Volatility targeting

Definition

Volatility targeting sizes positions inversely to their current volatility so each contributes a constant expected risk: target risk ÷ estimated volatility = exposure. When volatility doubles, size halves. It stabilizes the RISK a portfolio runs — instead of constant notional with wildly varying risk — and is the sizing engine of systematic futures programs (strategy-futures-trend-following).

How it works / structure

• Formula: exposure_i = (target vol_i × equity) / (estimated vol_i × unit value); estimates from realized vol (exponentially weighted, 20-60 day) or ATR (atr_14_pct) — pinned per program.
• Parameters (engine-executable): per-position and portfolio vol targets, estimator and window, rebalance trigger (estimate drift threshold vs schedule — continuous resizing pays friction), leverage cap (low-vol regimes otherwise demand unbounded size — the cap is structural), and floor/ceiling on the estimate.
• Why it can help returns, not just risk: volatility clusters and is negatively correlated with returns in equities — cutting size when vol spikes historically avoided part of the worst returns (the Moreira-Muir mechanism).
• Cross-asset equalization: vol-scaled positions make a bond future and an equity future comparable risk units — the precondition for port-diversification-math to work across a futures book.

When it applies

Multi-asset systematic books (its native habitat); any strategy whose raw signal says direction but not size; drawdown-budget implementations (risk-max-drawdown-budget translates budgets into vol targets). Equity-index overlay versions are the academic evidence’s subject.

Risk profile & failure modes

• Estimate lag at breaks: vol targeting sizes on trailing estimates; the first shock arrives at full pre-shock size — it manages the second week of a crisis, not the first day.
• Leverage in the quiet: low-vol regimes produce maximum leverage exactly when compressed volatility can expand violently (the cap and floor exist for this; uncapped vol-targeting into a vol spike is the 2018 short-vol shape — regime-volatility).
• Friction from resizing: constant re-estimation means constant trading; the rebalance threshold is a real P&L parameter.
• Correlation blindness: equal per-position vol does not equal portfolio risk control when correlations jump to one (risk-correlation-exposure owns that layer).

Evidence & limits

Moreira-Muir (2017) found volatility-managed equity portfolios improved Sharpe ratios and reduced crash exposure in US data; Harvey et al (2018) found vol targeting reduced drawdowns and improved risk-adjusted returns for risk assets (equities, credit) but added little for assets without the vol-return correlation (bonds, currencies) — the honest boundary of the evidence. Both are backtest literatures with period dependence; the risk-stabilization property is arithmetic, the return improvement is regime-conditional.

Falsifiable-thesis examples

Illustrations only, not signals:

• “Vol-targeting strategy S at 10% (60-day estimator) keeps its realized 12-month vol between 8% and 12% in replay” — falsified by the realized number.
• “The vol-targeted version of S beats the constant-notional version on replay Sharpe this year” — falsified by the paired replay.

Cross-references

• Base rule it refines: risk-fixed-fractional; estimator: indicator-atr
• The phenomenon it exploits: regime-volatility (clustering, vol-return correlation)
• Its native programs: strategy-futures-trend-following, strategy-momentum (crash-filtered variants)
• The layer above: port-diversification-math, risk-correlation-exposure