The Application of Machine Learning Algorithm to SRO's Fraud Detection and Fine Prediction

Dr. Lokanan received a SSHRC Insight Development Grant to study the MFDA’s ability to detect probability of fraud before it occurs and to ensure that penalties are proportionate to the harm caused.

The Mutual Funds Dealers Association of Canada (MFDA) is a self-regulating oversight institution that is responsible to regulate and police the Canadian mutual fund industry as it relates to the sale of mutual funds. Of late, the MFDA has been accused of light touch regulation for its failure to litigate and prosecute mutual fund dealers who peddle investments and other cases of financial misconduct. Central to these concerns are issues related to fraud detection and inconsistent application of financial penalties for rule violations. All of this is to say that the MFDA has not been successful in having a deterrent effect due to weak fraud detection and inconsistent application of financial penalties.
These concerns have led to calls for a self-regulatory organization (SRO) framework that would work to enhanced investors’ protection and to streamline the investment industry regulatory regime. To heed these calls, the Canadian Securities Administrators (CSA) in June of 2020 released the “CSA Consultation Paper 25402 Consultation on the Self-Regulatory Organization Framework” seeking input from relevant stakeholders to review the current regulatory framework of the Investment Industry Regulatory Organization of Canada (IIROC) and the MFDA. A key part of the consultation is to examine the existing framework of IIROC and the MFDA to create a single more powerful SRO that would work to enhance investors’ protection. Given that the MFDA is the oversight body for certain aspects of market operations, this study will employ machine learning algorithms to evaluate the MFDA’s ability to detect the probability of fraud before they occur and to ensure that penalties imposed for rule violations are proportionate to the harm caused to investors.
The proposed study makes several important contributions to the literature on financial market regulation. First, in the context of regulatory enforcement, machine learning algorithms can be leveraged to reduce false positives and improve fraud detection and consistencies in the application of fines. Fraud detection can be more effective when machine learning can be used to build algorithms that can receive input data and use statistical analysis to predict the probability of fraud from new entries.
Second, the application of inconsistent fines has been a problem for SROs operating in the securities industry. One of the issues is that MFDA's hearing panels rely on precedents to set fines; but, quite often, the material facts of the cases are fundamentally different, which leads to inconsistencies in fines levied on registered representatives. Improvement in computing technology has made it possible to build new analytical techniques that can increase the accuracy in the imposition of fines. The objective here is to add some science to the fine imposition process and build an end-to-end solution that is capable of predicting fines imposed on MFDA’s registered representatives. The ultimate goal is to build a supervised predictive machine learning model that will help regulators to make data-driven decisions to predict the proportionate fines for offenses.
Third, the study has the potential to provide new insights and inform the current consultation process on SRO reforms. Successful self-regulation in Canadian finance is important because government regulation is so completely ineffective. Canada is unique in having its "patchwork" system of inept provincial regulators. It is also notable for lax criminal enforcement for crime in the sector. As such, there is a clear need to better understand the efficacy of SROs in the face of enforcement of securities fraud and transgression in financial markets/securities trading in Canada.