An Introduction to NumerAi

2023 can undeniably be termed the year of AI. The surge in popularity of ChatGPT brought AI into the spotlight, with many tech giants, including Bill Gates, acknowledging AI’s transformative power, likening it to the advent of the internet. U.S.’s BuzzFeed, in response to a mere statement from its CEO about incorporating AI into its core business, saw its stocks soar by 300%. Companies even slightly related to AI experienced substantial gains. Clearly, this era belongs to AI.

For those in the blockchain industry, understanding projects related to AI is essential. Enter NumerAi, an older AI project, potentially predating even ChatGPT. Founded by Richard Craib in San Francisco in 2015, by 2017, NumerAi issued its NMR tokens, distributing them to data scientists who had participated in their competitions. Its primary objective was to allow users to influence stock trades using AI, deep learning, and a crypto-economic model.

In essence, NumerAi operates as a hedge fund built on AI technology. Its core is a free dataset comprised of clean, standardized, and anonymized high-quality financial data. The platform aims to decentralize the data science sector, facilitating competition in crafting effective machine learning prediction models.

Image source: NumerAi official website

From the short introduction on the official website we can easily see that although AI technology is developing incredibly fast, it is still quite difficult to get it to predict the stock market, so how exactly does it predict the stock market?

How does NumerAi work?

Image credit: NumerAi’s official website

As you can see in the picture above.NumerAi amalgamates all predictions through a competitive process, selecting reliable forecasts by token incentives. The procedure for producing these predictive models encompasses six stages.

Step One

By integrating the predictions of thousands of data scientists, the investment forecast model is continuously refined. Why are so many scientists involved in forecasting for this model? This brings us to the role of NMR tokens. NMR tokens are awarded to scientists who contribute to the forecasting ecosystem.

Step Two

The official team transforms accumulated financial data into Machine Learning (ML) problems. After standardizing this data, they present these challenges to the global data science community. Weekly contests are held to encourage high-quality contributions. Scientists participate in these weekly contests, addressing and solving stock market-related problems, aiming for top-tier contributions.

Image Source: NumerAi Official Website

As shown above, all accumulated training data is available for download by the officials. Any scientist wishing to participate in the contest can download this data for their stock prediction deep learning training. However, it’s essential to note that this data is encrypted. Scientists can only use this data for model training and cannot directly copy the trading strategies from this data. After model training, if a scientist’s model is profitable in the actual market, they earn points. The more points they have, the more reward tokens they receive. Conversely, if the model incurs a loss, they shouldn’t expect any token rewards.

Steps Three and Four

In the competition, the best-performing solutions naturally receive the highest token rewards, while underwhelming ones are eliminated. Over time, this iterative process refines the entire investment model. Essentially, this system operates on a survival-of-the-fittest principle.

Step Five

After extensive machine training, the officials integrate high-quality crowd-sourced stock market prediction models to address the inefficiencies of the stock market. Ultimately, NumerAi incorporates the best-performing solutions submitted into a master model to execute stock market trades.

Image Source: NumerAi Official Website

From the image, it’s evident that, over the years of NumerAi’s operation, approximately 25 million USD in rewards has been distributed to scientists, offering a total of 5,567 stock trading strategies to users.

Step Six

Ultimately, users can fork the official GitHub and utilize their stock prediction models using Python or R scripts. These operate similarly to traditional quantitative funds, except these strategies are derived from data trained by AI.

Throughout this ecosystem, while blockchain’s primary role is token-based incentives, the actual prediction models still rely on the collective efforts of numerous scientists. The officials primarily reward top-performing prediction models. The platform would be incomplete without these scientists.

An intriguing aspect of the platform is “Staking.” Every user can stake to signify their confidence in their predictions. If forecasting were free of cost, the platform would be inundated with superfluous data since users could submit any prediction, and some would inevitably be correct. By introducing a cost to malicious behavior and using staking to vouch for prediction accuracy, the platform can operate more efficiently.

NumerAi Token and Funding Status

As of July, NumerAi announced that it had raised $100 million in new funds for its hedge fund. The current total fund size has reached $320 million. To gauge the reliability of its forecasts, we can look at the size of the fund — clearly, this is not an insignificant amount.

Regarding token distribution, the total supply of NMR is 11 million tokens, issued on February 21, 2017. Currently, 57% of these tokens are in circulation, and at the time of writing, its price is $14, ranking 261st on CoinMarketCap. As an ancient AI project, it’s still far from its peak in the 2017 bull market, indicating it has yet to garner significant attention.

The tokenomics of NMR are relatively simple. The primary role of NMR is to incentivize AI scientists within the ecosystem. Users can stake their forecasting models with NMR tokens. After training with official data, based on actual performance, top strategies can earn NMR token rewards. This tokenomic approach allows the platform to allocate funds to strategies based on their risk profile.

Performance of NumerAi Fund

[Image Source: NumerAi official website]

As depicted in the above image, based on official data, the fund’s performance surpasses the average stock market performance. An official return rate of 29% is considered a very high metric in quant models. New users also enjoy some perks. By registering an official account and following instructions to build and upload a machine learning model, they can earn a reward of 0.1 NMR tokens, worth approximately $1.4.

Conclusion

Readers might have noticed that this project, due to its early inception, leans more towards using blockchain technology as an incentive mechanism. It encourages scientists within the ecosystem to compete in stock prediction, selecting the best models through natural selection, rather than focusing purely on AI technology.

However, this approach is innovative. We’re all aware of the challenges in predicting stock market trends. While its size may be small compared to traditional funds, as the first project combining blockchain, AI, and stock prediction, NumerAi has performed commendably.

While it may seem outdated compared to the myriad of new AI projects in 2023, the project remains promising. We all understand the importance of training data in the AI era. Whoever compiles the most extensive and highest-quality training data might produce the best AI product. NumerAi’s token incentives have accumulated a vast amount of stock prediction data from numerous scientists. In the realm of stock predictions, NumerAi can surely be considered a pioneer. While its reliability requires time to validate, we must recognize its contributions to the entire field of stock predictions.