Network double autoregression

Abstract

Modeling high-dimensional time series with simple structures is a challenging problem. This paper proposes a network double autoregression (NDAR) model, which combines the advantages of network structure and the double autoregression (DAR) model, to handle high-dimensional, conditionally heteroscedastic, and network-structured data within a simple framework. The parameters of the model are estimated using quasi-maximum likelihood estimation, and the asymptotic properties of the estimators are derived. The selection of the model's lag order will be based on the Bayesian information criterion. Finite-sample simulations show that the proposed model performs well even with moderate time dimensions and network sizes. Finally, the model is applied to analyze three different categories of stock data.