Trend Pullback Re-entry (趋势回调再入场): 50 of 90 segments profitable in testing; median +3.4% underperforms buy-and-hold

Bets the trend pullback is over: enters when price reclaims the fast MA with RSI not overheated; a shift to chop or a deeper retracement brings repeated stop-outs.

The “Trend Pullback Re-entry (趋势回调再入场)” strategy won only half of its 90 splits: 50 profitable segments and 40 losing segments, with a median segment return of +3.4%, while over the same period the average segment return of simply holding the underlying was +29.3%. This is the first thing to look at in this replication—under most splits, it did not beat “doing nothing.” Let’s lay out the data clearly. The 90 backtests come from the same BTCUSDT 1h dataset split into 15 rounds × 6 segments; the windows do not overlap, but the 90 runs are highly correlated with one another and are not 90 independent samples. The median segment return was +3.4%, the highest +130.7%, and the lowest -37.2%; the distribution is clearly right-skewed, with a few favorable intervals contributing most of the result. The deepest drawdown was 50.0%, deeper than the worst segment’s segment return of -37.2%, indicating that losses were not spread evenly but amplified within certain segments. Of the 90 segments, 0 were liquidated. The average profit per trade was $3,544 and the average loss per trade was $2,658; the profit side is slightly thicker than the loss side, but that thickness needs to be realized through the win rate—once stop-losses are triggered consecutively in a ranging market, this small edge can easily be worn away. What is it betting on? It bets that once a trend forms, it will continue, and that a pullback is just a pause on the way up; therefore it enters when the trend direction and momentum indicators both signal a “pullback within strength.” Its best segment, 2023-09-01 to 2024-03-28, gained +130.7% with a maximum drawdown within the segment of only 12.9%—exactly how it looks most comfortable in such a one-sided upward market. Its weakness lies in the same place: its worst segment, 2025-02-13 to 2025-11-23, had a segment return of -37.2% and an intra-segment drawdown of 42.0%; that was a period when price repeatedly stepped down, and every so-called “pullback” was actually a continuation of the decline, yet the strategy still caught it according to the logic of trend continuation. Ranging and low-volatility markets are equally unfriendly—when there is no trend to continue, stop-losses get repeatedly hit, and segment returns are naturally flattened. Parameter sourcing is the biggest uncertainty in this replication. The original text did not provide values such as the moving-average period, RSI period and pullback threshold, or stop-loss percentage; we filled them in according to industry conventions and copied the rest from the original. If the parameters change, the median segment return and the deepest drawdown will change accordingly, so the numbers above only describe performance “under this set of values” and cannot be treated as inherent properties of the strategy. The sample covers the given historical interval; the random seed for splitting has been stored, and anyone using the same seed can reproduce exactly the same 90 splits—but what is reproduced is the split method, not the future. The three numbers readers should really watch are: profitable segments 50/90, median segment return +3.4%, and deepest drawdown 50.0%. The first and second tell you that most of the time it does not make money; the third tells you how badly it can lose at its worst. If you stare only at the +130.7% segment, the conclusion flips completely—the same data with a different split yields a different story, which is exactly the signal that it is sensitive to the splitting method. This content is a replication validation of historical data and does not constitute investment advice; this platform does not recommend, rate, or rank.