Photo via Fast Company
Prediction markets like Kalshi and Polymarket have generated considerable buzz among traders and social media enthusiasts, many claiming AI-powered trading systems can generate substantial returns. However, recent research from Arcada Labs challenges these optimistic narratives with concrete data from a controlled experiment conducted earlier this year.
Arcada Labs researchers tested six leading AI models by allocating $10,000 to each for trading on prediction markets over a 57-day period. The results were sobering: all models lost money during the study, with losses ranging from 16% to 30.8% on Kalshi, according to Grace Li, co-founder of Arcada Labs. The discrepancy between platform performance suggests that market selection plays a critical role—models given broader market access on Polymarket performed better than those restricted to 26 predetermined markets on Kalshi.
Despite the disappointing initial results, Li indicates that more recent testing shows promise. She points to improved performance on Polymarket, where models have greater autonomy to identify and pursue trading opportunities. This suggests that as AI systems become increasingly sophisticated at processing real-time information and making autonomous decisions, their trading capabilities may improve significantly over time, potentially establishing what she calls 'AI hedge funds' as a norm in financial markets.
The broader implications extend beyond immediate financial gains. Li emphasizes that researchers are less focused on short-term economic returns and more interested in understanding what expanded AI autonomy means for society. For Nashville-area investors and business leaders evaluating AI-driven financial technology, the research underscores the importance of cautious evaluation and realistic expectations as autonomous trading systems continue evolving.