Every signal channel advertises a percentage. Our replay of 3,227 published signals shows why that number decides almost nothing on its own, and what has to sit beside it before it means anything.

What does a win rate actually count?

A win rate is the share of trades that ended in profit, and channels also sell it to you as accuracy or hit rate. That sounds like a definition, though it does not work as one: everything depends on what counts as a trade, what counts as ended, and who decides. Two of the largest exchanges publish incompatible formulas under the same word.

Binance computes the win rate for a copy-trading lead as the number of profitable closed positions divided by the total number of positions, and states that partial closes are not counted as closed positions (Binance copy trading performance indicators). OKX, under a heading that uses the same phrase, divides the number of days with profit by the number of days leading trades (OKX Learn).

Those two numbers cannot be compared with each other, and neither can be compared with a Telegram channel's own figure, which is usually computed by the channel from its own posts. Ask what the number counted before you ask whether it is impressive.

Why does the same percentage mean opposite things?

Because a win rate says how often, and money depends on how much. Both halves are needed, and the relationship between them is arithmetic anyone can redo on a calculator.

Call the distance from entry to stop one unit of risk. If the target is the same distance away, the reward-to-risk ratio is 1:1. The share of trades you need to win just to break even is 1 divided by 1 plus that ratio:

Reward to risk Win rate needed to break even
0.25 to 1 80.0%
0.5 to 1 66.7%
1 to 1 50.0%
2 to 1 33.3%
3 to 1 25.0%

At 1:1 that division is 1 over 2, so 50%, the figure everyone already treats as neutral. Every other row is the same sum with a different denominator.

Read the top row again. A strategy whose target is a quarter of the size of its stop needs to win four trades in five before it earns anything at all. That structure is easy to build and easy to advertise, because a small target close to the entry gets hit often. The losses are rare and large; the percentage stays high while the account goes down.

The reverse holds too. A channel winning a third of its trades with targets three times the size of its stops is doing better than a channel winning 79% at a quarter-size target, even though nobody would buy the first one.

What do our own numbers say?

Our corpus loses money, and it does so at a percentage that sounds ordinary. Replaying every signal the way its channel published it, with the full ladder of targets and the channel's own stop, 1,514 outcomes carry a profit or loss figure as of August 2026. Of those, 475 finished in profit, 728 finished in loss and 311 finished flat.

As-published replay, as of August 2026 Value
Outcomes with a result 1,514
Finished in profit 475 (31.4%)
Average result on a winner +5.17%
Average result on a loser -4.73%
Reward to risk implied by those averages 1.09 : 1
Win rate needed to break even 47.8%
Average result per outcome -0.65%

The payoff here is nearly symmetrical, so the break-even line sits just under half. Taking these calls at equal size, you would have needed to be right about 48% of the time. The channels were right 31% of the time, and that gap is where the -0.65% per outcome comes from.

Now change one rule. Score the same signals under our public one-take metric, which closes at a 2% move instead of following the channel's whole ladder, and the hit rate for calls carrying the channel's own stop comes out at 60.5%. Same channels, same messages, same market — and a number nearly twice as high, because the target moved closer. Neither figure is wrong and neither is a measure of skill on its own. Our full rule set is on the Methodology page.

How many trades before a percentage means anything?

Fewer trades produce more extreme percentages, and small samples are where the impressive numbers live. If a channel shows you 70% across 20 signals, the range of true win rates consistent with that result runs from roughly 48% to 85%, which covers "worse than a coin flip" and "genuinely good" at once. That range is a confidence interval, the band you get when you ask what underlying rate could plausibly have produced the sample you saw.

The construction used here is the Wilson interval, which NIST recommends precisely because the simplest textbook formula misbehaves on small samples (NIST/SEMATECH e-Handbook, section 7.2.4.1). In practice that means separating a true 55% from a coin flip takes several hundred trades, not several dozen.

Selection makes it worse. Take a hundred channels that are all guessing, each publishing 20 signals. A handful will finish above 70% by chance alone, and those are the ones with a screenshot to post.

We rate a channel from 10 scored verdicts, and by the standard above that threshold is far too low to prove anything. Ten is where a percentage stops being pure noise; it is nowhere near where it becomes reliable. At 14 scored trades the number is barely a hint, and at 337 it is worth arguing about, which is the range our rated channels occupy as of August 2026.

Why do the best-looking numbers carry the least evidence?

Because the same conditions that produce a high percentage also produce a small number of resolved trades. In our data the relationship is visible directly: under the one-take rule, the best hit rates sit on samples of a couple of dozen trades and stops nothing touches.

Two mechanisms drive it. The first is the wide stop. When a channel publishes no stop, our replay applies a 10% fallback so the call can be scored at all, and a stop that far away is rarely touched. Signals scored under our fallback show a hit rate of 81.3%, against 60.5% where the channel published its own stop, so our own choice of fallback is what lifts those channels. The second is the unfilled entry: of the 1,639 outcomes in that fallback group, only 390 ever produced a scored verdict, because most were cancelled when price ran past the target before the entry filled.

So the channels supplying the least information about risk end up with the healthiest-looking percentage and the smallest evidence base behind it. On our own pages, a rate of 91.3% on 23 scored trades sits beside a rate of 56.9% on 297, and the first is the one that catches the eye. Those two numbers measure different things, and the first is mostly the sample talking. What the missing stop does to all of this is covered in Why half of all crypto signals come without a stop-loss.

How is a high win rate manufactured?

Six choices about what to count, every one of which raises the percentage, and not one of which requires falsifying a single post:

  1. Publish no stop. With no defined loss, a trade that goes wrong stays open and never enters the record as a loser.
  2. Publish a stop nothing will reach. The loss is real when it lands, and it lands rarely, so the ratio of wins to losses looks strong right up until one arrives.
  3. Close manually, by announcement. When the exit is decided in chat rather than by a published level, every outcome is the author's to name.
  4. Count only the first target. A ladder of six targets lets "TP1 hit" be announced as a win on a trade that later reversed through the entry.
  5. Discard the entries that never filled. Calls the market ran away from vanish from the denominator, and they are numerous: 49% of the outcomes in our corpus end that way, though one channel supplies most of them and the rest average about 22%, as detailed in Half the trades never happen.
  6. Report gross. Fees, funding on a held perpetual and slippage all come out of the follower's result and none of them appear in a percentage.

The CFTC lists several of these as markers of a fraudulent trading system in its own advisory to consumers, including the warning to "be alert for the possibility that the system promoter manufactured results by selecting historical trades that would have yielded the greatest returns" and the reminder that in a paper record "a stop-loss order might be executed at a better or worse price, or not be executed at all" (CFTC, Commodity Trading Systems Sold on the Internet).

What does holding losers do to the number?

It inflates it, and the effect has been measured. Traders close their winners and keep their losers open, so the record of closed trades looks better than the portfolio it came from.

Terrance Odean studied 10,000 discount brokerage accounts between 1987 and 1993 (Are Investors Reluctant to Realize Their Losses?, Journal of Finance, 1998). Those investors realised 23.3% of their gains against 15.5% of their losses, a gap with a t-statistic of -32 in his Table II. The behaviour was not vindicated by results:

"For winners that are sold, the average excess return over the following year is 3.4 percent more than it is for losers that are not sold." — Terrance Odean, Are Investors Reluctant to Realize Their Losses?, Journal of Finance, 1998

The arithmetic of his own position counts shows the size of the distortion. Those accounts held 47,000 positions in gain out of 107,978 in total, which is 43.5%, while their closed trades came in at 53.6% winners. Ten percentage points of apparent accuracy came from which trades were allowed to close.

That is US equities in the 1990s, not crypto signals, and it describes individual investors rather than channels. The mechanism transfers because the incentive does: any record built from closed positions can be improved by leaving the bad ones open.

Does a high win rate mean skill?

Winning for a while is common. Winning repeatedly is rare, and the gap between those two facts is where most performance claims live.

The largest study of the question followed Taiwanese day traders from 1992 to 2006, a population averaging around 450,000 a year, and found roughly 20% earning profits net of fees in a typical year:

"While approximately 20% earn profits net of fees in the typical year, the results of our analysis suggest that less than 1% of day traders (4,000 out of 450,000) are able to outperform consistently." — Barber, Lee, Liu and Odean, The Cross-Section of Speculator Skill, Journal of Financial Markets, 2014

A Brazilian study of 19,646 people who began day trading between 2013 and 2015 found 1,551 who persisted beyond 300 days, of whom 97% lost money and eight individuals earned more than an entry-level bank teller (Chague, De-Losso and Giovannetti, Day trading for a living?). For crypto specifically, the Bank for International Settlements estimated that between 73% and 81% of crypto app users likely lost money, with the median investor down 48% (BIS Working Paper 1049).

Against those base rates, a channel's 90% is an extraordinary claim arriving with ordinary evidence. Professional records cluster near 50% and earn their money on the size of the winners.

What would a regulated firm have to publish?

A much narrower number than a channel does, with disclosure attached. None of these rules binds a Telegram channel, which is exactly why the comparison is informative: they show what a regulator considers the minimum for an honest percentage.

Requirement Rule Who it binds
Performance figures must be representative of all reasonably comparable accounts, and net of all commissions, fees and expenses NFA Compliance Rule 2-29(b)(5) NFA member firms in the US futures markets
Profit may not be mentioned without an equally prominent discussion of the risk of loss NFA Compliance Rule 2-29(b)(3) Same
Backtested results are prohibited once three months of actual results exist NFA Compliance Rule 2-29(c)(4) Same
Extracted or selected performance requires the full portfolio it came from SEC Marketing Rule 17 CFR 275.206(4)-1(d)(5) Registered US investment advisers, for securities
An inappropriately short performance period is itself potentially misleading ASIC Regulatory Guide 234.90 and 234.99 Australian licensees

The Australian regulator states the sample-size half of this outright. RG 234.90 treats an advertisement as potentially misleading if it uses past performance from "an inappropriately short time period". For something operating under a year, RG 234.99 adds, such a period "would usually be inappropriate and may be misleading" (ASIC RG 234, June 2026).

The National Futures Association, the self-regulatory body for US futures firms, has also examined whether disclaimers fix the problem, and concluded they do not. Its interpretive notice on hypothetical results opens by saying the practice "has repeatedly produced misleading promotional material", and adds that "the use of the mandated disclaimer has not prevented recurring abuses" (NFA Interpretive Notice 9025).

What to read instead of the headline percentage

Six things, in the order they change your view of the number:

  1. The number of trades behind it. Under a few dozen, treat the percentage as unmeasured.
  2. The average win and the average loss. Without both, no percentage can be turned into money.
  3. Whether a stop was published. No stop means no defined loss and no scoreable record.
  4. What counted as a win. First target, full ladder, or a manual close announced in chat.
  5. What was excluded. Unfilled entries, deleted calls and trades still open are all denominators quietly removed.
  6. Whether costs are in it. Fees and funding are paid by the follower, not by the percentage.

Where those checks sit in a full pre-purchase process is set out in How to verify a crypto signal channel before you pay, and the limits of any replayed record, ours included, are in What a backtest can and cannot prove.

What a win rate can legitimately tell you

It is incomplete rather than worthless. A win rate computed on a known rule set, over a stated number of trades, against published levels, tells you how often a channel's calls reached a defined target before a defined stop. That is a real fact about the past, and it is also the limit of what our own ratings can claim.

What no percentage carries on its own is the size of anything: how much a win paid, how much a loss cost, how much of the record never got counted. With those figures beside it you have evidence; alone, you have a headline. For traders rather than channels, the equivalent full-history measure is described in How to read Life Score.

Current channel ratings are at Signal Providers, and what our verification does and does not claim is set out in the Verification Guidelines. Nothing here is financial advice.

Sources