We revisit the stock market return predictability using the variance risk premium and conditional variance as predictors of classical predictive regressions and time-varying coefficient predictive regressions. Also, we propose three new models to forecast the conditional variance and estimate the variance risk premium. Our empirical results show, first, that the flexibility provided by time-varying coefficient regressions often improve the ability of the variance risk premium, the conditional variance, and other control variables to predict stock market returns. Second, the conditional variance and variance risk premium obtained from varying coefficient models perform consistently well at predicting stock market returns. Finally, the time-varying coefficient predictive regressions show that the variance risk premium is a predictor of stock market excess returns before the global financial crisis of 2007, but its predictability decreases in the post global financial crisis period at the 3-month horizon. At the 12-month horizon, both the variance risk premium and conditional variance are predictors of stock excess returns during most of 2000-2015.