Labeling firms’ financial stability from auditors’ opinions: Textual use of Machine Learning Techniques

Abstract

Keywords are valuable information for characterizing texts. In order to extract information of company-level financial stability, we use a sample of manufacturing firms to propose a comprehensible and robust approach based on textual representations included in financial statements. We use Auditors’ Opinion to characterize financial stability of a firm under certain uses of audit expressions and words combinations, describing key factors in auditors’ evaluation acclaimed in financial reports. Our study’s scope is to promote a solid “pre-assessment” framework using textual information from financial statements to provide adequate evidence either to potential investor in taking appropriate decisions or to analysts who wish to form a quick and reliable opinion about a company. For the purpose of this evaluation, we use an approach of frequency-based keyword extraction (exemplary text collection: 6.686 pdf documents in English). Also, we validate -manually- some 504 pdf reports from the initial sample in order to validate methods’ accuracy. Results of the study reveal, among others, the identification of specific clusters of words within the firms’ financial statements that verbally reflect their financial condition, characterizing also the “quality” of their financial statements to a significant extent. Furthermore, we find that there are special words describing inconsistencies in financial statements, such as "change auditor" or "we have not received all the information" or "detect misstatements", or "draw attention", that may be important issues for the examing firms to be considered. Also, where there are satisfactory comments in Auditors’ Opinion, but they highlight the companies’ weaknesses, whether to accurately determine financial figures or to calculate potential impairments, auditors use specific arrays of words that can immediately characterize the financial position of each company. Finally, the results indicate a significant perspective for possible evolution in machine learning techniques regarding the assessment of quality characteristics in financial statements using auditors’ findings which open the floor for further field research for analysts and auditing firms.

Keywords: Document labeling, auditors’ opinion, financial statements’ analysis, auditing technics, quality of financial statements, machine learning technics, feature selection.