The short answer
Over a small number of trades, luck is louder than skill. Before you trust a result, good or bad:
- Count the trades. Under 30 is a story, not evidence. 100–300 across different market conditions is a reasonable start.
- Calculate the range, not just the number. A 56% win rate over 50 trades really means “somewhere between about 42% and 70%”.
- Forward-test the exact rules on new data, with real costs.
- Check you followed the rules. A strategy can’t be judged on trades it didn’t take.
Then use the expectancy calculator and risk of ruin calculator to see what the numbers mean for your account.
Why ten good trades prove almost nothing
Flip a fair coin ten times. Getting seven heads happens all the time. Nobody would say the coin “has an edge”. Trading results behave the same way over small samples. That’s why the backtesting guide treats under 30 trades as an anecdote.
How often luck alone looks like an edge
Take a strategy with no edge: it wins 40% of the time with +1.5R wins and −1R losses, which is exactly break-even (1 ÷ 2.5 = 40%).
- Over 20 trades, the chance it shows a 60% win rate or better by luck alone is about 6%.
- So out of 100 traders using a no-edge method, about 6 would see a great 20-trade record and believe they’d found something.
- Over 50 trades, the chance it shows 56% or better drops to about 1.6%.
(Calculated by script with the binomial formula.) More trades make luck much less convincing. That’s the whole point of sample size. It’s also why screenshots of a few big wins mean little: see comparing yourself to other traders’ screenshots.
The win-rate range you can calculate yourself
A win rate from a sample is an estimate. A simple way to see how wide it is:
Range ≈ win rate ± 2 × √( win rate × (1 − win rate) ÷ number of trades )
(The exact multiplier is 1.96 for about 95% confidence; 2 is close enough by hand.)
Example: your record shows a 56% win rate. With +1.5R wins and −1R losses, you need more than 40% to break even.
| Trades | Likely range of the true win rate | Clearly above 40%? |
|---|---|---|
| 50 | about 42% to 70% | Only just |
| 200 | about 49% to 63% | Yes |
| 500 | about 52% to 60% | Yes, with a narrower range |
(Calculated by script.) The same 56% means something very different after 50 trades than after 500.
How many trades does a small edge need?
Most real edges are small. Say your method wins 48% of the time at +1.5R / −1R:
- Expectancy: 0.48 × 1.5 − 0.52 × 1 = +0.20R per trade.
- But each single trade swings a lot: the standard deviation of one trade’s result is about 1.25R.
- To be fairly sure (about two standard errors) that the true expectancy is above zero, you need roughly (2 × 1.25 ÷ 0.20)² ≈ 156 trades.
(Checked by script.) So if you trade three times a week, that’s about a year. That isn’t a reason to give up. It’s a reason to judge slowly, keep risk small while you collect evidence, and not switch strategy every month.
Forward testing: the step between backtest and real money
A backtest runs your rules on past data. A forward test runs the same rules on new data, live, with real costs. Our Strategy Lab lets you backtest simple rules on real gold data with costs and a curve-fit check; the forward test is what comes next.
How to forward-test a gold strategy
- Freeze the rules. Write them so someone else would take the same trades. No changes during the test.
- Decide the size of the test in advance, for example 50 trades or 3 months, whichever comes later.
- Use a demo, or the smallest real size. A small live account adds real slippage and real feelings. See our demo practice plan.
- Record every trade in R, including the ones you skipped. Our journal guide and the free trade journal help.
- Include all costs: spread, commission, swap. See spreads, commissions and swaps.
- Compare with the backtest’s noise range, not just its average.
Is the forward result “too bad”?
Example: your backtest says +0.20R per trade. After 40 forward trades you have +0.05R.
- With a per-trade spread of 1.25R, the noise band after 40 trades is about ± 2 × 1.25 ÷ √40 ≈ ± 0.39R.
- So anything from about −0.19R to +0.59R is consistent with the backtest. +0.05R is inside. It doesn’t prove the strategy works, but it doesn’t prove it’s broken either.
Keep going. If after 100–150 trades the result is still far below the backtest, look for causes:
- Costs higher than you assumed (often on fast strategies).
- Execution: late entries, skipped trades, moved stops. Check rule-following first: see grade decisions, not outcomes.
- Curve-fitting: the backtest was tuned too closely to past data. See curve-fitting and walk-forward analysis.
- A different market regime. Gold trends some months and ranges in others. See gold market regime detection.
Warning signs that a good run was luck
- Fewer than 30 trades, or all from one week or one big move.
- One or two trades made most of the profit.
- The rules changed during the run.
- Results only in one market regime (for example a strong trend).
- No costs in the numbers.
- You feel you’ve “cracked it”. See overconfidence after a winning streak.
Warning signs that a bad run is real
A losing run isn’t proof either, but these deserve a closer look:
- The forward result is outside the noise range after a decent number of trades.
- Average win has shrunk, or average loss has grown, not just the win rate.
- Costs eat most of the average win.
- The market has clearly changed character since the backtest.
If the strategy might really be broken, stop or cut size while you check. The risk of ruin calculator shows how fast a negative edge damages an account.
Your review card
- ☐ Number of trades in this sample: ____ (minimum before judging: ____)
- ☐ Win rate ____%, range ____% to ____%
- ☐ Break-even win rate for my average win/loss: ____%
- ☐ Expectancy after costs: ____ R
- ☐ Rules followed on ____% of trades
- ☐ Market conditions covered: trend ☐ range ☐ news-heavy ☐
- ☐ Decision: keep ☐ keep testing ☐ change one thing ☐ stop ☐
Common mistakes
- Judging after a week. Decide the sample size before you start.
- Changing three things at once after a bad run. Change one, then test again.
- Only counting the trades you took, not the setups you skipped.
- Ignoring costs in a backtest of a fast strategy.
- Trusting a big win rate with no range around it.
Sources
- All win rates, R values and trade counts are example numbers. The win-rate ranges (normal approximation), the binomial luck probabilities, the trade count for a +0.2R edge and the forward-test noise band were calculated by script.
- For a fuller treatment of testing, sample size tables and curve-fitting, see our backtesting lesson and trading maths and expectancy.
Educational content only, not financial advice. Trading gold with leverage carries a high risk of loss.
Frequently asked questions
How many trades do I need to know if my strategy works?
More than most people think. Under 30 trades tells you almost nothing. Around 100 to 300 trades across different market conditions is a reasonable starting point. A small edge, such as +0.2R per trade, can need about 150 trades just to be fairly sure it is above zero.
What is forward testing?
Trading your exact, written rules on new data that wasn’t used to build them, either on a demo or with very small live size, and recording every trade. It checks whether a backtest result survives real spreads, slippage and your own decisions.
Can a strategy with no edge look profitable?
Yes, easily over a small number of trades. In our example, a no-edge system shows a 60% win rate over 20 trades about 6% of the time by luck alone. Out of 100 traders, about 6 would think they found an edge.
My forward test is worse than my backtest. Is the strategy broken?
Not necessarily. Some drop is normal because of costs, slippage and how you execute. Compare the forward result with the range you’d expect from noise for that number of trades. If it is far outside that range, or worse in every way, re-check the rules, costs and data.
Should I change my strategy after a losing week?
Usually not. A week is too few trades to judge. Decide in advance how many trades you’ll take before reviewing, for example 50, and judge whether you followed your rules first.
Comments
No comments yet.