Win rate by itself has almost no reference value when the number of trades is small. A strategy that wins 60% of 30 trades could have a true win rate anywhere from 42% to 75% — a range that completely includes the 50% "no better than a coin flip" mark. To narrow the margin of error to plus or minus 5 percentage points, you need roughly 369 trades. So if you see a strategy report claiming a "65% win rate" based on only a few dozen trades, it is most likely describing a stretch of luck rather than a repeatable edge.
A leading global cryptocurrency platform,suitable for both beginners and experienced traders.
New user benefit: 20% off trading fees upon registration!！
Why Small Sample Win Rates Are Unreliable
Win rate is a proportion, and proportions fluctuate wildly when the sample size is small. A sample size tool on TradingView based on the Wilson interval gives a reproducible example: with 30 trades, a 60% win rate, and a 95% confidence level, the true win rate could fall between 42% and 75%. That is a range wider than 30 percentage points, which means you cannot tell whether the strategy actually has a positive edge — because it might just be random fluctuation around a 50% win rate over a month.
The same tool also gives the number of trades needed to narrow the range: about 369 trades to control the error within plus or minus 5 percentage points, about 92 trades for plus or minus 10 percentage points, and more than 2000 trades for plus or minus 2 percentage points. These numbers are purely statistical references, and they assume each trade is independent and the win rate is stable. In real trading, market regimes change and trades cluster together, so the actual sample size needed is usually larger than these figures, not smaller.
Why Different Sources Give Different "Minimum Trade Counts"
You will see recommendations ranging from 30 to 400 trades. Trade Ideas considers 30 to 50 trades enough for a preliminary comparison, and 100 trades enough to have some confidence in win rate and profit factor. A statistical discussion on the StrategyQuant forum uses the Cochran formula and estimates that, at a 95% confidence level and 5% margin of error, a sample from 10 years of 4-hour candles requires about 376 trades (around 638 trades at a 99% confidence level). Another analysis from a Russian trading community gives an even stricter threshold: at 400 trades, a profit factor below 1.75 should be treated as noise; at 1000 trades, anything below 1.42 could still be noise.
These numbers differ because they use different standards. Thirty trades only means "you can start looking." One hundred trades means "you have initial confidence." Three to four hundred trades means "the statistical range of the win rate can barely be narrowed to a meaningful degree." They do not contradict each other — they simply correspond to different judgment criteria. My view is this: a win rate based on fewer than 100 trades is only useful for ruling out obviously broken strategies, not for proving a strategy works. If you plan to trade with real money, 300 trades is a more reasonable lower reference point.
A More Easily Overlooked Problem Than Trade Count: Number of Parameters
Trade count alone is not enough. It needs to be considered together with the number of parameters in the strategy. A source on trading system research gives a "sample size to parameter ratio" rule: the ratio of trades to parameters should be at least 10, meaning you should optimize at most 1 parameter for every 10 trades. Under this rule, a strategy with 1 parameter needs at least 10 trades, with 50 recommended; 5 parameters suggest 250 trades; 10 parameters suggest 500 trades; 20 parameters suggest 1000 trades.
This perspective explains a common phenomenon: a strategy using seven or eight parameters (moving average periods, stop loss ratios, take profit targets, filter conditions, etc.) performs well over 200 trades, but the ratio of parameters to sample size is seriously out of balance. It is very likely fitting the noise in those 200 trades rather than capturing a repeatable pattern. If your strategy has many parameters, the 300-trade baseline needs to be raised accordingly.
Win Rate Is Not Everything: It Does Not Tell You How Much You Lose When You Lose
Even if a win rate holds up statistically, it is still not enough on its own to judge whether a strategy is worth executing. A 60% win rate can correspond to a profitable strategy or a losing one, depending on how much you earn on average when you win, how much you lose on average when you lose, and the cost of each trade. The TradingView sample size tool states this explicitly in its "failure scenario" section: win rate does not tell you about the size of profits and losses, costs, or drawdowns. A strategy that is right most of the time can still lose money.
A practical approach is to look at expected value after costs, not just win rate. If a strategy report only gives a win rate without the average profit/loss ratio and per-trade costs, you cannot tell whether that win rate corresponds to a positive expectation. Another easily overlooked point is survivorship bias and selection bias: if you tested 50 parameter sets and only reported the one with the highest win rate, that win rate has already been inflated by the selection process, and no confidence interval can fix that.
What You Can Check Yourself
When you receive win rate data for a strategy, ask three questions in order. First, were these trades generated in-sample or out-of-sample? If they were calculated on the same data used for parameter tuning, the win rate already contains curve fitting. Second, how many parameters are there? Divide the number of trades by the number of parameters. A ratio below 10 means a high risk of overfitting. Third, are the profit/loss ratio and costs provided? If not, the win rate cannot tell you anything on its own.
For strategies you plan to execute with real money, a conservative threshold is: at least 100 out-of-sample trades, no more than 5 parameters, and a reported expected value after costs. If you are only looking at win rate, any number below 300 trades is better treated as a clue that "this strategy might have something" rather than evidence that "this strategy works." The follow-up signal you really need to watch is whether the win rate range still excludes 50% after out-of-sample trades accumulate beyond 300.
A leading global cryptocurrency platform,suitable for both beginners and experienced traders.
New user benefit: 20% off trading fees upon registration!！
References
- TradingView · Taught to Trade - Sample Size Panel, 2026-10-29
- Trade Ideas · How to Evaluate a Trading Strategy, 2026-10-29
- StrategyQuant Forum · What is your minimum number of trades per year?, 2026-10-29
- T‑Bank · Сколько сделок нужно, чтобы поверить своей стратегии?, 2026-10-29
- Zenodo · Spiral Trading: Risk Control First, Probability Thinking and Trend Second (Sample Size to Parameter Ratio Rule), 2026-10-29
