MACD Histogram Trend Short: Standardised Momentum, Not Crossovers

The MACD histogram is usually watched for a crossover. This strategy reads it as a momentum gauge instead, and it standardises the reading before acting on it. Raw histogram values scale with price, so a threshold that means "strongly negative" on BTC is meaningless on XRP. Dividing by a rolling standard deviation turns the histogram into a z-score that carries the same meaning on every pair — which is what makes a single tuned threshold defensible across a six-pair basket. It sells when the MACD line is below zero (a bearish regime, not merely a bearish crossing), the standardised histogram sits at or below the entry threshold, and it is still falling relative to a few candles ago: downside momentum accelerating, not just present. It covers when the standardised histogram has climbed back above the exit threshold and is still rising, so upside momentum has taken over and the short thesis is spent. The 12 / 26 / 9 MACD periods are deliberately fixed at their canonical values, and the tunable parameters are the entry and exit thresholds and the slope lookback. Risk, exactly as the pinned backtest executed it: a hard -10% stoploss, which a short position needs because its loss is otherwise unbounded, and a flat +10% ROI target that closes any single trade reaching it. This is a short-only design built for a falling market. Over the pinned in-sample window buy-and-hold was about -62%, and a long-only strategy could not post a positive absolute return there at all. The published thresholds come from a 100-epoch hyperopt run against the in-sample leg only, never against the holdout. It moved in-sample return from +0.11% to +2.23% and profit factor from 1.02 to 1.92 — and, more to the point, the held-out out-of-sample leg improved too: return +0.03% to +0.25%, profit factor 1.04 to 1.59. Tuning that helps only the leg it was fitted to is overfitting; this one helped both. One caveat stated plainly, because the page will not state it for you: tuning made the strategy far more selective, so the out-of-sample leg contains only 21 trades. That is a small sample and a profit factor computed on it is a weak estimate, however good it looks. The in-sample leg's 115 trades carry the weight here. Size accordingly, and re-run the backtest over your own window before committing capital.