Measuring the consistency of quantitative and qualitative information in financial reports: A designing science approach

Abstract

This study uses a design science approach to examine the consistency between quantitative financial ratios and qualitative narrative disclosures in the annual reports. To extract information on the tone of unstructured qualitative textual data, we first use the term frequency/inverse document frequency (TFIDF) text mining technique to classify each company`s narrative disclosure as either “Positive” or “Negative.” For the quantitative information, we use the K-means method to cluster each company`s financial performance data into “Good” or “Poor” groups. Consistency is said to occur when the textual and numerical data form either a “Positive-Good” pair or a “Negative-Poor” pair. The design model is presented in a stepwise fashion and therefore is transparent for evaluation and validation. Our evaluation process demonstrates the feasibility of the design model. The evaluation was conducted using listed semiconductor companies in countries with different levels of market development. The results show that U.S. firms are less likely to exaggerate in their narrative disclosures and are more likely to understate their performance in MD&As compared to companies in other markets such as China and Taiwan.