Working Paper: NBER ID: w10914
Authors: Eric Ghysels; Pedro Santaclara; Rossen Valkanov
Abstract: We consider various MIDAS (Mixed Data Sampling) regression models to predict volatility. The models differ in the specification of regressors (squared returns, absolute returns, realized volatility, realized power, and return ranges), in the use of daily or intra-daily (5-minute) data, and in the length of the past history included in the forecasts. The MIDAS framework allows us to compare models across all these dimensions in a very tightly parameterized fashion. Using equity return data, we find that daily realized power (involving 5-minute absolute returns) is the best predictor of future volatility (measured by increments in quadratic variation) and outperforms model based on realized volatility (i.e. past increments in quadratic variation). Surprisingly, the direct use of high-frequency (5-minute) data does not improve volatility predictions. Finally, daily lags of one to two months are sucient to capture the persistence in volatility. These findings hold both in- and out-of-sample.
Keywords: Volatility forecasting; MIDAS regression; High-frequency data
JEL Codes: G1
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| Daily realized power (L94) | Future volatility (G17) |
| Daily ranges (R12) | Future volatility (G17) |
| High-frequency 5-minute data (Y10) | Future volatility (G17) |
| Past volatility (G17) | Future volatility forecasts (G17) |