Why a 70% win rate is a myth — and 58% is already profitable
The break-even math, the '70%' configs that died in our validation, and the real ceiling the data supports. The article course sellers hope you never read.
Start with the math nobody does
At 87% payout, breaking even requires 53.5% accuracy (100 ÷ 187). Which means: an honest 58% strategy is already strongly profitable long-term. If you understand the math you don't need 70% — whoever promises 70% is betting you won't do this division.
Where the "70%" numbers come from
In our own studies, 70%+ figures showed up several times. All of them died in validation. Real examples from our research log:
| Configuration | In train it looked like… | In validation… |
|---|---|---|
| 'With-the-tide' gate on the champion | 66.9% | 51.2% — pure overfit |
| DR + 3-candle streak filter | 63.0% (n=109) | tiny sample = noise; unstable across pairs |
| 4pm hour cherry-pick | 65% | 40% in the next window |
The recipe for fake 70% is always the same: tuning parameters on the same window where results are measured (overfit), tiny samples, or lucky OTC slices. Course sellers stop at the first column; we publish the second.
The honest ceiling: 58–61%
After 12 rounds of study, the ceiling that survives out-of-sample sat consistently between 58% and 61% (Double Reversal and Exhaustion+Band, JPY pairs, the right hour windows). This matches quantitative trading literature: real short-timeframe edges are small, unstable and demand discipline — they're not magic.
Is 58% enough? Yes, if: payout is ≥ 80%, risk management caps bad days (we use a level-based daily target with a stop), and you accept variance — losing weeks happen even with real edge.
Break-even win rate for every payout
Two formulas are all you need. Break-even accuracy = 1 ÷ (1 + payout). Expected result per trade = p × payout − (1 − p), where p is your real win rate. Payout changes by asset and by hour, so the target moves with it — this is why the bot has a minimum-payout filter.
| Payout | Break-even win rate | Result per $100 staked at 55% | at 58% | at 70% (the marketing claim) |
|---|---|---|---|---|
| 70% | 58.8% | −$6.50 | −$1.40 | +$19.00 |
| 75% | 57.1% | −$3.75 | +$1.50 | +$22.50 |
| 80% | 55.6% | −$1.00 | +$4.40 | +$26.00 |
| 85% | 54.1% | +$1.75 | +$7.30 | +$29.50 |
| 87% | 53.5% | +$2.85 | +$8.46 | +$30.90 |
| 90% | 52.6% | +$4.50 | +$10.20 | +$33.00 |
| 95% | 51.3% | +$7.25 | +$13.10 | +$36.50 |
Two things jump out. First, 58% loses money at a 70% payout and makes +$7.30 per $100 at 85% — the same strategy flips sign depending on the payout you accept. Second, a true 70% would earn about $30 on every $100 traded at normal payouts. Anyone holding that would not need to sell signals.
How much a win rate is worth depends on the sample
A win rate is an estimate, not a fact. The standard way to show its uncertainty is a 95% confidence interval; we use the Wilson interval because it behaves well with small samples. Here is the same observed 60% at different sample sizes, compared with the 54.1% break-even at 85% payout:
| Observed | Trades | 95% range (Wilson) | Proves it beats 54.1%? |
|---|---|---|---|
| 60% | 20 | 38.7% – 78.1% | No |
| 60% | 50 | 46.2% – 72.4% | No |
| 60% | 150 | 52.0% – 67.5% | No |
| 60% | 500 | 55.6% – 64.2% | Yes |
| 60% | 1,000 | 56.9% – 63.0% | Yes |
| 63% (our log, above) | 109 | 53.6% – 71.5% | No |
Even 150 trades at 60% cannot rule out a real rate of 52% — below break-even. The number of trades needed before the lower end of the interval clears 54.1% is: 613 at an observed 58%, 270 at 60%, 152 at 62%, 80 at 65% and about 40 at 70%. So a real 70% would be easy to prove with a short, public, time-stamped log. Sellers almost never publish one.
You can compute the lower bound yourself (z = 1.96, p = wins ÷ n): (p + z²/2n − z·√(p(1−p)/n + z²/4n²)) ÷ (1 + z²/n).
Why small samples lie — in both directions
Take a strategy with zero edge (a coin flip, 50%). Exact binomial probabilities of it looking good:
- ≥ 60% over 20 trades: 25.2% of the time.
- ≥ 70% over 20 trades (14 wins): 5.8%.
- ≥ 60% over 50 trades: 10.1%.
- ≥ 60% over 150 trades: 0.88%.
- ≥ 60% over 500 trades: about 0.0004%.
Now test 20 random configurations on 50 trades each: the chance that at least one shows ≥ 60% is 88.2%. Test 20 configurations on 20 trades each and the chance that at least one shows ≥ 70% is 69.5%. That is how a "70% backtest" is manufactured without any cheating in the code — just by trying many settings and keeping the best.
The opposite also happens. A strategy with a genuine 58% ends at 50% or worse over 20 trades 30.6% of the time, and shows 53.3% or less over 150 trades 14.1% of the time. A good strategy can look broken for a week; that is not a reason to switch settings every day.
Overfitting: why the study uses three windows
Our study (45+ days of real M1 candles, 8 forex pairs, Deriv feed) splits the data in three. Train: the window where parameters were tuned — every number there is suspect. Test: the next window, no re-tuning. Fresh: data downloaded after the study was frozen, which no tuning ever saw. Only the fresh number is published, and results with fewer than 150 signals are discarded.
The table at the top of this page shows what happens without this: 66.9% in train became 51.2%; a 4 pm "best hour" of 65% became 40% in the next window. Every extra filter (hour, pair, candle streak) shrinks the sample and multiplies the number of combinations tried — exactly the conditions in which the coin-flip numbers above produce fake winners. Rule of thumb: if a setting was chosen after looking at a window's result, that window is no longer evidence.
How to audit a Telegram signal seller
- Ask how a "win" is counted. Many channels count a win "in G1" or "in G2" (after one or two martingale steps) as a win. A pure coin flip then shows 75.0% wins with 1 gale and 87.5% with 2 gales; a 57% entry shows 81.5% and 92.0%. "90% with gale" is compatible with zero edge. The money result per dollar staked does not change: at 50% and 85% payout, the 2-gale cycle still loses 7.5% of everything staked (see the martingale math).
- Log every signal yourself, before the result: time, pair, direction, expiry and the payout at that moment. Screenshots published afterwards prove nothing — deleted losses are invisible.
- Collect at least 150 signals, ideally 300+. Compute the Wilson lower bound and compare it with the break-even for your payout (tables above).
- Check the market. Signals on
_otcpairs: our study measured OTC at ~50% — the broker's synthetic price (why OTC is random). - Measure your own execution in demo. On M1, entering a few seconds after the signal changes the entry price. Your fills are what count, not the channel's.
- Look for the losing weeks. A 58% edge produces them; a record with none is edited.
What our study actually measured (nothing near 70%)
| Strategy | Out-of-sample accuracy | Beats 54.1% (85% payout)? |
|---|---|---|
| Double Reversal (Bollinger + RSI) | ~57–61% | Yes — best in catalog |
| Exhaustion + Band | ~57–59% | Yes — orthogonal to DR |
| RSI Reversal | ~52–55% | Borderline — auxiliary vote only |
| CCI / Z-Score / Stoch Reversal | ~52–56% | Borderline — redundant with Bollinger |
| Pure Bollinger / RSI+MA / Triple | ~51–55% | Weak on their own |
| Strong Trend / Pullback / MACD / SuperTrend / Ichimoku | ~46–49% | No — they lose on M1 |
| Any strategy on OTC pairs | ~50% | No |
The best improvement we found was not a magic filter but the expiry: the same Exhaustion+Band signal measured 2 candles ahead won 60.8% versus 57.5% at 1 candle. The 9–10 am and 6–11 pm (UTC-3) windows concentrate the edge (hours study). At 58% and 85% payout the expectation is +$7.30 per $100 staked — real, small, and easy to destroy with a low payout or a bad hour.
🤖 Get the free trading bot
Open-source bot for Quotex & IQ Option (13 strategies). Edit it with ChatGPT/Claude — no coding. Free, no spam.
FAQ
I saw a video with 20 straight wins! At 58%, the chance of 5 straight wins is ~6.6% (0.58⁵) — it happens constantly; record 50 sessions and publish the best one. 20 in a row with edit cuts is evidence of nothing.
Doesn't AI change this? AI helps test more hypotheses faster — including discarding the false ones. It doesn't repeal payout math.
So what's the path? Out-of-sample validated strategy + right hours + stop-based management + demo before real. Less glamorous than "70%" — and it's what the data supports.
⚠️ Binary options carry a high risk of losing all your capital. Educational content based on demo-account data — not investment advice. Past results do not guarantee future results. No bot or strategy guarantees profit.