Module 10 of 14 · Building a portfolio
Trading through volatility regimes
Volatility targeting already shrinks positions when markets get rougher. But there is a second effect: in many futures markets the quality of the signals also changes with the regime. Trend rules tend to work best when volatility is low relative to its own history and worst when it is unusually high — turbulent markets whip back and forth, and trends get interrupted.
To use that, you need a measure of “unusually high” that works in every market:
- Relative volatility = today's volatility estimate ÷ its own average over the last ten years.
- Its quantile within its own history so far: 0 for the calmest reading ever seen, 1 for the most turbulent. Use only past data.
- A forecast multiplier that falls smoothly from above 1 at quantile 0 to below 1 at quantile 1 — for example 2× down to 0.5×.
- Smooth the multiplier over a couple of weeks so it doesn't generate trades by itself, multiply the raw forecast, then re-scale and cap as usual.
Use a continuous multiplier rather than switching the strategy off above a threshold. Switching on and off is expensive, and the regime boundary is fuzzy anyway.
🎛 Regime scaler lab
Sharpe, unscaled
0.65
Sharpe, regime-scaled
0.74
Skew, unscaled
1.29
Skew, regime-scaled
1.35
compare with unscaled
Relative volatility = today's volatility estimate ÷ its trailing 10-year average; its quantile within its own past (0 = calmest ever, 1 = most turbulent) is computed with no look-ahead. The forecast multiplier moves linearly from the “calm” value at quantile 0 to the “turbulent” value at quantile 1, is smoothed over 10 days, and the result is re-scaled to an average absolute forecast of 10. Averaged over four 20-year synthetic markets in which, by construction, trends weaken in turbulent spells.
A modest, robust improvement
The bar chart is the motivation: trend Sharpe falls step by step from calm to turbulent regimes. Leaning into calm periods and backing off in storms raises the overall Sharpe ratio. Watch skew as well: in real portfolios this adjustment often trims trend following's positive skew, because some of its biggest wins come from violent breakouts that you have now chosen to hold smaller. In this synthetic market the effect on skew is small — check it on your own data before relying on either.
In real data the same adjustment tends to help trend rules more than carry rules, and it helps mean-reversion strategies a lot, because those lose most when volatility spikes (Module 11). Keep the mapping simple and symmetric: the more knobs you fit to history, the less likely the improvement survives.
Things to try
- • Set both multipliers to 1×: the scaled and unscaled results become identical, a useful sanity check.
- • Push the turbulent multiplier to 0 (switch trend off entirely in the top regime). Does the Sharpe keep improving, or does lost upside start to cost you?
- • Invert it: calm 0.5×, turbulent 2×. The deterioration tells you the effect is real in this market, not an artefact of re-scaling.
Original interactive lesson. Labs run on a seeded synthetic market in your browser, built to illustrate mechanics, not to predict any real market. Contract figures are illustrative; check exchange specifications. Educational analysis, not investment advice.