This paper develops a new modeling technique for credit risk by applying the self-exciting threshold autoregressive (SETAR) model with quadruple thresholds to the credit spread in Japan. Using this technique, we successfully
reveal that the credit spread dynamics in Japan are divided into five risk levels (categories) by four threshold values. Our
investigations also clarify that the credit spread in Japan is highly persistent when the spread is in the second-lowest credit
risk level, and that it moves faster without showing persistent dynamics when it is in the middle- and higher-credit risk
levels. Furthermore, the levels of the boundary values that specify the lowest-and highest-credit risk regimes can be interpreted
as a type of value-at-risk measure; it is considered to be an extreme case when the spread is in the lowest- or highest-
credit risk levels suggested by our model.