Should Mutual Fund Investors Time Volatility?

Abstract

Increasing (decreasing) investment in an actively managed mutual fund when fund volatility has recently been low (high) leads to a significant improvement in investment performance. Specifically, volatility-scaled fund returns exhibit significantly higher alphas and Sharpe ratios than the original (unscaled) fund returns. Scaling by past downside volatility leads to even greater performance improvement than scaling by total volatility. The superior performance of volatility-managed mutual fund trading strategies is attributable to both volatility timing and return timing. Fund flows are negatively related to past fund volatility, suggesting that fund investors are aware of the benefit of volatility management.

Disclosure: The authors report no conflicts of interest.

Editor’s Note

Submitted 3 March 2020

Accepted 7 September 2020 by Stephen J. Brown

This article was externally reviewed using our double-blind peer-review process. When the article was accepted for publication, the authors thanked the reviewers in their acknowledgments. Claude B. Erb, CFA, and one anonymous reviewer were the reviewers for this article.

Acknowledgements

We thank Stephen J. Brown and Steven Thorley, CFA, for helpful comments. Lingling Zheng acknowledges financial support from the National Natural Science Foundation of China (Project No. 71703164).

Notes

¹ We excluded from our sample load funds and funds that charge redemption fees. In addition, we removed index funds from our analysis because these funds tend to impose strict trading restrictions.

² The Investment Company Institute (2020) has reported that actively managed mutual funds accounted for 70% of the total net assets of all equity funds at the end of 2019.

³ We used net returns in all of our analyses in order to focus on the performance experienced by actual investors. The results based on gross returns are slightly stronger than those based on net returns.

⁴ Kenneth French’s website is http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html.

⁵ The fact that c is not known in real time is not an issue. The volatility-scaled strategy specified in Equation 2$r_{\sigma ,t}=\frac{c}{{\sigma }_{t-1}}{r}_t\text{, }$ (2) is equivalent to a strategy in which investors choose a target level of volatility ex ante. We show in the section “Target Volatility” that setting a target level of volatility produced qualitatively the same results.

⁶ Although Federal Reserve Board Regulation T limits leverage to 1.5 for retail investors, we argue that these investors might have access to alternative funding or financing beyond their brokers—for example, home equity loans. More importantly, institutional investors can take on much higher leverage, and according to the Investment Company Institute (2020), they own a large percentage of US equity mutual funds. Institutional share classes accounted for nearly 40% of the total assets of long-term US mutual funds at the end of 2019. That said, our results were qualitatively similar if we imposed a maximum leverage ratio of 1.5 instead of 2.

⁷ We note that the mean difference in alphas is identical to the difference in mean alphas, but the median difference in alphas is not the same as the difference in median alphas.

⁸ Specifically, in each simulation run, we kept the fund returns intact and redrew fund volatility. We then reconstructed volatility-managed returns with the simulated data. We used the same asset pricing models as used previously to evaluate the funds’ performance. By redrawing fund volatility in the simulated data, we altered the dynamics of volatility and the volatility–return relationship. Therefore, by construction, the value of volatility timing should be zero in the simulated data. We compared the actual alphas (estimated from actual data) to the distribution of alphas obtained from simulated data. This procedure allowed us to obtain the bootstrapped p-value for the hypothesis that the actual mean and median alphas are zero.

⁹ The intuition for volatility timing is that if investment weight (i.e., beta) is positively correlated with volatility, beta will tend to be high at extreme return levels. This tendency would result in an overestimate of the unconditional beta of the strategy and push down the estimate of the unconditional alpha (Lewellen and Nagel 2006).

¹⁰ Two potential explanations exist for why past fund volatilities negatively predict future fund returns. First, if fund volatility and contemporaneous fund returns are negatively correlated (i.e., volatility is higher in down markets) and fund returns are persistent, then high past fund volatility will be negatively related to low future fund returns. Second, high fund volatility may induce fund outflows. If funds incur a significant cost when meeting investor redemptions (e.g., the need to sell stocks at inopportune times), then fund outflows will drag down fund performance. This result would imply a negative relationship between past fund volatility and future returns. In untabulated tests, we found evidence consistent with both of these explanations.

¹¹ Another, related issue is margin interest rate. In our baseline analysis, as in most prior academic studies in this literature (e.g., Moreira and Muir 2017), we implicitly assumed that investors could borrow and lend at the risk-free rate. In practice, the borrowing rate could be significantly higher than the risk-free rate. To examine the impact of margin interest rate on the profitability of our strategies, we calculated the level of margin rate that would drive the alpha of the volatility-scaled strategy to zero. Our analysis shows that this margin rate would be, on average, 11%–12% per year.

¹² We obtained active share data from Martijn Cremers’s website: https://mcremers.nd.edu/.