Numerai is a hedge-fund that is crowdsourcing the distributed development of trading algorithms via weekly competitions that incentivize data scientists to create models that perform well on new data.
Numerai is a hedge fund crowdsourcing the distributed development of trading algorithms via weekly competitions with prizes that incentivize data scientists to create models that perform well on new data. By curating a community of thousands of anonymous data scientists who continually crunch market data, Numerai seeks to improve its own investment calculations. Numerai leverages the power of distributed intelligence, machine learning, blockchains, and tokenization to create a new type of management model for a hedge fund.
The project was founded by Richard Craib. The project has been operating as early as 2015, originally using bitcoin to pay data scientists for contributions, though released its own token in 2017, the NMR, which could be better integrated into the project’s economy. NMR was distributed via airdrops and as payment to financial data modeling tournament participants (bitcoin was used as payment prior to launch.) No token sale was conducted. In 2018, Numerai announced a reduction in total NMR supply, from 21 million to 11 million, coupled with plans to decentralize the currency’s issuance. These changes were paired with a re-design of the NMR token, intended to fix perceived shortcomings with the currency and make it more suitable for general use within ‘Erasure Bay’, an information marketplace developed by the Numerai team that uses the hedge fund’s Erasure smart contracts.
Traditionally, hedge funds are highly competitive and rarely engage in collaborative activities or information sharing. Their data are proprietary, expensive, and seldom shared. Likewise, hedge funds tend to prefer a full-time, hands-on relationship with hired analytical talent. For these reasons, hedge funds seem structurally opposed to crowdsourcing and decentralization.
Numerai has introduced innovations to build outside of these limitations. One is tokenization: Numerai’s tokenization of their prediction-model-building tournament for data scientists compensates participants for developing high performing models, allows Numerai 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 decision. The key elements of such tournaments are as follows:
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 are not shared beyond hedge fund staff. However, removing metadata about them allows Numerai to distribute them and eliminates any incentive for their hoarding. Every data scientist accesses the same data, but they do not know what the data refer to.
Tournament. Different data is used as a basis for a tournament to create the highest performing prediction model. While some provided data are historic, with the predicted outcomes already known so that data scientists may better hone their models, the predictions that inform the Numerai hedge fund are based on live data, where the outcome is not yet known.
Scoring. Predictions are scored for concordance when they are uploaded into a ‘meta-model’ run by Numerai, and for their AUC-ROC curve once the round resolves. This meta-model constantly updates based on the predictions, and data scientists are rewarded for how well their prediction improves the meta-model. Earnings from the weekly prize pool—6000 USD and 2000 NMR— are paid out based on live performance, with the outcome of predictions being settled after one month, and a leaderboard displaying top performers.
Intellectual property. Tournament participants own the rights to any models they submit and continue to earn rewards for as long as their model helps improve the quality of the Numerai meta-model. At time of writing, a total $12.6 million combined USD/NMR has been paid out to tournament participants, with $1.6 million staked and $8.6 thousand burned.
Numerai’s token, NMR, is issued using the ERC-20 standard and was originally designed as a cryptographic token purely for use in weekly tournaments— the project claims the token is not a cryptocurrency as it cannot be exchanged for goods and services, has no use outside tournaments, and is not intended to be traded on any known exchanges. With the development of Erasure, a decentralized predictions data marketplace, Numerai redesigned NMR, positioning it as both a cryptocurrency and the native token of the Erasure Protocol. As part of the re-design, Numerai cut the supply from 21 million to 11 million, relinquished their ability to mint tokens, added governance rights to the token, and expanded the token’s uses beyond Numerai tournament staking— the token is used to subscribe to prediction feeds and stake to feeds for quality attesting purposes.