Walk-Forward Backtested · No Data Leakage

ML Signals You Can Actually Verify

Get directional bias on 60+ stocks and ETFs from four machine learning models — measured on data they never trained on. Honest accuracy, not curve-fitting.

Get a Free Signal → View Accuracy Leaderboard
60+
Tickers scanned
Daily bot scans
4
ML models
~400
Backtested samples
7
Time horizons (1d–63d)

What is Walk-Forward Backtesting?

Most "ML trading signals" are trained on a full historical dataset and then tested on that same data. That's called overfitting — the model memorizes the past rather than learning to predict the future. A backtest on training data will always look great, even for a completely useless model.

Walk-forward backtesting is different. The dataset is split into multiple sequential windows. For each window, the model is trained on everything before the test period, then evaluated on the test period it never saw. Repeat 5 times across the full history. The resulting directional accuracy is a genuine out-of-sample measurement.

Step 1
Historical Split
Up to 10 years of daily price data per ticker is divided into 5 walk-forward windows, each roughly 80 trading days (~400 total test samples).
Step 2
Blind Training
For each window, the model trains on data from before the test period only. It has no access to the prices it's about to predict.
Step 3
Directional Accuracy
The model predicts UP or DOWN. Each prediction is checked against actual prices. Accuracy across all 5 windows is the displayed DA%.
How to read the DA gauge
Below 50%
40–49%
Coin flip
50%
Meaningful edge
55%
Strong edge
60–65%
50% = pure coin flip. ≥55% = statistically meaningful directional edge over ~400 samples.

Four Models, One Consensus

Every signal is backed by four independent ML models that each predict the direction separately. When they agree, conviction is higher. When they diverge, the signal is rated Weak regardless of the raw accuracy number.

📈
Linear Regression
The simplest baseline — useful for seeing whether a ticker has any detectable linear trend. Low weight in conviction scoring; mainly serves as a reality check against the more complex models.
🌲
Random Forest
An ensemble of decision trees that captures non-linear interactions between indicators. Uses OOB (out-of-bag) scoring for an additional internal accuracy check, with no extra train/test split needed.
LightGBM
Gradient-boosted trees — the primary model used by the live paper-trading bot. Handles high-dimensional indicator data efficiently and tends to have the highest walk-forward directional accuracy across most tickers.
🎯
Adaptive LGB
LightGBM with per-ticker feature selection using Mutual Information scoring. Different stocks respond to different signals — NVDA may be momentum-driven while utilities are mean-reverting. The adaptive model drops the 10 weakest features for each ticker before training.

Built for Swing Traders & Options Buyers

WalkForwardEdge is a directional bias tool, not a brokerage. Use the signal to establish your pre-trade directional lean, then open your broker's options chain to size and execute.

🤖
Live Paper-Trading Bot
Scans 60+ tickers 3× daily (9:35, 12:30, 15:55 ET) using LightGBM. Enters paper trades on Strong/Moderate signals with ≥52% DA. Stop-loss, take-profit, and horizon expiry all tracked automatically.
🏆
Accuracy Leaderboard
Every prediction is logged with its entry price. When the horizon elapses, outcomes are scored automatically. The leaderboard shows per-ticker DA% across all 4 models so you can see where the edge is strongest.
📋
22 Technical Indicators
RSI, MACD, Bollinger Band distance, 200-DMA distance, momentum across 5/10/20/63-day windows, VIX regime, earnings proximity flag, news sentiment (TextBlob), gap %, and more.
🕐
7 Horizons
1-day (next session), 3-day, end-of-week, 10-day, 21-day (standard monthly options), 42-day, and 63-day (quarterly) predictions — aligned to standard options expirations.
📰
News & Market Context
Latest headlines with sentiment analysis, put/call ratio, ATM implied volatility, and VIX reading — all shown alongside the signal so you have full context for your trade decision.
🔗
Broker Deep Links
One-click links to Robinhood, TradingView, Yahoo Finance, and MarketWatch for any ticker. Get the signal here, execute in your broker of choice.
Not investment advice. This is a directional bias filter for educational and informational purposes only. Directional accuracy is measured on historical out-of-sample data — past performance does not guarantee future results. All paper trades are simulated with a fixed $1,000 notional per trade. No real money is managed.

See Where the Edge Is Today

Free to use. No login required. Just enter a ticker and get the signal.

Open Signal Dashboard → View Leaderboard