Accounting Fraud Detection Using The Fraud Triangle Factors
Master thesis
Permanent lenke
https://hdl.handle.net/11250/3158970Utgivelsesdato
2024Metadata
Vis full innførselSamlinger
- Master Thesis [4505]
Beskrivelse
Full text not available
Sammendrag
This thesis examines whether the factors of the fraud triangle influence a firm's likelihood of committing financial fraud. According to Cressey’s theory, opportunity, pressure, and rationalization are always present in fraud situations. We develop variables serving as proxies for these elements and employ a logit regression analysis on a data sample of U.S. publicly listed firms from 2000 to 2019. Our analysis identifies three opportunity proxies, four pressure proxies, and one rationalization proxy significantly related to financial statement fraud. Our findings suggest that an increase in the Beneish M-score and institutional ownership intensifies the likelihood of fraud, while increased shareholder activism reduces it. Additionally, CEO duality is significantly positively related to the likelihood of fraud, whereas the average tenure of board members and board gender diversity are significantly negatively related. Forced CEO turnover is also linked to a higher likelihood of accounting fraud. Utilizing fraud triangle factors, we build machine-learning models for identifying fraudulent activities within financial statements. The performances of predictive models are cross-validated and compared using four evaluation metrics: Accuracy, Recall, Specificity, and AUC-ROC (Area Under the Curve – Receiver Operating Characteristics). Among these models, XGBoost emerges as the best-performing model. To further explore the implications of our model's predictions, we investigate the relationship between the predicted probabilities of financial fraud and firms' annual ESG scores. Our findings indicate that firms with a higher likelihood of accounting fraud tend to exhibit lower ESG performance. This thesis contributes to the existing literature by introducing new proxies for each element of the fraud triangle using updated data through 2019, offering valuable insights to help regulatory bodies, auditors, and investors identify conditions under which fraud is likely to occur. Furthermore, it explores the broader implications of fraud predictions on firm governance and performance through the use of machine-learning models.