Using data from Moody's, we examine three sources of sovereign credit ratings persistence: true state dependence, spurious state dependence and serial error correlation. Accounting for ratings persistence, we also examine whether ratings were sticky or procyclical for two major crises: the European debt crisis and the East Asian crisis. We set up a dynamic panel ordered probit model with autocorrelated disturbances and nonparametrically distributed random effects. An efficient Markov chain Monte Carlo algorithm is designed for model estimation. We find evidence of stickiness of ratings and of the three sources of ratings persistence, with the true state dependence being weak.