Performance Measurements for Machine-Learning Trading Systems

Abstract

This study presents several insightful performance measurements for machine-learning trading systems. Machine-learning trading platforms are presumed to operate in a continuous time domain, whereas their learning configurations prompt recurring yet bounded improvements over time. The study provides practical estimation guidelines for the relevant parameters and further illustrates the functionality of the proposed scheme through some conjectural examples. The recommended performance measurements aim to help internal auditors of trading departments and regulatory institutions to better track errors at these automated systems.