Reading gold's volatility regimes

Two regimes, repeated

Gold alternates between compression — narrow ranges, mean-reverting behaviour, frequent false breaks — and expansion, where trends persist and pullbacks are shallow. Neither regime is permanent, and the transition between them is where most systematic strategies take their worst losses.

Identifying the regime without hindsight

• —Compare short-window realised volatility against a longer baseline. A ratio persistently above 1.3 marks expansion.
• —Track the proportion of closes near the extremes of their bar. Trending regimes produce more of them.
• —Watch how often breakouts of the prior day's range hold. A hold rate below 40% is a compression signature.
• —Use a lookback long enough to avoid flipping regimes weekly, since regime whiplash is worse than a wrong regime.

What should and should not change

Adapting stop distance and position size to volatility is sound engineering. Adapting the actual entry logic to the regime is usually how a robust system becomes an overfitted one — you have just doubled your parameter count and halved your sample size for each branch.

Adapt your sizing to the regime. Be extremely reluctant to adapt your logic to it.

Sizing to the regime instead of the calendar

If risk is defined in currency and stops are defined in volatility units, position size adapts to regime automatically — which is the entire point. The operator who keeps a fixed lot size across regimes is not maintaining constant risk; they are taking twice the risk in expansion and half the intended risk in compression.

The mechanic in one line

• —Measure realised volatility on the trading timeframe.
• —Set the stop at a structural level beyond a volatility multiple.
• —Derive lot size from the currency risk divided by the stop distance.
• —Never derive the stop from the lot size — the causality only runs one way.

Transitions are where systems break

Most of the damage a regime-sensitive system takes is not inflicted inside a regime but during the handful of sessions when one becomes the other. Compression logic entering an expansion gets run over; expansion logic entering compression bleeds on false breaks. Detecting the change quickly is worth more than optimising behaviour within either state.

Detect early, act conservatively

You do not need to predict the regime. You need to stop pretending the last one is still running.