Forecasting stock volatility using time-distance weighting fundamental’s shocks

Abstract

The effect of low-frequency fundamentals on high-frequency volatility is assumed to be constant over a low-frequency period in the GARCH-MIDAS model, but this is unrealistic, especially for datasets with large differences in sampling frequencies. We propose a new GARCH-MIDAS model, which innovatively employs a time-distance weighted (TDW) function that can generate time-varying shocks to high-frequency volatility from fundamentals within the same low-frequency period. The empirical results for the Chinese stock market show that the GARCH-MIDAS-TDW model outperforms the GARCH-MIDAS model in both in-sample fitting and out-of-sample forecasting performance, and this finding is also robust to the S&P 500 stock index.