Classifying the influence of negative affect expressed by the financial media on investor behavior

Prior text mining studies have documented a causal link between human emotions and stock market patterns, yet relatively little research exists into what triggers these emotions. This paper aims to bridge the gap by empirically testing a social psychology theory of human behavior. Underlying our approach lies Attribution Theory, which addresses how observers form causal inferences and moral judgments to explain human behavior, particularly those with negative outcomes. The system presented here works in three stages. The first phase computes a measure of media pessimism by counting negative terms from the General Inquirer dictionary to detect acts of corporate irresponsible behavior. The second phase extends the term-counting approach to capture contextual information. Emotion topic priors are incorporated in a Latent Dirichlet Allocation (LDA) model to infer the financial media's expression of negative affect. Finally, the system combines the two components in an ensemble tree to classify the impact of financial media allegations on a company's stock market patterns. The paper underlines the potential benefit of text mining technology for the support of investor strategies, and more generally demonstrates the power of combining multiple methods for applications in specific domains.