From Apparent Profit to a Reproducible Null Result: A Cost-Aware Evaluation of Directional-Change Reversal Trading in Cryptocurrency Perpetual Futures

Directional-change (DC) methods declare a reversal once price moves a set distance from its running extreme. This paper tests whether DC reversals form a tradable strategy in Bybit ETHUSDT and XRPUSDT perpetual futures (2021-2024), starting from a TradingView strategy that appeared profitable in a short window. Exact Python replication showed the full-history strategy lost about 96%. The redesigned study ran each threshold as an independent causal event stream with next-bar execution, volatility normalization, regime maps, a dynamic machine-learning threshold model and two preregistered filters, using chronological walk-forward folds, block bootstrap inference and family-wise correction. Regime-conditioned selection performed worse than a fixed detector even before costs. The best fixed multiplier earned +4.93 basis points per trade before execution costs but -8.97 after taker fees and slippage, and its simultaneous 95% lower bound was negative after correcting for selection among nine multipliers. Neither a confirmation-impulse filter nor a BTC lead-lag filter passed its preregistered gates. Under taker execution, DC reversal did not provide a statistically established, cost-surviving edge.