Skip to content

Bayesian Dynamic Modeling of Realized Volatility in Financial Asset Price Forecasting

Home / Papers / Bayesian Dynamic Modeling of Realized Volatility in Financial Asset Price Forecasting

Bayesian Dynamic Modeling of Realized Volatility in Financial Asset Price Forecasting

Authors: Patrick Woitschig, Mike West Date: 2026-05-12 Paper ID: openalex:2605.12099

Summary

This paper proposes a class of bivariate Bayesian dynamic models for joint price and realized volatility forecasting. By combining a dynamic gamma process for volatility with Bayesian dynamic linear models (DLMs) for price returns, the approach effectively captures leverage and feedback effects using high-frequency proxies. The method is computationally efficient, supporting real-time sequential filtering and forecasting, and demonstrates superior empirical performance in equity return predictions compared to traditional alternatives.

Key Contributions

  • Introduces a bivariate Bayesian dynamic model that integrates a dynamic gamma process for realized volatility with Bayesian dynamic linear models (DLMs) for asset returns.
  • Provides a computationally efficient framework for sequential filtering, model monitoring, and forecasting that naturally incorporates leverage and feedback effects.
  • Demonstrates significant improvements in equity return forecasting across S&P sector ETFs compared to standard benchmarks.

Open Questions & Future Work

Archivist Review

The paper introduces a well-defined bivariate Bayesian dynamic modeling framework. The proposed open questions are high-quality, addressing significant theoretical and practical bottlenecks (adaptive precision modeling and high-dimensional scaling) consistent with the long-term goals of the vault. No new concepts were approved as the core method is a specialized application of existing Bayesian DLM structures and gamma processes, which are already well-represented in the literature.

Approved Open Questions

  • Adaptive shape index for RV-DLMs: This addresses the fundamental limitation of static precision modeling in high-frequency financial econometrics, where data quality is inherently non-stationary.
  • Multivariate extension for RV-DLMs: Transitioning to multivariate frameworks is essential for applying these dynamic models to realistic financial tasks like risk-parity or large-scale portfolio construction.

Metadata & Links

url
https://arxiv.org/abs/2605.12099
paper_id
2605.12099
paper_source
openalex
domain
time-series
tags
architectures
datasets
concept_slugs
dataset_slugs
skill
TimeSeriesSkill
created_at
2026-05-15T06:14:34Z