Bucknell Professor Co-authors Study on New Analytical Roadmap to Improve Stock Investment Decisions

For years, Freeman College of Management Professor Kate Suslava, accounting, has examined the coded language executives use in corporate disclosures — the euphemisms, subtle signals and narrative choices that often reveal more than the numbers alone. Drawing on that expertise, along with her experience as a CPA and former accounting practitioner, Suslava has now helped develop a new analytical roadmap for investors navigating the fast-growing world of artificial intelligence and text-based data.

Suslava is one of three authors of "The Next Accounting Frontier: Bringing Structure and Reliability to NLP," published in the November issue of The Journal of Portfolio Management. The study shows how a structured approach to natural language processing (NLP) can help investors extract reliable, financially relevant insights from qualitative corporate communications across global markets.

She collaborated with Jacob Pozharny, co-chief investment officer at Bridgeway Capital Management, and Joshua Livnat, professor emeritus at New York University's Stern School of Business. Together, the team demonstrates how organizing corporate text using accounting's established taxonomy — the "language of business" that has guided financial reporting for centuries — gives AI models the structure they need to produce consistent and decision-useful signals.

"What we're basically saying is that we used NLP to train AI — but we're giving it structure," Suslava says. "Without structure, AI can hallucinate or return random information. Our dataset gives it parameters based on accounting disclosures, helping analysts focus on financially relevant items like earnings, inventory or accounts receivable."

Central to the study is Computext, a dataset developed by Suslava and Livnat that maps text from corporate filings, earnings calls and reports to specific financial-statement categories. Pozharny has applied the system across international markets, allowing the researchers to evaluate its performance globally as well as in the U.S.

"Computext extracts managers' statements on topics such as profit margins, R&D and capital expenditures, enriching traditional numerical data with qualitative insights," Suslava says. "Portfolio managers can use these signals to refine stock screens, design custom strategies and compare changes in tone across quarters."

Pozharny adds that textual analysis fills an increasingly important gap.

The study finds that sentiment signals created from structured NLP categories can generate positive returns, or alpha, while showing low correlation with traditional investment factors — meaning the insights are additive rather than overlapping.

Beyond performance, Suslava says the approach offers practical benefits for companies wary of "black box" AI systems that require uploading sensitive information.

"We're giving analysts a roadmap for using AI more effectively, without exposing confidential data," she says. "Because accounting's taxonomy is so well established, it provides a reliable foundation for both human and machine interpretation."

The dataset can support in-house AI training or function on its own as a sentiment "scorecard" to help analysts quickly review which accounting topics companies emphasize — and how positively or negatively they discuss them.

For Suslava, whose scholarship reveals how executives strategically frame financial narratives, the study represents a natural extension of her work.

"Accounting disclosures are dense, but they’re incredibly informative," she says. "Our goal was to translate that complexity for the machine — and for the investor."