Skip the worst trades, don't pick the best ones
When you add a filter to an idea, leaving out the trades with the worst value of a data point fits the past much less than keeping only the trades with the best value. In 34,223 filter choices on NQ, it carried about a third less luck and kept 92% of the trades. It protects you; it rarely adds profit by itself.
The question
Traders rarely trade a raw idea. They add conditions: only above VWAP, only when volume is high, not in the first minutes. Each condition can be written two ways. Keep only the trades where a data point has its best value, or drop the trades where it has its worst value.
Both are chosen by looking at past trades, so both can fit the past instead of the market. Which one fits the past more?
What we ran
- NQ futures, 285 sessions, September 2025 to November 2026. Longs and shorts apart. Each test pairs a starting condition with a second data point, takes the first moment of the day, holds 30 minutes, and pays costs.
- 34,223 pairs per side. Filters were chosen on the first 142 sessions and judged on the next 143, which they never saw. The top 10% by past improvement counted as chosen.
- The luck part = how much better a filter looked on the sessions it was chosen from, minus how much better it did on the sessions it never saw. Zero means no fitting at all.
Results
| Rule | Better where chosen | Better on new sessions | Luck part | Trades kept |
|---|---|---|---|---|
| Keep only the best value, longs | +15.0 | −0.6 | 15.6 | a slice |
| Leave out the worst value, longs | +10.0 | +0.1 | 9.9 | 92% |
| Keep only the best value, shorts | +13.5 | −1.6 | 15.1 | a slice |
| Leave out the worst value, shorts | +9.9 | −0.9 | 10.8 | 92% |
| Sessions used to choose | Longs | Shorts |
|---|---|---|
| 71 | 19.5 | 20.1 |
| 142 | 15.9 | 14.1 |
| 213 | 12.5 | 10.3 |
Rules often said to protect you did not change the luck part: the filter must help in both halves of the past (16.0 for longs), or must also help on the S&P (15.7).
A strict luck test, beating the best of 200 shuffled tries, let through only 26 filters. Those lost 4.3 points a trade on new sessions: the strongest-looking filters were the most fitted.
Why it might be so
Keeping only the best value keeps a small slice of trades. A small slice can look great by chance, and the more data points you look at, the more such slices you will find.
Leaving out the worst value keeps most trades, so a wrong choice costs little. Bad conditions also tend to have a reason that lasts: a trade against a strong move stays a poor trade for years.
What it means when you test
Ask first which trades to skip, not which to keep.
Count the data points you looked at before choosing. Ten looked at and one kept is mostly luck.
Be wary of any filter that leaves fewer than about 100 trades, and of any filter without a trading reason.
Prefer more history to more conditions. That is why our main research now uses nine years of NQ and judges each finding once on three years it never saw.
Limits
One market, about 14 months, a fixed 30-minute hold. On shorts no rule beat zero on new sessions, so leaving out the worst lowers the damage rather than creating an edge. Past results, not a promise.
Simulated results on past market data. Research only, not trading signals or advice. Past results do not promise future ones; see the hypothetical performance disclosure at the bottom of the page.
Get this note as a PDF
Leave your email and the PDF downloads straight away. Easy to keep, print or pass on.
Test your own ideas the same way
Analyze runs any idea on NQ, ES, gold, oil and Bitcoin with costs, both halves and a luck check, so you can see whether it beats chance before you trade it.
Past results on market data, not a forecast. No advice and no signals.
More research
We test the ideas traders talk about
Popular trading ideas tested after costs and against luck, including the ones that failed.
How to read the QuantLogic NQ indicator
What the green, red and gray strip under the NQ chart means, how it held on three unseen years, and what it does not do.
What AI can and can't find in trading
AI finds patterns, including ones that aren't there. Used honestly, it kills bad ideas before they cost you money, and it can still find small edges that last.
Don't fade into a trend
Over nine years of NQ, fades against a trend the indicator saw lost far more than fades on grey. The useful rule is the skip.
Educational content, not investment advice. Futures trading involves substantial risk of loss. See the risk disclosure.