Some recent research shows that macro variables, despite their enormous in-sample predictive ability, do not forecast stock returns out of sample. Specifically, Brennan and Xia (2002) cast doubt on the forecasting power of the consumption-wealth ratio (cay) because it is negligible in out-of-sample tests. In this paper, we argue that Brennan and Xia's results reflect an unstable relation between cay and stock returns. After we add lagged stock market variance, which follows from a limited stock market participation model by Guo (2000), and the stochastically detrended risk-free rate to the forecasting equation to partially control for the aforementioned unstable relation, the predictive power of cay improves dramatically. Also, simple trading strategies based on documented predictability tend to generate higher mean returns with lower volatility than the buy-and-hold strategy, and this difference appears to be economically important.