The Idiosyncratic Volatility Puzzle – Anomaly or Data Mining?

In this study, I investigate the robustness of the idiosyncratic volatility puzzle to the configuration of the research design. Using the regression- as well as the portfolio-based concept, I start with the replication of the idiosyncratic volatility puzzle approving the findings of Ang et al. (2006). However, when idiosyncratic volatility is estimated from monthly data and a time window spanning 1 or 5 years, the puzzle vanishes, regardless of the research method employed. Similar result hold if only stocks with a market capitalization above the cross-sectional median or those with a price higher than 10$ are used. Independent of the weighting scheme, the puzzle is also absent in the regression-based context when the risk premia are estimated by generalized least squares weighting returns by the inverse of their variance estimates. The same finding is derived in the portfolio-based context by extending the holding period to 12 months or controlling for the past month maximum daily return.