Inferring the Financial Materiality of Corporate Social Responsibility News

Abstract

The intangible nature of Corporate Social Responsible (CSR) issues has typically hindered financial analysts' abilities to integrate such information into investment models. To address this limitation, we employ a probabilistic topic model known as Latent Dirichlet Allocation (LDA) to infer contextual information and semantic meaning in text. Using a sample of 105,983 CSR articles from newswires, newspapers, blogs and magazines over the period 1980-2014, the model detects CSR allegations based on ethical grounds versus those that attribute corporate difficulties and litigation risk. Our findings indicate a statistically significant and negative correlation between material CSR concerns and firms’ future earnings surprises, and a statistically insignificant correlation for more ethical concerns.

Keywords: Corporate Social Responsibility, news, text analysis, probabilistic topic models

JEL Classification: G12, G14