A model can read thousands of documents, identify relationships in data and produce convincing analysis. That does not establish lasting stock market performance. Several stages between well-processed information and a financial result can change the outcome.
A recent critical review
Linsen Zhu and Mengqing Cai submitted a preprint on AI in equity and cryptoasset markets on 4 September 2026. Their review examines publicly available research through 31 August 2026. Its preprint status and scope must be kept in mind.
The authors describe advances in prediction, text processing and workflow organisation. They find the evidence less robust when it comes to sustained net performance. This does not amount to claiming that no market participant ever makes money using AI.
The past can make a model look artificially convincing
A historical evaluation becomes misleading if it uses information unavailable at the time being evaluated. Repeatedly selecting the best results can also make a model chosen after numerous trials appear robust.
The review identifies these difficulties among the obstacles between experimental performance and economic outcomes. It offers no method guaranteed to overcome them.
Markets impose further constraints
Costs, liquidity and execution conditions can change the result of a prediction. An advantage observed in historical data does not prove it will persist across other periods or larger amounts of capital.
Our analysis: the financial AI debate benefits from separating three questions: what the model understands, what it predicts and what the complete system actually demonstrates.
Automation does not remove supervision
In its reminder of 27 June 2024, FINRA states that its rules continue to apply to member firms using generative AI. The notice concerns the US framework for its members, not universal approval of a service. The Financial Stability Board's consultation paper of 10 June 2026 also examines the benefits and risks of AI adoption.
Conclusion: evidence should match the promises
AI can improve financial analysis without guaranteeing a return. Recent research encourages examining the quality and limitations of evidence, rather than equating technical sophistication with economic certainty. This article provides general analysis, with no recommendation to buy or sell.
Comments 0
Be the first to comment on this article.