The use of generative AI for stock selection

We study whether generative artificial intelligence (genAI) can automate feature discovery in US equities. Using large language models with retrieval-augmented generation and structured/programmatic prompting, we synthesize economically motivated features from analyst, options and price–volume data. These features are then used as inputs to a tabular machine-learning model to forecast short-horizon returns. Across multiple data sets, AI-generated features are consistently competitive with baselines, with Sharpe ratio improvements ranging from 14% to 91% depending on the data set and configuration. Retrieval quality is pivotal: better knowledge bases materially improve outcomes. The AI-generated signals are weakly correlated with traditional features, supporting combination. Overall, generative AI can meaningfully augment feature discovery when retrieval quality is controlled, producing interpretable signals while reducing manual engineering effort. The comparison covers alternative prompting routes and combinations of baseline and generated features, with separate analyses of implementation lags and trading costs. These findings are retrospective: the overlap between large language model training information and the backtest period remains a limitation, and the reported statistical tests do not rule out this source of bias.