Machine-Learning Trading Signals
for MetaTrader 5
Subscribe once and our models trade for you — copied straight into your own MT5 account via MQL5. Built on 11 years of market data, walk-forward validated on 5 years of unseen data, and forward-tested live. Transparently reported. No black boxes.
MindTheSpread Forex Majors
Machine-learning signals for the major FX pairs, executed automatically in your MetaTrader 5. Direction, entry, stop-loss and take-profit are all model-driven, gated by a calibrated confidence score so only high-conviction setups ever trade. ATR-based sizing adapts to volatility, and returns are net of real spread — the cost is baked into the model itself.
What sets it apart is the validation. The model was fully trained and frozen on data through 2023, then traded forward on 2024–2026 — a true out-of-sample record with no look-ahead and no curve-fitting. No martingale, no grid, no averaging down. Just a disciplined statistical edge, copied straight to your account.
True Forward Out-of-Sample
Metrics reflect the frozen-model forward evaluation. Past performance does not guarantee future results.
Six Production Signal Models
Each model targets a single macro driver — JPY carry, CAD oil, GBP policy — validated with annual walk-forward cross-validation across five years of unseen data. All six are 5/5 years positive. Forex Majors is live on MQL5 now; the rest are rolling out as their own signals.
JPY Crosses
3.24 Sharpe across 7 yen pairs — all 5 folds positive, 12.6 trades per week
Trades seven JPY pairs using a 3-class multiclass architecture with isotonic calibration and regime features. Driver: risk-off/carry trade dynamics — JPY moves in the same direction against all counterparties during risk events. Stop-loss at 3× ATR, take-profit at 10× ATR. Walk-forward CV (2020–2024): 3,269 trades, 51.5% win rate, +3.24 Sharpe — all 5 folds positive, MinFoldSharpe +2.20. DD circuit breaker and inverse-ATR lot sizing applied, matching live EA behaviour.
CAD Crosses
3.01 Sharpe across 7 CAD pairs — highest Sharpe in the suite, 5/5 folds positive
Trades seven CAD pairs using a 3-class multiclass architecture with isotonic calibration and regime features. Driver: WTI crude oil prices + BoC monetary policy + US economic strength. Stop-loss at 3× ATR, take-profit at 10× ATR. Walk-forward CV (2020–2024): 1,322 trades, 53.9% win rate, +3.01 Sharpe — all 5 folds positive. Highest win rate and Sharpe of any model in the suite.
GBP Crosses
2.00 Sharpe across 6 GBP pairs — 5/5 folds positive, BoE policy driver
Trades six GBP pairs using a 3-class multiclass architecture with isotonic calibration and regime features. Driver: Bank of England monetary policy + Brexit/geopolitical risk premium embedded in sterling. Stop-loss at 3× ATR, take-profit at 10× ATR. Walk-forward CV (2020–2024): 1,523 trades, 46.5% win rate, +2.00 Sharpe — all 5 folds positive. Uses atr_010 precision label tuned for GBP's high-volatility moves.
USD Crosses
1.90 Sharpe on 6 USD pairs — 5/5 folds positive, 1:4 payoff ratio
Trades six USD pairs using a 3-class multiclass architecture with isotonic calibration and regime features. Driver: Federal Reserve monetary policy + USD safe-haven demand. Stop-loss at 2.5× ATR, take-profit at 10× ATR, delivering a 1:4 payoff — break-even at just 20% win rate. Walk-forward CV (2020–2024): 2,107 trades, 42.1% win rate, +1.90 Sharpe — all 5 folds positive.
EUR Crosses
1.93 Sharpe across 7 EUR pairs — all 5 folds positive
Trades seven EUR pairs using a 3-class multiclass architecture with isotonic calibration and regime features. Driver: ECB monetary policy + European growth cycle. Stop-loss at 3× ATR, take-profit at 10× ATR. Walk-forward CV (2020–2024): 1,823 trades, 47.2% win rate, +1.93 Sharpe — all 5 folds positive. Uses atr_010 precision label tuned for EUR's more measured movements relative to high-volatility pairs.
CHF Crosses
1.85 Sharpe across 6 CHF pairs — 52.2% win rate, safe-haven signal engine
Trades six CHF pairs using a 3-class multiclass architecture with isotonic calibration and regime features. Driver: CHF safe-haven flows — the franc strengthens during risk-off episodes and weakens during risk-on periods, creating consistent cross-pair co-movement. Stop-loss at 3× ATR, take-profit at 10× ATR. Walk-forward CV (2020–2024): 857 trades, 52.2% win rate, +1.85 Sharpe — all 5 folds positive, MinFoldSharpe +0.97. DD circuit breaker and inverse-ATR lot sizing applied.
Side-by-Side Comparison
Sharpe ratios from walk-forward cross-validation on unseen data. Major Forex uses a hard train/test boundary (trained through 2023, tested 2024–2026).
| Metric | JPY Crosses | CAD Crosses | GBP Crosses | USD Crosses | EUR Crosses | CHF Crosses |
|---|---|---|---|---|---|---|
| Sharpe Ratio | 3.24 | 3.01 | 2.00 | 1.90 | 1.93 | 1.85 |
| Win Rate | 1 : 3.3 RR | 1 : 3.3 RR | 1 : 3.3 RR | 1 : 4 RR | 1 : 3.3 RR | 1 : 3.3 RR |
| Net Return (bps/trade) | 15 | 15 | 11 | 8 | 8 | 13 |
| Validation Trades | 3269 | 1322 | 1523 | 2107 | 1823 | 857 |
| Instruments | 7 | 7 | 6 | 6 | 7 | 6 |
| Overfitting Risk | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 | 0.00 |
| Positive Years (CV) | 5/5 | 5/5 | 5/5 | 5/5 | 5/5 | 5/5 |
Sharpe ratios annualized from walk-forward CV (2020–2024). Major Forex Sharpe from true OOS (train through 2023, test 2024–2026). Returns net of spread costs. Green highlights indicate category leader.
How Our Models Work
The technology is the product. We combine proven machine learning with rigorous financial engineering — no neural-network hype, just robust, interpretable models validated forward on unseen data. This is the engine behind every signal we publish.
Frozen-Model Forward Testing
The flagship model was trained and frozen on data through 2023, then traded forward on 2024–2026 it had never seen — 473 live-style trades with a hard boundary. This is forward performance, not a re-fitted backtest: the truest test of a real edge.
Calibrated Probability Engine
Raw model scores are mapped to true probabilities by an isotonic calibrator, so a 60% signal has historically won 60% of the time. Trading is gated on that calibrated confidence — the engine acts only when the odds are genuinely in its favour.
Gradient Boosting Ensemble
Our models use an ensemble of decision trees trained with gradient boosting — the same class of algorithms that wins ML competitions on tabular data. Robust, interpretable, and battle-tested.
50+ Engineered Features
Each prediction synthesizes over 50 carefully engineered features spanning momentum, volatility, market microstructure, and more — distilled from decades of quantitative finance research.
Hourly Signal Generation
Models process H1 (hourly) OHLC candles to generate directional predictions with a 48-hour horizon. Each signal includes a calibrated confidence score — only trades above the confidence threshold are executed.
Built-in Transaction Costs
Every prediction accounts for real spread costs. Our net return figures are after deducting actual bid-ask spreads — no inflated backtests, no hidden slippage.
Walk-Forward Validation
Every model is validated with 5-fold annual walk-forward cross-validation (2020–2024). Each fold trains on expanding history and tests on the next full year — exactly how the model runs in production.
Confidence-Gated Trading
Each prediction is scored by an isotonic calibrator that converts raw model output into a true probability. Only signals above the model's confidence threshold are traded — filtering out noise and low-conviction predictions.
ATR-Based Risk Management
Stop-losses and take-profits are set as multiples of ATR (Average True Range), so they automatically adapt to current market volatility. Tight in calm markets, wider in volatile ones.
Risk-First Architecture
ATR-based stop-losses and take-profits adapt to market volatility. Combined with per-symbol limits and circuit breakers, drawdown is structurally constrained.
Signal Generation Pipeline
Rigorous Validation. No Shortcuts.
In quantitative trading, trust is built through process. Every model passes a multi-stage validation pipeline designed to catch overfitting before it reaches your account.
11 Years of Training Data
Models are trained on data spanning 2015—2026, encompassing multiple market regimes: COVID crash, rate hike cycles, carry trade unwinds, and sustained low-volatility periods.
5-Fold Annual Walk-Forward CV
Performance is measured across 5 annual folds (2020–2024). Each fold trains on expanding history and tests on the next full year — exactly how the model runs in production.
True OOS Boundary (Major Forex)
The Major Forex model was fully trained and frozen on data through 2023, then evaluated on 2024–2026 data it had never seen — 473 trades on a hard out-of-sample period with no look-ahead.
Overfitting Risk Scoring
Each model receives a quantitative overfitting risk score (0.0—1.0). Models that show strong backtests but degrade on unseen data are permanently rejected from production.
Anti-Overfitting Checklist
Transparency: Our Rejected Model
We believe in full transparency — including our failures. During development, one model (Elite CAD) showed exceptional backtest results (81% win rate, Sharpe 7.0) but completely failed out-of-sample validation:
This model was permanently rejected from production. We share this because it demonstrates our validation pipeline works: impressive backtests alone don't earn a place in production. Only models that perform on unseen data make the cut.
Frequently Asked Questions
Everything you need to know about our models and methodology.
Up and Running in Minutes
No software to install. No coding required. Subscribe once and the models trade for you — copied straight into your MT5 account.
Open MetaTrader 5
Use any MT5 broker account you like. Nothing to install from us — no EA, no plugin, no API keys.
Subscribe on MQL5
Subscribe to MindTheSpread on the MQL5 Signals marketplace and link it to your MT5 terminal in a couple of clicks.
Trades Auto-Copy
MT5 mirrors every signal into your account automatically — entries, stops and take-profits — while you stay in full control.
Ready to Trade with Confidence?
Join traders who trust rigorous validation over flashy backtests. Subscribe to our Forex Majors signal on MQL5 — transparent, forward-tested performance, copied straight to your account.
Works with any MT5 broker. Cancel your subscription anytime.