Predicting Market Reactions to Bad News

Our Applied Finance Project aims to develop a framework to predict short-term and medium-term market reactions to bad news shocks. The study is based on a sample of 18,497 bad news articles and time series of 1,008 Russell 3000 stocks returns during the period 2005 to 2017. Our research proposes a three-stage model for the analysis. Firstly, given a dataset of bad news events and stock prices, we employ time series clustering techniques on cumulative abnormal returns of stocks, by which the news articles related to those stocks are grouped into different clusters. Secondly, we apply Natural Language Processing and multi-class classification algorithms on relevant news articles to extract features of each cluster. Then, by applying Support Vector Machine model, whenever specific bad news is released, we can predict the subsequent short-term, and medium-term market reactions post negative news. Finally, we develop long/short trading strategy for both short-term and medium-term horizons that asset managers in the real world can apply every day