Constructing a Volatility Factor by Removing Cross-Period Dependence

Summary

The document summarizes Chinese equity research on refining a traditional idiosyncratic volatility factor. The baseline measure is built from a Fama–French three-factor model. The report links volatility clustering in individual stock returns to persistence in the factor across periods, arguing that monthly stock selection can then recycle past information and obscure the signal’s current contribution. It proposes adding a regression step to remove this cross-period dependence.

The summary reports tests on all A-shares from January 2005 through April 2020. The adjusted factor had a mean monthly IC of -0.055 and a mean RankIC of -0.077; a quintile long-short portfolio had an annualized return of 18.89%, information ratio of 2.17, monthly win rate of 78.26%, and maximum drawdown of 8.29%. The adjusted factor was also less correlated with turnover and retained some selection ability after turnover was removed. These historical results do not establish future performance; the report flags market change and factor-return variability as risks.

Key ideas

• The baseline volatility factor is derived from a Fama–French three-factor model.
• Volatility clustering may make volatility signals persist across monthly cross-sections.
• A regression adjustment is proposed to remove this persistence and isolate current information.
• The reported A-share tests show the adjusted factor retaining stock-selection ability after turnover adjustment.
• Backtest statistics cover a historical period and do not guarantee future returns.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.