Numerai (NMR — Numeraire) is a San Francisco–based AI-driven hedge fund and data science tournament platform, founded in 2015 by Richard Craib, that aggregates machine learning predictions from thousands of competing quant data scientists globally by providing encrypted, obfuscated stock market training data (so that data scientists cannot infer which stocks they are predicting), accepting their weekly prediction submissions, combining them using a proprietary ensemble/meta-model (the Numerai Meta-Model), and running the resulting signals as live equity market-neutral trading strategies — with NMR tokens serving as economic stakes that data scientists lock up when submitting predictions, earning more NMR for accurate models and losing their stake for inaccurate ones, creating a live skin-in-the-game incentive that distinguishes Numerai from pure data science competitions.
| Stat | Value |
|---|---|
| Ticker | NMR |
| Price | $8.37 |
| Market Cap | $58.97M |
| 24h Change | +0.9% |
| Circulating Supply | 7.04M NMR |
| Max Supply | 11.00M NMR |
| All-Time High | $93.15 |
| Contract (Ethereum) | 0x1776...6671 |
| Contract (Energi) | 0xd729...27a2 |
How It Works
- Encrypted data — Numerai provides participants with obfuscated financial feature datasets. Data is normalized and encrypted so that participants cannot identify which specific stocks they are predicting — removing the ability to use insider knowledge or to overfit to known tickers; participants only know “feature_X” correlates with targets.
- Model training — Data scientists build machine learning models (any architecture: neural nets, gradient boosting, linear regression) to predict the target variable (a stock ranking signal). They train on historical data and validate performance.
- Submission + staking — Participants submit weekly predictions for the live universe (current stocks). On the same submission, they stake NMR on the prediction. The stake is locked for the 4-week scoring window.
- Meta-Model — Numerai combines all submitted predictions using a stake-weighted ensemble (the Numerai Meta-Model). Higher-staked, long-running models with a track record have more influence on the final combined prediction.
- Scoring (Corr and MMC) — Predictions are scored on Correlation (Corr — how well the model ranks stocks relative to the target resolution) and Numerai Meta-Model Contribution (MMC — whether the model provides unique alpha beyond what the existing meta-model already has).
- Rewards and burns — Stakes in accurate models earn NMR bonuses. Stakes in underperforming models lose NMR (burned/penalized). The burning mechanism makes NMR deflationary over time as losing stakes are destroyed.
Tokenomics
| Parameter | Value |
|---|---|
| Ticker | NMR (Numeraire) |
| Chain | Ethereum ERC-20 |
| Contract | 0x1776e1F26f98b1A5dF9cD347953a26dd3Cb46671 |
| Max Supply | ~11,000,000 NMR |
| Distribution | Week-by-week tournament payouts and burns; Numerai retains a treasury |
| Deflationary | Yes — losing stakes are burned, reducing supply over time |
Use Cases
- Tournament staking — Data scientists stake NMR on model predictions to earn or lose based on live stock market performance (4-week lag).
- Signals product — Numerai Signals allows participants to submit predictions derived from their own proprietary external data (not limited to Numerai’s obfuscated features). Uses same NMR staking.
- Erasure Protocol — Numerai launched Erasure, a generalized protocol for any prediction marketplace with stake/burn mechanics. Erasure Bay enabled anyone to request any information for a stake (though this product was quieter in market traction than Numerai Tournament).
History
- 2015 — Richard Craib (South African quant/ML practitioner) founds Numerai in San Francisco. The core idea: encrypt hedge fund data, release it to data scientists globally, use their predictions to trade. Unconventional from the start — most quant funds are secretive, Numerai’s entire dataset is public (but encrypted).
- 2016-02 — Numerai publishes academic paper in science journal Nature Communications titled “Crowdsourcing investment research with Numerai.” The paper describes the tournament design and encrypted data approach. Significant academic and quant finance press coverage.
- 2017-06-21 — NMR token launched via an unusual airdrop: 1,000,000 NMR airdropped to 12,000 active Numerai tournament participants at the time — no ICO, no sale. Data scientists who had already been using and trusted the platform receive free NMR. This is one of the earliest major crypto airdrops to a user community.
- 2017–2018 — NMR listed on exchanges. The token begins trading. NMR price reaches over $160 in early 2018 bull market — a remarkable appreciation given the $0 initial distribution cost for recipients.
- 2019 — Erasure Protocol launched. Erasure generalizes the Numerai mechanism (prediction + stake + burn) into an open protocol others can build on. Erasure Quant (leaderboard), Erasure Predictions (personal stake-on-predictions), and Erasure Bay (request information, provider stakes an answer) launch as Erasure sub-products.
- 2020-2021 — Numerai grows to 1,500+ active staking data scientist during the DeFi bull market. NMR token achieves a new ATH of ~$160+ in 2021 driven by renewed interest. Numerai’s live stake-weighted meta-model reportedly runs billions in AUM.
- 2022 — Numerai introduces new performance metrics (MMC contribution, feature neutralization). Bear market pressure reduces NMR price significantly. Tournament continues uninterrupted.
- 2023 — Numerai v4 dataset updates. The obfuscated feature dataset is refreshed with higher-dimensional financial features. The team is small (~10 people) but the tournament has thousands of active participants submitting weekly.
- 2024 — Numerai remains one of the longest-running crypto-adjacent AI projects with a documented live trading history spanning years. The fund’s performance is not publicly disclosed, consistent with normal hedge fund practice.
Common Misconceptions
“Numerai is a crypto trading algorithm.”
Numerai trades traditional equity markets (US stocks), not cryptocurrency. The NMR token and Ethereum infrastructure are used for the staking tournament incentive mechanism, not for the underlying trading strategy.
“Data scientists win by having good predictions each week.”
Numerai’s scoring has two components. The Correlation score (how well your model ranked stocks relative to outcomes) matters, but so does the Meta-Model Contribution (MMC) — whether your predictions add unique alpha the meta-model doesn’t already capture from all the other data scientists. A data scientist whose predictions are highly correlated with everyone else’s contributes less incremental value.
Social Media Sentiment
Numerai has a deeply engaged niche following of quant data scientists, ML researchers, and crypto-financial crossover enthusiasts who appreciate the intellectual novelty of the mechanism design and the genuine “skin in the game” requirement. The project is cited frequently in machine learning and finance circles as a legitimate use case for blockchain incentive mechanisms (staking on prediction quality). Criticism focuses on the opacity of Numerai’s actual trading performance (as a hedge fund, the returns are not publicly audited), and some users note that the encrypted data prevents domain expertise from providing any edge, making all participants reliant purely on ML methodology. NMR price history reflects both the speculative boom-bust of crypto in general and the token’s genuine deflationary mechanism from burned losing stakes.
Last updated: 2026-04