
CrunchDAO: Community-Owned Hedge Fund & ML Competitions Explained
The world of quantitative finance has long been an exclusive club dominated by PhDs at elite institutions with access to proprietary data and billion-dollar infrastructure. CrunchDAO is changing that equation. Founded in 2021 and now backed by a recent $5 million raise in late 2025, in total funding from Galaxy Ventures, Multicoin Capital, VanEck, and Road Capital, this decentralized autonomous organization has built a global brain trust of over 10,000 machine learning engineers and 1,200+ PhDs spanning more than 100 countries, all competing to solve the same prediction problems that hedge funds pay millions to crack internally.
Unlike traditional hedge funds that operate behind closed doors, CrunchDAO has created an open marketplace where anyone with the right skills can contribute and get paid for their insights. The platform has already delivered tangible results: achieving double-digit improvements in financial forecasting for ADIA Lab (the research arm of the Abu Dhabi Investment Authority) and outperforming internal benchmarks in cancer research with the Broad Institute of MIT and Harvard. The platform pays out substantial prizes - for the ADIA challenge alone, over $100,000 was awarded to top-performing data scientists in a single challenge.
How CrunchDAO Actually Works: A Marketplace, Not a Fund
The first thing to understand is that CrunchDAO is not itself a hedge fund. It is a “global collective intelligence network” that powers hedge funds, financial institutions and other organizations with collective intelligence. CEO Jean Herelle describes it as "a marketplace connecting data scientists with financial institutions". Here's the model:
1. Institutions bring the problems: Hedge funds and trading firms submit datasets and prediction challenges to the platform.
2. The community builds the models: Hundreds of data scientists compete to extract "alpha" from these datasets, building predictive models that forecast asset returns.
3. The platform aggregates the intelligence: CrunchDAO blends the best models into ensemble forecasts, sometimes called "meta-models", that capture the wisdom of the crowd.
4. Institutions pay for the insights: Clients pay subscription fees to access these collective predictions, and the revenue flows back to the community.
This "collective intelligence protocol" structure is fundamentally different from a traditional asset manager. Crunch serves as a collective intelligence protocol that distributes machine learning challenges (called “Crunches”) across a global community, where anyone with the right skills can compete.
The platform runs four types of challenges:
- Limited in time competitions where the client buys winning models outright;
- Continuous competitions where clients pay to query the community model via API every hour, day, or week;
- Rallies that serve as test runs before launching a full continuous competition; and
- Code Competitions where participants submit their actual model code rather than just predictions.
The Equity Tournament: DataCrunch (Weekly Cross-Sectional, 3,000 US Stocks)
The flagship challenge on CrunchDAO is DataCrunch Equity Market Neutral #2, a weekly tournament focused on predicting the expected returns of the 3,000 most liquid US equities.
Here's how it works in practice:
The Problem: Every Friday at 8 PM UTC, the platform releases a new dataset covering thousands of publicly traded US companies. Participants have until the following Tuesday at 12 PM UTC to submit their predictions—ranking which stocks will outperform and which will underperform over specific forward horizons.
The Data: Each row represents a single stock at a given weekly timestamp ("moon"). The dataset includes approximately 1,150 anonymized features that describe the state of each asset, everything from technical indicators to fundamental metrics, all scrambled so participants don't know exactly which company they're analyzing. This obfuscation protects the hedge fund's proprietary edge while still allowing the community to extract statistical signals.
The Target: A single target derived from 28-day forward returns. The goal is to rank the constituents of the investment universe from most attractive to least attractive relative to their peers.
The Payout: $1,000 USDC is distributed every week.
The reward calculation is intentionally progressive: only participants who rank above the median receive any payout, and the distribution curve is steeply weighted toward the top performers. The formula uses an exponent of 20, meaning the top 1% of participants receive dramatically larger shares of the prize pool than those just above the median. With roughly $52,000 in yearly rewards available across various targets, this remains one of the platform's most competitive and highest-signal challenges.
The Metric: Predictions are scored using Spearman rank correlation against live market outcomes. This emphasizes ordinal accuracy—getting the relative ranking of stocks correct—rather than precise price predictions.
The underlying hedge fund, DataCrunch, uses this quantitative research to manage a systematic market-neutral portfolio. The long-term strategic goal is capital appreciation by capturing idiosyncratic return at low volatility, in other words, making money whether the market goes up or down by betting on relative performance differences between stocks.
The Crypto Tournament: Synth (Probabilistic Price Forecasting)
While DataCrunch focuses on traditional equities, Synth represents CrunchDAO's expansion into crypto and tokenized assets. This competition is fundamentally different in both structure and methodology.
The Problem: Instead of ranking stocks, participants must build probabilistic forecasting models for nine assets: Bitcoin (BTC), Ethereum (ETH), Solana (SOL), Tether Gold (XAUT), and tokenized stocks including Apple (AAPLX), Nvidia (NVDAX), Tesla (TSLAX), Alphabet (GOOGLX), and the S&P 500 ETF (SPYX).
The Twist: Synth doesn't ask for simple price predictions. It demands full probability density functions (PDFs) over future price changes, effectively modeling the entire distribution of possible outcomes, not just a point estimate. As the documentation explains:
The Challenge: Participants submit 100 potential price paths for the next 24 hours at 5-minute intervals, effectively modeling the full distribution of possible outcomes. All required forecasts for a given prediction round must be generated within 40 seconds, so model inference speed is a critical constraint.
The Scoring: Predictions are evaluated using the Continuous Ranked Probability Score (CRPS) , which measures how well the predicted probability distribution aligns with the actual observed outcome.
A lower CRPS indicates a more accurate forecast. The leaderboard uses a 7-day rolling average of CRPS scores across all assets and horizons, normalized relative to other participants.
The Payout: The competition features a 50,000 per month. Rewards are distributed every 7-8 days, and only the top 10 participants receive 100% of the pot. This is a winner-take-most structure that rewards elite performance.
Synth operates as a decentralized prediction subnet on Bittensor, meaning it's not just a standalone competition but part of a broader blockchain-based machine learning network. The subnet's roadmap includes expansion to the top 25 crypto tokens, which would encompass the majority of the market's activity.
How to Participate in CrunchDAO: From Zero to Submission
Getting started on CrunchDAO is remarkably straightforward, even for those without prior quant finance experience:
1. Create an account on the CrunchDAO hub
2. Join a Crunch—each challenge has structured rules, data, and a prize pool clearly defined
3. Build and submit models—beginners can start with Quickstarter notebooks that provide boilerplate code and baseline models, while advanced users can bring their own custom implementations
The platform provides a local testing environment (via `crunch.test()` in Python or `crunch test` via CLI) that simulates the submission process before you commit to the live tournament. A successful local test usually guarantees no errors during execution on the submission platform.
For the equity tournament specifically, participants submit their model code (not just predictions) and the platform runs inference on out-of-sample data. This "code competition" format ensures that the hedge fund can query the model directly for ongoing predictions rather than relying on static forecasts.
Rewards and Governance: The CrunchDAO Incentive Structure
CrunchDAO operates a dual-incentive system that combines immediate cash rewards with long-term governance participation.
USDC Rewards
The primary, tangible payout for competition performance is USDC, a regulated, fiat-collateralized stablecoin pegged 1:1 to the US dollar. DataCrunch distributes 1,000 USDC every week to top-performing models, with a steeply progressive payout curve that heavily rewards the top 1% of participants. The Falcon (Synth) crypto forecasting competition features a $30,000 USDC prize pool, with weekly payouts claimable directly on Solana mainnet. All rewards are paid on-chain and withdrawable to personal wallets.
CRNCH Governance Token
CrunchDAO has announced a native governance token called CRNCH. According to the platform's fundraising documentation and investor materials, members will earn CRNCH tokens by participating in competitions, and these tokens will grant governance rights, i.e. the ability to vote on platform rules, strategic direction, and treasury management.
The token was announced as part of Crunch Lab's 5 million strategic round. However, based on our research, CRNCH is not yet fully distributed or tradable; the token generation event and full rollout are still in development.
This governance structure is what will make CrunchDAO a genuine DAO rather than a traditional company. Decisions are executed through smart contracts, rules that cannot be changed without collective approval, ensuring transparency and preventing centralized interference
CrunchDAO vs. Numerai: Two Visions of Crowdsourced Alpha
For those familiar with Numerai, the question inevitably arises: How is CrunchDAO different?
Numerai operates as a single hedge fund that crowdsources predictions on obfuscated data, combines them into a proprietary meta-model, and trades its own capital. Data scientists stake Numeraire (NMR) tokens on their individual predictions, earning or burning tokens based on performance.
CrunchDAO, by contrast, is a platform that powers multiple hedge funds and institutions. It pays out rewards in USDC, regulated, fiat-collateralized stablecoin pegged 1:1 to the US dollar, eliminating the problems of volatility and liquidity that plague the NMR token.
So, unlike Numerai's "staking" model, where data scientists lock NMR tokens at risk of being burned, CrunchDAO's CRNCH token is designed for governance participation, not performance-based staking. The immediate, tangible rewards on CrunchDAO are paid in USDC. CRNCH represents a longer-term alignment mechanism for community governance.
The key differences between the two platforms are:
| Numerai | CrunchDAO | |
|---|---|---|
| Business Model | Single proprietary hedge fund | Marketplace serving multiple institutional clients |
| Data Model | Fully obfuscated; users never know the assets | Varies by competition; some obfuscation but more transparency in Synth |
| Staking | Stake NMR on individual model performance | Earn USDC or CRNCH through competition success; governance-focused |
| Payouts | NMR tokens based on stake-weighted performance | USDC stablecoin plus CRNCH tokens |
| Competition Scope | Weekly equity tournament only | Equity, crypto, biomedical research, causal discovery, and more |
| Community Scale | ~5,000 active data scientists | 10,000+ ML engineers, 1,200+ PhDs across 100+ countries |
CrunchDAO's marketplace approach allows it to expand into domains far beyond finance. The platform already hosts competitions in biomedical research (predicting genes driving obesity and metabolic disease), causal discovery challenges, and time series anomaly detection—all using the same crowdsourced intelligence model.
The Bigger Picture: Intelligence as a Commodity
Crunch’s mission and vision is bigger than just forecasting returns or generating alpha. It is building the infrastructure for collective intelligence, working to commoditize predictive insights in the world's largest markets. This is based on the fundamental idea that talent is everywhere, but opportunity is not.
The investment thesis draws a parallel to the Netflix Prize competition, which proved that groundbreaking technical innovations can emerge from any corner of the world when the right incentives exist. Crunch champions decentralized, open, and meritocratic systems for solving complex problems. Its philosophy is that intellectual progress and superior solutions arise not from top-down directives, but from competitive, diverse, and well-incentivized ecosystems.
For data scientists, CrunchDAO offers something rare: the ability to apply machine learning skills to real financial markets without needing capital, compliance infrastructure, or a Wall Street network. For institutions, it offers access to a global talent pool that would be impossible to assemble in-house.
Getting Started
CrunchDAO competitions are always live. Whether you're interested in the weekly equity tournament (DataCrunch), the probabilistic crypto forecasts (Synth), or cutting-edge biomedical challenges, the platform offers a low-barrier entry point into the world of quantitative research.
- Create an account at hub.crunchdao.com
- Browse active Crunches to find a competition that matches your skills
- Download a Quickstarter notebook and make your first submission within hours