A Robust Multivariate Outlier Detection Method for Detection of Securities Fraud

Insider trading is one the most common deceptive trading practice in securities markets. Data mining appears as an effective approach to tackle the problems in fraud detection with high accuracy. In this study, the authors aim to detect outlying insider transactions depending on the variables affecting insider trading profitability. 1,241,603 sales and purchases of insiders, which range from 2010 to 2017, are analyzed by using classical and robust outlier detection methods. They computed robust distance scores based on minimum volume ellipsoid, Stahel-Donoho, and fast minimum covariance determinant estimators. To investigate the outlying observations that are likely to be fraudulent, they employ event study analysis to measure abnormal returns of outlying transactions. The results are compared to the abnormal returns of non-outlying transactions. They find that outlying transactions gain higher abnormal returns than transactions that are not flagged as outliers. Business intelligence and analytics may be a useful strategy for detecting and preventing of financial fraud for companies.