Evaluating Algorithmic Trading Signals by Giorgos E. Siligardos, PhD
Ahead Of The Curve
Are you using an objective, unbiased, and globally applicable method to evaluate the performance of your algorithmic trading signals? If not, then this article may be your cup of tea.
There are two main reasons why algorithmic trading systems (ATS) are favored by many traders. The first is the absence of emotions in the world of computers. The second is that when an ATS is organized properly, it maximizes the earnings/labor ratio — a direct application of the “If you work for a living, don’t kill yourself working” adage from the 1966 movie The Good, The Bad And The Ugly.
In this article, I will first explain the need to differentiate the evaluation of the trading signals of a system from the evaluation of the system as a whole. I will then elaborate on the most appropriate method to construct an unbiased performance curve.
TRADING SIGNALS VS. TRADING SYSTEM
Every ATS is eventually based on two pillars. The first one provides the type of orders (buy/sell, sell short, cover) and the time those orders should be taken (that is, timing). The second pillar provides the amount (volume) of the tradable to trade. When your testing is a mixture of these two pillars, you can’t be sure of the reason behind the profitability of the results. In other words, there is a key difference between testing the performance of your algorithmic trading signals and testing the performance of a system that trades those signals with specific money allocation rules.