What type of study is this?

September 10, 2025Open Access

Using an LSTM Neural Network to Improve Symmetric and Asymmetric GARCH Volatility Forecast

SRSetya Budi Rahmawanto DNDidit Budi NugrohoSatya Wacana Christian University STSuryasatriya TrihandaruSatya Wacana Christian University

Key Points

The Extended LSTM hybrid model enhances volatility forecasting accuracy, outperforming GARCH-type models.
Results indicate that integrating LSTM with GARCH frameworks captures complex volatility patterns effectively.
The model architecture, including ReLU and GRU layers, addresses over-smoothing and adapts to market fluctuations.
This approach suggests valuable insights for investors and risk managers in financial risk management.

Abstract

Volatility forecasting is crucial for financial risk management, yet traditional models like GARCH struggle with nonlinearities and asymmetric effects. This study leverages Long Short-Term Memory (LSTM) neural networks to enhance symmetric and asymmetric GARCH models, addressing these limitations. By integrating LSTM with GARCH, GARCH-X, and Realized GARCH frameworks, we propose hybrid models (Baseline and Extended versions) to improve forecasting accuracy. Using daily data from FTSE 100, Nikkei 225, and SP 500 indices (2000–2020), we compared hybrid models against traditional models. Results show that the Extended LSTM hybrid model outperforms both traditional GARCH-type models and the Baseline LSTM, capturing complex volatility patterns more effectively. The Extended model’s architecture, featuring ReLU, GRU, and dropout layers, mitigates over-smoothing and enhances responsiveness to market fluctuations. This research demonstrates LSTM’s potential to refine volatility forecasting, offering valuable insights for investors and risk managers.

Demander à l'IA

Bookmark

View Full Paper