Abstract: Traditional models used to describe the dynamic behavior of credit spreads didn't concentrate on different time scales of the mean-reversion process. This stud investigutes this proposition in the Chinese corporate bonds context by proposing a kind of compound state credit modd.Based on the mixture normal distribution, a kind of multi-step MCMC simulation is used to deal with the model's Bayesian inference. On different pay period indexes we find that the credit spreads indices with different pay periods have different structures of heteroscedasticity and the mean-reversion states can include the long and short time scales by using compound state credit modd.The results show that the long scale process captures the trends, while the short one captures the extreme events. At last, the superiority of compound state credit modd is illustrated by being compared with the classical mean-reversion model.

Key words: credit spread, mean-reversion, compound state, bay esian inference, Markov chain