What Makes a Good Trading Signal? 4 Tests Every Signal Should Pass
"This signal has a 73% win rate." "Members made $4,200 last week." "Up 412% YTD."
These are the marketing headlines that sell signal services. None of them tell you whether the signals are actually good. A 73% win rate could be a real edge or a complete fabrication. The "$4,200 last week" claim ignores the losing weeks. "Up 412%" cherry-picks the best stretch.
A good trading signal has four properties — properties that very few commercial signal services can actually demonstrate. This article describes those four tests, walks through how to apply each one to any signal service you're evaluating, and shows what passing all four looks like.
If a service can't pass all four, it doesn't mean their signals are useless. It means you can't verify whether they're useful, which is functionally the same thing.
Test 1: Measurable on a fixed horizon
The first test is the most basic and the one most signal services fail.
A good signal has a defined measurement window. "BUY NVDA at $148" is not a complete signal. The complete signal is "BUY NVDA at $148, target +2% within 4 hours." The horizon turns a bet into something testable.
Without a horizon, "right" and "wrong" are undefined. The signal service can claim victory whenever the price eventually crosses the target, even if it took three weeks. They can quietly drop signals that don't work out and only show you the ones that did.
When evaluating a service, ask: at what horizon do you measure whether a signal was correct? If the answer is vague, or "we just call out the entries and let you manage the trade" — the service has no measurable signal. You can't tell whether it's working because there's nothing to measure.
The honest version is what SultraxAI publishes: every signal is checked at 1 hour, 4 hours, and 24 hours after it fires. You can see the win rate at each horizon. The service can't selectively forget the losers because the back-check is automated.
Test 2: Reproducible from the same inputs
The second test: a good signal should fire on the same data, every time, in the same way. If two people running the same signal service look at the same chart, they should see the same signals.
This rules out anything human-discretionary. "Our analysts review the chart and call out the best setups" is not a reproducible signal. It's an opinion. Opinions can be valuable but they're not signals — they can't be back-tested honestly because there's no rule to back-test.
A reproducible signal is one where you could describe the firing conditions in code. "RSI on the 1-hour chart crosses above 30 from below, AND volume on that candle exceeds the 20-period average by 1.5x, AND price is within 2% of the 50-period moving average." That's reproducible. You can write the logic, run it over historical data, and check whether the win rate matches the live claim.
When evaluating a service, ask: what are the rules that fire a signal? If they can't or won't tell you, the signal isn't reproducible. The whole "secret sauce" framing is usually a way to hide the absence of rules.
Test 3: Survives transaction costs
The third test is the one that quietly kills most retail signal edges.
A signal with a 55% win rate, average win 0.8%, average loss 0.7% has a positive expectancy of roughly +0.13% per signal. That looks profitable until you account for round-trip transaction costs. On a typical retail brokerage with 0.2% round-trip cost (commission + slippage + spread), the +0.13% edge becomes -0.07% — net negative.
A good signal has positive expectancy AFTER realistic transaction costs for its target audience. A signal that requires institutional-grade execution to be profitable is not useful for retail traders, no matter how high the gross win rate is.
When evaluating a service, ask: have you computed net expectancy at the slippage and fees a retail trader would actually pay? If they hand you a gross win rate and shrug at the cost question, they haven't done the math. The signal might still be profitable; you just have no way to know without testing it yourself.
The right back-of-envelope test: take their advertised win rate, advertised average win, advertised average loss, and subtract 0.2% from each return (round-trip retail cost). If the system is still positive, it's worth investigating. If it's not, you're paying for marketing.
Test 4: Tracked over a meaningful sample
The fourth test is statistical: a signal's win rate is meaningless without a sample size to anchor it.
A "92% win rate over 11 signals" is not statistically distinguishable from random. The 95% confidence interval on 11 trials is roughly 30 percentage points wide. The true win rate could be anywhere from 60% to 100% — or, more relevantly, anywhere from genuine skill to pure luck.
The rough rule of thumb: under 100 resolved signals, treat any reported win rate above 60% with skepticism. From 100 to 500 signals, you can start to trust the number directionally. Above 500 signals, the win rate is statistically meaningful and small differences become real.
When evaluating a service, ask: how many resolved signals does your published win rate include? If the answer is small (under 50), the number isn't a stable estimate of true performance. If they refuse to disclose the sample size, the published win rate is functionally meaningless.
The SultraxAI dashboard currently shows a 50.4% win rate at the 1-hour horizon across 371 signals. The 95% confidence interval on that number is roughly 45% to 56% — wide enough that we explicitly tell users not to read it as a "the platform's edge is 0.4 percentage points." With more signals, the interval narrows. With fewer, it widens. The point is that the sample size is published alongside the rate.
Putting the four tests together
| Test | Pass if... | Fail if... |
|---|---|---|
| Measurable | Win rate is published at a fixed horizon | "We don't track it that way" |
| Reproducible | Firing conditions are explicit rules | "Our analysts use their judgment" |
| Survives costs | Net expectancy positive after 0.2% round-trip | Only gross numbers advertised |
| Meaningful sample | 200+ resolved signals reported | "Selected recent winners shown" |
A signal service that passes all four is rare. Most fail at least two. Many fail all four and replace verification with marketing — testimonials, screenshots, vague performance claims.
The cost of failing these tests isn't that the signals are necessarily bad. It's that you can't tell. You're paying for something you can't verify. That's a worse position than paying for something verified-mediocre, because at least with verified-mediocre you can compute whether it fits your strategy.
What a passing service looks like
A signal service that passes all four tests has these features:
- A public dashboard showing the live win rate
- Win rate broken down by fixed horizon (1h, 4h, 24h or similar)
- Sample sizes shown alongside every rate
- Disclosure of average win, average loss, and standard deviation
- Acknowledgment of where the system fails (rare but the hallmark of honest services)
- An option to export the raw signal history so you can verify the claims yourself
Very few services in retail trading have all of these. The ones that do are worth investigating seriously. The ones that don't are worth a quick check against the four tests and then a polite "no thank you."
SultraxAI publishes the dashboard with all of the above. We get this right not because we're virtuous but because we believe the audience that wants this kind of transparency is the audience worth serving. Marketing-only signal services churn through retail traders who realize the numbers don't match. Verified signal services don't have that churn problem because the numbers are public from day one.
The four-test framework above will save you money on the next signal service you evaluate. It won't tell you whether to use the service. But it will tell you whether you can trust what they're claiming — which is the more important question.
Try the platform (free)