Risk Management

Volatility Targeting: Sizing Is a More Reliable Lever Than Timing

Saral Team · 1 August 2026

Almost all retail strategy effort goes into deciding what to buy and when. Very little goes into how much. That allocation of attention is close to backwards, because the two questions have very different levels of statistical support.

The asymmetry

Tomorrow’s return is nearly unpredictable from past returns. If it were not, the prediction would be arbitraged away. This is the sense in which markets are efficient, and it is why direction-forecasting strategies have small, fragile edges.

Tomorrow’s volatility is highly predictable from recent volatility. Volatility clusters: turbulent days follow turbulent days and calm follows calm. Engle won a Nobel Prize for formalising this with ARCH models in 1982, and the effect is one of the most robust regularities in financial data — present in every market, every asset class, every frequency anyone has looked at.

So one input to a portfolio decision is nearly unforecastable and the other is quite forecastable. Building a strategy that leans on the forecastable one is not a clever trick; it is the obvious response to what the data supports.

What volatility targeting does

The rule is simple. Pick a target annualised portfolio volatility. Estimate the portfolio’s current realised volatility from recent returns. Scale total exposure by the ratio:

exposure = target_volatility / realised_volatility

If realised volatility is 50% and the target is 25%, hold 50% of capital in the strategy and 50% in cash. If realised volatility falls to 20%, exposure can rise back toward full.

This is a gross exposure decision, and it is deliberately separate from the relative weighting decision. The weighting scheme decides how capital is split between holdings; the volatility overlay decides how much capital is deployed at all. Keeping them separate matters because they answer different questions and are wrong in different ways.

The evidence

Harvey, Hoyle, Korgaonkar, Rattray, Sargaison and van Hemert’s “The Impact of Volatility Targeting” (2018) is the most thorough study. Across equities, credit, commodities and currencies over long histories, they found volatility targeting improved risk-adjusted returns, with the largest gains in assets that have the strongest volatility clustering and the most negative relationship between volatility and returns — which is to say, risk assets.

The mechanism behind the second effect matters more than the Sharpe number. Volatility spikes and drawdowns coincide. Markets get turbulent as they fall. A rule that mechanically cuts exposure when volatility rises therefore reduces exposure going into drawdowns — without forecasting them, without a view, without being right about anything.

That is a genuinely unusual property. Most drawdown-reduction techniques require a correct prediction. This one requires only that volatility clusters, which it reliably does.

The critical choice: can it lever?

Volatility targeting has an upside and a downside application, and they are very different strategies wearing the same name.

De-risk only (max leverage 1.0). When realised volatility exceeds the target, cut exposure and hold cash. When it falls below, exposure returns toward — but never above — 100%. The book never borrows.

Full targeting (max leverage above 1.0). When realised volatility is below target, lever up to reach it. This is what institutional risk-parity and managed-futures funds do, and it is where a substantial part of the academic Sharpe improvement comes from, because it deploys more capital during calm periods.

For a retail portfolio the first is almost always correct, for a reason that has nothing to do with the mathematics. Low realised volatility is common immediately before a volatility shock. A leveraged position sized during a calm stretch is exactly the position you do not want when the shock arrives, and the gap risk on a leveraged crypto position over a weekend is not something a daily-rebalanced rule can manage.

Capping leverage at 1.0 gives up some of the documented improvement and removes the failure mode that ends portfolios. That is the right trade for money you cannot replace.

Estimating volatility

Two parameters, both trade-offs.

Lookback. A short window (30 days) reacts quickly to regime changes and is noisy — the exposure will jump around and generate turnover on estimation noise rather than real changes. A long window (180 days) is stable and slow, meaning exposure stays high well into a volatility spike. Sixty days is a common middle for a weekly-rebalanced strategy.

Target. This is a preference, not an optimum. Setting it below the asset’s typical volatility means holding cash most of the time; setting it above means the overlay rarely binds and does nothing.

For context: the Nifty 50’s long-run annualised volatility is broadly in the mid-teens. Bitcoin’s has often been in the 60–80% range. A 25% target on a crypto book is therefore aggressive in absolute terms — still well above equity-market volatility — while representing a substantial reduction relative to the underlying asset. It is a risk control, not a safe strategy, and calling it either without the comparison is misleading.

Where it disappoints

It does not help with gaps. Volatility targeting reduces exposure after volatility rises. A market that is calm on Friday and gaps 20% on Monday gives the rule no opportunity to act. It manages turbulence, not surprise.

It underperforms in calm bull markets. Whenever realised volatility sits below target, a leverage-capped implementation holds full exposure and no more, so it matches the underlying. Whenever volatility is above target, it holds less and lags. Over a long calm rally, the overlay costs return and delivers nothing visible.

It generates turnover. Exposure changes at every rebalance as the volatility estimate moves. In crypto for an Indian investor that turnover meets a 1% TDS on every transfer, which is a direct and material cost of running the overlay frequently.

The estimate is backward-looking. Realised volatility from the past sixty days is a good predictor of near-term volatility precisely because volatility clusters — but at a regime break, it is wrong in the most expensive direction.

Try it

The Crypto Volatility-Targeted Core template holds Bitcoin and Ethereum continuously with inverse-volatility relative weights, and scales gross exposure to a 25% annualised portfolio volatility target with leverage capped at 1.0. Weekly rebalance.

Compare it against the dual momentum template over the same window. Both reduce drawdown; they do it differently. Dual momentum makes a binary in-or-out decision on trend. Volatility targeting stays invested and continuously varies size. Neither is strictly better, and running both makes clear that “reduce risk” is not one decision but at least two.