How Many Trades Does a Backtest Need?
What a sample size buys
The table below shows the 95% interval around an observed 50% win rate — the range the true rate could plausibly sit in given that many trades.
| Trades | 95% interval | True rate could be |
|---|---|---|
| 30 | ±18.0 points | 32.0% – 68.0% |
| 50 | ±13.9 points | 36.1% – 63.9% |
| 100 | ±9.8 points | 40.2% – 59.8% |
| 200 | ±6.9 points | 43.1% – 56.9% |
| 500 | ±4.4 points | 45.6% – 54.4% |
| 1,000 | ±3.1 points | 46.9% – 53.1% |
At thirty trades a strategy showing 60% wins is entirely consistent with a true rate of 42%. That is not a reason to distrust backtesting; it is a reason to keep going past the point where the result first looks good.
Sample is not the constraint here
| Instrument | Bars | From | To |
|---|---|---|---|
| AUDCAD | 1,888,257 | 2021-01-03 | 2026-04-30 |
| AUDUSD | 1,909,057 | 2021-01-03 | 2026-04-30 |
| BTCUSD | 2,627,549 | 2021-07-24 | 2026-07-23 |
| ETHUSD | 2,270,819 | 2021-01-01 | 2026-08-03 |
| EURGBP | 1,905,973 | 2021-01-03 | 2026-04-30 |
| EURJPY | 1,890,388 | 2021-01-03 | 2026-04-30 |
| EURUSD | 1,911,141 | 2021-01-03 | 2026-04-30 |
| GBPJPY | 1,939,025 | 2021-01-03 | 2026-04-30 |
Across fifteen instruments there are 28,075,555 one-minute bars available. A five-hundred-trade sample is a matter of continuing rather than of finding more data.
The subtler problem
More trades fix randomness, not selection. A thousand trades all taken in the 33.1% of days that trend will report an edge belonging to those conditions. Sample size answers "is this noise"; it does not answer "will this hold when conditions change", and only out-of-sample testing speaks to the second.
FXAbsolute free backtester
Test the finding on the chart—not just on the page
Open FXAbsolute and replay S&P 500 candle by candle. Turn the observation into explicit entry, stop, target and exit rules, then measure the result yourself.
Questions
How many trades make a backtest reliable?
As a rough guide, 100 trades narrows the 95% interval around a 50% win rate to about ±10 points and 500 trades to ±4.4. Thirty trades leaves it at ±18, wide enough to contain almost any conclusion.
Is 30 trades enough to test a strategy?
No. At thirty trades an observed 60% win rate is statistically consistent with a true rate of 42%, so the result cannot distinguish a good strategy from a lucky one.
Does a large sample guarantee the strategy works?
No. Sample size addresses randomness, not selection. A large sample drawn entirely from favourable conditions still reports an edge that belongs to those conditions.