How Do You Know if Your Trading Strategy Actually Has a Real Edge?
Direct answer
Edge test: how many trades do you need?
You know a strategy has an edge when its results, after all costs, are positive across enough trades that luck is an unlikely explanation. Sample size is the part most traders skip. With 100 trades, a 55% win rate has a 95% confidence range of roughly 45% to 65%, so it cannot be told apart from a coin flip. Separating 55% from 50% takes about 385 trades; separating 52% from 50% takes about 2,400. The same logic applies to average profit per trade in R multiples. Testing many variations on the same data also creates fake edges, a problem research calls backtest overfitting, so keep some data or live trades aside to confirm.
- Win-rate uncertainty ≈ √(p × (1 − p) ÷ n)
- 100 trades at 55%: 95% range about 45% to 65%
- Telling 55% from 50%: about 385 trades
- Telling 52% from 50%: about 2,400 trades
- Average +0.1R with 1R spread per trade: about 385 trades
Almost every trader believes their method works. The honest test is not whether you can explain the logic, or whether last month was green, but whether the numbers could plausibly be luck. That is a statistics question, and it has a fairly simple answer.
What people on X are asking
In October 2026 one trader wrote that a strategy doesn't have an edge because you can explain why it works, but when its rules produce a favourable outcome across enough trades, after costs. A year earlier a forex educator argued that if you have 300+ backtested trades showing profit, you have an edge. And in 2024 another account shared a method to check whether your edge has stopped working. The first point is right. The 300-trade rule is a reasonable rough floor, but whether 300 is enough depends on how big the edge is, as the table below shows.
What people in the thread said
- @thedrliquidity replied that logic you never logged is just a story, and that around 100 journaled setups with costs included is where an edge starts to show. Logging with costs is the right habit; 100 trades only reveals large edges.
- @XAUtism_fx said being able to explain a strategy had fooled them many times, and that sample size is where it falls apart. The statistics agree.
These are community comments, summarised in our words; the data and sources below come first.
The fact-checked answer
Step 1: define the edge in money, after costs. Win rate alone is not an edge. You need expectancy: average win times win rate, minus average loss times loss rate, minus spread and commission. Our page on win rate vs expectancy covers the formula. Expressing results in R, profit divided by the amount risked, makes trades comparable.
Step 2: check whether the sample is big enough. The NIST engineering statistics handbook gives the standard error of a proportion as the square root of p × (1 − p) ÷ n. Roughly 95% of the time, the true win rate lies within 1.96 standard errors of what you measured. The handbook also gives more accurate intervals for small samples, but the simple version shows the scale.
How uncertain is a measured win rate? (true rate near 50%)
Normal approximation, SE = √(0.5 × 0.5 ÷ n), range = 55% ± 1.96 × SE. Trades needed to separate a true rate of p from 50% at 95%: about (0.98 ÷ (p − 0.5))², so 385 for 55% and 2,401 for 52%. UXS calculation from NIST formulas. Tap a column header to sort.
| 30 | 9.1% | 37% to 73% | No |
| 100 | 5.0% | 45% to 65% | No |
| 300 | 2.9% | 49% to 61% | Borderline |
| 385 | 2.5% | 50% to 60% | Just |
| 1,000 | 1.6% | 52% to 58% | Yes |
The same logic for average R. If your trades average +0.1R and individual results vary by about 1R (a typical spread for fixed-risk trades), the standard error of the average is 1 ÷ √n. To be 95% sure the average is above zero you need 1.96 ÷ √n below 0.1, about 385 trades. Halve the edge to +0.05R and you need about 1,540. Small edges need very large samples.
Step 3: guard against overfitting. If you try 50 indicator settings and keep the best, the winner will look good partly by chance. A 2016 Journal of Computational Finance paper showed that standard checks often fail to catch this in investment backtests and proposed a way to estimate the probability that a backtest is overfit. The practical fix: decide rules before testing, hold back a slice of data you never tune on, then confirm on that slice and on paper or small live trades. Our page on why strategies fail live covers the other usual causes.
UXS decision diagram
Does your strategy have an edge?
Text version of the diagram
- Is expectancy positive after spread, commission and slippage? If no: No edge yet: fix costs or the rules
- Is the sample large enough for the size of the edge (see table)? If no: Keep logging trades before trusting it
- Were the rules fixed before the test, not tuned on the same data? If no: Retest on data you did not tune on
- Does it hold on out-of-sample or live trades? If no: Likely overfit or the market changed
- Otherwise: Evidence of an edge: size positions conservatively
Diagram drawn by UXS. A simplified decision aid for education, not a recommendation.
Step 4: keep checking. Edges can fade as markets change. Track a rolling sample, for example your last 100 trades, against the original expectancy, and treat a drop well outside the expected range as a signal to stop and review. Losing streaks alone are not proof; our page on how many losses in a row are normal shows how long they can run by chance.
Decision helper
| Your situation | Next step |
|---|---|
| Under 100 trades | Treat results as anecdotes; keep logging |
| 100–400 trades, small edge | Not proven; reduce size or keep testing |
| 400+ trades, positive after costs | Confirm out of sample before scaling |
| Live results far below backtest | Check costs, slippage and overfitting |
Related reading: how to keep a trading journal, what is a good Sharpe ratio for a trading strategy and how to backtest a forex strategy.
Common mistakes
- Judging an edge on 20 or 30 trades.
- Ignoring spread, commission and slippage.
- Tuning settings until the backtest looks good.
- Using win rate without average win and loss.
- Abandoning a sound plan after a normal losing streak.
FAQ
How many trades do I need to prove an edge?
It depends on the edge size: about 385 trades to separate a 55% win rate from 50%, and about 2,400 for 52%.
Is 100 trades enough to test a strategy?
Only for large edges. At 100 trades a 55% win rate is within normal noise of 50%.
What is a trading edge?
A positive expected result per trade after all costs, measured over a statistically meaningful sample.
What is backtest overfitting?
Choosing settings that fit past data by chance, so the backtest looks better than future results.
Is win rate enough to judge a strategy?
No. You also need average win and loss sizes, and costs, to get expectancy.
How do I know if my edge has stopped working?
Compare a rolling recent sample with your expected results and review when it falls outside the normal range.
Sources
- NIST/SEMATECH e-Handbook: Confidence intervals for proportions
- NIST/SEMATECH e-Handbook: Confidence limits for the mean
- Journal of Computational Finance (2016): The probability of backtest overfitting
Facts checked November 10, 2026. X posts are cited as evidence of the question, not of the answer.