Volatility scaling: sizing positions by inverse volatility for stable risk exposure
An equal-dollar allocation to a high-volatility and a low-volatility asset gives the high-volatility asset roughly all the portfolio's risk. Volatility scaling addresses the asymmetry by sizing each position in proportion to the inverse of its volatility. The result is a portfolio whose risk contribution is balanced across assets and (with rolling rescaling) constant through time.
What volatility scaling is
Volatility scaling is a position-sizing rule that sets each asset's portfolio weight in proportion to the reciprocal of that asset's recent realised volatility. For an asset with volatility σᵢ, the weight is wᵢ = (1 / σᵢ) / Σⱼ(1 / σⱼ), which produces a portfolio where each asset's volatility contribution is approximately equal at the time the weights are set.
The technique is the foundation of risk-parity-style allocations, time-series momentum implementations, and many systematic trading strategies that operate across multiple asset classes. The motivation is the same in each case: a uniform application of a signal across very different assets requires that the position sizes be normalised by something other than capital, because capital weights produce risk contributions that depend on the assets' volatility differential.
Volatility scaling can be applied at the portfolio level (allocating capital across assets) or at the strategy level (within a single asset, scaling position size with rolling realised volatility to maintain a constant volatility target). The two applications share the same underlying logic but operate at different layers.
How it works
For portfolio-level scaling, the inputs are the most-recent estimates of each asset's volatility. The weights are computed once at each rebalancing date and held until the next rebalancing. For a 60-day rolling volatility window and monthly rebalancing, the position in an asset whose 60-day volatility doubled since the last rebalancing would be roughly halved at the next rebalancing.
For strategy-level scaling, the position size is updated more continuously. The standard time-series-momentum implementation, for instance, holds a position in an asset only when the trailing return is positive and sizes the position so that each asset contributes a fixed amount of volatility—typically 10% annualised per asset for a portfolio of 50–100 assets, where the 10% target is adjusted for cross-asset correlation to produce the desired total portfolio volatility.
The volatility input is typically a rolling realised standard deviation over a window of 30–90 days for daily data, or 12–24 months for monthly data. Shorter windows produce more responsive scaling but introduce more noise into the position sizes; longer windows are smoother but slower to adapt to genuine regime changes.
What the evidence shows
Moskowitz, Ooi, and Pedersen (2012) implemented volatility scaling as part of their canonical time-series momentum study across 58 futures markets. The study showed that volatility scaling delivered a meaningful Sharpe ratio improvement over equal-weight implementations of the same momentum signal: 1.8 vs 1.4 in their primary specification. The improvement was attributable to the more uniform risk contribution across assets, which prevented any single high-volatility market from dominating the portfolio's behaviour.
Asness, Frazzini, and Pedersen (2012) extended the analysis to factor portfolios in equities, showing that volatility-scaled factor portfolios produce more stable risk profiles than capital-weighted equivalents over multi-decade samples. The factor premium is similar in magnitude under both implementations, but the volatility-scaled version delivers it with materially smaller drawdowns and faster recovery from regime-change episodes.
For multi-asset portfolios, the implication is that volatility-scaled allocations have outperformed equal-capital allocations on risk-adjusted measures across most evaluation windows. The exception is regimes where one asset class dominates the cross-section by a large margin (the equity bull markets of the late 1990s and 2010s), where the volatility-scaled portfolio's underweight to the leading asset class produces underperformance.
Limitations and trade-offs
Volatility scaling depends on the historical volatility being a useful estimate of forward volatility. The two are correlated but not identical—volatility is itself volatile, and the rolling estimate lags structural changes in the underlying volatility regime. The technique therefore tends to under-react to genuine regime shifts (size positions too large for the new regime) and over-react to transient volatility spikes (size positions too small for the calmer aftermath).
The technique is also vulnerable to leverage at the strategy level. A volatility-scaled time-series-momentum implementation that targets 10% annualised volatility per asset will use leverage when an asset's realised volatility is below the target. The leverage exposes the strategy to gap risk: a one-day large move can produce a loss many multiples of the target daily volatility. Risk-management overlays (drawdown caps, gap-risk hedges) are typically required to address this.
For investors comparing volatility-scaled strategies with capital-weighted alternatives, the relevant comparison is risk-adjusted rather than absolute. The headline returns of a volatility-scaled portfolio look different from those of an equal-weight equivalent because the risk profiles differ; presenting both side by side requires consistent volatility-targeting or other risk-normalisation.
Volatility scaling in pfolio
Volatility scaling is implicit in pfolio's portfolio construction methods. Hierarchical Risk Parity and risk-parity approaches naturally scale weights by inverse volatility; mean-variance optimisation produces the same effect through the covariance structure. The construction methodology is documented at how we build portfolios.
Related articles
Disclaimer
Get started now
