The transition matrix exhibits migration probability of credit rating depending on both the current grade and the distance between the current and next grades in a monotonic decreasing way. These distributions of credit rating and transition matrix vary over obligors and are often required on the individual basis of obligor. The individual-based distribution and transition matrix are both cross-sectionally and longitudinally correlated. It is also interesting to figure out that under what financial conditions a stress occurs. In order to accommodate all these issues, we consider a hierarchical Bayesian approach. This approach involves dynamic latent variables to avoid the intractable estimation problem arising from MCMC sampling and to accommodate cross-sectional and longitudinal correlations as well. Threshold variables are also introduced in the Bayesian model to determine the cut-off points for credit ratings and to keep the transition matrix decreasing in a monotonic way by accumulating the previous information on credit ratings. We apply this Bayesian approach to the yearly data from 2000 to 2006 of Korean construction corporations.