Liquidity Sweep Reversal: 63 of 90 Slices Profitable, Median Slice Return +6.1%, Max Drawdown 19.9%

Bets on reversal after a false breakout: only enters once price sweeps a Donchian boundary and snaps back; a genuine breakout that keeps running leaves it trapped.

This 'Liquidity Sweep Reversal' beat 'do nothing' holding over the same period across 90 different cuts, but the edge is thin and the drawdown is not small. One-line conclusion: it is a strategy that can make small money but requires enduring large drawdowns; it is not a low-risk tool. To lay out the numbers clearly: in the 90 backtests, 63 slices were profitable; median slice return +6.1%, maximum +23.4%, minimum -11.1%, deepest drawdown 19.9%, 0 slices liquidated; average profit per trade $462, average loss $250, and the profit/loss ratio is not bad. Over the same period, the mean slice return of simply holding the underlying was +4.2%. The best slice was 2026-06-27 to 2026-08-02, with slice return +23.4% and a deepest drawdown in that slice of only 5.8%; the worst slice was 2026-08-09 to 2026-09-08, with slice return -11.1% and a deepest drawdown in that slice of 19.2%. Note that these 90 are not 90 independent samples, but 90 ways of slicing the same BTCUSDT 5-minute data—15 rounds, 6 non-overlapping windows per round, randomly jittered boundaries per round; the random seeds have been stored, and the same seed can reproduce exactly the same 90 splits. What does it bet on? When price breaks through key liquidity levels such as prior highs and prior lows, it is often not a true breakout, but a sweep of a batch of stop losses before reversing. The strategy earns the price difference from 'price returning to the range after the sweep'; the core assumption is that mean reversion after false breakouts will prevail. This also determines that it likes markets with clear liquidity levels and sweep activity. The weaknesses are equally clear. When the market really enters a one-sided trend and, after a sweep, does not reverse but continues, the reversal assumption fails; in ranging markets, repeated small false breakouts create consecutive small losses. The worst slice was -11.1%, but that slice's drawdown reached 19.2%, showing that drawdown clearly accumulates in losing slices; what readers really need to watch is drawdown, not slice return. Another sensitive point is parameters: donchian_period and stopPercent were not given in the original text and were filled in according to industry convention; if the parameters are changed, the spread between the best slice +23.4% and the worst slice -11.1% may widen further—this is the area that most needs caution. There are two more honest limitations. First, the 90 cuts are not 90 independent samples; on the same data, the best slice earned 23.4% and the worst slice lost 11.1%, showing the conclusion is quite sensitive to 'which day you start counting from'. Second, the median slice return +6.1% and the mean +4.2% of holding across slices are not exactly the same metric (one is median, one is mean), so the comparison can only indicate rough direction. The sample interval and parameter source are the two hard constraints of this replication; the conclusion is only for directional reference. This content is a replication and verification of historical data and does not constitute investment advice; this platform does not recommend, rate, or rank.