Building an AI Stock Basket Strategy for News-Explained Momentum
Summary
The document describes a research framework that uses a language model to classify equity contracts into overlapping AI subtheme baskets, summarize current news, and map potential leader–follower relationships. It filters out ETFs and other non-company instruments, then detects stocks whose returns diverge from their basket using Z-scores. When a leader breaks away, news analysis must support the move and its direction, while historical pair statistics must validate any proposed follower relationship before the system signals a trade. The intended opportunity is delayed movement in followers, rather than chasing the leader.
The prototype combines model-generated company profiles, web-search news, price data, basket statistics, execution logic, and hard and trailing stops. The author reports a working research loop and a notify-first default, but gives no performance results establishing an edge. The document warns that company classifications and news can be noisy, past lead–lag behavior may not persist, and account-level exposure, daily loss limits, execution checks, and other safeguards remain unfinished. It presents the system as experimental and recommends continued signal review before considering live deployment.
Key ideas
- Filter the equity universe by instrument type so ETFs and indices do not contaminate company baskets.
- Represent companies with multiple AI subtheme exposures, roles, and possible relationships.
- Use basket-relative price anomalies to trigger investigation, with news serving as an explanation rather than a standalone entry signal.
- Require historical source-to-target statistics to support a proposed follower trade.
- Treat the framework as experimental until signal quality and broader risk controls are validated.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.