In a groundbreaking study published in the Journal of Economic Literature, researchers Mihail Velikov and Robert Novy-Marx have unveiled the extraordinary potential of artificial intelligence (AI) and large language models (LLMs) to revolutionize academic finance scholarship. Their work demonstrates that AI-driven tools can fully automate the generation of publication-ready finance research papers, producing outputs that are virtually indistinguishable from those authored by human scholars. This development heralds a transformative shift in how academic knowledge is created, shared, and evaluated, raising profound questions about the future of scientific discovery.
Velikov, an associate professor at Penn State’s Smeal College of Business, along with Novy-Marx, a distinguished professor at the University of Rochester, constructed a sophisticated pipeline integrating data mining with advanced LLM capabilities. Over the course of approximately 12 hours, they produced nearly 400 comprehensive finance papers. The AI was responsible for hypothesis generation, manuscript writing, and constructing detailed empirical analyses that conformed to scholarly standards congruent with human academic productions. This endeavor not only showcased the raw computational power of AI but also highlighted its ability to simulate critical thinking elements traditionally thought unique to human expertise.
Originally embarking on a project to identify new signals within corporate accounting data that could predict stock market outperformance, the researchers sifted through more than 30,000 potential market anomalies. Through rigorous validation methods, including benchmarking against 200 well-documented financial anomalies, they distilled their findings down to 95 genuinely novel predictive signals. This massive data excavation laid the foundation for AI-generated narrative scholarship. The team developed a bespoke web tool that synthesized these signals into structured report templates closely resembling typical anomaly discovery papers yet lacking interpretative depth.
The turning point came when Velikov realized that large language models excel at constructing plausible causal stories and theoretical explanations for observed data patterns—an essential aspect of scientific publishing known as hypothesis formulation. By integrating this ability with the quantified signals, they enabled the LLM to generate coherent and varied research papers expanded from their initial templates. Specifically, using Anthropic’s Claude Opus 4.1, the latest AI model at the time, the AI produced four independent research manuscripts per signal, each proposing distinct hypotheses and theoretical frameworks to explain the underlying financial phenomena.
This feat resulted in a corpus of 380 artificially produced manuscripts encompassing all standard academic sections: abstracts, introductions, methodology, empirical results, discussions, conclusions, and citations. The full codebase and papers were openly shared on GitHub, promoting transparency and reproducibility. The implications are staggering: AI can now produce dense and credible academic content at a scale and speed far exceeding human capacities, fundamentally challenging the economics of academic publishing and peer review systems.
The study’s authors sound a cautionary note concerning the broader impacts of this AI-driven proliferation. Academic journals and conferences have seen submission volumes soar in recent years, already burdening peer reviewers. AI’s capacity to mass-produce research threatens to exacerbate this overload, making traditional peer-review processes increasingly untenable. Velikov emphasizes the necessity for the scientific community to adapt and innovate peer evaluation methods to maintain research integrity and quality in an era dominated by agentic AI technologies.
One particularly troubling dimension highlighted by the study is the widespread use of hypothesizing after results are known (HARKing). The AI-generated papers derived their hypotheses post hoc — after identifying statistical anomalies. While HARKing is an established but frowned-upon practice in academia when conducted by humans, its automation and scale through AI complicate ethical and epistemological considerations about what truly constitutes novel scientific contribution. Moreover, AI’s propensity to hallucinate—producing factually incorrect or misleading information—compounds concerns about the reliability of AI-authored research outputs.
Although the researchers concentrated their efforts on quantitative finance, Velikov and Novy-Marx acknowledge that their findings have far-reaching consequences across disciplines. Any field reliant on data-mining followed by interpretive narrative generation could witness a similar influx of AI-fabricated scholarship. This convergence of computational power and natural language generation is poised to reshape the entire landscape of scientific research, potentially democratizing knowledge creation but also demanding new frameworks for quality control and intellectual attribution.
Despite these seismic changes, the authors affirm that AI will not supplant human researchers wholesale. Rather, they foresee a profound evolution in academic roles. Researchers must deepen their understanding of AI systems, leveraging these tools to enhance their investigations and insights rather than fearing obsolescence. The integration of AI could make research more efficient, nuanced, and comprehensive, provided scholars maintain critical oversight and ethical standards in employing such technologies.
This work, funded by INQUIRE Europe, represents a timely and urgent clarion call for the scientific community to confront the disruptive potentials of AI-generated scholarship. The balance between accelerating knowledge production and preserving rigorous scientific validation and originality lies at the heart of this transformation. As AI technologies continue to advance, the researchers argue that proactive adaptation in peer review, publication standards, and academic training is not merely prudent but essential.
In conclusion, the study by Velikov and Novy-Marx delineates a future where AI is not an auxiliary tool but an autonomous agent in generating academic research. This proffers both an unprecedented opportunity to scale scientific inquiry and a profound challenge to the principles and processes that have historically underpinned scholarly work. As universities, journals, and funding bodies grapple with these changes, the overarching question remains: how will science preserve its integrity and meaningful contributions in a world where machines can conceptualize and communicate research as humans do?
Subject of Research: Not applicable
Article Title: Artificial Intelligence–Powered (Finance) Scholarship
News Publication Date: 1-Mar-2026
Web References:
- Journal of Economic Literature article: http://dx.doi.org/10.1257/jel.20251821
- GitHub repository with AI-generated papers: https://github.com/velikov-mihail/AI-Powered-Scholarship
References: Journal of Economic Literature, DOI: 10.1257/jel.20251821
Keywords: Artificial intelligence, Generative AI, Machine learning, Neural net processing, Finance, Academic publishing, Peer review, Scientific publishing, Scientific community
