Network Vector Autoregressive Model for Dyadic Response Variables

TL;DR

This paper introduces the NVARD and VCNVARD models to analyze dyadic response data with network dependencies, capturing dynamic interdependencies and heterogeneity for improved prediction of bilateral trade flows.

Contribution

The paper develops novel network vector autoregressive models specifically for dyadic data, incorporating time-varying coefficients to handle heterogeneity and improve predictive accuracy.

Findings

Models effectively capture dependencies among dyadic responses.

Application to world trade data shows improved prediction accuracy.

Models handle heterogeneity across time and pairs.

Abstract

For general panel data, by introducing network structure, network vector autoregressive (NVAR) model captured the linear inter dependencies among multiple time series. In this paper, we propose network vector autoregressive model for dyadic response variables (NVARD), which describes the dynamic process of dyadic data in the case of the dependencies among different pairs are taken into consideration. Besides, due to the existence of heterogeneity between time and individual, we propose time-varying coefficient network vector autoregressive model for dyadic response variables (VCNVARD). Finally, we apply these models to predict world bilateral trade flows.