Please use this identifier to cite or link to this item:
Title: VAR-GRU: A Hybrid Model for Multivariate Financial Time Series Prediction
Authors: Lkhagvadorj Munkhdalai
Meijing Li
Nipon Theera-Umpon
Sansanee Auephanwiriyakul
Keun Ho Ryu
Keywords: Computer Science
Issue Date: 1-Jan-2020
Abstract: © 2020, Springer Nature Switzerland AG. A determining the most relevant variables and proper lag length are the most challenging steps in multivariate time series analysis. In this paper, we propose a hybrid Vector Autoregressive and Gated Recurrent Unit (VAR-GRU) model to find the contextual variables and suitable lag length to improve the predictive performance for financial multivariate time series. VAR-GRU approach consists of two layers, the first layer is a VAR model-based variable and lag length selection and in the second layer, the GRU-based multivariate prediction model is trained. In the VAR layer, the Akaike Information Criterion (AIC) is used to select VAR order for finding the optimal lag length. Then, the Granger Causality test with the optimal lag length is utilized to define the causal variables to the second layer GRU model. The experimental results demonstrate that the ability of the proposed hybrid model to improve prediction performance against all base predictors in terms of three evaluation metrics. The model is validated over real-world financial multivariate time series dataset.
ISSN: 16113349
Appears in Collections:CMUL: Journal Articles

Files in This Item:
There are no files associated with this item.

Items in CMUIR are protected by copyright, with all rights reserved, unless otherwise indicated.