Machine-Learning-Generated Synthetic Data for Regulation Best Interest Compliance

12 Pages Posted: 11 Jul 2019

See all articles by J.B. Heaton

J.B. Heaton

J.B. Heaton, P.C.

Jan Witte

University College London - Department of Mathematics

Date Written: July 9, 2019

Abstract

Regulation Best Interest requires broker-dealers to understand potential risks, rewards, and costs associated with recommendations to retail customers. For those charged with this legal duty, historical financial data provides little comfort. There is too little historical financial data and too many ways to overfit it with investment strategies that may have poor risk-return performance in the future. We develop a method for generating synthetic data to test investment products for investor best interest. We illustrate our method by showing that the Dow Jones Industrial Average is highly outlier-dependent, such that forecasts of the strategy's future performance based on historical data would be overstated by over 80%. Our method can better screen out investment recommendations that are unlikely to be viewed by a later factfinder (in litigation/arbitration or a regulatory enforcement action) as not being in a retail client's best interest.

Keywords: Synthetic Data, Machine Learning, Best Interest, Compliance

Suggested Citation

Heaton, J.B. and Witte, Jan, Machine-Learning-Generated Synthetic Data for Regulation Best Interest Compliance (July 9, 2019). Available at SSRN: https://ssrn.com/abstract=3417199

J.B. Heaton (Contact Author)

J.B. Heaton, P.C. ( email )

20 West Kinzie
17th Floor
Chicago, IL 60654
United States
(312) 487-2600 (Phone)

HOME PAGE: http://jbheaton.com

Jan Witte

University College London - Department of Mathematics ( email )

Gower Street
London, WC1E 6BT
United Kingdom

Register to save articles to
your library

Register

Paper statistics

Downloads
7
Abstract Views
33
PlumX Metrics