Detecting Financial Statement Fraud through Multidimensional Analysis of Text Readability

Abstract

This study uses Coh-Metrix to analyze multiple dimensions of readability of the MD&A section of the SEC Form 10-K. We incorporate the five main Coh-Metrix components of text easability (word concreteness, syntactic simplicity, referential cohesion, deep cohesion, and narrativity) into a logistic model to test their predictive power for financial misreporting. We find that compared to the MD&As of nonfraud firms, the MD&As of fraud firms connect clauses and sentences less coherently, use more story-like language, and show a higher number of vague and abstract words. Thus, referential cohesion, narrativity, and word concreteness significantly enhance predictive ability in fraud detection. The Coh-Metrix readability measures enhance the linguistic complexity assessment beyond traditional readability measures, such as the Fog Index and the Flesch Indexes. Financial analysts and investors can utilize the Coh-Metrix readability measures to supplement traditional readability measures and common financial statement variables in predicting financial misreporting. Data Availability: Data are available from the public sources cited in the text. JEL Classifications: G32; K42; M41; M48.