A new Twitter based credit rating model methodology

Abstract In this paper we propose a novel way to predict corporate credit ratings by showing how a new type of data, Twitter, can be extracted and used for this purpose. We make three contributions to knowledge. First, we relate tweets from the companies themselves and tweets about the companies to the probability of a credit rating level. Second, we transform the tweets into two different sentiment scores which are used as predictors for credit rating levels and compare their predictive performance. The sentiment scores are calculated by using each of two alternative word-lists. Third, we propose two approaches how alternative information from Twitter, linguistic features in tweets, can be selected and incorporated into credit rating models. We compare the performance between the models containing the sentiment scores and the linguistic features. We analyze data relating to NASDAQ and NYSE listed companies over 2011-2019. Overall, we find that including information from Twitter gives a higher predictive performance compared to those of models that omit them.