How many trades for a reliable backtest?
Far more than most backtests have. With the spread of results typical for a 1:3 strategy, a strategy with no edge at all can show up to 0.32 R per trade over 100 trades by luck, and it takes about 1,137 trades to tell an edge of 0.10 R per trade apart from zero. In our grid of 100 backtests the median one had 26 trades, and 0 of the 18 profitable ones were distinguishable from chance.
Why the number of trades matters
Each trade of a 1:3 strategy ends at about minus one R or about plus three R, where one R is the amount risked. Results that jump this much need many trades before their average settles. In our pool of 6,722 trades from three 1:3 strategy templates, the standard deviation of one trade was 1.72 R. The noise in the average result falls only with the square root of the number of trades: four times as many trades halve it, a hundred times as many cut it by ten.
That is why a short backtest with an impressive result means little. The average of a few dozen trades can sit far from the strategy's true edge in either direction, and a trader who picks the best-looking short backtest is mostly picking the luckiest one.
How far luck alone takes a strategy with no edge
We took the real trades of the three 1:3 templates, subtracted their average so the edge is exactly zero, and drew random backtests of a given length 2,000 times. The table shows the result per trade that only one zero-edge backtest in forty reaches or exceeds.
| Trades in the backtest | Result per trade reached by luck alone |
|---|---|
| 30 | 0.64 R |
| 100 | 0.32 R |
| 300 | 0.20 R |
| 1,000 | 0.10 R |
A backtest of 30 trades that made 0.64 R per trade says almost nothing: a strategy with no edge does that by chance. At 1,000 trades the band narrows to 0.10 R, which is still larger than the cost of a typical trade.
Trades needed to detect an edge
To separate an edge from zero with 95% confidence, the average has to stand about two standard errors away from zero. With the measured spread per trade, an edge of 0.20 R per trade needs about 285 trades, and an edge of 0.10 R needs about 1,137. Halving the edge you want to detect multiplies the trades you need by four. A strategy that trades a few times a month on one asset does not reach those numbers in a decade.
What this means for a grid of backtests
Our grid contains 100 backtests: five strategy templates, each run on separate assets over four years. The median backtest had 26 trades, and 81 had fewer than 100. Of the 56 backtests with at least 20 trades, 18 ended in profit after costs, and 0 of those passed the test against zero for their own number of trades. By chance alone about 1.40 of the 56 would be expected to pass.
In other words, every profitable result in the grid is compatible with no edge at all. That does not prove the strategies have no edge; it means the backtests are too short to show one, which is the normal situation for single-asset backtests of setups that trade a few times a month.
Practical rules
- Look at the number of trades before you look at the result.
- Compare the result per trade with the zero-edge band for that number of trades; inside the band it is not evidence.
- To get more trades, test the same rules on more assets or a longer period rather than loosening the rules.
- Keep in mind that testing many variants raises the chance that one of them clears the band by luck.
How we measured
- Engine backtests on the production server, five strategy templates on separate assets, period 2022-01-01 to 2026-01-01, fixed risk of 1% per trade, commission and swap of one commission-charging broker, entries and exits at bid and ask prices.
- Zero-edge band: net results in R of all closed trades of the three templates with a 1:3 target, centred on zero; 2,000 random backtests of each length drawn with replacement; the value shown is the level that one backtest in forty reaches or exceeds.
- Trades needed = (two standard errors × standard deviation per trade ÷ edge)², rounded up. Test of a profitable backtest: its average net result per trade divided by its own standard error, against the same two-standard-error threshold.
Backtests run on October 30, 2026.
Questions
Are 100 trades enough for a backtest?
Enough to see whether a strategy is grossly broken, not enough to confirm an edge. A zero-edge strategy with a 1:3 target can show up to 0.32 R per trade over that many trades.
Does a strategy with a higher win rate need fewer trades?
Yes. When the target is close to the stop, results per trade vary less, so the average settles sooner. The edge per trade is usually smaller too, which works the other way.
Can I combine several assets to get more trades?
Yes, if the rules are the same. Keep in mind that assets that move together, such as several dollar pairs, add fewer independent trades than their count suggests.
For education, not investment advice. Past results, including backtests, do not guarantee future results.
← All articles