We Backtested 13 Telegram Signal Channels on Real Price Data. Four Made Money. (2026)

Of the 13 Telegram signal channels our users backtested this year, ten produced enough completed trades to score and four had positive expectancy. The median scored channel loses 0.14R on every trade it posts. Copying every completed trade in the sample at equal risk would have returned -9.32R across 134 trades - the sort of slow, unspectacular bleed that a wall of green screenshots hides perfectly.

That is the whole answer. The rest of this article is the evidence, the method, and the four things the profitable channels did differently.

Snapshot: 15 September 2026. 120 channels observed, 13 backtested, 235 signals, 134 completed trades. Every number on this page comes from a query against our production database. We publish that full query set, and the scoring formula behind it, in our leaderboard methodology article on 30 September - so you can check the arithmetic rather than trust us.

What we tested, and what we did not

Our users connect their own private Telegram channels so their signals can be copied to MT5. When a user runs a channel backtest, we take that channel's stored message history, parse the signals out of it, and replay each one against 1-minute price data for the symbol - Dukascopy tick files aggregated to minutes for most of them, MT5 bridge candles for the rest.

The population, as of the snapshot:

The gap between 120 and 13 is worth explaining, because it is the honest limit of this dataset. A backtest runs when a user asks for one, and 107 of the 120 monitored channels have never been asked about. Nobody has checked. That is the single most common state of a Telegram channel that someone is actively copying money into: unexamined.

What this test does not include: spread, commission, swap, or the slippage between the moment a signal is posted and the moment your order fills. Every result below is therefore the best case. Your real outcome is this, minus costs.

What happened to 235 signals

Two rows deserve more than a line each.

22.6% of signals had an entry that price never touched. This is the quietest form of misleading performance in the whole category. A channel posts a limit order below the market, price never gets there, price then runs to where TP1 would have been, and the channel posts a green checkmark. It is not a lie exactly - the level was reached - but nobody owned that trade. A backtest catches this because it insists the entry fills before the outcome counts. Screenshots do not.

14.5% could not be tested at all. Being straightforward about our own gaps: these are rows where we had no usable 1-minute history for the symbol and the period. Most date from a stretch when our market-data cache was empty after a restart; a Dukascopy fallback has since been added and the rate is falling. Crucially, these rows are excluded from scoring rather than counted as zeroes - a channel is never punished for our missing data.

The TP1 myth

Here is the part that changes how you read a channel's feed. Of the 134 trades that actually completed:

TP1 was reached on 57.5% of completed trades. A full win happened on 35.8%. The 22-point gap between those numbers is the entire business model of a certain kind of signal channel: post a message the moment TP1 prints, and never post again about that trade.

And look at what those partial wins were actually worth. Twenty-nine trades reached TP1 and then gave the rest back. Their average result was -0.079R, and only 6 of the 29 finished positive. Four out of five trades that had already gone far enough to bank a first target ended as net losses. The screenshot said TP1. The account said otherwise.

There is a practical lesson in that number and it is not subtle: if your copier does not move the stop to breakeven when TP1 fills, you are systematically converting small wins into small losses. We wrote that up separately in partial close and breakeven explained.

R, not win rate

Distribution of the 134 completed trades, in R:

The shape is the familiar one: a wall of full stop-outs at exactly -1R, a scattering of small wins, and a thin tail of real winners. Nothing beyond +2.48R exists in this sample, because these are channels posting defined targets, not letting winners run indefinitely.

The median scored channel won 31.0% of its trades. Four channels were still profitable, and none of them got there by winning more often - they got there by winning bigger. The best channel per trade in the sample (+0.52R) won 58.3% of the time with an average reward-to-risk of 1.26. A different channel won 92.9% of the time and made less than it per trade, because its average winner was under half an R.

If you take one habit from this article, make it this: when a channel advertises a win rate, ask what the average win and the average loss were. You can do that arithmetic on our free expectancy calculator in about twenty seconds.

The ranked eight

These are the channels with at least 10 completed trades, in rank order. Labels are generated by our system from the dominant symbol and the median holding time; names are never published.

Three things in that table are worth more than the ordering.

The best channel in the sample is ranked second. Channel #99 won 92.9% of its trades and made +0.349R each. On pure trading quality it is the standout. It last posted a signal on 28 January 2026. Our rank multiplies trust by a recency factor - 1.00 if the channel posted within 30 days, 0.85 within 90, 0.70 beyond - and #99 takes the 0.70. That is deliberate. A channel that has gone quiet for eight months is not a channel you can copy; it is a historical record.

The top-ranked channel edits and deletes a fifth of its messages. Channel #117 has the best combination of recency and results, and an honesty score of 66 out of 100 because 22.7% of its 128 stored messages were edited and 22.7% were deleted. It ranks first while telling you, in the same table, to go and look at why.

The bottom of the table is not random. The top four ranks are the four channels with positive expectancy; the bottom four are the four that lose money per trade. Within each half the ordering is shuffled by sample size, honesty and recency - Channel #111 sits last on a better expectancy than #115 because it has the fewest trades, the lowest honesty score and no signal since August. The scoring is doing its job.

Deleted messages: how common is it really?

Every article about spotting fake channels asserts that they edit their history. We can measure it, because we store every message a monitored channel posts along with whether it was later edited or deleted.

Across 92 channels with at least 20 stored messages - 41,303 messages in total:

Editing is normal. A majority of channels edit most of their messages, and much of that is innocent - a provider who updates the original post to say "TP1 hit" has edited it. Our honesty score counts edits regardless of intent, which makes a high edit rate a question rather than a verdict.

Deleting is not normal. The median channel deletes nothing at all. Here is the spread among the backtested channels:

Channel #111 deleted 655 of its 1,243 messages. More than half its published history is gone. Its backtested expectancy on what remains is -0.202R. Channel #28, at the other end, edited one message out of 209 and deleted none - and has the best per-trade expectancy in the entire sample.

That correlation is not proof of anything on its own; the sample is small. But it is the reason our honesty score weighs a deletion twice as heavily as an edit: honesty = 100 - 50 x edit rate - 100 x delete rate, floored at zero. An edited signal is usually an updated signal. A deleted signal had one job, and it is not there any more.

Gold is the whole story

Three out of four signals in this dataset are gold. Six of the eight ranked channels are gold channels, and three of those six had positive expectancy. Whatever "Telegram signals" means as a category, in practice it overwhelmingly means somebody calling XAUUSD. If you are evaluating this space, evaluate gold channels - and hold them to gold's volatility when you size the trades. Our roundup of gold signal Telegram providers goes through the named ones.

Who is still posting?

A dead channel with a beautiful record is a museum piece. Of the ten scored channels:

• Seven posted a signal within the last 30 days.
• Two of those seven are profitable per trade.
• Of the eight ranked channels, five are still active, and two of those five are profitable.

Two out of seven is the number to carry out of this article if you carry only one. Among the channels people are actually copying right now, in a sample assembled by those same people, fewer than a third made money per trade on their own posted signals, before costs.

What this does not prove

The honest limitations, in the order they matter:

1. The sample is small and self-selected. Thirteen channels, 10 to 28 signals replayed each, and 10 to 20 completed trades apiece among the eight that ranked. Users add channels they already like, which if anything biases the sample upward. A channel with 17 completed trades can be lucky or unlucky by a wide margin.
2. No costs are modelled. No spread, no commission, no swap, no slippage between the post and your fill. Every figure here is the ceiling.
3. Partial wins carry a modelling assumption. We assume a portion of the position closes at TP1 and the rest runs to the stop. A real copier's partial-close ratio changes the R.
4. INSUFFICIENT_DATA rows are excluded, not zeroed. That is the right call, but it means a channel whose signals we could not test is scored on the subset we could.
5. Past results, as always, are a description of the past. A positive expectancy over 17 trades is weak evidence of an edge. It is much stronger evidence than a screenshot.

The full formula, the factor weights and the reasons behind each of them are in how we score Telegram signal channels.

What to do with this

Copying is the easy part - there are a dozen tools that will do it, and we compare them in the best Telegram signal copiers roundup. Verification is the part almost nobody does, and it is the part that decides whether copying makes you money.

The check that matters takes one afternoon: get the channel's real history, replay it on real prices, and look at the expectancy rather than the win rate. If the channel will not let you see enough history to do that, you have your answer already. Channel backtests are included on the Pro plan (5 per week) and Advanced (20 per week). The anonymised leaderboard those backtests feed is currently in tester preview; the backtest itself is available today.

Snapshot 15 September 2026. Every number on this page is produced by one of the SQL queries we publish in the methodology article. We re-run them monthly and update this page when the snapshot changes.

See pricing or start a free trial to backtest the channels you are already copying.

Frequently Asked Questions