Zero-Shot LLMs Fail to Beat Simple Baselines on Stock Prediction

A structured zero-shot NLI pipeline with temporal aggregation and multi-layer explainability consistently fails to outperform simple baselines for short-term stock movement prediction across multiple models and time horizons. The result holds despite explicit recency weighting and event-impact modeling.