AI-Powered (Finance) Scholarship

Summary

A demonstration that LLMs can be used to automatically generate hundreds of complete academic finance papers on stock-return predictability. The authors mine over 30,000 candidate predictor signals from accounting data, run them through the Novy-Marx and Velikov (2024) "Assaying Anomalies" protocol to identify 96 that pass standardised criteria, generate template performance reports, and then use state-of-the-art LLMs to produce three distinct complete paper versions per signal — each with a different creative name, custom introduction, theoretical justification, and citation set.

Contribution

Two contributions: a working LLM pipeline that turns a passing-signal report into a fully written finance paper at scale, and a cautionary demonstration of "industrialised HARKing" (Hypothesizing After Results are Known) — including LLM-generated citations to "existing (and, on occasion, imagined) literature." Companion artefacts (code, generated papers, an AI-generated podcast) are released on GitHub.

Method

Empirical pipeline demonstration: data mining + the Assaying Anomalies protocol + LLM generation of multiple paper versions per surviving signal.

Relevance to RISE

A canonical "what could go wrong" reference for RISE: it shows that end-to-end generation of plausible, well-formatted research papers is already feasible in a structured domain, and that hallucinated citations are a built-in failure mode. Directly relevant to RISE catalog projects building end-to-end pipelines such as data-to-paper and sakana-ai-scientist, and a test case for the autoresearchclaw evaluation agenda.

Critique / open questions

The pipeline is by construction post-hoc rationalisation of mined signals; the paper itself frames this as a feature for the cautionary argument rather than as a research-discovery system. No human-quality or peer-review evaluation of the generated papers is reported in the visible excerpt.

Key quotes

"This experiment illustrates AI's potential for enhancing financial research efficiency, but also serves as a cautionary tale, illustrating how it can be abused to industrialize HARKing (Hypothesizing After Results are Known)."

"The different versions include creative names for the signals, contain custom introductions providing different theoretical justifications for the observed predictability patterns, and incorporate citations to existing (and, on occasion, imagined) literature supporting their respective claims."