Using data from Moody's, we examine two sources of persistence in the determination of sovereign credit ratings for the period 2000-2011; true and spurious state dependence. Accounting for ratings' persistence, we also examine whether the ratings were sticky or procyclical before and during the European debt crisis. We set up a dynamic ordered probit model with random effects, which are modelled nonparametrically, while we also address the initial conditions problem. To estimate the proposed model an efficient Markov chain Monte Carlo algorithm is designed. We find evidence of stickiness of the ratings and weak true state dependence.