The Low-Risk Anomaly Revisited on High-Frequency Data

Under the capital asset pricing model assumptions, the market capitalization-weighted portfolio is mean-variance efficient. In real-world applications, it has been shown by various authors that low-risk portfolios outperform the market capitalization-weighted portfolio. We revisit this anomaly using high-frequency data to construct low-risk portfolios for the S&P 500 constituents over the period 2007-2012. The portfolios that we consider are invested in the 100 lowest risk stocks and apply equal weighting, market capitalization weighting, or inverse risk weighting. We find that the low-risk anomaly is also present when using high-frequency data, and for downside risk measures such as semivariance and Cornish-Fisher value at risk. For the portfolios considered, there does not seem to be any statistically or economically significant gain of using high-frequency data.