Evaluating Algorithmic Performance

Abstract

This article outlines some of the challenges in evaluating algorithmic performance. The root of the problem is the lack of a good clean data set that would enable apples-to-apples comparison between various algorithms. Given the data limitations, traders should exercise caution before accepting any algorithmic performance study at face value. The article also offers some suggestions on ways traders can get a better understanding of the algorithms they are using with the data they have.

TOPICS: Statistical methods, exchanges/markets/clearinghouses, performance measurement