Forecasts of risk prices at alternative time scales can be used to consolidate history dependence in asset return time series. The resulting Markovian structure identifies a martingale component in the latent transition dynamics. I apply the model to U.S. stock markets and find the concentration of return volatility on the martingale component - the spectral gap - is countercyclical, and predicts annual market returns out-of-sample (o.o.s.) with an R-squared of 10.8%. Value (HML) predictability is concave and front-heavy, peaking at a one-year 14.7% o.o.s. R-squared. In contrast, the momentum predictability term structure is convex, insignificant on the short end, but accelerates to 31.4% o.o.s. R-squared at the three-year horizon. I form t...