You've seen it before. A forex signal provider posts a 78% win rate. Bold. Confident. Usually accompanied by a screenshot of a green equity curve.
Here's the problem: that number might be statistically meaningless.
Not fraudulent (though sometimes that too). Just — meaningless. And the Wilson Interval is the tool that proves it.
---
What Is a Win Rate, Really?
A win rate is a proportion. You take your winning trades, divide by total trades, and get a percentage.
Simple enough. But proportions have uncertainty baked into them. The smaller your sample, the wider that uncertainty gets.
Flip a coin three times, get three heads. Your "win rate" is 100%. Does that mean the coin is rigged? Obviously not. You haven't flipped it enough times to know anything useful.
The same logic applies directly to trading signals.
---
Enter the Wilson Interval
The Wilson Interval (a confidence interval formula developed by statistician Edwin Wilson in 1927) gives you a statistically honest range for a true win rate, accounting for sample size.
Instead of saying "our win rate is 68%," the Wilson Interval says: "based on this many trades, the true win rate is somewhere between X% and Y%, with 95% confidence."
That range — called the confidence interval — is the honest version of the number. A single point estimate without a confidence interval is marketing, not statistics.
The formula looks intimidating. You don't need to memorise it. What you need to understand is what it produces.
---
Why Small Samples Are Dangerous
Here's a concrete example. Imagine a signal provider shows a 70% win rate across 20 trades.
Run that through Wilson at 95% confidence. The interval comes back as roughly 46% to 88%.
That's not a win rate. That's a guess with a wide enough range to include catastrophic failure.
Now run 70% across 200 trades. The interval tightens to about 63% to 76%. Suddenly the number means something. You can actually use it to size positions and model expected returns.
This is why we publish trade counts alongside every performance metric at Trade Forge. A win rate without a denominator is noise.
| Sample Size | Stated Win Rate | Wilson Lower Bound | Wilson Upper Bound |
|---|---|---|---|
| 20 trades | 70% | ~46% | ~88% |
| 50 trades | 70% | ~55% | ~82% |
| 100 trades | 70% | ~60% | ~79% |
| 200 trades | 70% | ~63% | ~76% |
| 500 trades | 70% | ~66% | ~74% |
The math doesn't lie. Small samples produce intervals so wide they're useless for decision-making.
---
What This Means for Algorithmic Trading Signals
In algorithmic trading signals, this matters more than most providers want to admit.
Models get backtested over thousands of historical trades. Live deployment is a different story. A signal strategy might go live and fire 30 signals in its first month. Providers announce results. Subscribers react.
But 30 signals is nowhere near enough data to evaluate anything. At 70% stated win rate across 30 trades, your Wilson lower bound is around 51%. That's barely above a coin flip.
We've seen this pattern repeatedly when evaluating competitor track records. The marketing number sits at 70-80%. The statistically honest number — once you apply Wilson — includes outcomes that would wipe most retail accounts.
Our calibration engine retrains weekly on 2,400+ live signals across all asset classes. That's the sample depth where Wilson intervals start tightening into ranges you can actually trade around.
---
Win Rate Isn't Even the Right Metric
Here's the hot take: win rate is the most overused and least useful metric in retail signal marketing.
What actually matters is expectancy — which combines win rate with R-multiples (R-multiples being how much you made relative to what you risked, where 1R means you made exactly what you had at stake).
A system with a 45% win rate and average 2.5R winners beats a system with a 70% win rate and average 0.6R winners. Every time. On a long enough timeline.
Win rate without R-multiple context is an incomplete picture. R-multiple without Wilson-validated sample size is an incomplete picture. You need all three numbers, and you need them over enough trades to be statistically meaningful.
Most signal providers only show you one of the three. Usually the one that looks best.
---
When This Doesn't Apply to You
Not everyone needs to care about Wilson Intervals.
If you're a long-term investor in index ETFs or passive strategies, you're not evaluating signal win rates. This doesn't apply to you.
If you're using signals purely for trade ideas and applying your own discretionary judgment, statistical validation of the signal source matters less. You're filtering anyway.
This matters most if you're directly copying algorithmic trading signals — sizing positions based on published performance, trusting the stated metrics, or comparing forex signal providers on headline numbers.
If that's you, Wilson Intervals are one of the most important tools you're probably not using.
---
How We Apply This at Trade Forge
Every signal style we publish — scalp, day trade, swing — gets Wilson-tested before we report performance publicly.
We set a minimum trade threshold before any win rate gets published externally. Below that threshold, we report sample size and interval width instead of a point estimate.
We also apply Platt calibration (the technique that turns raw model scores into actual probabilities, so "70% confidence" actually means the signal wins 70% of the time) to our underlying signal scores before Wilson validation runs.
The goal is honest numbers. Not impressive numbers. If our best forex signals in 2026 have a validated 61% win rate with tight Wilson bounds, that's what we publish. Not a cherry-picked subset that hits 78%.
---
FAQ
How many trades do I need before a win rate is meaningful?
As a rough guide, you want 100+ trades before the Wilson Interval tightens enough to be useful. 200+ is better. Under 50, treat any stated win rate with genuine scepticism.
Can I calculate a Wilson Interval myself?
Yes. Search "Wilson score interval calculator" — there are free tools online. Input your wins, total trades, and confidence level. Takes about 30 seconds.
Does a high win rate guarantee profitability?
No. A 75% win rate with small winners and large losers loses money consistently. You need win rate, average R-multiple, and statistically valid sample size together.
Where can I see Trade Forge's live performance data?
Our track record and methodology are published at tradingforge.io. We show trade counts, Wilson-bounded intervals, and R-multiple distributions — not just headline percentages.
---
If you want to see how statistically grounded forex signals actually look in practice, join our free Telegram channel. Same signals as VIP, slight delay, no credit card. See the data for yourself before you decide anything.
---
This article is for educational purposes only. It does not constitute financial advice. Trading involves risk of loss.