[Paper Review] Volatility around the clock: Bayesian modeling and forecasting of intraday volatility in the nancial crisis

Jonathan Stroud, Michael Johannes|arXiv (Cornell University)|Nov 13, 2012

Financial Risk and Volatility Modeling28 references1 citations

TL;DR

This paper proposes a Bayesian hierarchical model using 5-minute intraday returns to estimate multiple stochastic volatility factors, jumps, seasonality, and market microstructure effects during the 2007–2009 financial crisis. By employing an integrated MCMC and particle filter approach without realized volatility aggregation, it improves out-of-sample volatility forecasts by up to 50% compared to benchmarks.

ABSTRACT

High frequency data provides a rich source of information for understanding nan- cial markets and time series properties of returns. This paper estimates models of high frequency index futures returns using 'around the clock' 5-minute returns that incorporate the following key features: multiple persistent stochastic volatility factors, jumps in prices and volatilities, seasonal components capturing time of the day pat- terns, correlations between return and volatility shocks, and announcement eects. We develop an integrated MCMC approach to estimate interday and intraday parameters and states using high-frequency data without resorting to various aggregation measures like realized volatility. We provide a case study using nancial crisis data from 2007 to 2009, and use particle lters to construct likelihood functions for model comparison and out-of-sample forecasting from 2009 to 2012. We show that our approach improves realized volatility forecasts by up to 50% over existing benchmarks.

Motivation & Objective

• To model intraday volatility dynamics during the 2007–2009 financial crisis using high-frequency data.
• To incorporate multiple persistent stochastic volatility factors, jumps in prices and volatilities, and time-of-day seasonal patterns.
• To estimate correlations between return and volatility shocks and the impact of macroeconomic announcements.
• To develop a likelihood-based model comparison and forecasting framework using particle filters and MCMC.
• To improve volatility forecasting accuracy beyond realized volatility benchmarks using full high-frequency data without aggregation.

Proposed method

• Uses 5-minute returns from index futures to model intraday volatility with multiple stochastic volatility factors.
• Employs a Bayesian hierarchical model that jointly estimates interday and intraday parameters and unobserved states.
• Applies an integrated Markov Chain Monte Carlo (MCMC) algorithm to sample from the joint posterior distribution of parameters and states.
• Incorporates time-of-day seasonality through deterministic seasonal components in the volatility equation.
• Models jumps in returns and volatility using latent Poisson processes and jump size distributions.
• Utilizes particle filters to construct likelihood functions for model comparison and out-of-sample forecasting from 2009 to 2012.

Experimental results

Research questions

• RQ1 How do multiple stochastic volatility factors and jumps affect intraday volatility dynamics during the financial crisis?
• RQ2 To what extent do time-of-day seasonal patterns and announcement effects improve volatility modeling?
• RQ3 Can a fully Bayesian, non-aggregated approach to high-frequency data outperform realized volatility benchmarks in forecasting?
• RQ4 How well does the model capture the joint dynamics of return and volatility shocks during periods of market stress?
• RQ5 What is the predictive performance of the model in out-of-sample forecasts from 2009 to 2012?

Key findings

• The proposed model improves out-of-sample volatility forecasts by up to 50% compared to existing benchmarks.
• Incorporating multiple stochastic volatility factors and jumps significantly enhances model fit and predictive accuracy.
• The inclusion of time-of-day seasonal components and announcement effects captures intraday volatility patterns more effectively than models without them.
• The particle filter-based likelihood construction enables reliable model comparison and out-of-sample evaluation.
• The MCMC framework successfully estimates both interday and intraday parameters and unobserved states without relying on realized volatility measures.
• The model demonstrates strong predictive power during the post-crisis period (2009–2012), indicating robustness beyond the crisis phase.