Volatility-Regime Short Momentum

Momentum only pays inside a productive band of volatility. When realised volatility sits far below its own long-run level the market is chopping and momentum signals are mostly noise; when it sits far above, price gaps around and entries get whipsawed out before the move arrives. This strategy measures realised volatility as the rolling standard deviation of returns over 29 candles, divides it by its own slow baseline, and only opens a position while that ratio sits between 0.99 and 2.43. The rest of the time it holds nothing at all. The directional core is deliberately plain, so that the regime filter is what is actually being tested: a 14-period rate-of-change reading below -0.9%, confirmed by price trading under a 104-period trend EMA that is itself declining. Demanding the slope as well as the side is what stops the strategy from selling into every bounce. It is short-only by design, and that was an evidence-led decision rather than a stylistic one. An earlier symmetric version of exactly this logic was tested over the same window and the long branch lost 1.31% across 254 trades while the short branch made 0.47% - volatility expansion is not direction-neutral in a falling market, so the symmetric version spends its productive regime buying bounces into supply. Removing the long side is the honest expression of the edge the regime filter actually finds. Positions close when price crosses back over the trend EMA or when volatility leaves the productive band, because at that point the condition the trade was opened under no longer holds. Risk is capped hard: every position carries a -10% stoploss, and a flat 10% take-profit closes anything that reaches it first. A hard stop is not optional here, because a short position has unbounded loss if price rallies against it. Every signal parameter is exposed as an IntParameter or DecimalParameter, so the strategy can be re-optimised on your own pairs and window. The published defaults are the winning values from a 100-epoch hyperopt run over the buy space, executed against the in-sample leg only and never against the out-of-sample data. Tested on six Binance USDT-margined perpetual futures - BTC, ETH, ATOM, DOT, LINK and XRP - on the 1h timeframe in isolated-margin mode. Past backtest performance is not a prediction of future returns. This window was a severe bear market, which is the regime this design is built for; it has not been validated across a sustained bull market.