Numerai Explained: Tournament vs. Signals vs. Crypto
You've heard of Numerai, the crowdsourced hedge fund where thousands of data scientists compete to predict the stock market and earn cryptocurrency rewards. But if you visit Numerai's website, you'll quickly notice something confusing: there isn't just one Numerai. There are three.
Numerai Tournament, Numerai Signals, and Numerai Crypto all operate under the same banner, all use the same NMR token for staking and rewards, and all feed into some form of meta-model that drives real trading activity. Yet each track demands a fundamentally different skill set, a different relationship with data, and a different tolerance for complexity.
This guide breaks down exactly what each track offers, what it demands from you, and how to decide where you belong.
The Numerai Tournament: The Pure ML Competition
The Numerai Tournament is where most people start. It's the original, flagship competition that put Numerai on the map.
How it works: Every week, Numerai releases a clean, regularized dataset covering roughly 5,000 global stocks with over 1,000 obfuscated features. You download this data for free, train a machine learning model, and submit predictions. Numerai then evaluates your predictions against live market performance over the following four weeks.
The key differentiator: The data is fully obfuscated. You have no idea which stock is which. You don't know what any feature represents. This is intentional. By blinding the data, Numerai creates a pure machine learning problem. You don't need to know anything about finance - no P/E ratios, no earnings reports, no macroeconomics. You just need to build models that extract predictive signal from noise.
The trade-off: Because the data is obfuscated, models built for the Tournament cannot be used outside of Numerai. You're solving a problem that exists only within Numerai's ecosystem. If you want to monetize your own proprietary data or trade independently, this isn't the track.
Staking and rewards: As with all Numerai tracks, you may optionally stake NMR (Numerai's native cryptocurrency) on your model. Positive scores earn additional NMR; negative scores result in a portion of your stake being permanently burned. This "skin in the game" mechanism ensures that only confident, high-quality predictions are staked.
Ideal for: Data scientists who want to focus purely on machine learning without any finance domain knowledge. Beginners who want to learn the Numerai ecosystem with the lowest barrier to entry. Anyone who wants a clean, pre-built dataset and doesn't want to source their own data.
Numerai Signals: Bring Your Own Edge
Numerai Signals flips the Tournament model on its head. Instead of Numerai providing the data, you bring your own.
How it works: Signals asks you to submit a "signal" - a feed of numerical data about publicly traded stocks. This could be technical indicators (MACD, RSI), alternative data (credit card transactions, satellite imagery, social media sentiment), or blended factors (Barra risk models, Fama-French factors). You source the data, you build the model, and you submit predictions on Numerai's universe of roughly 5,000 global stocks.
The key differentiator: Unlike the Tournament's obfuscated dataset, Signals deals with real tickers. You know exactly which stock you're predicting. This means you can apply domain expertise, use public or proprietary datasets, and—crucially—build models that could theoretically be used outside of Numerai. However, Numerai only receives your final predictions, not your source code or underlying data, protecting your intellectual property.
What you're evaluated on: Signals are scored against Numerai's custom targets, with target_factor_feat_neutral_20 being the primary metric for payouts. More importantly, Numerai measures originality, i.e. how much your signal differs from the thousands of others already submitted. A signal that merely replicates what everyone else is doing won't earn significant rewards, regardless of its standalone accuracy.
Staking and rewards: Staking works identically to the Tournament: stake NMR to earn or burn based on performance. However, Signals has a one-week round resolution compared to the Tournament's four weeks, meaning faster feedback and more frequent payout opportunities.
Ideal for: Quants who already have access to stock market data or alternative datasets. Data providers who want to monetize unique features. Anyone who wants to apply finance domain knowledge rather than work with obfuscated features. Advanced users comfortable sourcing, cleaning, and modeling their own data.
Numerai Crypto: The New Frontier
Launched in 2024, Numerai Crypto extends the Signals model to the cryptocurrency market. If you understand Signals, you already understand Crypto, but with a few important twists.
How it works: Like Signals, Numerai Crypto asks you to bring your own data and submit predictions. But instead of stock tickers, you're predicting token symbols. The universe covers roughly the top 300 largest tokens globally, updated daily, with BTC, ETH, and NMR always included.
What you're predicting: You can submit signals derived from technical indicators, on-chain metrics, sentiment data, or any other source. Numerai scores submissions against a 20-day target and evaluates contributions using MMC (Meta Model Contribution) and correlation metrics.
The big difference - and it's significant: Unlike the Tournament's private meta-model or Signals' hedge fund integration, the Numerai Crypto Meta Model is released publicly, free of charge. Numerai blends all staked submissions into a single forecast and gives it back to the community. You can use this meta-model to inform your own trading—a benefit that doesn't exist in the other tracks.
Staking and rewards: Staking works identically to other tracks, with NMR locked on Ethereum and subject to earn/burn based on performance. The minimum stake to be included in the weighted meta-model is 1 NMR.
Ideal for: Crypto-native quants who want to contribute to a crowdsourced prediction engine and receive a professionally blended meta-model in return. Data scientists interested in the crypto market who already have access to token data or on-chain metrics. Anyone who wants to apply Numerai's incentive model to digital assets.
Numerai Tournaments: Side-by-Side Comparison
| Tournament | Signals | Crypto | |
|---|---|---|---|
| Data Source | Provided by Numerai (free, obfuscated) | User-provided | User-provided |
| Universe | ~5,000 global stocks (obfuscated) | ~5,000 global stocks (real tickers) | ~300 tokens (real symbols) |
| Finance Knowledge Required | None | Helpful | Helpful |
| Model Portability | Cannot be used outside Numerai | Can be adapted for own trading | Can be adapted for own trading |
| Round Resolution | 4 weeks | 1 week | Varies (target dependent) |
| Key Metric | MMC, CORR | MMC, originality, target_20d | MMC, CORR (20D2L target) |
| Meta Model Access | Private | Private | Public, free |
| Primary Challenge | Pure ML on obfuscated data | Sourcing unique data and generating original signals | Sourcing unique crypto data |
| Barrier to Entry | Lowest | High | Medium-High |
Can You Do More Than One?
Yes. Many top Numerai participants compete across multiple tracks. In fact, the skills are largely transferable:
- Tournament experience teaches you how to model on obfuscated data and understand Numerai's scoring mechanics, which is valuable context for Signals and Crypto.
- Signals experience forces you to source and clean real-world financial data, a skill directly applicable to Crypto.
- Crypto experience exposes you to a different asset class and the unique challenges of token-level prediction.
If you're new to Numerai, start with the Tournament. It has the lowest barrier to entry and lets you focus on modeling without worrying about data sourcing. Once you're comfortable with the weekly cadence and staking mechanics, Signals or Crypto become logical next steps.
Which Track Should You Choose?
Ask yourself these three questions:
1. Do you have your own stock market or crypto data?
- No → Start with the Tournament. You'll get free, high-quality data and can focus entirely on modeling.
- Yes, stock data → Consider Signals. You can monetize your existing research pipeline.
- Yes, crypto data → Consider Crypto. You'll contribute to a public meta-model you can actually use.
2. Do you want to apply finance domain knowledge?
- No, I just want to do ML → The Tournament is perfect. The obfuscation means finance knowledge is irrelevant.
- Yes, I understand markets → Signals or Crypto let you apply that expertise.
3. What's your goal?
- Learn the ecosystem with minimal friction → Tournament.
- Monetize proprietary data or unique signals → Signals (stocks) or Crypto (tokens).
- Get a free, crowdsourced crypto prediction feed → Crypto (the public meta-model is a unique benefit).
Final Thought
Numerai isn't one platform. It's three, each designed for a different kind of contributor. The Tournament democratizes machine learning on financial data. Signals rewards those who bring their own edge. Crypto opens the same model to a new asset class and gives the results back to the community. Overall, the ecosystem is designed to let you grow from pure ML on obfuscated data to sophisticated signal generation on real assets.