Tabular Foundation Models in Financial Predictions

This technical note introduces tabular foundation models (TFMs) to accounting research and evaluates their performance in one common benchmark task: predicting firms' next-year earnings changes. TFMs produce numeric predictions and learn in context, without post-release training or tuning, making them natural and easy to use for forecasting. The TFMs we study are pretrained on synthetic tabular data, so they avoid look-ahead bias from memorizing firms' outcomes. We compare Google Research's TabFM with a fine-tuned open-weight large language model developed in recent accounting research and with several tree-based models. TabFM is more accurate than all comparison models. It is also faster and less computationally intensive than the large language model. Several tree-based specifications produce predictions more quickly than TabFM, so choosing between them involves a trade-off between accuracy and speed. We conclude that TFMs have promise as a new tool for forecasting in accounting and finance. However, our evaluation is based on a single benchmark task, and no class of models is likely to dominate in all settings.

1. 27 Oct 2025

   FinCARE: Financial Causal Analysis with Reasoning and Evidence

   KG+LLM for Financial Causal Discovery: Combines SEC knowledge graphs, LLM reasoning, and causal discovery to build more accurate finance‑grounded causal models.

   arXiv

   2cites
2. 13 Sep 2025

   FinZero: Launching Multi-modal Financial Time Series Forecast with Large Reasoning Model

   Multi-modal Forecasting: FinZero, a pre-trained model fine-tuned by the Uncertainty-adjusted Group Relative Policy Optimization method, is introduced in the article, enhancing the accuracy, adaptability, and scalability of financial time series forecasting.

   arXiv

   5cites
3. 29 Aug 2025

   FinCast: A Foundation Model for Financial Time-Series Forecasting

   Time-Series Forecasting Model: FinCast, a new model for financial time-series forecasting, outperforms existing methods by effectively capturing diverse patterns without needing domain-specific adjustments.

   arXivIn Proceedings of the 34th ACM International Conference on Information and Knowledge Management

   27cites
4. 12 Mar 2025

   CoFinDiff: Controllable Financial Diffusion Model for Time Series Generation

   CoFinDiff, a new synthetic financial data generation model, is designed to overcome the shortcomings of traditional statistical models, producing synthetic data that matches any conditions.

   arXivIn International Joint Conference on Artificial Intelligence

   23cites
5. 3 Oct 2024

   Soft Measures for Extracting Causal Collective Intelligence

   A study uses large language models to extract fuzzy cognitive maps from text, but emphasizes the need for specific soft similarity measures for this process.

   Machine learningIn NLP4SCIENCE

   3cites
6. 2 Nov 2023

   Interpretive Earnings Forecasts via Machine Learning: A High-Dimensional Financial Statement Data Approach

   Using machine learning models and comprehensive Compustat financial statement data for earnings forecasting can yield predictions that are up to 13% more accurate than traditional linear approaches.

   SSRN

   2cites