How to detect illegal corporate insider trading? A data mining approach for detecting suspicious insider transactions

Abstract

Only in the U.S. Stock Exchanges, the daily average trading volume is about 7 billion shares. This vast amount of trading shows the necessity of understanding the hidden insights in the data sets. In this study, a data mining technique, clustering based outlier analysis is applied to detect suspicious insider transactions. 1,244,815 transactions of 61,780 insiders are analysed, which are acquired from Thomson Financial, covering a period of January 2010-April 2017. In order to detect outliers, similar transactions are grouped into the same clusters by using a two-step clustering based outlier detection technique, which is an integration of k-means and hierarchical clustering. Then, it is shown that outlying transactions earn higher abnormal returns than non-outlying transactions by using event study methodology.