Can News Predict the Market? Limits of Zero-Shot Financial NLP and the Role of Explainable AI

Zero-shot AI models cannot reliably predict short-term stock movements from news, despite advances in large language models. The research found that these models "consistently fail to outperform simple baselines" across multiple prediction horizons, indicating fundamental limitations in translating news sentiment into actionable trading signals.

AI systems perform particularly poorly at predicting negative market movements, suggesting that the relationship between news and stock prices is more complex than current models can capture. This asymmetric failure is important for investors relying on these tools, as it means predictions may be systematically biased.

Explainability can add value even when prediction accuracy is limited. The paper's key finding is that transparency features—which show why a model made a prediction through token-level, article-level, and natural language explanations—"reliably distinguish between trustworthy and unreliable predictions," offering practical decision-support value independent of raw accuracy.

Financial AI systems should prioritize transparency and uncertainty awareness over pure accuracy claims. Rather than deploying models that claim high prediction rates, the research advocates for systems that help users understand model confidence and reasoning, enabling better human judgment in high-stakes financial decisions.

Domain-specific training may be necessary for financial AI applications. The consistent failure of zero-shot approaches (models without specialized financial training) suggests that responsible deployment in finance requires either domain-adapted models or explicit acknowledgment of the technology's fundamental limitations.