CRT Liquidity Sweep Reversion Strategy: Only 29 of 90 Segments Profitable in Test, Deepest Drawdown 38.3%

Bets on reversion after a liquidity sweep: enters when price crosses back above the lower Donchian band and exits on a cross below; in a sustained downtrend the sweeps keep failing.

Let’s start with the conclusion: across 90 different splits, this strategy was profitable in only 29, with a median segment return of -4.6%, while the average segment return from simply holding the underlying over the same period was +3.5%—it did not beat doing nothing. Laying out the 90-segment backtest: median segment return -4.6%, highest +51.3%, lowest -31.5%, deepest drawdown 38.3%, and 0 blow-up segments. The most telling case is round 5, segment 5 and segment 6: the two are back-to-back; the former returned +51.3% (its deepest drawdown 17.1%), and the latter returned -31.5% (its deepest drawdown 32.7%). Same logic, same data—only the window boundary shifted, and the conclusion went from a gain of fifty percent to a loss of thirty percent. This is not a stability problem; it is a split-dependence problem. Its core assumption is actually clear: it bets that price will revert after a liquidity sweep. Price sweeps upward through stop orders near the prior high, creates a false breakout, and then returns into the range; the strategy enters on the opposite side of the sweep direction to capture this reversion. So it earns from false breakouts and is essentially betting that the market has no real trend continuation. Its weakness lies exactly here: once the market enters a one-way trend and does not revert after the sweep but continues in that direction, this logic will be stopped out repeatedly; in a narrow, low-volatility regime, the sweep magnitude is insufficient and the reversion room cannot cover costs. Average profit per trade $1,506, average loss $488; the profit/loss ratio looks close to 3:1, but profitable segments are only 29/90, showing that it relies on a few large winning segments while continuously grinding down in most periods, and the 38.3% deepest drawdown means this grinding is no small matter. There are two places where we need to be honest. First, parameters: range_period, stopPercent, and atr_period were not given in the original text and were filled in according to industry convention; with a different set of values, the distribution above could very likely change shape, so this result can only show how this logic performed under this parameter set and cannot be taken as the full picture of the strategy. Second, the sample: the 90 backtests are 90 splits of the same dataset, not 90 independent samples; they are highly correlated and cannot be read as 90 independent validations; fortunately, the random seed is stored in the database, so anyone using the same seed can reproduce exactly the same 90 splits. The two numbers readers should watch most when judging are the median segment return -4.6% and profitable segments 29/90: the former shows that in the typical case it loses money, while the latter shows that most of the time it is not making money. The highest segment +51.3% is eye-catching, but it sits right next to the worst segment -31.5%; this dispersion is better used as evidence of sensitivity to the split method, not as evidence of potential. This content is a reproduction and verification of historical data, does not constitute investment advice, and this platform does not recommend, rate, or rank anything.