Migration matrices are considered a major determinant for credit risk management. They are widely used for credit value-at-risk determination, portfolio management or derivative pricing. It is well known that migration matrices show strong variations and cyclical behavior through time. We compare a factor model approach and numerical adjustment methods for estimation and forecasting of conditional migration matrices. Our findings show that the methods may lead to quite different forecasting results. Although the numerical adjustment methods fail to outperform the naive approach of taking previous year's migration matrix as an estimator, the one-factor model provides significantly better in-sample and out-of-sample results. Additionally, on the basis of a chosen risk-sensitive goodness-of-fit criteria, we are able to interpret the results in terms of risk.