The Sentiment Augmented GARCH-LSTM Hybrid Model for Value-at-Risk Forecasting

Abstract

In this paper, we present a new media and social media sentiment-based hybrid GARCH-LSTM model that can more accurately forecast volatility and Value-at-Risk of individual stocks than models proposed by previous studies. Various families of GARCH models and their hybrid extensions have been developed to achieve more accurate conditional volatility forecasts. However, in this paper, we demonstrate that the performance of these models can be significantly enhanced by incorporating different sentiment indicators on external media platforms. We emphasize the nonlinear relationship between variance and sentiment indices and how this relationship can be integrated into various volatility forecasting models. By leveraging the refined predictive capabilities of our extended models, we consolidate them with empirical applicability in the realm of financial risk assessment. Our approach enhances the traditional Value-at-Risk methodology, facilitating a more precise estimation of potential financial losses and offering a robust foundation for strategic risk management decisions. To evaluate our models, we conducted an empirical study on the logarithmic returns of individual stocks comprising the S&P 500 index from 2019 to 2024. In conjunction with the standard Value-at-Risk statistical tests, our study incorporates different loss functions to examine prospective loss magnitudes. Our frameworks demonstrate superior performance in comparison to the traditional GARCH model for a considerable subset of equities, as determined by conventional Value-at-Risk statistical evaluations. Furthermore, comparative analysis indicates that the proposed model ensures the most accurate conditional volatility forecast and its Value-at-Risk estimation achieves the smallest expected loss while satisfying all the statistical tests. These results underscore the superiority of our proposed model in predicting financial risk and volatility efficaciously, although we make further proposals to improve the dissemination of forecasts. © The Author(s) 2025.