Volatility targeting normalizes risk across time by adjusting position size to a target volatility. Regimes matter: estimation error, clustering and market microstructure can make naive targets over‑ or under‑react. A regime‑aware approach blends short‑ and long‑horizon estimators, adds stress overlays and enforces a drawdown budget so leverage shrinks when it should.
Framework
- EWMA realized vol with asset‑specific half‑lives.
- Implied/forward inputs (e.g., options, spreads) for anticipation.
- State labels via z‑scores and simple HMM‑style filters.
- Leverage translation with floors, caps and ROC limits.
Regime detection
States (calm, volatile, stressed) are inferred from realized vs. implied vol and trailing drawdown. Transition smoothing avoids flip‑flopping. Stressed states cap leverage and tighten risk limits; calm states revert to long‑run targets while respecting turnover.
Sizing and drawdown budget
Target exposure uses target_vol / est_vol within a band. A portfolio‑level drawdown budget adds a hard stop: if drawdown breaches the limit, cut gross until recovery. This prevents compounding losses during spikes.
Implementation notes
- Robust estimators and winsorization of outliers.
- Throttle updates and integrate with execution.
- Backtest with realistic costs and borrow limits.