Predicting the influence of event news transferring between countries using LSTM-Graph Neural Networks

Abstract

In financial markets, there is a lot of news coming out every day and affecting asset prices. To understand how information about specific events in news articles propagates from a country to other countries, we focus on predicting the change of the amount of news articles in each country. While previous studies utilized GAT (graph attention networks) to capture cross-country dependencies, they aggregated past information and did not consider temporal structures. In this paper, we extend GAT model to LSTM-GAT for modelling the change of information propagation across time. Our experiment shows that LSTM-GAT improves the prediction accuracy compared to other baseline methods, which capture only one of cross-country and temporal dependencies.