Modeling volatility with multivariate GARCH models through the integration of deep Learning: A literature review

Abstract

This paper presents a literature review that examines how deep learning methodologies have been integrated into

multivariate GARCH models. While traditional GARCH-type models have long been used to capture time-varying volatility and spillover effects, their parametric and linear nature limits their capacity to reflect the increasing complexity and non-linearity in financial markets. Recent advances in deep learning, notably recurrent neural networks (RNNs) and long short-term memory (LSTM) architectures, provide powerful tools for modeling volatility dynamics and incorporating exogenous information. Based on a critical synthesis of the recent literature, this review outlines the conceptual foundations of hybrid deep learning–GARCH models, compares their methodological benefits and limitations, and identifies key research gaps. The aim is to offer a clear overview of the current state of research and to guide future studies in developing more flexible and accurate volatility forecasting models.

Keywords : Volatility modeling, Multivariate GARCH, Deep learning, Hybrid models, Financial econometrics.

Classification JEL : C58; C45; C53; C22; G17

Paper type : Theoretical Research or Empirical Research