A trader runs a strategy through 20 trades, sees a 70% win rate, and concludes they've found an edge. Three months and 150 trades later, that win rate has drifted down to 48% and the account is underwater. Nothing about the strategy changed โ what changed is that 20 trades was never enough data to know anything in the first place. Sample size is the quiet variable behind most "my strategy stopped working" stories, and it's almost never actually the strategy.
A 70% win rate over 20 trades and a 70% win rate over 200 trades are not the same claim โ one is a coin flip that went your way, the other is evidence.
30
trades โ the loose statistical floor for any win-rate estimate to start stabilizing
100+
trades typically needed before expectancy and profit factor stop swinging wildly
ยฑ20%
how far a true 50% win rate can realistically wander over just 20 trades
200-300
trades often cited as a reasonable sample for evaluating a full strategy, across market conditions
1
regime change that can quietly invalidate even a large historical sample
Why Small Samples Lie
Flip a fair coin 20 times and it's not unusual to land 13 or 14 heads โ a 65-70% "win rate" from pure randomness, with a coin that has no edge at all. Trading has the same problem, compounded by the fact that most setups don't have a fixed, known probability the way a coin does. A short run of trades can hand you a flattering win rate, a beautiful equity curve, and total confidence โ none of which tells you anything reliable about what happens over the next 200 trades. The math term for this is variance, and small samples have a lot of it.
The Law of Large Numbers, in Trading Terms
The law of large numbers says that as you repeat a random process more times, the observed results converge toward the true underlying probability. Flip a coin 20 times and you might see 70% heads; flip it 2,000 times and you'll land very close to 50%. Trading works the same way: your first 20-30 trades on a setup are mostly noise wrapped around a signal you can't yet see clearly. Your next 200 start to reveal what the setup's real win rate, average win, and average loss actually are.
A 70% win rate over 20 trades is a story. A 70% win rate over 300 trades is a strategy.
How Many Trades Is "Enough"? A Practical Scale
There's no single magic number โ the honest answer depends on how much confidence you need and how much variance your setup naturally has. But there are useful reference points traders actually use:
| Sample Size | What You Can Reasonably Say | What You Still Can't |
|---|---|---|
| Under 20 trades | Almost nothing โ treat this as a pilot run, not a verdict | Anything about win rate, expectancy, or whether the edge is real |
| 20-50 trades | Rough directional read; worth continuing if results aren't disastrous | Confidently size up or market the strategy as "proven" |
| 50-100 trades | Win rate and R:R start resembling their true values, with wide error bars | Rule out a lucky or unlucky stretch entirely |
| 100-300 trades | Expectancy and profit factor become meaningfully stable, especially if spread across varied conditions | Guarantee performance holds if the market regime changes |
| 300+ trades, multiple regimes | Reasonable statistical confidence the edge is real, not noise | Assume it survives forever โ edges still decay |
Sample Size Isn't Just Trade Count โ It's Trade Variety
300 trades taken in a single trending month don't carry the same statistical weight as 300 trades spread across trending, ranging, high-volatility, and low-volatility conditions. A setup can look bulletproof through a strong uptrend and fall apart the moment the market chops sideways for six weeks. When you're judging whether you have enough data, ask not just "how many trades" but "how many different kinds of market I've actually tested this in." A large sample from one narrow regime is still a small sample in the way that matters.
Why This Matters More for Some Trading Styles Than Others
- Scalpers and high-frequency day traders can reach 100+ trades in weeks โ sample size builds fast, but so does the temptation to over-optimize on noise
- Swing traders may need several months to accumulate 100 trades โ patience matters more than speed here
- Position traders might place 5-15 trades a year, meaning true statistical significance can take a decade โ journaling every trade's thesis in depth matters more than the trade count ever will
- Options and futures traders juggling multiple simultaneous positions should count independent setups, not just filled orders, when estimating real sample size
The Trap: Optimizing a Strategy Based on a Small Sample
The most expensive mistake sample size causes isn't misplaced confidence โ it's over-optimization. A trader sees a losing stretch after 15 trades, tweaks the entry rule, sees three wins, and concludes the tweak fixed it. In reality, three trades proves almost nothing, and the "fix" may have just curve-fit to recent noise. Every rule change resets your effective sample size back toward zero for that rule. Traders who change their strategy every 10-15 trades never actually accumulate enough data on any single version of the strategy to know if it works.
How Your Trading Journal Fixes This
The reason sample size trips traders up so often isn't that the math is complicated โ it's that most traders don't have a clean, tagged record of exactly which trades belong to which version of a setup. Without that, "how many trades have I taken on this exact setup, with this exact rule set" is a guess, not a number. This is what tagging trades by setup is really for: it lets you filter your history down to a specific strategy and see, honestly, whether you're looking at 12 trades or 150.
- 1Tag every trade with the exact setup or rule version it belongs to, not just a general strategy name
- 2Treat any read on win rate or expectancy under ~30 trades as provisional, not conclusive
- 3Wait for 100+ trades on a specific rule set before increasing size meaningfully based on its results
- 4When you change a rule, mentally reset the sample size clock for that version โ don't blend old and new results
- 5Check whether your sample spans multiple market regimes before trusting it fully, not just whether it's numerically large
None of this means you should freeze and wait 300 trades before trusting any data โ early signal still matters, and a strategy that's clearly bleeding money after 20 trades doesn't need 280 more to prove it's broken. The point is calibration: hold small samples loosely, treat early win rates as hypotheses rather than facts, and let your confidence grow only as the trade count โ and the variety of conditions behind it โ actually grows with it.
TradeEdge tags every trade by setup automatically, so you can filter your history and see exactly how many trades back a given strategy โ before you decide whether to trust it.