Writing Styles Social Conformity, Firm Performance, and Earnings Persistence – A Deep Learning Approach

This paper examines the relation between annual report boilerplate and firm performance and earnings persistence. Utilizing a neural network algorithm inspired by word2vec, we compute the annual report boilerplate, defined as the average probability of the occurring of each sentence in a given report based on the order of words. We find that: (1) the annual reports of firms with lower earnings are more standardized and their writing styles display stronger social conformity behavior (i.e., they have a higher level of boilerplate); and (2) firms with annual reports with less boilerplate have more persistent positive earnings