Key takeaways
- CrunchDAO uses crowdsourced machine learning predictions from 2,000 data scientists across 50 countries to generate trading signals.
- The CRUNCH token incentivizes community participation by distributing fees earned from selling signals to institutional investors.
- Crowdsourced ensemble predictions eliminate outliers and noise better than individual models, achieving 3X industry average performance.
- Data scientists from non-finance fields can contribute their machine learning skills on weekends without leaving primary employment.
- On-chain voting and smart contracts enable transparent decision-making and automated payment distribution to thousands of contributors simultaneously.
From crunching numbers to a global DAO
Ashton Addison: What is your background in blockchain, and what led you to start CrunchDAO?▶ 00:47
Arnaud Castillo: I was in the tech industry, so friends were always talking to me about Bitcoin and Ethereum, but I didn't use them much at first. What drew me to CrunchDAO was the idea of crunching numbers. In finance today you have a huge amount of data, and a community can crunch a lot of it. CrunchDAO is a team of research data scientists leveraging collective intelligence and web3 to produce next-generation financial insight. Every weekend we produce a data set, and around 2,000 data scientists from 50 countries can take it and make predictions. The DAO then ensembles those predictions into a market signal. We also have a token, the Crunch. Whenever we sell the signal to an asset manager or institutional investor, they pay fees that we convert into Crunch and give to the community. The better your predictions, the more tokens you receive as a share of the DAO's performance.
Ashton: How long have you been building this, and where does it stand today?▶ 03:14
Arnaud: We started about a year and a half ago with a simple hackathon with a French IT school. Since then we've run weekly tournaments and grown the community to 2,000 data scientists. We're already trading the signal, and performance is very good. Compared with the average in the industry, we're doing three times the performance.
When hundreds of data scientists bring you predictions and you ensemble them, you eliminate outliers and noise
Why the crowd beats the PhD hedge fund
Ashton: Why do we need alternative hedge fund models, and how does yours improve on what exists?▶ 03:40
Arnaud: Using machine learning in finance isn't new. Many quantitative funds do it. The added value is the crowd. When hundreds of data scientists bring you predictions and you ensemble them, you eliminate outliers and noise, and the predictions are much more accurate. The second part is that traditional hedge funds recruit the best PhDs from the best schools in the world. These are people who could have found a Nobel Prize or a breakthrough in physics or chemistry, and the very high salaries tempt many of them into finance. In our system, we prefer that data scientists work in their field. One person on our team works on brain imagery and machine learning during the week, and on the weekend applies what he learned to our data set. Once he has set up his model, he can put it in a Docker instance on AWS and it runs every weekend. It grabs the data, creates the predictions and sends them back to the DAO. So you get work from thousands of data scientists in different industries who don't need to know anything about finance.
Ashton: Do you see other hedge funds taking an interest in what CrunchDAO is doing?▶ 09:25
Arnaud: We have a lot of token holders from the hedge fund industry who invested in Crunch, so they're very interested in what we do. I think they believe there is value here, in both machine learning and the wisdom of the crowds. We also try to share our research as much as possible, because the more we share, the more people bring their skills. That's not only machine learning but also portfolio optimization, which is a big part of asset management. We plan to share that research too. It's like open science: anybody can bring value, and contributors will also earn tokens.
How web3 actually runs the DAO
Ashton: Is the web3 technology you use actually functional, rather than just a DAO-style structure?▶ 07:03
Arnaud: It's totally functional. Decisions are taken by votes, and the votes are on chain. Anyone in the community is a token holder, and data scientists receive tokens as they work, so they have real stake in the game. The Crunch token is also our payment mechanism. At the end of a tournament we need to pay fees to, say, 1,000 data scientists, and we have a smart contract for that. It's like an airdrop, but to 1,000 data scientists in one go. Transparency matters too, because you can see what decisions are proposed, and anybody can propose something. Right now we're working on a staking mechanism. A few people from the community proposed different ways to manage staking, and the vote happens this weekend.
Ashton: Does geography matter for a DAO, and where is CrunchDAO based?▶ 10:32
Arnaud: In traditional business it can take years before you go international. In our case, within six months we had data scientists from 50 countries. Before we moved to crypto, paying people in places like India or Africa was quite challenging. It was a mess. With crypto it's much easier. We started the business out of France, and we're moving the DAO to Switzerland as a non-profit organization. We believe it's a great, very transparent place, and the Ethereum Foundation is there too. We'll set up the Swiss non-profit as soon as early 2023, and its members will be the token holders.
It's like open science: anybody can bring value, and contributors will also earn tokens.
Freelance work, regulation and getting involved
Ashton: Some people still find crypto and DAOs too complex or are wary of digital assets. How do you see that barrier?▶ 12:37
Arnaud: We see the DAO as an organization where people work as freelancers from anywhere in the world. It's different from being an employee at a bank. Everybody is remote and free. They earn in token, but it can be exchanged for USDC or whatever, so for us it's a payment mechanism, and it's totally cross-border. It's much easier to pay somebody in Nigeria this way than to send them a wire. A colleague of mine said he doesn't see why Uber isn't a DAO. The drivers should be able to finance through a DAO the developments they want on the platform and earn the benefits, and they wouldn't need any shareholders. There are many peer-to-peer businesses where a DAO could be a much better system than the existing mechanism. Sharing is the best way to share value with all the stakeholders, not just the shareholders, whether you bring assets, energy, skills, knowledge or time.
Ashton: Do you expect regulation of DAOs, as with exchanges and stablecoins, and could it hinder innovation?▶ 15:21
Arnaud: It depends on what happens next with regulation in this industry. Today we don't do KYC. If somebody wants to participate, they can come, create predictions and be rewarded for that. We've excluded some countries because of regulation, so unfortunately some countries can't access the platform. If regulation changes so that anybody who receives crypto or a token has to go through KYC, we'll have to follow the rules for sure.
Ashton: How can people learn more and get involved?▶ 17:11
Arnaud: Go to the website and the tournament section, where you can see what you have to do to participate. You can download the data set, try a first sample model and see how easy it is. If you have data science or machine learning skills, you can make your first predictions and push them to the tournament every weekend. There's a leaderboard, like on Kaggle, and you start to earn Crunch based on your performance and your contribution.
This interview has been edited for length and clarity. Recorded December 2022.
