Learnijoy
ResearchAI ResearchAI Investing

Forma AI Model Forecasts Full Financial Statements 20 Quarters Ahead

Travis L. Johnson, Jiannan Jiang, Soumyabrata Chaudhuri, Yihao Chen, Lauren Falvey, Donal O'Cofaigh· August 13, 2026 View original

Key takeaways

  • Specialized AI models like Forma outperform generalist LLMs for long-horizon financial statement forecasting.
  • Forma can forecast 78 financial line items up to 20 quarters ahead with high accuracy.
  • The model's tuple interface allows for flexible scenario analysis without retraining.
  • Accurate long-term financial forecasts are crucial for robust discounted cash flow valuations.

Who benefits

FinanceInvestment BankingAsset ManagementCorporate FinanceConsulting

Summary

A new transformer model, Forma, significantly outperforms existing methods in forecasting complete financial statements up to five years into the future. It uses a novel tuple-based interface and a masked-tuple Gaussian likelihood to achieve high accuracy and robust predictive intervals.

Traditional financial forecasting often struggles with long horizons and the joint prediction of complete financial statements beyond a single year, despite the critical need for such data in discounted cash flow valuations. Researchers have introduced Forma, a transformer-based model designed to address this gap. Forma processes financial statements as sets of (account, quarter, value) tuples, optimizing for a masked-tuple Gaussian likelihood. The model was evaluated on ProForma-20Q, a new benchmark dataset for forecasting 78 line items 1 to 20 quarters ahead. Forma demonstrated superior performance compared to classical machine learning, gradient boosting, zero-shot time-series models, and even frontier large language models, with its accuracy advantage increasing over longer horizons. Furthermore, Forma's forecasts largely adhere to accounting identities, and its Gaussian predictive intervals consistently maintain proper coverage.

Why it matters

Professionals in finance and investment can leverage this model for more accurate long-term financial projections, improving valuation models and strategic planning.

How to implement this in your domain

  1. 1Evaluate existing financial forecasting tools against Forma's reported capabilities for long-horizon accuracy.
  2. 2Explore integrating similar transformer-based architectures into proprietary financial modeling systems.
  3. 3Utilize the ProForma-20Q benchmark to test and validate internal forecasting models.
  4. 4Develop scenario analysis capabilities by leveraging Forma's tuple interface to pin future revenue paths.

Original post by Travis L. Johnson, Jiannan Jiang, Soumyabrata Chaudhuri, Yihao Chen, Lauren Falvey, Donal O'Cofaigh

"arXiv:2608.11327v1 Announce Type: new Abstract: Specialist training beats generalist scale when forecasting financial statements. To our knowledge, no prior work jointly forecasts complete financial statements beyond one year, yet in a discounted-cash-flow valuation most firm val…"

View on X

Originally posted by Travis L. Johnson, Jiannan Jiang, Soumyabrata Chaudhuri, Yihao Chen, Lauren Falvey, Donal O'Cofaigh on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses

More in AI Research

AI ResearchAI Engineering & DevTools

Task-Vector Interference in Merged LLMs Driven by Orientation, Not Magnitude.

This research reveals that interference in merged language models, often attributed to magnitude, is primarily driven by the orientation of task-vectors. It demonstrates that erasing interference along specific directions causally removes its effects, while magnitude-based interventions are insufficient and inconsistent.

Chencheng ZhuAug 13, 2026
AI Engineering & DevToolsAI Research

New Method Detects Gradual GNSS Spoofing in Autonomous Driving.

This paper proposes a causal high-order liquid evidence framework to detect gradual GNSS spoofing attacks in autonomous driving. By modeling the evolution of GNSS-motion inconsistency with multiple evidence streams and adaptive liquid encoders, the method achieves high F1-scores in detecting subtle spoofing.

Muhammad Ayub Sabir, Junbiao Pang, Fatima AshrafAug 13, 2026
AI ResearchAI Engineering & DevTools

MOON Improves Multitask Learning with OrthoNormalized Gradient Updates.

This paper introduces MOON (Multi-Objective OrthoNormalized Updates), a novel approach for multi-task learning that addresses limitations of Euclidean gradient manipulation in multi-objective optimization. MOON performs gradient manipulation under spectral-nuclear norm geometry, leading to more efficient optimization and improved performance in modern architectures like Transformers.

Shiji Zhou, Kunlin Lyu, Lei Zhang, Ruodong Wang, Yifan SunAug 13, 2026