Complex Systems and Goodman's Law of Trading Systems

Complex systems involve up to a dozen or more indicators and a rule set that can be followed only with a computer program. As you experiment with the simple system, changing the inputs and adding new indicators and rules, consider the following critique.

Indicator parameters and filters are a type of what is called curve-fitting. Given enough jiggling of parameters, filters, and a few indicators, I can find a set of rules for the system or for the money management scheme that will generate an enormous profit and lose very little over a given data set.

The question is—will it work in the future as well as it did in the past? The answer is no. Because the filters and rule set are curve-fitted so tightly to that specific data, a small change in the market environment will quickly change the winning system to a losing system. This is why so many systems look good on paper, but once they go live, disaster strikes.

It is truly best to keep it simple when developing a trading method based on indicators. If it requires too many parameter adjustments and additional filters and rules, it is too delicate and will break hard and probably quickly once it is set to real-time trading. This author has long been skeptical of any trading method based on curve-fitting, even if it is a minor component such as in Ichimoku clouds or Drummond point and line. Why should parameters that worked over a limited data set continue to work over a different data set?

Tip: Always ask two questions: (1) What is this indicator measuring? and (2) What is the buyer-seller tug of war logic to this indicator?

Goodman's Law of Trading Systems

This is all summarized here. I take time to restate it because I have seen too many new traders afflicted with the curve-fitting disease in a search for the free lunch trading system.

The more complex the curve-fitting, the less likely the system will be successful long-term, real-time.

This is because the complex system depends highly on the very specific characteristics of the data set over which it was designed. A small change in those specifics in the future will cause the system to run aground.