AI Not a 'Winning Formula' in Crypto Investment...Evidence of Sustained Outperformance Lacking

Artificial intelligence (AI) agents' ability to predict cryptocurrency prices and develop investment strategies is improving, but there remains a lack of evidence demonstrating their capacity to deliver long-term, stable outperformance when factoring in actual trading costs and risks, according to new research.
The study published on arXiv on April 4, titled "Artificial Intelligence in Stock and Digital Asset Markets: Advances, Evidence of Profitability, and Limitations of Automated Investment," reviewed AI investment-related research published through August 31, 2026. The analysis included publicly traded stocks, exchange-traded funds (ETFs), centralized digital asset spot markets, perpetual futures, and on-chain markets.
Researchers acknowledged technological advancements in machine learning (ML), time-series foundation models, finance-specific large language models (LLMs), reinforcement learning (RL), and AI agents for price forecasting, news analysis, and portfolio construction. However, they emphasized that such technical progress does not necessarily translate into actual investment returns.
The core finding is the distinction between AI's "predictive ability" and its "actual profitability." Even if AI accurately forecasts market movements, transaction fees, spreads, market impact, and capital costs during order execution can erode profits. As investment scale increases or more investors adopt similar strategies, historical returns may diminish.
These challenges are amplified in digital asset markets. Centralized exchanges operate 24/7 with fragmented liquidity and constantly changing supported assets, raising concerns about custody and data reliability. Perpetual futures introduce additional costs like funding rates, leverage, margin requirements, and liquidation risks. On-chain markets face gas fees, smart contract risks, transaction ordering issues, and maximum extractable value (MEV), all of which impact profitability.
Researchers noted that while digital asset markets offer investment opportunities—such as price discrepancies between exchanges, carry from futures markets, and momentum factors—their potential returns may reflect compensation for volatility, limited capital, custody risks, or liquidation dangers. The presence of such opportunities does not prove AI can consistently monetize them after accounting for competition and costs.
Actual AI investment research has yielded mixed results: some studies showed improved historical performance but saw predictive power weaken over time or lose out when adjusting data points. Other research using AI portfolio strategies recorded during the investment period failed to demonstrate statistically significant outperformance.
The study concludes that, based on publicly available research up to March 31, there is insufficient evidence that general AI models or agents can consistently generate positive returns across different market conditions after accounting for trading costs and scale. Digital asset-related AI research also lags behind stock market studies in terms of evidence.
Researchers clarified this does not mean AI cannot produce outperformance—it simply means the current body of published research is insufficient to generalize that specific AI investment approaches will consistently profit across varying market environments.
Korean Source
This article is an English localization of a Korean-language crypto news report. Original headline: AI, 코인 투자에서 '필승 공식' 아니다..."지속적 초과수익 증거 부족"