Imag(in)ing Volatility Dynamics

This paper predicts future stock return volatility based on historical volatility term structure (HVTS). The forecasting volatility on target name, is a weighted average of historically future volatilities from all up-to-forecasting-day firm-day observations, with weights determined by the difference of HVTS between target name and historical observations.With only HVTS information, our forecasting technique generates a competitive out-of-sample performance (OOS) on future 1-month volatility compared with two-dimension conditional pooling estimation (2DCP) by Wu and Xu (2022) - ’risk is local’, but strongly beats pooled estimation by Bollerslev et al., (2018) - ’risk is everywhere’. We further find, the distance of HVTS of target against historical ones matters the forecasting performance at both cross-section and time-series. Combining additional information of implied volatility, our technique beats the above two, slightly better than 2DCP. A further long-short option investment shows its advantage compared with traditional volatility mispricing factors and other volatility estimators.