
Man Group (AI & Quant Platform)
Institutional-grade systematic trading and AI-driven alpha generation.

The world's most sophisticated crowdsourced hedge fund driven by encrypted machine learning.

Numerai is a decentralized hedge fund that utilizes a unique meta-model architecture to aggregate predictions from thousands of independent data scientists worldwide. Built on the Ethereum blockchain, Numerai provides obfuscated financial data to participants, ensuring that proprietary trading signals remain encrypted while allowing for high-performance machine learning model development. The technical core of the platform is the 'Meta-Model,' which uses a stake-weighted ensemble of all user submissions to execute trades in global equity markets. By 2026, Numerai has further evolved its 'Signals' and 'Tournament' platforms to incorporate multi-modal data streams and advanced feature neutralization techniques that mitigate regime-change risks. Data scientists are incentivized via the NMR token, a cryptographic asset used to stake on model performance, effectively aligning the interests of the fund with the accuracy and originality of the contributors' models. The architecture is designed to solve the 'overfitting' problem inherent in financial time-series data by rewarding models that generalize well on live, unseen data rather than historical backtests.
Numerai is a decentralized hedge fund that utilizes a unique meta-model architecture to aggregate predictions from thousands of independent data scientists worldwide.
Explore all tools that specialize in ensemble modeling. This domain focus ensures Numerai delivers optimized results for this specific requirement.
A linear algebraic process that removes the component of a prediction that can be explained by a set of features, reducing exposure to known market factors.
A scoring metric that measures how much unique information a model adds to the existing ensemble of all submissions.
A cloud-infrastructure-as-code solution that automates model execution via AWS Lambda triggered by Numerai's weekly pipeline.
Proprietary method of transforming real-world stock market data into a 0-1 range where the underlying asset identities are hidden.
A mechanism where NMR tokens are permanently destroyed if a model performs poorly on live data.
A refined metric introduced to measure a model's expected contribution to the hedge fund's actual returns.
A separate platform allowing users to upload their own signals for a predefined universe of stocks using Tickers.
Create an account on the Numerai platform and generate API keys via the account settings.
Install the 'numerapi' Python library using pip to interact with the GraphQL API.
Download the latest encrypted dataset (v4.3 or v5.0 'Rain' dataset) containing features and targets.
Perform local data exploration and feature engineering using the provided 'feature_stats' file.
Train a machine learning model (XGBoost, LightGBM, or Neural Networks) on the 'train' and 'validation' data.
Use 'Feature Neutralization' scripts to reduce model exposure to known risk factors like volatility and country risk.
Generate predictions for the 'live' data segment and format them into the required CSV/Parquet structure.
Upload predictions via the API or Web Dashboard to the specific tournament round.
Stake NMR tokens on the model to qualify for payouts based on Correlation (Corr) and Meta-Model Contribution (MMC).
Deploy the model to 'Numerai Compute' using the provided Docker templates for automated weekly submissions.
All Set
Ready to go
Verified feedback from other users.
"Highly praised by data scientists for its unique data and fair reward system, though criticized for the high barrier to entry and market risk of NMR."
Post questions, share tips, and help other users.