Financial fraud detection

Abstract

Data science plays a significant role in the decision-making process of businesses in today’s era. The vast amount of data being generated and collected, and the ever-evolving mechanisms and software, are transforming the way the accounting and auditing operations are performed, as professionals in these fields are now utilizing data analytics techniques to analyze financial data and identify trends and hidden patterns. A critical application of data science that has attracted the attention of both the academic and the business community is the detection of Financial Statement Fraud. The integration of advanced analytics, machine learning platforms, and automated models in this area allows organizations to more efficiently determine potentially fraudulent activity and thus, be more proactive and attentive during the financial reporting practices. To this end, this study proposes two approaches for detecting fraudulent financial statements, both of which are based on the Management Discussion and Analysis section of the annual SEC company fillings. The first methodology utilizes linguistic variables, related to the context, the structure, and the sentiment of the document, whereas the second one uses the full textual information of the MD&A section in the form of words and phrases (N-grams). Natural Language Processing (NLP) tools and the Random Forest classification algorithm are employed in both models. With regards to the metrics, Accuracy, Sensitivity, Specificity, Precision and F1-score are calculated to evaluate the performance of the models. In addition to achieving the best possible predictive results, this research aims to provide specific “red-flag” indicators at a word and phrase level, which could assist the auditing decision-making procedures. In conclusion, this dissertation forms a complete, competent and interpretable solution to the Financial Statement Fraud Detection problem. It can serve as a foundation for internal or external financial reporting audits, as well as a thorough tool for detecting fraud with the appropriate adjustments to suit the specific needs of a business based on its size, industry, and operating environment.