A 2021 paper published by MDPI claimed that an AI model predicted Bitcoin (BTC), Ethereum (ETH), and Litecoin (LTC) prices with Mean Absolute Percentage Error (MAPE) of 0.2454%, 0.8267%, and 0.2116%, respectively. The model was trained on 80% of historical data from Jan. 2018 to Oct. 2020 and tested on the rest. Researchers admitted that performance in ever-changing markets remains uncertain.
Earlier Research: 1.84% Error in 2015
AI-driven price prediction is not new. A 2015 paper in the International Journal of Computer Applications described a neural network and fuzzy inference system that simulated stock prices for BEXIMCO Ltd. with an average error of 1.84%. The authors claimed that a stronger knowledge base and more training data could yield virtually error-free forecasts.
Google AI Lead: You Can Never Predict the Future
Laurence Moroney, AI Lead at Google, told crypto.news in October 2023: “You can never predict the future. There’s nobody out there who can predict prices correctly.” He warned that if many traders use the same AI model simultaneously, its verifiability and accuracy would collapse because the model itself moves the market. Moroney believes AI will augment human advisors, not replace them.
CFTC: AI Trading Bots Often Used for Scams
The U.S. Commodity Futures Trading Commission (CFTC) issued a Jan. 24, 2024 advisory warning that AI crypto trading bots are frequently a cover for fraud. Melanie Devoe, Director of the CFTC’s Office of Customer Education and Outreach, stated: “AI has become another avenue for bad actors to defraud unsuspecting investors.”
Existing AI Trading Tools Show Mixed Results
Despite skepticism, several AI-powered trading bots are on the market. OctoBot Cloud’s ChatGPT Trend Master lost 0.812% in the past month but gained 26.22% over six months; ChatGPT Crypto Tactics dropped 2.2% monthly yet rose 26.12% semi-annually. Other tools like PionexGPT and CryptoHopper use AI to generate trading scripts or rank strategies, but none guarantee consistent profitability.
Hedge Funds and Numerai’s Impressive Returns
A BNY Mellon report noted that hedge funds began integrating AI as early as 2019. Fully AI-driven funds like Numerai, Aidiyia Holdings, and Cerebellum Capital rely entirely on machine learning. Numerai crowdsources strategies from external data scientists; its platform’s average one-year return stands at 40.18%. Top performer Houdini achieved a 15% three-month gain, 67.8% yearly, and 516.7% all-time. Models are evaluated using Meta Model Contribution (MMC) and Benchmark Model Contribution (BMC).