A volatility-managed portfolio changes its exposure when estimated risk changes; it does not make the underlying return predictable. The hard question is whether the reduction in risk arrives early enough, survives costs, and works outside the sample that inspired it.
The mechanism: scale risk, not conviction
Volatility management is a sizing rule: estimate recent variance, then scale portfolio exposure inversely to that estimate. When measured volatility rises, exposure falls; when it subsides, exposure can rise. This is different from forecasting whether an asset will rise or fall.
A simple schematic is:
base portfolio return × risk scale from past volatility = managed returnThe formula is not a free lunch. A volatile observation can arrive after a sharp loss, and an estimate based on recent data can react late or overreact. A backtest must specify the estimator, lag, scaling function, and any cap before results are inspected.
Why the literature disagrees
The published evidence is mixed, not a settled endorsement. Moreira and Muir's influential approach motivated scaling exposure down when factor volatility is high. But the summary on the London Business School research record describes conflicting findings: earlier work reported out-of-sample failure and transaction-cost problems, while the authors' conditional multifactor approach reports performance relative to its unconditional counterpart out of sample and net of costs.
The disagreement matters. The conditional multifactor result is about a particular portfolio construction and model, not proof that any trader can improve any strategy by dividing by volatility. Different assets, estimators, rebalancing rules, and cost assumptions answer different questions.
Evidence outside equities adds another caution. A University of Strathclyde repository record reports that volatility management in commodity-futures portfolios succeeded in sample for specified portfolios but did not improve real-time out-of-sample performance. A later international equity study in the University of Liège repository reports promising findings for some managed factors, yet says transaction costs leave only market and momentum strategies partially robust. Neither abstract supports a universal rule.
Turning the idea into a test
A fair test begins with rules written before the results are visible. Choose the base portfolio and volatility estimator, calculate the risk signal only from data available at that time, and define the exposure cap and rebalance schedule. Keep an unmanaged version as the control.
Then compare both portfolios on the same dates and with the same return definition. Include turnover, spreads, commissions, and any financing costs relevant to the instruments. Separate model selection from evaluation: tune on an earlier period, then freeze every choice for a later out-of-sample period. Repeat across assets or regimes only if those comparisons were part of the original question.
Do not grade the method by Sharpe alone. Inspect the path of drawdown, tail losses, exposure changes, and net returns; a smoother curve can still be less useful after trading costs. For a complementary lens on uncertainty in strategy estimates, see confidence intervals for trading expectancy and the studio's approach to verifiable testing.
The useful conclusion
Volatility scaling is a portfolio-weighting method, not a market signal. The evidence depends on construction and costs: published studies report both failures and conditional successes, including weaker results after costs or outside the sample.
A test worth trusting is one where the volatility estimate could have been computed at the time, the rules were frozen before evaluation, and the comparison still holds after realistic costs. If the edge needs tomorrow's volatility estimate, it was never available today.