Redefining Financial Manipulation Detection: a New Model's Performance Against the M-Score

Abstract

Financial manipulation negatively affects market participants. Therefore, various techniques are developed to detect manipulation. One of the most widely used techniques in the literature is the M-score. The study investigates whether the M-score is an effective model across different periods and aims to provide perspective to researchers and market players. In this study, we developed a model to detect financial manipulation. We compared our model results with the M-score results. In the study, we first scanned the US Securities and Exchange Commission's "Accounting and Auditing Practice Bulletins" and identified fifty-one companies (red flags) from the reports. Then, we calculated some ratios from the financial reports of these companies (red flags) and companies with control variables and created the prediction model. The results show that the success of the created manipulation prediction model is eighty-three per cent. The M-score results are lower. The model in the study is the most up-to-date model with many explanatory variables to detect financial manipulation. The study showed that established manipulation detection models may lose effectiveness over time, and the effects of the variables in the model may change. Therefore, the study offers a novel perspective to the literature. In future studies, researchers could investigate how imbalances in the proportion of "1" versus "0" observations affect the detection model.