AI Stock-Picking Faces Scrutiny as Study Finds LLM Trading Near Random

Large language models have increasingly been used to analyze financial statements, news and market sentiment, but their stock-picking ability still needs to be validated through out-of-sample testing. Alex Izydorczyk, a former head of data science at a hedge fund, said professional quantitative investing also involves alpha-signal research, transaction costs, position sizing and risk management, and cannot rely solely on model-generated trading recommendations.

Based on information compiled as of July 20, 2026, Izydorczyk said most LLM trading strategies performed little better than random trading on data not used in training. AI is currently better suited to helping write parts of research code and is not yet capable of replacing the full investment process. The related report did not disclose the research institution, testing period, returns or amount invested.