We examine the role of a belief-based momentum indicator, measured by conditional past returns (CPR), in the realized volatility (RV) predictability of equity markets. Based on the week- and month-horizon CPR, we construct the HAR-CPR and HAR-LCPR models on the basis HAR-RV model. Here, the HAR-LCPR model additionally includes the daily leverage factor in the absence of daily CPR. In China, we find that: 1) week- and month-horizon CPR have significantly positive impacts on one-, five-, and 22-days-ahead RVs; 2) our out-of-sample results further indicate that the HAR-LCPR model performs best in forecasting one- and five-days-ahead RVs, whereas the HAR-CPR model is a more reliable forecasting model for 22-days-ahead volatility; 3) the performance also passes various robustness tests, including sub-period performance testing, alternative training rolling window, and alternative RV estimation. We show the economic mechanism underlying the predictive role of CPR from the perspective of investors’ trading activities.