We test the conditional CAPM with time-varying forward-looking betas, assuming a two-state model for the market risk premium. For market state identification we employ a recursive Markov-switching model based on a forward-looking Sentiment factor. The empirical results for our sample of S&P500 constituents for the period from 1996 to 2007 show that in ‘good’ states of the economy the classical CAPM with just the market factor is able to explain the cross-section of expected returns very well, while in ‘bad’ states firm characteristics like size and book-to-market become relevant.