Trick 1: parameter hunting
Fire up the optimizer and test thousands of combinations of periods, thresholds and filters until you discover that a 47-period EMA plus RSI at exactly 62.7 'magically' catches every turn of the last five years. Big winners, tiny losers, PF 10-20 instantly. You haven't found an edge — you've memorised history.
Trick 2: over-segmentation and filter stacking
Split the data into bull, bear and range regimes and fit separate 'best' parameters to each. Stack ten filters — volume spikes, specific hours, news-day exclusions — until only the perfect trades remain. The strategy now only trades when history was kind, because the ugly periods have been filtered out of existence.
Trick 3: brutal fitting on cherry-picked samples
Optimise aggressively on one or two years of friendly trending data, or throw a kitchen sink of indicators at the optimizer overnight. It will always find a unicorn parameter set. Always.
Why it breaks the moment it goes live
All three tricks teach the strategy to predict the past. The first regime change — a volatility shift, a macro turn, a black swan — and the magic 47-period stops working. Backtest PF 15 becomes live PF 0.6. Professionals call this data-mining bias, and they test for it directly: does it hold out-of-sample? Does PF survive small parameter changes? Does it work across regimes? Overfit strategies fail all three instantly. Real edge is simple, robust and explainable — and survives the future, not just the past.