This Is The World’s First Cryptocurrency Issued By A Hedge Fund
By Laura Shin
With performance falling, investors fleeing and offices closing, hedge funds had a rough 2016.
But San Francisco-based Numerai was just getting started.
Instead of relying on the genius of a single person or small team, the firm uses encrypted data sets to crowdsource stock market prediction models built with artificial intelligence from 12,000 anonymous data scientists worldwide who compete to win bitcoin.
And if that flew over your head, the main takeaways are that this hedge fund has orders of magnitude more data scientists trying to improve its investing strategy than the biggest name hedge funds today, and what makes the firm possible is a confluence of technologies that didn’t even exist a few years ago — primarily in cryptography and artificial intelligence.
Now the one-year-old company, founded by 29-year-old South African Richard Craib (a Forbes 30 Under 30 designee) and funded by the likes of Union Square Ventures, is launching its own cryptocurrency, the Numeraire.
It is the first virtual currency issued by a hedge fund and one of the first released by a company (rather than a group of developers or a non-profit).
Craib says the Numeraire helps solve a fundamental problem in finance. “Finance is totally competitive right now. There’s no reason why you’d want to help your friend with a hedge fund if you have a hedge fund,” he says. “But if you both held the same cryptocurrency, and it went up in value together, maybe there would be more collaboration.” He believes the currency will incentivize the data scientists to make better stock market prediction models, thereby improving the hedge fund’s performance and returning more wealth to its anonymous workforce.
While Craib won’t reveal performance numbers or assets under management, USV was sufficiently impressed to lead a $6 million round this past fall. The company also received $1.5 million in seed funding, led by Howard Morgan, founder of the highly successful hedge fund Renaissance Technologies, in April.
Calling Numerai “a social network or software company whose business model happens to be a hedge fund,” USV partner Andy Weissman says, “Network effects get more valuable with every participant that uses them. What’s an example of that in a financial institution? We’ve never seen one before except for this.”
Numerai’s mix of data science, cryptography, artificial intelligence, crowdsourcing and bitcoin has given the fledgling business a sexier flair than most hedge funds. (Numerai won a 2016 Forbes Fintech 50 award.) And now Craib is risking the company’s nascent success with a real-world test of two hot but so far unproven concepts. The first is app-coins, which are cryptocurrencies that can have some attributes that remind people of shares in companies and can therefore run afoul of securities law. The other is smart contracts, which are software programs that execute contractual obligations. The biggest test of a financial smart contract so far, the DAO, which was set up like a venture fund, failed spectacularly last summer, when an unknown person or group found a loophole in the code that allowed them to pilfer $50 million.
But several industry players are looking beyond these risks and surmising that Numerai with the Numeraire could be an intermediary step to an entity that may someday exist: a fully decentralized, or leaderless, software-powered hedge fund.
So despite the risks, Craib is fairly confident about throwing app-coins and smart contracts into an already heady mix of technologies. “We’re going to be the first decentralized smart contract powering a hedge fund,” says Craib, wearing a brown hoodie and gray sweatpants in his coworking space in the trendy SoMa district. “It’s never been done before and there’s no one who’s going to be able to do it like this.”
How Numeraire Works
Data scientists entering Numerai’s tournament currently receive, weekly, an encrypted data set that is an abstract representation of stock market information that preserves its structure without revealing details. The scientists then create machine-learning algorithms to find patterns in that data, and they test their models by uploading their predictions to the website, which so far has received about 40 billion predictions.
The models get ranked, with the top 100 earning bitcoin. Numerai caps its payouts in a U.S. dollar-denominated amount, currently $15,000 a month with the top user able to earn $54,000 a year, but the company uses bitcoin as the payment mechanism to preserve anonymity and to solve the logistical nightmare of making payments to foreign countries. It’s so far paid out $200,000.
Numerai, which currently receives about 100 million predictions from its users per day, then creates a meta-model from all submissions to make its investments. Problem is, because the models are built using historical data, the data scientists sometimes “over-fit” it, so their model does better in the tournament and thus earns them more bitcoin, but those predictions then perform less well in the live stock market.
“We want to make a system where our incentives are aligned with user incentives,” says vice president of engineering Geoffrey Bradway, “so if we do really well, the users do really well, and if the users do well, we do really well.”
That’s where the Numeraire comes in. Tuesday morning, Numerai released one million Numeraire, a French word that is also an economic term meaning unit of account, to its 12,000 users who will now be able to express how confident they are that the model will work well on live data by “staking” Numeraire every time they submit.
“We wanted to give users a channel to express, ‘I did this trick, I don’t think it’s good trick to do but it really helped my performance’ vs. a user who is like, ‘There are no tricks going on to the best of my knowledge,’” says Bradway.
If the user is confident, he will stake some of his Numeraire, which causes the Numeraire smart contract to hold that sum for a certain period during which it judges the model’s performance on new data. If the predictions indeed performed well, he will earn some bitcoin, plus get his Numeraire back. Those whose predictions were less accurate have their Numeraire destroyed. Each week, everyone who participates in the tournament will receive some portion of the new Numeraire being minted that week (number TBD though Craib says the number of Numeraire will be capped at 21 million like bitcoin and the release will take place over a period, also TBD, of about four or five years). Numerai itself may retain some to incentivize users to perform other tasks.
Though Craib said it is impossible to quantify just how common overfitting is, it appears that getting users to stake Numeraire could serve as a decent proxy for how confident they are in their model. One participating data scientist, Phillip Culliton, a senior machine learning researcher at Multimodel Research who has earned about $700 on Numerai, says, “You can definitely be more confident [when submitting a model] — but it’s hard to be fully confident that you’re generalizing [to live data] well.”
Christian Catalini, a professor at MIT Sloan School of Management, says the Numeraire is a clever way to create an even better meta-model: “This information they weren’t getting is actually valuable, which is, ‘How confident you are about your model?”
Craib believes that over time, the meta-model will give more weight to the submissions with the biggest stakes.
“What we think this will do is incentivize the most important part of the hedge fund, which is to have your models work on live data,” says Craib. “Over time, the Numeraire will perfectly incentive the users to do the one thing we want them to do, which is to make the fund go up.”
A Road Full Of Risks
With the Numeraire, Numerai is entering somewhat uncharted territory in several respects: It could run into trouble with securities law. The token could accidentally create perverse incentives or outcomes. And because the Numeraire will exist on an immutable smart contract, the whole enterprise could go awry if a malicious hacker discovers some weakness in the code to exploit.
However, experts in at least the first two of these areas seem to believe the Numeraire won’t fall victim to these outcomes. Craib is trying to prevent mishaps with the smart contract by turning it on only after an initial feedback and comment period. (The code is live on Github for anyone to review.)
First: securities law. The Numeraire is one of a slew of new cyptocurrencies that have been released in the last year or so that have piqued the interest of Andreessen Horowitz, USV and other venture capital firms and entrepreneurs — and they’re generating real money for their developers. In 2016, according to cryptocurrency research firm Smith + Crown, 64 tokens were released in what are commonly called “initial coin offerings,” raising $103 million, enabling these efforts to bypass venture funding.
However, Peter Van Valkenburgh, research director at Coin Center, an advocacy group focused on the public policy issues facing cryptocurrencies, says that, unlike some other new tokens, the Numeraire would likely not be considered a security. “If they’re giving them out to people who submit trading strategies or perform some sort of useful work, it’s not a speculative investment relationship between the putative issuer and putative investor,” says Van Valkenburgh. “It’s really more like someone being paid for contributing something, more like a wage in a giant decentralized corporation. … It wouldn’t be that different from rewarding people with points.”
Olaf Carlson-Wee has evaluated many new cryptocurrencies in his role as the CEO and founder of Polychain Capital, a hedge fund that invests in blockchain-based assets. (He personally invested in Numerai but declined to say whether Polychain would invest in Numeraire.) As far as the economic system goes, he says the theory behind Numeraire seems solid: “If the Numerai fund performs well, that should lead to more assets under management, which should lead to more revenue for Numerai’s general partner which should lead to higher payouts for the data scientists, which implies a higher Numeraire valuation.” He says a good proxy for what a Numeraire should be worth (to either holders or speculators on secondary exchanges) is the value of all future payouts to data scientists on the Numerai platform.
As for whether the smart contract could blow up, Carlson-Wee says that it’s much simpler than the DAO and so can be attacked in fewer ways. Plus, he points out, “a token has been released a number of times with a smart contract.”
But given that this is the first time a cryptocurrency is being released to a non-cryptocurrency enthusiast crowd, will users take it up? Numerai data scientist Culliton hasn’t been thrilled to receive bitcoin payouts, because he considers cryptocurrencies risky. And he doesn’t believe that if you own a cryptocurrency and persuade others to buy it, then the value of your holding goes up. “If you’re looking at a situation in which participation is necessary, then you have to worry about participation falling off,” he says.
But Catalini, who has studied how various release methods affected the uptake of bitcoin, doesn’t think Numeraire will be unpopular because the initial recipients aren’t cryptocurrency fans. “What really matters is these are people who are enthusiasts about this new model of work,” he says. “They may not be cryptocurrency enthusiasts but these are data scientist enthusiasts.”
The Uber of Hedge Funds
If the experiment works, Numerai could cement its status as it one of a new breed of hedge funds, says 30-year hedge fund veteran Jeffrey Tarrant, chief executive officer and founder of asset management firm Protege Partners and a Numerai general partner. If the first two types were the “stock picker” hedge funds and the second were quantitative, using mathematical models, Numerai is part of a wave Tarrant calls autonomous learning investment strategies or ALIS firms.
The combination of unstructured data, data science platforms that allow Numerai to structure that data, and artificial intelligence creates “a next generation of investment process and management, not just related only to the hedge fund world. It redefines the investment process as well,” says Tarrant. “It’s kind of like Uber — it was the stacking together of GPS, a payment system and a rating system. When you put together three different technology stacks, it has a chance to be next generation.”
Noting that the jailing of hedge fund trader Mathew Martoma and the $1.8 fine levied on Steven Cohen’s former firm SAC Capital had caused many hedge fund managers to pull back from an edge they may have previously pursued by golfing with CEOs, Tarrant says to succeed now, hedge funds will need to turn to information from sources such as Twitter, GPS and SEC filings. “What type of people will be analyzing these exponentially growing sources of data? It’s data scientists, it’s physicists, it’s mathematicians, it’s people like Richard.”
Craib, a Cornell math major, came up with the idea for Numerai when he was working at an asset manager in South Africa and wanted to see if a friend who worked on neural networks might be able to make better predictions than he could with a particular data set. “My boss was like, ‘I’ll kill you if you do that. You can’t give way our data to someone we don’t even know,’” he recalls.
Others see Numerai as a first step toward something even more radical. “If we take this to the extreme, you could imagine a totally distributed version of this,” says Catalini. “You could have a cooperative of data scientists who get the data, work on it and share the returns.”
Fred Ehrsam, cofounder of cryptocurrency exchange Coinbase who recently left the company, says Numerai’s first achievement has been to create a marketplace for artificial intelligence, and that the addition of cryptocurrency creates decentralized economic incentives. “This is like step 2 towards a fully decentralized hedge fund powered by cryptocurrency and the token model…. It’s another step towards really creating an AI hive mind where the application happens to be trading on financial markets.”
Weissman and Carlson-Wee say the Numeraire could help Numerai become the first investment firm with a network effect. “The web and internet services has figured out how to harness these network effects to build Goliath companies like Uber and Facebook, but the financial services industry never figured out how to do that,” says Carlson-Wee. “So Richard may have created this unprecedented mechanism to get network effects around money management.”
If that happens, Numerai could someday amass enough investment to move the market with its trades. But Craib says because it encrypts data, Numerai won’t bump up against the same issues as quickly other hedge funds that had to stop taking money from investors because “they had hit that capacity where they were moving the market themselves and changing the data themselves.” Since the data in Numerai is encrypted, it could release data on different sectors of the market — currencies as opposed to equities — “and suddenly we can manage an extra $50 billion or whatever it is.”
Whether that’s the case remains to be seen. The tokens, code, white paper and explainer blog post and video have all been released. Numerai will take comments and may revise the Numeraire system before launching the smart contract.
But once that’s done, then Craib turns to the task of building the more traditional part of Numerai’s business — the hedge fund, which, after starting with $1 million of Craib’s own money as well as seed funding, hasn’t raised money in earnest.
“We also need to become a hedge fund that has lots of [limited partner] money and does a lot of normal things that big hedge funds do,” says Craib.
Make that the only normal hedge fund thing Numerai will do.