The Network of Firms Implied by the News

Abstract

Measuring aggregate risks is challenging due to the lack of firm interconnections data. Our paper changes this by introducing a natural language processing methodology that leverages financial news to construct firm networks. Applying our methodology to Reuters between 2006 and 2013, we extract a vast network of 8,400 links between 2,000 firms and make it openly accessible. Its connectivity statistics predict aggregate risk measures, like S&P 500 returns and industrial production growth, by revealing information about financial uncertainty and contagion. Our work complements existing research by providing more comprehensive and timely firm networks, and more predictive signals of aggregate fluctuations.

Keywords: Networks, contagion, predictability, risk measurement, machine learning, natural language processing. JEL codes: E32

JEL Classification: E32, E44, L11, G10, C82