A1988
Title: Forecasting realized volatility: A hybrid model integrating BiLSTM with HAR-type models
Authors: Yi Luo - Xian Jiaotong-Liverpool University (China) [presenting]
Marwan Izzeldin - Lancaster University Management School (United Kingdom)
Abstract: A hybrid methodology is proposed that combines both Heterogeneous Autoregressive (HAR)-type models and Deep Feedforward Neural Network (DFN) model as well as the Bidirectional Long Short-term Memory (BiLSTM) model in predicting realized volatility. On the one hand, Neural Network architecture naturally deals with many important stylized facts of realized volatility (e.g., long-memory and nonlinearity, etc.), complementing the linear HAR-type models. On the other hand, the interpretation of results produced by Neural Networks can be improved from the aspect of high-quality input features generated by HAR-type models. Empirical results show that BiLSTM-based hybrid model outperforms all other models in the out-of-sample forecasting across all forecasting horizons. Additionally, both the performance of DFN-based and BiLSTM-based hybrid model significantly beat their single-model counterparts, indicating HAR-type components can be considered as effective features in Neural Networks.
