Global News Networks and Return Predictability by Gustavo Freire, Ali Moin, Alberto Quaini, Amar Soebhag :: SSRN

We employ a novel, high-dimensional global news corpus (GDELT) to examine the effects of news-based sentiment on international equity market returns. Leveraging over 520 million articles from 14 developed countries, spanning 260 major themes and classified by source and target country, we construct granular sentiment linkages that capture how one country’s media reports on another. Using machine learning, we construct out-of-sample market-timing strategies based on local and global sentiment measures and document significant economic gains. Further, global sentiment measures contain distinct, incremental information compared to local measures, indicating market integration via sensitivity to shared news information and sentiment.