Nov 2026· Proceedings of the 7th ACM International Conference on AI in Finance· 0 citations· 39 references
TL;DR
These findings establish multi-source fine-tuning as a practical framework for coordinating structured, semantic, and market supervision in financial language models whose outputs are empirically evaluated against subsequent market outcomes.
Abstract
We introduce FinMFT, a multi-source fine-tuning framework for transforming an open-weight language model into a structured, semantically informed, and market-aligned financial-news model. FinMFT addresses two central challenges in financial language modeling: (i) transforming general linguistic fluency into structured financial judgments supported by the underlying text and (ii) aligning those judgments with economically meaningful outcomes. It progressively aligns a shared representation with three distinct sources of supervision: vendor annotations establish financial-event structure; teacher-generated semantic assessments encode mechanisms, evidence, directional impact, and uncertainty; and realized market outcomes align the representation and a compact outcome head with multi-horizon, industry-neutral return ranks and direction classes. We assess predictive validity out of sample using rolling model fits and a strict pre-close evaluation design. At each of the one-, five-, and twenty-day horizons, FinMFT achieves higher rank IC than every external benchmark on the common evaluation population, including commercial vendor sentiment, dictionary-based methods, FinBERT, a return-supervised TF–IDF model, and prompted variants of the same backbone. These findings establish multi-source fine-tuning as a practical framework for coordinating structured, semantic, and market supervision in financial language models whose outputs are empirically evaluated against subsequent market outcomes.
This work proposes an end-to-end pipeline for evaluating and improving LLMs on European financial regulatory question answering and detects a blind spot that matters wherever answers must be verifiable.
Aulia Kharis Rakhmasari, Fang-Rong Yu, Alexander Miserlis Hoyle et al.· arXiv.org· 0 citations
The use of large language models (LLMs) in finance has expanded rapidly, yet general-purpose systems face recurring problems in high-risk and regulated applications: Unsupported factual generation, weak reasoning across text, tables, charts, and numerical data, and limited traceability of model-assisted decisions. This...
Fu-Ming Niu, Feng-Shuo Liu, Ze-Kang Zhang et al.· FinTech and Sustainable Inno...· 0 citations
Recent advances in Generative AI have substantially improved financial sentiment analysis through post-trained financial large language models (LLMs). However, existing approaches remain confined to predicting the sentiment of each article in isolation and operate as black boxes, which limits contextual understanding a...
Giorgos Iacovides, Ole Hannibal, Wu-Yang Zhou et al.· 0 citations
Financial workflows need fast, calibrated, auditable decisions, not just generated text. We train System-1 decision models, small language models fine-tuned to answer typed questions (choice/score/noul) with a probability in a single forward pass, across two model families (K2-Horizon and Qwen3.5/3.8) and nine rungs fr...
Financial sentiment analysis remains challenging due to the complexity of natural language, specialized financial terminology and the need for interpretable predictions in high-stakes applications. Although FinBERT provides strong contextual language understanding, its predictions do not explicitly represent the reason...
Sarah Anis, Mohamed Mabrouk, M. Aref· Computers· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.