The Moscow Exchange summarized the results of the virtual trading MOEX AI Hackathon, where 50 teams tested AI agents with a capital of 1 million virtual rubles each. The main conclusion turned out to be unexpectedly human: trading more does not mean earning more.
The most unprofitable agents were those with maximum activity. One of them conducted almost 7,700 transactions with a turnover of 1.8 billion virtual rubles and lost a significant part of its capital in commissions. The leaders of the rating acted more calmly: their turnover ranged from 13 to 40 million virtual rubles.
The best algorithms did not just monitor charts, but analyzed more than 30 news sources and RSS feeds. When the news background contradicted technical signals, the agent could refuse a deal or change its position. Risk management also proved to be important – limits prevented the AI from spending capital too aggressively.
For example, the worst-ranked AI agent had a module with a well-defined risk contour, but it played a decorative role.
Testing took place from May 13 to June 10. Now, the Moscow Exchange plans to create a machine-readable data layer, a separate AI protocol, and an agent verification system to safely allow autonomous systems into real trading in the future.
Комментарии