ProbRes: Volatility Learning for Probabilistic Time-Series Forecasting
ProbRes: Volatility Learning for Probabilistic Time-Series Forecasting
Authors: Tingting Wang, Yunyi Zhang, Benyou Wang Date: 2026-06-01 Paper ID: openalex:2606.02117
Summary
ProbRes is a post-hoc probabilistic calibration method designed to improve the reliability of time-series forecasts by explicitly modeling volatility dynamics. By separating the learning of conditional mean and conditional volatility, the framework effectively addresses challenges posed by heteroskedastic and non-Gaussian data. During inference, ProbRes generates well-calibrated predictive distributions through a residual resampling process that is agnostic to the underlying point-forecasting architecture. Experimental results confirm that the method achieves superior calibration of prediction intervals across both synthetic and real-world financial datasets.
Key Contributions
- Introduced ProbRes, a post-hoc calibration framework that explicitly decouples conditional mean and volatility modeling for robust probabilistic forecasting.
- Demonstrated that ProbRes effectively handles heteroskedasticity and non-Gaussian error distributions in both univariate and multivariate settings.
- Provided theoretical validity for the approach and showed superior calibration of prediction intervals compared to standard techniques on various benchmarks.
Key Concepts
- probres: A post-hoc probabilistic calibration framework that decouples conditional mean and volatility modeling to handle heteroskedasticity in time series.
Archivist Review
I approved ProbRes as it provides a distinct, architecture-agnostic approach to probabilistic calibration via volatility-mean decoupling. The proposed method represents a reusable mechanism for addressing heteroskedasticity in time-series forecasting, aligning with the criteria for archival. No open questions or datasets met the rigorous standards for novelty and standalone importance.
Approved Concepts
- ProbRes: Introduces a decoupling strategy for mean and volatility in probabilistic forecasting, offering a flexible post-hoc calibration mechanism.
Links
Metadata & Links
- url
- https://arxiv.org/abs/2606.02117
- paper_id
- 2606.02117
- paper_source
- openalex
- domain
- time-series
- tags
-
- architectures
-
- datasets
-
- concept_slugs
- probres
- dataset_slugs
-
- skill
- TimeSeriesSkill
- created_at
- 2026-06-04T06:33:37Z