The low-volatility anomaly is the empirical finding that stocks with high idiosyncratic volatility earn abnormally low future returns. This directly contradicts both the CAPM (higher risk should mean higher returns) and theories predicting a positive idiosyncratic volatility premium (Merton 1987, Malkiel and Xu 2002).
Two dimensions: aggregate and idiosyncratic
Ang, Hodrick, Xing, and Zhang (2006) distinguish between two separate volatility effects:
Aggregate volatility risk: stocks with high sensitivity to innovations in market-wide volatility have low average returns. The price of aggregate volatility risk is approximately -1% per annum. This is theoretically motivated: in Campbell’s (1993) and Chen’s (2002) ICAPM, investors want to hedge against rising volatility (which signals deteriorating investment opportunities), so they bid up prices of stocks that do well when volatility rises, lowering their expected returns.
Idiosyncratic volatility: stocks with high firm-level volatility (relative to the Fama-French model) have “abysmally low” average returns. The quintile spread is -1.06%/month (strongly significant). This is not explained by aggregate volatility exposure, nor by size, value, momentum, or liquidity effects. It persists across bull/bear markets, recessions/expansions, and volatile/stable periods.
U.S. evidence
- Data: NYSE, AMEX, NASDAQ stocks, 1963-2000
- Idiosyncratic volatility measured as standard deviation of daily residuals from FF3 regressions over the prior month
- Q5-Q1 spread: -1.06%/month
- Robust to controlling for size, value, momentum, liquidity, volume, analyst dispersion
- Robust across different formation and holding periods
International evidence
Ang et al. (2009) confirm the puzzle is global:
- 23 developed markets (MSCI Developed Country Index), January 1980 to December 2003
- Q5-Q1 Fama-French alpha spread: -1.31%/month across all markets
- Individually significant in every G7 country
- Fama-MacBeth pooled coefficient: -1.54 (t=-5.82, value-weighted)
- Stronger under value-weighting (not a micro-cap artifact)
- International volatility portfolios strongly comove with U.S. equivalents, suggesting a common global driver
What doesn’t explain it
The 2009 paper systematically rules out:
- Private information (PIN measure)
- Transaction costs (zero-return proportion)
- Analyst coverage and institutional ownership
- Price delay (Hou-Moskowitz measure)
- Return skewness
- Leverage/option effects (Johnson 2004)
After all controls, the spread remains at -0.67 to -0.69%/month. The puzzle remains open.
Construction in MAC3
In the MAC3 model, residual volatility is the EWMA volatility of the residuals from the market beta regression:
where are residuals from regressing stock returns on the ESTU return. The half-life parameter is calibrated to maximize explanatory power over a one-month horizon.
MAC2 combined this with market beta into a single “volatility” factor; MAC3 separates them because they capture different risk dimensions. Large-cap stocks tend to cluster at the low end (exposure ~ -1), while small caps dominate the high end (exposure up to ~ +4).
Relationship to quality
Low volatility is one component of the safety dimension in QMJ. Asness, Frazzini, and Pedersen (2019) include low market beta and low earnings volatility in their safety composite. The QMJ framework implicitly captures some of the low-volatility effect through its safety score.
Absorption by the five-factor model
Fama and French (2016) show that the five-factor model substantially absorbs the volatility anomaly through RMW and CMA loadings. Low-volatility stocks have strongly positive RMW and CMA slopes, meaning their returns behave like those of profitable firms that invest conservatively. High-volatility stocks have strongly negative RMW and CMA slopes, resembling unprofitable aggressive investors.
The five-factor model shrinks the three-factor intercepts considerably. For example, the three-factor intercept for microcaps in the lowest residual variance quintile is +0.34%/mo (t=5.20); the five-factor model reduces it to +0.22%/mo (t=3.41). For microcaps in the highest residual variance quintile, the three-factor intercept of -1.23%/mo (t=-7.66) shrinks to -0.85%/mo (t=-5.63). The model’s remaining problems concentrate in small high-volatility stocks that share the “lethal combination” of negative RMW and CMA exposures.
Theoretical explanations
Why would high-volatility stocks have lower returns?
- Lottery preferences: investors overpay for high-volatility stocks as lottery tickets (Barberis and Huang 2008)
- Leverage constraints: investors who cannot leverage up are forced to buy high-beta/high-vol stocks to reach desired returns, bidding up their prices (Frazzini and Pedersen 2014)
- Aggregate volatility hedging: high-vol stocks tend to covary positively with aggregate volatility, providing a hedge that investors value (Ang et al. 2006)
- The puzzle remains unsettled; the anomaly is “a substantive puzzle” per the authors