Numerai: A Profile in Tokenization - Smith + Crown
Feature |

Numerai: A Profile in Tokenization

Numerai leverages the power of distributed intelligence, machine learning, blockchains, and tokenization to create an entirely new type of management model for a hedge fund--one that is informed by the intellectual contributions of thousands.

Numerai is the first hedge fund trying to crowdfund the distributed development of trading algorithms. They curate a community of thousands of anonymous data scientists continually crunching market data, providing predictions, and helping improve Numerai’s own investment calculations. Traditionally, hedge funds are highly competitive and rarely engage in collaborative activities or information sharing. Their data are proprietary, expensive, and can’t be shared, and when they hire analytical talent, they prefer a full-time, hands-on relationship. Hedge funds seem structurally opposed to crowdfunding and decentralization.

Numerai has introduced many innovations to get around these limitations. One is tokenization: Numerai has issued one million Numerai crypto-tokens over the Ethereum blockchain to their 12,000+ data scientists. The tokens are used to determine rewards in a prediction-model-building tournament that pays out Bitcoin for high performing models. Numerai’s tokenization of their tournament allows them to manage incentives and coordinate the activity of many self-interested parties. The result is the aggregation of thousands of uncorrelated models into a ‘distributed intelligence’ that can inform investment decisions. And Numerai is still iterating.

Tokenization: It’s even more complicated than it sounds

“Application Tokenization” refers to turning some feature in an application into a tradeable cryptographic token. It is one of the most powerful tools that blockchain technology enables. Tokenization has been implemented in a diverse array of industries.

  • The Loyalty rewards blockchain start up, Incent, launched a crypto-token that represents tradeable consumer rewards points.
  • The venture capital firm Blockchain Capital recently launched a sale for a tokenized share in a venture capital fund.
  • The social media platform Steem implemented a three token economy that uses token holdings as a proxy for reputation to weight up-votes for content curation.

The challenge

The problem is that tokenization sits at a complex intersection of economics, psychology, and product development. A token must have a functional role in a platform, provide sufficient incentives to existing and potential users, and be part of a sustainable economy. Tokenization is no less than a mini-economy as a backend. This is one of the reasons we at Smith + Crown enjoy researching token sales so much: they are on the bleeding edge of how systems can be redesigned around open blockchains.

Tokenization has the added benefit of helping projects raise funds: they can sell these tokens to early adopters or speculators. This is probably one of the reasons tokenization has become so popular, despite the difficulty of developing a functioning token economy from scratch. Numerai is also rare in not launching a token sale (ICO): they introduced the Numeraire into an already working platform.

How Numerai Works

Numerai is a hedge fund that facilitates a massive global community of pseudonymous data scientists to make predictions about future prices. Numerai is a complicated undertaking, given the sheer number of technological and business model innovations it is combining. It leverages the power of distributed intelligence, machine learning, blockchains, and tokenization to create an entirely new type of  management model for a hedge fund–one that is informed by the intellectual contributions of thousands.

In brief, Numerai works like this:

  • Data distribution: Data scientists are given contextless data to serve as inputs to their own predictive model. These data could be macroeconomic indicators, commodity prices, exchange rates–data that are usually proprietary and can’t be shared beyond hedge fund staff. However, removing meta-data about them allows Numerai to distribute them.  Every data scientist gets the same data, but they don’t know what the data refer to.
  • Tournament: The data is used as a basis for a tournament to create the highest performing prediction model. Some data are historic, with the predicted outcomes already known, to help data scientists hone their models, but the predictions that matter to Numerai are based on live data, where the prediction isn’t yet known.
  • Scoring: These predictions are scored when they are uploaded into a ‘meta-model’ run by Numerai. This meta-model constantly updates based on the predictions, and data scientists are awarded for how well their prediction improves the meta-model.
  • Intellectual property: Tournament participants own the rights to any models they contribute and continue to earn rewards for as long as their model helps improve the quality of the Numerai meta-model.

Numeraire token

Up until recently, Numerai compensated data scientists entirely with Bitcoin based on the contribution of the prediction to the meta-model. They have now introduced a twist: a cryptographic token called the Numeraire.

All participating data scientists get a portion of Numeraire. An initial payout of 1 million was made based on historic performance, and each week, new Numeraire are minted and distributed according to performance. Numerai retains control over the emission method and the right to introduce new features related to Numeraire. The eventual maximum supply is 21 million Numeraire, similar to Bitcoin’s maximum supply, and Numerai doesn’t plan on hitting the maximum soon.

Role in the tournament

Each week when data scientists upload predictions, they make a bet that their model will be successful by wagering a portion of their Numeraire. The wager is made in two parts: the first is the quantity of Numeraire they are willing to stake and the second is their degree of confidence, measured as the number of Numeraire they are willing to give up in order to earn one dollar USD. Prediction models are compared to the meta-model and given a binary score, indicating whether they are valuable or not. If not, their numeraire are burned. If so, they get paid out based on their staked numeraire divided by their confidence.

The weekly payout pool is a constant amount, and they are distributed prediction by prediction, with high-confidence predictions being evaluated for payout first. When the payout pool is depleted, remaining data scientists no longer receive payouts or have their staked numeraire burned, though their prediction will influence how much numeraire they receive in the weekly minting. High-confidence will lower the payout amount but increase the likelihood of payout before the pool is depleted. The reader can review the full mechanics of staking payouts in Numerai’s white paper.

Numeraire thus give holders the ability to earn more dollars in these Numerai-hosted tournaments. Numeraire are not currently tradable on exchanges, but Numerai hopes they will be in the future, giving Numeraire-holders another way to earn fiat.

The Advantages of Tokenization

In introducing the Numeraire, Numerai earns some advantages that would have been costly or impossible otherwise.

  • Securing the tournament: Tokenization is one method of preventing sybil attacks on the tournament process; that is, uploading similar predictions from fake accounts to earn more rewards for the same model. It is not unreasonable to think multiple data scientists will upload the same prediction value, but it’s not clear how to determine whether they came from the same person. Numerai’s data scientist community likely includes some of the savviest hackers on the planet and they could spoof attempts at filtering duplicate accounts. But they can’t duplicate Numeraire holdings.
  • Aligning incentives to foster a trusted exchange of information: A data scientist’s confidence in his or her prediction is quite valuable information to Numerai. However, it would be difficult to elicit an accurate number if there was nothing at stake in revealing it. Wagering actual bitcoin would be problematic because it could discourage data scientists from wagering at all, and if Numerai took the wagered Bitcoin, it would seem that the company profited from both good models and bad models, potentially souring what is otherwise a productive relationship. In contrast, Numeraire have little intrinsic value outside of the context of the Numeraire tournament. Numeraire serve as a potential source of value for participants in the tournament but aren’t worth much to Numerai, in and of themselves.
  • Leveraging smart contracts and distributed ledgers: Using an Ethereum meta-token lets Numerai use smart contracts and maintain an auditable trail of submissions and payouts. It also lowers the degree to which data scientists trust Numerai to not arbitrarily modify Numeraire–something they could do if Numeraire were just a data point in a Numerai database.
  • Liquidity and flexibility: Numeraire will eventually support a secondary market, giving data scientists another form of compensation that Numerai doesn’t actually have to fund.

Numerai: A lesson in tokenization

The case study of Numerai emphasizes how tokenization can potentially disrupt and transform traditional industries and even create markets for new kinds of tradeable commodities. First, tokenization is a powerful tool for managing incentive structures. Tokenization is an opportunity for entrepreneurs to align incentives to produce outcomes that would have been impossible otherwise. The Numeraire token incentivized mass collaboration in the development of price prediction technology, something that had previously never been achieved. This shows how tokenization can be implemented to open up new business models built around collaboration.

In massive collaborative networks or in other peer-to-peer networks, it becomes necessary to prevent bad actors from unfairly gaming the system. Tokens can be designed to discourage bad actors in several ways. The first is to ration network resources in a way that limits potential avenues for attack. Numerai did this by requiring Numeraire tokens to be used to participate in the tournament. This prevents participants from spamming the system with duplicate accounts. Tokenization can also be used to provide carrots and sticks to users, promising them rewards for acting in a way that benefits the network and retribution for attempting to undermine the network. One example of this is the Cosmos network. Validator nodes on Cosmos are required to stake tokens in order to earn block creation rewards. However, if they are found to be acting maliciously, they will have their staked tokens burned and their accounts disabled.

Tokenization also represents an opportunity for entrepreneurs to create economic ecosystems that are fully transparent and require a minimum of trust between users. By taking advantage of smart contracts and blockchain technology, entrepreneurs can create a new source of value that can enrich their users and their platform alike.