How Many Trades Do You Need to Know If a Strategy Works?
Each faint trail is one simulated future for this strategy. The bold line is the median — half of runs finished above it, half below. The shaded band spans the lucky 10th to unlucky 90th percentile, so an outcome inside the band was reasonable to expect; outside, less so.
- After 20 trades, a measured 50% win rate could really be anywhere from about 28% to 72%
- After 100 trades the range narrows to roughly ±10 percentage points — still wide
- After 500 trades it narrows to about ±4 points, enough to make real decisions
- Small samples are why most strategy changes are reactions to noise, not signal
- Measure expectancy over the sample, not just win rate — both carry uncertainty
A trader takes 20 trades, wins 13, and concludes the strategy has a 65% win rate. Another takes 20 trades, wins 7, and abandons a strategy as broken. Both are drawing conclusions from a sample that is too small to support them. The math explains why.
The confidence interval for a win rate
A win rate measured from a sample is an estimate. The 95% confidence interval tells you the range the true win rate probably sits in:
95% range ≈ Measured win rate ± 1.96 × √( p × (1 − p) ÷ n )Where p is the measured win rate and n is the number of trades. The interval is widest when p is near 50%, which is where most strategies sit.
How wide the range really is
| Trades (n) | ± range at 50% win rate | True win rate could be |
|---|---|---|
| 20 | ±21.9 pts | 28% – 72% |
| 50 | ±13.9 pts | 36% – 64% |
| 100 | ±9.8 pts | 40% – 60% |
| 200 | ±6.9 pts | 43% – 57% |
| 500 | ±4.4 pts | 46% – 54% |
| 1,000 | ±3.1 pts | 47% – 53% |
At 20 trades, a measured 50% win rate is compatible with a strategy that is clearly losing and one that is clearly winning. That is not a measurement. At 100 trades it is still a ten-point band either side. Only around 500 trades does the estimate become precise enough to separate a 1.5R strategy that loses (below 40%) from one that wins comfortably.
Why this matters for your edge
Break-even at 1.5:1 reward:risk is a 40% win rate. If your measured win rate after 50 trades is 48%, the true rate could be as low as 34% — well below break-even. You cannot yet claim an edge. After 300 trades at the same 48%, the lower bound rises to about 42%, and the edge starts to look real.
Practical rule: treat anything under 100 trades as a hypothesis, 100–300 as early evidence, and 300+ as a measurement worth making decisions on.
How to get to a meaningful sample faster
- Backtest bar by bar on historical data to build a large sample before going live
- Forward test on a demo account at the same frequency you plan to trade live
- Log every trade in R so samples from different account sizes can be combined
- Do not change rules mid-sample — each change resets the count