Abstract
We develop a two-state hidden Markov model where the process driving market returns transitions between turbulent and calm states. A cross-sectional momentum strategy embeds a call option on the market, inducing a state-contingent convex relation between market and momentum returns. In turbulent states, the short side of the momentum strategy has high beta and convexity with respect to the market, as a result of higher effective leverage of the past-loser securities, making momentum crashes more likely. A momentum timing strategy based on this model avoids momentum crashes and achieves superior out-of-sample risk-adjusted performance.