What is it about?
Financial forecasting models are usually tested on historical market data. But when a model performs poorly, it is often difficult to know **why**, because real financial markets mix many different behaviors together—such as changing volatility, extreme shocks, market regime changes, persistent patterns, and sudden jumps. **FinStressTS** creates controlled synthetic financial time series in which these different market behaviors can be varied systematically. We build **30 testing environments covering six common financial mechanisms** and use them to compare **15 forecasting models**, ranging from traditional statistical methods to modern deep-learning models. This controlled setting allows us to identify which types of forecasting models work best under different financial conditions, how well they predict both future outcomes and uncertainty, and how much data they need to perform well. In this sense, FinStressTS provides a **stress-testing laboratory for financial forecasting models**.
Featured Image
Photo by Adam Śmigielski on Unsplash
Why is it important?
Financial forecasting models are increasingly used in investment, risk management, trading, and financial decision-making. However, strong performance on historical datasets does not necessarily mean that a model will remain reliable when market conditions change. FinStressTS helps address this problem by testing forecasting models under **controlled and interpretable financial scenarios**. Because each environment is designed around a specific market behavior—such as volatility changes, extreme events, regime shifts, or persistent dynamics—we can identify not only **which model performs better, but also when and why it performs better or fails**. This matters for both researchers and practitioners. Researchers gain a reproducible benchmark for developing and comparing new forecasting methods, while financial institutions can better understand the strengths and weaknesses of different models before relying on them in real-world applications. Ultimately, FinStressTS can contribute to **more robust, transparent, and trustworthy financial forecasting**.
Perspectives
One motivation for this work was a simple question: **when a forecasting model performs well—or fails—do we really understand why?** In financial forecasting, researchers often compare models using a small number of historical datasets. These datasets are realistic, but they also combine many market characteristics at the same time. As a result, it can be difficult to tell whether a model succeeds because it handles volatility well, adapts to regime changes, captures persistence, or simply happens to fit the particular historical period being tested. With FinStressTS, we wanted to complement real-world evaluation with a more controlled approach. By generating financial time series with known and adjustable properties, we can isolate specific challenges and examine how different forecasting models respond to them. Our broader view is that model evaluation should go beyond asking **“Which model has the lowest forecasting error?”** We should also ask **“Under what conditions does this model work, when does it fail, and can we trust its uncertainty estimates?”** We hope FinStressTS encourages more diagnostic, transparent, and stress-oriented evaluation of financial forecasting models.
Associate Professor Kewei Huang
National University of Singapore
Read the Original
This page is a summary of: FinStressTS: A Parametric Synthetic Benchmark for Time-Series Forecasting in Finance, August 2026, ACM (Association for Computing Machinery),
DOI: 10.1145/3770855.3817578.
You can read the full text:
Resources
Contributors
The following have contributed to this page
