Order Block Pullback Strategy: Only 16 of 90 Segments Profitable, Median Segment Return -7.0%, Fails to Beat Holding

Bets on trend continuation after a moving-average breakout: entry only with above-average volume, exit on a cross back below; low-volume fakeouts in choppy ranges grind capital down.

The reproduction of the Order Block Pullback Strategy on BTCUSDT 15m leads to a rather unoptimistic conclusion: among 90 different splits, only 16 were profitable, the median segment return was -7.0%, and it did not beat “doing nothing.” Looking at the numbers: across the 90 backtest segments, the median segment return was -7.0%; the best segment was only +6.3% (its deepest drawdown 11.2%), while the worst segment was -23.5% (its deepest drawdown 24.2%); the deepest drawdown among all 90 segments reached 26.8%, and there was no liquidation throughout. With 16/90 profitable segments, this means that for more than eight out of ten times this rule set was losing money. Over the same period, the “buy and hold” benchmark had an average segment return of -4.4%, while the strategy’s median segment return of -7.0% was about 2.6 percentage points lower—across the same set of splits, simply holding actually lost a bit less. This is the first thing readers should know. Average profit per trade was 617 USD and average loss 210 USD; a profit/loss ratio near 3:1 looks good, but combined with the 16/90 profitable-segment ratio, its structure becomes clear: it relies on a few favorable segments to recover part of the losses, and most of the time small losses slowly accumulate. What exactly is it betting on? In one sentence, it bets that “price zones left by large orders will be remembered”: when price pulls back to the order block, it finds support or resistance and then continues in the original direction. This is essentially a variant of trend continuation plus key-level trading, and the return source is the trend-following move after the pullback. The weakness is also here. This assumption most easily fails in ranging and low-volatility environments: after price returns to the order block, it does not continue but repeatedly crosses through the zone; stop losses are triggered again and again, and the 210 USD small losses keep adding up. When a trend segment finally arrives, the 617 USD single-trade profit is not enough to cover the preceding consecutive losses—this is exactly where the 16/90 number comes from. Another risk is parameter sensitivity: the original text only gave the strategy logic; the four parameters ma_period, stopPercent, interval, and volume_expansion_period were not provided and were filled according to industry convention. These values directly determine entry timing and stop distance; with another set of values, the number of profitable segments very likely would not be 16. Methodological limitations also need to be stated clearly. The 90 runs are not 90 independent samples; rather, the same dataset was cut into 6 non-overlapping windows, repeated for 15 rounds, and the boundaries were randomly jittered to obtain 90 split variants. They are highly correlated with one another and can only show that “with this dataset, split this way, this is the result”; they cannot be treated as 90 independent pieces of evidence. The random seed has been stored in the database; the same seed can reproduce exactly the same 90 splits, but reproducibility is not the same as extrapolation. Each sample interval is only a few dozen days long, does not cover a full cycle, and the parameter values are also not from the original source. This content is a reproduction verification of historical data and does not constitute investment advice. This platform does not recommend, rate, or rank anything.