Is the Result Significant?

Q: Exactly how many trades do I need?

There’s no single magic number — it depends on the strategy’s win/loss distribution and how large an edge you’re trying to detect (smaller edges need more trades). As rough guidance, treat under ~30 as nearly meaningless, low hundreds as suggestive, and many hundreds across varied regimes as reasonably trustworthy. Sample *diversity* (regimes covered) matters as much as raw count.

advanced7 min read

Thirty trades prove nothing. How many you need before a result means anything.

A backtest with a handful of trades can show spectacular numbers that mean nothing — the result could easily be luck. Statistical significance is about asking: do I have enough trades for this result to be trustworthy, rather than a coincidence?

The principle is sample size: the fewer trades, the more easily luck alone explains a great result — so small-sample backtests are dangerously seductive. A strategy with 15 trades and a 70% win rate is essentially a coin-flip story; you could get that by chance with no edge at all. As the trade count grows into the hundreds, randomness averages out and the true edge (your real expectancy) emerges — this is the law of large numbers from Module 1, made quantitative. A useful rule of thumb: be deeply skeptical below ~30 trades, cautious into the low hundreds, and meaningfully confident only with a large, diverse sample. And crucially, more trades across more market regimes beats more trades from one quiet period. The discipline: before believing any backtest metric — CAGR, Sharpe, win rate — ask “across how many trades, over what variety of conditions?” A brilliant number on a tiny sample is noise wearing a suit.

Small samples lie — a few dozen trades can show great numbers purely by luck, with no real edge.
Law of large numbers — the true expectancy only emerges over a large number of trades.
Rough guidance — <30 trades: distrust; low hundreds: cautious; large & regime-diverse sample: more confident.
Quality of sample — trades spanning many regimes beat the same count from one calm period.

ExampleA strategy backtested over 18 trades shows a 67% win rate and a high Sharpe — impressive-looking. But 18 trades is far too few to distinguish skill from luck; flip a coin 18 times and you’ll often see streaks that look “significant.” The same rules tested over 600 trades across bull and bear markets would actually mean something.

Key takeawayA backtest’s metrics only mean something with a large enough sample — small samples are dominated by luck (a 70% win rate over 15 trades is noise). The true edge emerges over hundreds of trades across diverse regimes. Always ask “how many trades, in what conditions?” before believing any number.

FAQs

Exactly how many trades do I need?

There’s no single magic number — it depends on the strategy’s win/loss distribution and how large an edge you’re trying to detect (smaller edges need more trades). As rough guidance, treat under ~30 as nearly meaningless, low hundreds as suggestive, and many hundreds across varied regimes as reasonably trustworthy. Sample *diversity* (regimes covered) matters as much as raw count.