Enhancing empirical accounting models with textual information

Abstract

OF THE DISSERTATION Enhancing Empirical Accounting Models with Textual Information by Khrystyna Bochkay Dissertation Director: Dr. Carolyn B. Levine Rapid developments in information technologies and the increased availability of narrative disclosures in electronic form have provoked interest in textual analysis. In this dissertation, we survey research on textual analysis of mandatory and voluntary disclosures, describe methodologies for analyzing and incorporating text into quantitative models, and provide an analysis of MD&A text and earnings. Most empirical studies examine the association between text characteristics (e.g., tone and linguistic complexity) and future firm performance or market reactions. However, in-sample explanatory power is not equivalent to out-of-sample predictive power (Shmueli, 2010). We use regularized regression methods to examine whether textual disclosures in the Management Discussion and Analysis (MD&A) section of the 10-K report are helpful in predicting future earnings above and beyond traditional financial factors. We develop techniques to combine textual information from the MD&A section of the annual report with financial variables and generate explicit firm-level forecasts of future earnings. We employ the “bag-of-words” (BOW) approach to represent MD&A