Find Alpha Where the Market Breaks

If you’re building ML models for algorithmic trading, you already know the problem:

High-frequency market data is noisy, expensive to clean, and painfully slow to label.

The MagSeven High-Freq Anomaly Dataset skips all of that.

Instead of raw OHLCV files, you get a precision-engineered collection of real market stress events — moments where volatility and liquidity spike together, and models actually learn something meaningful.

We scan minute-level market data (1m and 5m aggregation) for the Magnificent Seven tech stocks and extract only statistically significant volatility + volume shock events, ready for immediate use.

This is not a general-purpose price dataset. It is a feature-engineered anomaly corpus built specifically for ML and quantitative research.

What You Get

200+ High-Quality Anomaly Events from:

• AAPL, AMZN, GOOGL, META, MSFT, NVDA, TSLA

Strict Dual-Factor Detection Logic:

• Volume Z-Score > 5.0
• Volatility Z-Score > 2.0
• (Triggered simultaneously)

ML-Ready Feature Engineering:

• Normalized price and volume (relative to event window start)
• Velocity and acceleration factors
• Money flow and volume momentum
• Bar intensity and market phase labels

Forward-Looking Outcome Labels Each event includes:

• future_return_30m
• max_upside_30m
• max_drawdown_30m
• Directional classification: BULLISH or BEARISH

Multi-Scale Context Windows

• 1-minute microstructure context
• 5-minute short-term trend context

Clean JSONL Format Stream directly into Python, Pandas, PyTorch, or TensorFlow dataloaders — no preprocessing required.

💡 Common Use Cases

• Anomaly Detection Train autoencoders or contrastive models on normal vs anomaly regimes.
• Directional Prediction Supervised Buy / Sell / Hold classification focused only on high-impact events.
• Market Microstructure Research Study liquidity shocks, volatility clustering, and intraday regime changes.
• Execution and Slippage Stress Testing Evaluate strategies under extreme but realistic market conditions.

🚀 Start Modeling Immediately

Get the full dataset with Forward-Looking Outcome Labels and Multi-Scale Context Windows.

Buy on Gumroad ($49)

Or preview the data structure:

View Sample on Hugging Face

📬 Contact & Support

If you have any questions about this dataset, licensing, or access to the full version, feel free to reach out:

📧 Email: [email protected]

Please note that this email is intended for dataset-related inquiries only. We aim to respond within 1–2 business days.