Most signal providers quote a win rate. We quote a win rate with confidence bounds. That distinction matters more than almost anything else in algorithmic trading signals.
Here's why — and how we do it.
---
Win Rates Lie. Statistics Don't Have To.
A 68% win rate sounds solid. But what does it actually mean if it's built on 12 trades?
Almost nothing. With a small sample, a 68% win rate is statistically compatible with a true underlying rate anywhere from 40% to 88%. You can't act on that range. It's noise dressed up as signal.
This is the problem we set out to solve when building Trade Forge's performance reporting layer.
---
What the Wilson Interval Actually Is
The Wilson interval (a method for calculating confidence bounds around a measured proportion — developed by Edwin Wilson in 1927) gives us upper and lower bounds on a win rate that account for sample size.
The formula looks intimidating. The concept isn't.
If we've fired 40 signals and won 28 of them, our observed win rate is 70%. But the Wilson interval at 95% confidence might tell us the true win rate sits somewhere between 54% and 82%. That's a very different story from "70% win rate."
As sample size grows, those bounds tighten. At 500 signals with a 70% observed rate, the Wilson interval narrows to roughly 65.7% to 74.0%. Now we're saying something meaningful.
We use the lower Wilson bound as our headline figure. Always. If the lower bound is 54%, we report 54%. Not 70%.
That's the Trade Forge standard.
---
Why We Don't Use the Simple ("Normal Approximation") Method
The textbook approach to confidence intervals uses the normal approximation. It works fine when your win rate is near 50% and your sample is large.
It breaks at the edges.
If your win rate is 90% or 12%, or your sample is under 30, the normal approximation produces intervals that bleed outside the 0–100% range. You get lower bounds below zero. Mathematically absurd, practically useless.
The Wilson interval stays bounded. It handles small samples without producing nonsense. For a forex signal provider operating across 24 pairs, multiple timeframes, and three signal styles (scalp, day trade, swing), small-sample edge cases appear constantly. We needed a method that didn't break.
---
How This Shows Up in Our Live System
Here's a concrete example from our data.
Our calibration engine retrains weekly on 2,400+ live signals. Early in the history of a new pair filter, we had 34 signals on a commodity pair. Observed win rate: 64.7%. Sounds fine. Wilson lower bound at 95% confidence: 47.1%.
We didn't promote that filter. We labeled it "developing — insufficient sample." It took another six weeks and 80+ additional signals before the lower Wilson bound crossed our internal threshold of 55%. Only then did we publish that filter's performance publicly.
That's the difference between honest reporting and marketing.
---
Comparing Reporting Methods
| Method | Handles Small Samples | Stays in \[0,1\] Range | Conservative Bias | Our Use |
|---|---|---|---|---|
| Raw win rate | No | Yes | No | Never headline |
| Normal approximation interval | Poorly | No | No | Deprecated |
| Wilson interval (95%) | Yes | Yes | Yes | Primary metric |
| Bayesian credible interval | Yes | Yes | Configurable | Secondary check |
We run Wilson as our primary. We cross-check with a Bayesian credible interval (a probability-based alternative that incorporates prior beliefs about win rates) on signal categories with fewer than 50 samples. If they diverge significantly, we flag the result as preliminary.
---
R-Multiples and Why Win Rate Alone Is Still Incomplete
Win rate, even properly measured, doesn't tell the whole story.
We also track R-multiples (R-multiples measure how much you made relative to what you risked — a trade that returns 2R made twice your risk; a trade that loses costs 1R). A strategy with a 45% win rate but an average winning R-multiple of 2.8 beats a 65% win rate strategy with an average win of 0.9R. The math is straightforward.
We publish both. Wilson-bounded win rate and average R-multiple, broken down by asset class, signal style, and session window. Our position: if a forex signal provider won't show you both numbers with sample sizes attached, that IS the track record.
---
Who Doesn't Need This
Not everyone needs to think about Wilson intervals.
If you're a long-term investor in ETFs or index funds, signal-level win rate statistics aren't relevant to your strategy. You're not making dozens of tactical entries per week. The statistical machinery we've built is for active signal followers — people making multiple decisions monthly where sample size and calibration actually matter.
If you're evaluating best forex signals 2026 options and you trade once a quarter, stop here. This framework isn't for you, and we'd rather tell you that than waste your time.
This level of statistical rigour is relevant when you're following high-frequency algorithmic trading signals at scale — scalp signals, short-duration day trades, or active swing setups across multiple pairs simultaneously.
---
FAQ
Why do you use the lower bound instead of the midpoint?
Because the midpoint flatters. The lower bound is the pessimistic, honest estimate. We'd rather under-promise and over-deliver than the reverse.
How many signals do you need before results are meaningful?
We set a minimum of 100 signals before any filter is considered statistically stable. Below that, we publish results with an explicit "preliminary" label and show the Wilson bounds visually.
Does this apply to crypto and commodity signals too?
Yes. We apply the same Wilson-based reporting across all asset classes — forex, crypto, commodities, indices, and ETFs. The mathematics doesn't care what the underlying asset is.
What's your current lower-bound win rate across live signals?
We publish this monthly on the Trade Forge website. It updates as new live results come in. We don't lock in a number here because it changes — and that's the point.
---
See It in Practice
The best way to understand how this works isn't to read about it. It's to watch live signals alongside their running statistics.
Our free Telegram and Discord channels carry the same signals as VIP (with a short delay). Join, watch the signals come through, and track the numbers yourself. The methodology is real. The data is public. Make your own assessment.
Join the free channel at tradingforge.io
---
This article is for educational purposes only. It does not constitute financial advice. Trading involves risk of loss.