ChatGPT, the large language model from OpenAI, racked up a 72% return in a simulated trading environment called 'Rallies AI Stock Market Arena.' The result, released by the platform's operators, shows how generative AI might one day reshape investment strategies — but also underscores the distance between a sandbox and the real market.
Inside the simulated arena
The Rallies AI Stock Market Arena is a virtual trading setup where AI models compete without real money at stake. Participants, including ChatGPT, are given a simulated portfolio and market data to make buy and sell decisions. The 72% return represents the model's performance over the course of the exercise. No details about the time frame, the number of trades, or the benchmark index were disclosed.
Why the result matters
That a general-purpose chat AI can generate a strong paper return catches attention, but it doesn't mean ChatGPT is ready to manage anyone's savings. Language models excel at pattern recognition and processing news sentiment, yet they lack the real-time adaptability and risk controls that professional traders rely on. The simulation removes human emotions, but it also removes real-world frictions: slippage, transaction costs, liquidity crunches, and the occasional black-swan event.
Real-world hurdles ahead
The same challenges that trip up every algorithmic trading experiment apply here. Regulators have yet to sign off on fully autonomous AI trading for retail clients. Market structure rules, data licensing, and liability questions remain unresolved. And a model that performs well in a closed test can fail when exposed to live order books and human counter-parties. The fact that the 72% return came from a simulated environment means it should be taken as a signal of potential, not as a proven strategy.
The Rallies platform has not announced whether it will release additional metrics or run a follow-up test. For now, the question hanging over the result is straightforward: Can ChatGPT's simulated success survive contact with real markets?
