← Back
Evidence source 4701Spot Checked

Can AI Do Financial Research? LLM-Guided Hypothesis Discovery in Asset Pricing

papers.ssrn.com2026-04-13Paper
Executive summary

Researchers have shown that large language models (LLMs), a type of artificial intelligence, can autonomously generate and test asset pricing hypotheses. In a transparent system using a symbolic accounting language, the AI proposed 280 investment signals. Of these, 38 showed unique predictive power even after comparison with 209 known financial anomalies. The study highlights AI’s potential to accelerate financial discovery while noting current limits, such as reliance on initial design and the scope of the accounting language.

What it examines

This paper explores if an AI agent, using a large language model, can independently discover and test new ideas in asset pricing. The AI works within a human-designed system, using accounting formulas and automated tests, to find and evaluate financial signals that predict stock returns.

What it concludes

The study shows that AI can help find new, meaningful financial signals, some of which are not explained by existing models. This approach could speed up financial research, improve investment strategies, and inspire future work on AI-driven scientific discovery in finance and other fields.

Extracted from this source

Evidence objects

Evidence 278982% extraction confidence
Researchers showed that large language models can autonomously generate and test asset pricing hypotheses, proposing 280 investment signals, with 38 demonstrating unique predictive power after rigorous comparison against 209 known financial anomalies.

key_findings bullet 1 · key_findings · validation V0

Evidence 279082% extraction confidence
The study introduces a transparent, fully automated, and interpretable AI-driven process for scientific discovery in finance, blending human oversight with machine autonomy to accelerate research and reduce human bias.

key_findings bullet 2 · key_findings · validation V0

Evidence 279182% extraction confidence
Despite its promise, the systems effectiveness depends on the quality of its initial design and accounting language, potentially missing complex patterns, highlighting both the potential and current limitations of AI-guided financial research.

key_findings bullet 3 · key_findings · validation V0

Evidence 279282% extraction confidence
This paper introduces a novel AI research agent using LLMs to autonomously generate, test, and refine asset pricing hypotheses in empirical finance. Its unique architecture integrates symbolic accounting formulas, automated validation, and a standardized empirical pipeline, enabling transparent, rigorous comparison against 209 published anomalies. The approach promises transformative impact on quantitative finance.

key_findings bullet 4 · key_findings · validation V0

Raw abstract and provenance

We study whether an AI research agent can autonomously execute the hypothesis discovery loop in empirical asset pricing. We place a large language model

Source row: 350 · abstract type: snippet