Moreira & Muir (2017) — Volatility-Managed Portfolios
Citation. Moreira, A., & Muir, T. (2017). Volatility-Managed Portfolios. The Journal of Finance, 72(4), 1611–1644.
One-line takeaway. Scale a factor’s exposure by the inverse of last month’s realised variance and you raise its Sharpe ratio and earn a positive alpha — a claim that says you can improve a portfolio by timing your own volatility, with no view on returns.
What the paper claims
Managing volatility — taking less exposure when recent realised variance is high and more when it is low — increases risk-adjusted returns across the major equity factors (market, value, momentum, and others). The headline evidence is a spanning regression: the volatility-managed factor f^\sigma_t = (c/\hat\sigma^2_{t-1})\,f_t regressed on the original factor f_t produces a positive, significant alpha, meaning the timing adds value beyond simply holding the factor. The mechanism is that volatility is persistent and forecastable while the equity premium is not strongly related to volatility at these horizons, so cutting exposure in high-variance months avoids risk without giving up much return.
How they show it
Monthly data, decades of factor returns; realised variance from daily returns within the prior month; the constant c chosen so the managed factor has the same unconditional volatility as the original (making the comparison risk-matched). Alphas are reported with standard errors; the effect is strongest for momentum, whose crashes are the most volatility-linked.
What I’d push on
- Out-of-sample and net of costs. The strongest challenge is Cederburg, O’Doherty, Wang & Yang (2020), who show that for a real-time investor estimating everything as they go, and after transaction costs, much of the alpha evaporates, and a naive implementation can underperform. The in-sample spanning alpha is not the same as a tradable edge.
- Leverage the normalisation hides. Matching unconditional volatility with a single constant c masks the fact that inverse-variance weighting demands enormous leverage in calm periods — capping it changes the result materially, which is an implementation choice, not the signal.
- Sharpe improvement vs significant alpha. These are different tests; a strategy can lift Sharpe while its spanning alpha fails to clear a bar, especially once you account for the specifications tried.
- Momentum-driven. Much of the aggregate result rides on the momentum factor, whose volatility-managed version is really managing momentum crashes (Barroso & Santa-Clara, 2015) — a narrower and better- understood phenomenon than “volatility management works.”
How it connects to the proposal — and what I found
This is a paper I could not just annotate but test, and did: my volatility-managed-portfolios experiment reproduces the debate on the Nasdaq with in-sample-vs-OOS, leverage caps, costs, and Newey–West alphas. The verdict lands between Moreira–Muir and Cederburg: volatility management improves risk-adjusted metrics (Sharpe up, drawdown roughly halved) and survives out of sample and costs once leverage is capped, but its spanning alpha is never statistically significant across eight specifications, and the naive uncapped version underperforms. The robust benefit is risk control, not alpha — which is precisely the empirical foundation of the proposal’s whole positioning, and a cleaner statement than either side of the original debate.