Does Higher-Frequency Data Always Help to Predict Longer-Horizon Volatility?

No. Conditional autocorrelation in realized shocks due to misspecification in expected return process affects the relative performance of longer-horizon volatility predictions of models using different frequencies of data. This is because, for multi-step forecasts of volatility, small violations of residual serial independence are compounded in temporal aggregation. In this paper, we show that the conditional autocorrelation in return residuals is a strong predictor of the relative performance of different frequency models of volatility. When the conditional autocorrelation is high, the higher frequency model performs markedly worse than its lower frequency counterpart. Empirically, we show that residual autocorrelation exists in the large cross section of stocks at any given point in time, and that this misspecification can substantially decrease the accuracy of multi-step forecasts generated by higher frequency models. Comparing the monthly volatility predictions from daily models and monthly models, we show that there is a trade-off between the gains from high-frequency data and the susceptibility of its multi-period ahead forecasts to returns model misspecification