Using a structural model of default, I derive rating characteristics if ratings are meant tolook ‘through the cycle’ as opposed to being based on the borrowers’ current condition.The through-the-cycle method, which is employed by most rating agencies, requires aseparation of permanent and cyclical components of default risk. In a time seriessetting, this can be done through the Kalman filter. The analysis shows that severalempirical irregularities of agency ratings could be the consequence of such a ratingmethod. The stability of through-the-cycle ratings is relatively high, while their defaultprediction power is low. Though not predictable in the usual sense, rating changesexhibit properties that call for a reconsideration of the existing evidence.