Sample Size: How Many Trades Before the Numbers Mean Something?
You took 20 trades and made 8R. Is the strategy good? The honest answer is that you cannot tell yet.
With small samples, luck dominates. Consider a coin that wins 50 percent. In 20 flips, getting 13 or more heads occurs in about one in eight tries. Applied to trading, a break-even strategy can show a handsome profit over 20 trades simply by chance.
Confidence grows with the square root of the sample size, which means that to halve your uncertainty, you need four times the trades. Rough guidance suggests at least 100 trades for a basic impression, and 200 to 300 for more reliable conclusions, especially for strategies with low win rates or high variance in outcomes.
Streaks of results can mislead in both directions. A good strategy can show a losing month, and a poor one can show a winning month.
This has practical implications. Do not abandon a promising strategy after ten losses, and do not scale up after ten wins. Decide in advance how many trades you will take before evaluating, and stick to that.
Lower-frequency strategies take longer to evaluate, which is a hidden cost of swing trading.
When assessing a vendor or signal service, ask for sample size before looking at returns.
Practical step: count the trades in your current track record and write down how many more you need before you will judge it.
Sample Size: How Many Trades Before the Numbers Mean Something?
Re: Sample Size: How Many Trades Before the Numbers Mean Something?
Worth saying out loud: 20 trades is about a month for a lot of part-time traders. So the honest timeline for knowing anything is more like four to six months, not a couple of weeks.
It’s Fairman 
Re: Sample Size: How Many Trades Before the Numbers Mean Something?
If waiting for 100 live trades feels too slow, backtest the same rules to get the sample, then use the live trades only to check you're executing them the way the test assumes. Two different jobs.
It’s Fairman 
Re: Sample Size: How Many Trades Before the Numbers Mean Something?
Low win rate setups need even more patience. With a 30 percent win rate, ten losses in a row is not rare at all, so a short sample can make a decent system look broken.
It’s Fairman 
-
Shadow Trader
- Posts: 5594
- Joined: Wed Sep 09, 2026 5:19 pm
Re: Sample Size: How Many Trades Before the Numbers Mean Something?
The coin example gets sharper with a payoff attached. Say a strategy wins 40 percent of the time and winners pay 1.5R, so expectancy is 0.4 times 1.5 minus 0.6, which is exactly zero.
Over 20 trades it needs 12 wins to show 8R or better, since 12 wins and 8 losses is 18R minus 8R, or 10R, while 11 wins only reaches 7.5R. The chance of 12 or more wins at a 40 percent hit rate is about 5.7 percent, roughly one trader in 18 by luck alone.
Stretch it to 100 trades and ask for the same 0.4R per trade, which is 40R. That needs 56 wins or more, and the odds drop to under 0.1 percent.
The chart shows the spread of average R per trade for that zero edge system at both sample sizes, with the 0.4R line marked. The 20 trade version is wide enough that a decent looking month sits comfortably inside it, while the 100 trade version barely reaches it.
So the question to ask of 20 trades is not whether the result is good but how many traders with no edge would have printed it. One in 18 is not rare on a busy forum.
Over 20 trades it needs 12 wins to show 8R or better, since 12 wins and 8 losses is 18R minus 8R, or 10R, while 11 wins only reaches 7.5R. The chance of 12 or more wins at a 40 percent hit rate is about 5.7 percent, roughly one trader in 18 by luck alone.
Stretch it to 100 trades and ask for the same 0.4R per trade, which is 40R. That needs 56 wins or more, and the odds drop to under 0.1 percent.
The chart shows the spread of average R per trade for that zero edge system at both sample sizes, with the 0.4R line marked. The 20 trade version is wide enough that a decent looking month sits comfortably inside it, while the 100 trade version barely reaches it.
So the question to ask of 20 trades is not whether the result is good but how many traders with no edge would have printed it. One in 18 is not rare on a busy forum.
-
FXS Prop Desk
- Posts: 1689
- Joined: Wed Oct 07, 2026 9:52 am
Re: Sample Size: How Many Trades Before the Numbers Mean Something?
Twenty trades can produce results that look like a strong edge from pure chance.
A rough guide is 100 trades or more before drawing firm conclusions, and even then the confidence range is wide. Tracking results in R rather than money also makes it easier to compare samples of different sizes.
A rough guide is 100 trades or more before drawing firm conclusions, and even then the confidence range is wide. Tracking results in R rather than money also makes it easier to compare samples of different sizes.
FXS Official Team | The Idea Lab & Strategy
-
Shadow Trader
- Posts: 5594
- Joined: Wed Sep 09, 2026 5:19 pm
Re: Sample Size: How Many Trades Before the Numbers Mean Something?
Fairman's 30 percent point deserves numbers, because the streaks run longer than people expect. Say a system wins 30 percent of the time and winners pay 3R, which gives an expectancy of 0.3 times 3 minus 0.7, or 0.2R per trade.
Any given run of ten trades all losing has a chance of 0.7 to the power of 10, about 2.8 percent. Over 100 trades, with so many places a run can start, a streak of ten or more shows up about 58 percent of the time. Eight in a row turns up somewhere in about 86 percent of 100 trade samples for that system, so a rule to stop after eight straight losses would shut down a working system most of the time.
Ten straight losses cost 10R, half of the 20R the system should make over those 100 trades. That drawdown is a normal part of the edge, not a sign it has gone.
So the stop rule for a low win rate system is better written in trades than in losses. Something like a review after 100 trades works, cutting early only if the losing run is far longer than these odds allow.
Any given run of ten trades all losing has a chance of 0.7 to the power of 10, about 2.8 percent. Over 100 trades, with so many places a run can start, a streak of ten or more shows up about 58 percent of the time. Eight in a row turns up somewhere in about 86 percent of 100 trade samples for that system, so a rule to stop after eight straight losses would shut down a working system most of the time.
Ten straight losses cost 10R, half of the 20R the system should make over those 100 trades. That drawdown is a normal part of the edge, not a sign it has gone.
So the stop rule for a low win rate system is better written in trades than in losses. Something like a review after 100 trades works, cutting early only if the losing run is far longer than these odds allow.
-
Shadow Trader
- Posts: 5594
- Joined: Wed Sep 09, 2026 5:19 pm
Re: Sample Size: How Many Trades Before the Numbers Mean Something?
A strategy that makes money over 20 trades proves very little. Randomness can produce impressive short streaks from a system with no edge. Judging a method requires enough trades for statistical confidence, which for scalping often means several hundred at minimum.
The reasoning is based on variance. With a win rate around 50 percent, the typical deviation from expected results shrinks only with the square root of the number of trades. Small samples leave wide ranges of possible outcomes.
Scalpers have an advantage in that they generate many trades quickly, allowing larger samples in weeks rather than years. The risk is drawing conclusions from a limited number of trades in a particular market condition. A strategy tested only during a trending month may fail in a ranging one. Aiming for samples that span different conditions, volatility levels, and sessions provides a more robust picture. Confidence intervals around expectancy, easily calculated in a spreadsheet, show how uncertain the estimate remains.
The reasoning is based on variance. With a win rate around 50 percent, the typical deviation from expected results shrinks only with the square root of the number of trades. Small samples leave wide ranges of possible outcomes.
Scalpers have an advantage in that they generate many trades quickly, allowing larger samples in weeks rather than years. The risk is drawing conclusions from a limited number of trades in a particular market condition. A strategy tested only during a trending month may fail in a ranging one. Aiming for samples that span different conditions, volatility levels, and sessions provides a more robust picture. Confidence intervals around expectancy, easily calculated in a spreadsheet, show how uncertain the estimate remains.