Abstract

An Agentic Framework for Financial Disclosure Intelligence TREO Talk Paper Kazim Topuz, Ph.D. The University of Tulsa kazim-topuz@utulsa.edu Tally Ferguson The University of Tulsa tally-ferguson@utulsa.edu Jessica Bedson The University of Tulsa jab1664@utulsa.edu Abstract Corporate disclosures in SEC 10-K filings are one of the most prevalent ways firms communicate forward-looking risk to investors, yet analytical methods for reviewing these disclosures have developed independently of one another. We introduce an agentic framework for financial disclosure intelligence in which lexicon-based and transformer-based models function as cooperative agents rather than opposing alternatives. The framework is applied to 10-K filings from a sample of 55 large-cap firms across all 11 GICS sectors from 2016 to 2025. From each filing, we extract the Risk Factors (Item 1A) and Management Discussion and Analysis (Item 7) sections and score paragraph-level tone using both the Loughran-McDonald dictionary and FinBERT. We also introduce a multi-layer validation design that addresses the absence of universal "ground truth" labels in financial disclosure text. We then examine systematic patterns of disagreement between the dictionary and transformer models to identify cases of contextual ambiguity, boilerplate language reuse, and legal hedging. Results show strong directional agreement between the two models in identifying positive and negative tone across firms and sectors, with clear differences in sensitivity. FinBERT produces a more extensive range of scores, while the Loughran-McDonald dictionary provides stable, interpretable readings at scale. The study contributes to information systems research by presenting a modular, extensible architecture for disclosure analytics that can incorporate additional agents, such as a large language model adjudication layer, without requiring redesign of the underlying system. Our agentic LLM agent approach should more comprehensively recognize how specific combinations of phrases disclosed in financial statements portend positive, negative, or neutral outlooks. Implementing such a system offers the potential to refine financial instrument valuation and enhance the precision of regulatory oversight and compliance monitoring. References Araci, D. (2019). FinBERT: Financial sentiment analysis with pre-trained language models. arXiv preprint arXiv:1908.10063. https://arxiv.org/abs/1908.10063 Loughran, T., & McDonald, B. (2011). When is a liability not a liability? Textual analysis, dictionaries, and 10-Ks. Journal of Finance, 66(1), 35–65.

Download

Share

COinS