A 76 Percent Win Rate That Still Loses Money: 1,259 Telegram Signals Replayed in MQL5
A gold channel on Telegram closes three trades out of four in profit. That is what it sells. I replayed 1,259 of its signals on real ticks in the Strategy Tester. It lost money in every way I tried to copy it.
The history runs from December 2024 to July 2026. Every message was read into side, entry zone, stop and targets, and the reposts were dropped. Then an EA traded them on 134.8 million real ticks: limit order in the zone, 12 hours to get filled, 48 hours in the trade at most.
The channel posts a zone, not a price. Where your order sits inside that zone changes everything.
| Where the order sits | Win rate | Win rate needed to break even | Profit factor | Average per trade, as a share of what you risk |
| Near edge of the zone | 75.8% | 78.3% | 0.93 | -1.7% |
| Middle of the zone | 49.7% | 51.0% | 0.98 | -0.8% |
| Far edge of the zone | 24.1% | 24.1% | 0.98 | -1.2% |
Figure 1. Win rate achieved against the win rate needed to break even
Read the second and third columns together. The win rate is not the channel's doing, it is yours. It comes from the entry you pick: 76 percent at the near edge, 50 percent in the middle, 24 percent at the far edge. All three land under the win rate that the stop and the target demand.
At the near edge the typical trade risks 7 points to make 2. That needs 78.3 percent winners just to break even. The channel gets 75.8 percent. Close, and on the wrong side of the line.
Nine ways to copy it: three entries by three exits (all out at the first target, thirds with breakeven, everything held for the last target). Nine losers, with profit factors between 0.90 and 0.98. None of them bleeds fast either. This is a channel with no edge, not a machine for losing.
One warning that cost me a day. My first test, written outside MetaTrader, used a fixed UTC+3 server clock. It said the channel was a clear loser and the numbers looked solid. But the server of this broker follows the New York close, so it runs on UTC+2 in winter. With the hour fixed, the measured loss shrank by two thirds and the certainty went with it. One hour, five months a year, invented the result.
Check your own broker in a minute. Gold reopens on Sunday at 18:00 New York time. On a New York close server, the first bar of the week sits at 01:00 server time, winter and summer alike. If your offset is hard-coded, that bar jumps twice a year, and so does every signal you time.
Before copying any channel: pull the whole history instead of scrolling, kill the reposts, check that the signals are not all on one side (that is a parser bug, not a bearish week), test at least three entry points inside the zone, and put the win rate next to the win rate your stop and target demand. On its own, the win rate tells you nothing.
The full write-up, with the MQL5 source of the parser and of the replay EA, is on its way as an article. My published tools for MetaTrader are on my profile.
