This paper studies the behavior of economic risk premiums over time. We use proxies for the state variables that are representative of earlier studies. A cross-sectional regression approach is used to decompose the predictable part of the portfolio returns to examine the portion “explained” by the model and to assess the relative importance of time-varying risk and time-varying risk premiums. We also decompose the variance of predicted returns to assess the relative importance of the economic risk variables. We find that the premium associated with stock market risk is the most important for purposes of capturing predictable variation of the common stock portfolios, while premiums associated with interest rate risks capture predictability of the bond portfolio returns. We also find that time variation in the expected risk premiums-not the betas-is the primary source of predictability at the portfolio level.