Accounting Variables, Deception, and a Bag of Words: Assessing the Tools of Fraud Detection

Abstract

We develop a data‐generated tool for distinguishing between fraudulent and truthful reports based on the language used in the management discussion and analysis section of annual and interim reports. Using this method, we are able to assign a probability of truth to each report which is then shown to be an effective indicator of fraud. Our work goes beyond the development of a tool alone, however, by conducting an extensive comparison of our probability‐of‐truth measure with eight alternative detection tools representing both quantitative and language‐based approaches. Comparisons are made across a variety of samples and show that our language‐based approach can be effective in both cross‐sectional and time‐series settings. It is useful both in distinguishing between fraudulent and truthful firms and in identifying fraudulent reports from a series of reports issued by a single firm. This second setting is one in which accounting‐based detection tools have frequently struggled. We establish that, not only is our probability‐of‐truth measure significantly associated with fraud, so too is the change in this measure from a firm's previous reports. Prior reports may serve an important benchmarking role in using language‐based tools to identify fraud.

Fraudedetectie door tekstanalyse van jaarverslagen

“…De afgelopen jaren is er in het wetenschappelijk onderzoek een verschuiving te zien van de focus op kwantitatieve informatie en risicofactoren naar de tekstuele informatie (Cecchini et al 2010;Glancy and Yadav 2011;Purda and Skillicorn 2015). Tekst is om verschillende redenen een interessante bron voor onderzoek naar methoden voor fraudedetectie.…”

“…Voor dit onderzoek gebruiken we twee machine learning-algoritmen die in eerdere text mining-onderzoeken succesvol waren, te weten Naive Bayes (NB) en Support Vector Machine (SVM) (Cecchini et al 2010;Conway et al 2009;Glancy and Yadav 2011;Goel et al 2010;He and Veldkamp 2012;Joachims 1998;Manning and Schütze 1999;Metsis et al 2006;Purda and Skillicorn 2015). Beide algoritmen kiezen een andere benadering.…”

Text mining to detect indications of fraud in annual reports worldwide

“…Several mathematical implementations exist that differ in how they learn the characteristics and construct the model. The Naïve Bayes classifier (NB) and Support Vector Machine (SVM) have been proven successful in text classification tasks in several domains (Cecchini et al, 2010;Conway et al, 2009;Glancy and Yadav, 2011;Goel et al, 2010;He and Veldkamp, 2012;Joachims, 1998;Manning and Schütze, 1999;Metsis et al, 2006;Purda and Skillicorn, 2015). Therefore, this research uses these two types of machine learning approaches to develop text mining models that can detect indications of fraud in the management discussion and analysis section of annual reports.…”

“…Examples of risk variables include the number of subsidiaries of a company, whether the company changed its CEO, and wether the company has a poor reputation (Fanning and Cogger, 1998;Bell and Carcello, 2000). In recent years, the focus of fraud detection research shifted from the quantitative information and risk variables to the use of textual information (Cecchini et al, 2010;Glancy and Yadav, 2011;Purda and Skillicorn, 2015) Various factors contribute to making texts an interesting subject for fraud detection methods. Textual information provides information that is complementary to the quantitative information and reaches a larger audience.…”

“…The machine learning methods exploited to this end are SVM, NB and Clustering (Cecchini et al, 2010;Goel et al, 2010;Glancy and Yadav, 2011;Purda and Skillicorn, 2015;Fissette et al, 2017a,b). In these models, the text is represented as a bag-of-words.…”