We find that factors explaining bank loan recovery rates differ depending on the state of an underlying credit cycle. Our modelling approach incorporates a two-state Markov switching mechanism to capture underlying economic conditions. This latent credit cycle variable helps to explain differences in observed recovery rates over time. Using US bank default loan data from Moody’s Ultimate Recovery Database and covering the pre-and post-global financial crisis (GFC) period, the paper develops and implements a dynamic model for bank loan recovery rates. We accommodate the distinctive empirical features of the recovery rate data, while incorporating a large number of possible determinants. We find that certain loan-specific and other variables hold different explanatory power with respect to recovery rates in ‘good’ versus ‘bad’ times in the credit cycle, i.e. depending on underlying credit market conditions. Our findings demonstrate the importance of accounting for counter-cyclical expected recovery rates when determining capital retention levels.