In 2015, Richard Craib founded Numerai with a radical idea: what if a hedge fund crowdsourced its investment strategy from thousands of data scientists globally, compensated them with crypto tokens, and used cryptography to prevent strategy leakage? The tournament model solved several problems simultaneously: accessing global data science talent without hiring them, incentivizing quality via financial stakes, and ensuring models are unique via obfuscated data. Numerai grew to manage $100M+ AUM using its AI-tournament approach.
How the Numerai Tournament Works
Data scientists download Numerai's 'encrypted' dataset — stock market features that have been obfuscated so competitors don't know which stocks they represent. This prevents reverse-engineering the strategy or gaming the system. Scientists build machine learning models predicting the 'target' — a measure of stock returns during the tournament period. The data has over 5,400 features covering fundamental, price, and alternative data.
Staking and payouts: participants stake NMR tokens on their model's predictions. If your model ranks highly (correlated with actual market outcomes), you earn NMR proportional to your stake. If it performs poorly, you lose a portion of your stake ('burnage'). This creates a strong incentive to submit genuinely predictive models — not random or manipulative predictions. The meta-model combines thousands of submitted models; Numerai trades based on this aggregate.
Numerai Signals: a complementary tournament where data scientists submit signals for specific tickers (not obfuscated data). Scientists can use their own alternative data, news sentiment analysis, satellite imagery, etc. The performance evaluation is relative to common risk factors — has your signal alpha beyond what Numerai's existing models capture? Signals enables more creative data sourcing.
- ✓Obfuscated dataset: stock features encrypted to prevent reverse engineering
- ✓ML model submission: predict target (stock returns) weekly
- ✓NMR staking: bet on your model's performance — earn or burn
- ✓Meta-model: Numerai combines thousands of models for actual trading
- ✓Numerai Signals: custom data signal submission for specific tickers
- ✓Real money: Numerai manages actual hedge fund capital with these models
NMR Token Economics and Erasure Protocol
NMR (Numeraire) token: fixed supply of 11 million NMR. Used exclusively for staking in Numerai tournaments. Burned when models underperform — deflationary mechanism. NMR's value is derived from: demand from data scientists wanting to participate (must hold NMR to stake), Numerai's growth (more participants = more demand), and speculation on Numerai's success as a hedge fund.
Erasure Protocol: a broader protocol built on Numerai's staking mechanism concept — allowing any data seller/buyer to stake on information accuracy. The concept: any information seller can stake NMR on their information's value; buyers can purchase with burn rights if the information proves wrong. This generalized staking-for-information-accuracy mechanism extended beyond stock predictions.
Earning NMR: top performers can earn substantial NMR. Payouts are proportional to stake × correlation with meta-model outcomes. Highly correlated models with large stakes can earn thousands of NMR annually. But the tournament is competitive — thousands of PhD-level data scientists compete. For casual participants, small consistent earnings from consistently predictive models are more realistic than large windfalls.
- ✓NMR supply: 11 million fixed — deflationary via burning
- ✓NMR demand: data scientists must buy NMR to participate
- ✓Burn mechanism: underperforming models lose staked NMR permanently
- ✓Erasure Protocol: generalized information staking beyond Numerai
- ✓Top earner potential: thousands of NMR/year for highly predictive models
- ✓Competition level: PhD data scientists and quant researchers — high barrier
Frequently Asked Questions About Numerai
Multiple Paths to Crypto Income
Numerai rewards sophisticated data scientists. MineXrpOnline rewards anyone with accessible daily XRP cloud mining — no machine learning PhD required.
Earn XRP Daily