We tested more than 20 binary options strategies on our Python bot. The overwhelming majority died in the test that actually matters: the fresh out-of-sample window — data the calibration process never saw. Exactly two survived. And with an inconvenient condition we did not expect: they only worked on the JPY pairs. In this article you get the complete rules of both, the real numbers (including those of the ones that failed), and the final configuration we are taking into the demo-account forward-test. No promise of riches — only what the data showed.
Methodology (read this before the numbers)
Base: 45 days of real M1 candles from Deriv, collected via the official API. We split into train/test and reserved a fresh out-of-sample window — a period that no calibration step ever touched. Break-even reference: with an 87% payout, the mathematical break-even sits at a 53.5% win rate. Below that, the strategy loses money even while “winning more than half”. And remember: win rate is not profit — payout, session hours and execution change everything. Everything here was measured on a DEMO account.
Strategy 1 — Double Reversal (Bollinger + RSI)
This is the classic mean-reversion strategy, with double confirmation. The logic: when price stretches too far from its recent average and the oscillator confirms the excess, the probability of the next candle moving back towards the mean increases a little. A little — not a lot. That is what this game is about.
Complete rules:
1. Bollinger Bands(20, 2) on the M1 chart.
2. RSI(14) on the same chart.
3. CALL: the candle closes below the lower band AND the RSI is at an oversold extreme.
4. PUT: the candle closes above the upper band AND the RSI is at an overbought extreme.
5. Entry at the open of the next candle, trading the return to the mean.
The two filters must fire together. A broken Bollinger band alone, or an extreme RSI alone, is not a signal. Now, the numbers in the fresh window:
| Pair | Win rate (fresh) | n | Verdict (break-even 53.5%) |
|---|---|---|---|
| USD/JPY | 56.9% | 334 | Passed |
| GBP/JPY | 54.6% | 273 | Passed (thin margin) |
| EUR/USD | 51.7% | — | Lost the edge in the fresh window |
| GBP/USD | 50.3% | — | Lost the edge in the fresh window |
| AUD/USD | 52.2% | — | Lost the edge in the fresh window |
| EUR/JPY | 46.4% | — | Failed |
The detail almost nobody publishes: in the train/test windows, the USD majors also appeared to work. It was in the fresh window that EUR/USD, GBP/USD and AUD/USD collapsed into the 50-52% zone. Our reading: the fresh window landed in a directional regime after a geopolitical shock, and mean reversion suffers precisely when the market decides to walk in one direction only. If we had published the training numbers, we would have been selling you an edge that no longer existed. And EUR/JPY proves that “JPY pair” is not an automatic pass: 46.4% is worse than flipping a coin.
Strategy 2 — Exhaustion + Band (our own creation)
This one was born inside the project, looking at the anatomy of the candle instead of classic indicators. The idea: an abnormally giant, directional candle that closes outside the Bollinger Band tends to represent exhaustion of the move — everyone who had to get in is already in. We trade the fade: against the direction of the giant candle.
Complete rules:
1. The candle is GIANT: total range ≥ 2× the ATR(14) calculated on the previous candles.
2. The candle is directional: body ≥ 60% of the range (not an indecisive double-wick).
3. The candle closes OUTSIDE the Bollinger Band (above the upper or below the lower).
4. All three conditions together → entry against the candle’s direction at the open of the next one (a giant bullish candle closing above the band = PUT; the mirror = CALL).
| Pair | Win rate (fresh) | n |
|---|---|---|
| USD/JPY | 56.6% | 488 |
| GBP/JPY | 58.5% | 340 |
Why we run both together: they are orthogonal
The obvious question: “isn’t it redundant to run two reversal strategies?” Not in this case. The Double Reversal measures distance from the mean (Bollinger + RSI); the Exhaustion + Band measures the anatomy of the candle (size vs. ATR and body proportion). In practice, they fire on different candles — they are orthogonal. The result of the union is the best of both worlds: practically double the signals, while keeping the win rate: 58.4% in the fresh window (n=498).
Another finding that changed the final configuration: the 2-candle expiry beat the 1-candle expiry in our tests — 60.8% and 58.7% versus 57.5% and 57.4%. Reversal on M1 seems to need a bit more time to complete than the 60 seconds of the next candle.
The ones that failed (the part nobody publishes)
A result only means something if you know what was discarded along the way. These are the ones we tested and rejected, with the numbers:
| Strategy | Win rate | Why it fell |
|---|---|---|
| Pin bar | 47-49% | On M1, a wick is continuation, not reversal — the opposite of the price action manual |
| Keltner Channel | 52.7% | Better than a coin flip, below the 53.5% break-even |
| Engulfing | 47.5% | Worse than random on M1 |
| RSI divergence | 54.4% | Clears break-even, but is redundant with the Double Reversal — adds no new signals |
| RSI-2 (Connors) / streaks | ~51-54% | On the break-even margin; no room for the real world |
The pin bar case deserves a highlight: it is probably the most-taught setup on the internet, and on M1 it won less than a coin toss. If you trade pin bars on 1-minute candles because you read it in an e-book, our data suggests you are on the wrong side of the statistics. We talk more about this realistic win-rate ceiling in “Is a 70% win rate a myth? The real ceiling of strategies”.
Final bot configuration
Asset: USD/JPY (GBP/JPY running in shadow mode, not trading)
Signal: union of the 2 strategies (Double Reversal ∪ Exhaustion + Band)
Candle: 60 seconds (M1)
Expiry: 120 seconds (2 candles)
Hours: 09:00-10:59 and 18:00-23:59 BRT, Monday to Thursday
Minimum payout: 80% (below that, the bot does not enter)
Honest expectation: 57-60% win rate — to be confirmed in forward-testing
Note the wording: to be confirmed. Backtesting and out-of-sample reduce the chance of self-deception, but do not eliminate it. The real verdict is the demo-account forward-test, which is already running — the first numbers are in Our bot’s demo account results: 3 brokers, real numbers.
Want to test with real data? This entire study used M1 candles from Deriv, the only one of the three brokers with an official, documented API.
Open a Deriv demo account →Affiliate link. Always test on demo first.
Read next
→ Our bot’s demo account results: 3 brokers, real numbers (July 2026)
→ Is a 70% win rate a myth? The real ceiling of strategies
→ Best hours and days for forex M1: real data
→ Python + IQ Option: the unofficial API explained
Dan Machado
Developer of the IA Trader Pro project. Builds and documents in public a binary options bot in Python, tested on demo accounts at three brokers (Deriv, IQ Option and Quotex) — publishing the numbers that work and, above all, the ones that don’t.
Disclaimer: binary options are extremely high-risk products and most retail traders lose money. All numbers in this article come from backtests and DEMO account tests — past and simulated performance does not guarantee future results. This content is strictly educational and does not constitute investment advice. Never trade with money you cannot afford to lose.