A Semiparametric Stochastic Volatility Model with Dependent Errors

TL;DR

This paper introduces a flexible semiparametric stochastic volatility model that relaxes Gaussian assumptions, capturing complex features in financial data and improving estimation accuracy through Bayesian MCMC methods.

Contribution

It develops a novel semiparametric SV model with dependent errors, enhancing robustness and flexibility over traditional Gaussian-based models in financial econometrics.

Findings

Lower bias and variance in parameter estimation

More accurate volatility estimates during large market fluctuations

Effective in modeling non-Gaussian, dependent return data

Abstract

This paper proposes a semiparametric stochastic volatility (SV) model that relaxes the restrictive Gaussian assumption in both the return and volatility error terms, allowing them to follow flexible, nonparametric distributions with potential dependence. By integrating this framework into a Bayesian Markov Chain Monte Carlo (MCMC) approach, the model effectively captures the heavy tails, skewness, and other complex features often observed in financial return data. Simulation studies under correlated Gaussian and Student's t error settings demonstrate that the proposed method achieves lower bias and variance when estimating model parameters and volatility compared to traditional Gaussian-based and popular Bayesian implementations. We conduct an empirical application to the real world financial data, which further underscores the model's practical advantages: it provides volatility estimates that respond more accurately to large fluctuations, reflecting real-world market behavior. These findings suggest that the introduced semiparametric SV framework offers a more robust and adaptable tool for financial econometrics, particularly in scenarios characterized by non-Gaussian and dependent return dynamics.