Numerai, founded in Dec. 2015, is building a blockchain based hedge fund that utilizes artificial intelligence and crowdsourced stock market predictions. The platform relies on a monthly machine learning and data science “tournament” where users, primarily data scientists, can submit trading algorithms about the stock market. Numerai then employs artificial intelligence algorithms to analyze the predictions, and the best are included in Numerai’s “meta-model” which combines all the underlying inputs on which it will trade the stock market.
The team believes there are inefficiencies in the stock market-related capital allocation and that creating a free and open dataset for machine learning will help solve this problem. Numerai claims that inefficiencies are in part due to assigning metrics to market data, such as stock price or dividends. Each month users are given an encrypted dataset that has standard market data labels removed. By removing the context of data the project believes users will submit more accurate predictions. For example, stock price could be labeled as “feature 1”, so users create stock prediction models objectively based on historical data as opposed to using personal biases about how stock price might affect a specific company. Numerai encrypts the dataset that tournament participants are given to prevent participants from copying the underlying trading strategy.
By giving the dataset away for free the team believes it will allow for open participation by users around the world, fostering a collaborative approach and generating better prediction models compared to typical hedge funds.
In Feb. 2017, the company announced its token, Numeraire (NMR), an ERC20 token used to participate in tournaments on the platform. The platform utilizes a “skin in the game” model where users are required to stake NMR to receive NMR for their accurate predictions. If the overall hedge fund performs well, they will distribute those returns in the form of NMR to users.
Numerai allows users to create their own prediction models that they can send to an internal smart contract along with staked tokens. After a pre-designated amount of time has passed, usually at least one month, Numerai evaluates the submitted predictions. Users that perform well will receive their staked NMR as well as an NMR reward. Poor predictions will result in the staked NMR being burned.
Each competition has a specified reward pool tied to the performance of the overall fund. As the fund’s performance increases the overall reward pool can grow as well. Rewards are given to users based on the performance of their models and the amount of NMR staked.
In order to prevent users from mimicking the fund’s trades themselves, Numerai encrypts its trading data before sharing it. The specific artificial intelligence algorithms that Numerai uses for trading are proprietary, even though the dataset is publicly distributed. Users are not submitting their actual models but instead their predictions, allowing them to keep their models should they decide not to participate. All submissions are anonymous, as Numerai does not require participants to disclose any personal information to submit stock prediction models.
The NMR token is used for staking in Numerai’s prediction tournament, holding NMR provides no other function.