System Optimization Techniques
by David S. Nicol
What is system optimization and is it good or bad? It's a little of both, system designers will tell you.
David Nicol explains the whys and wherefores.
System optimization is a controversial subject among market analysts, but it is also misunderstood by
analysts and traders. So what is system optimization? When designing a technical trading system, the
designer must use a mathematical formula or algorithm of some kind to produce trading signals. This is
usually referred to as an indicator and can be as simple as a single moving average or as complex as a fast
Fourier transform. Whatever the indicator may be, all indicators have one characteristic in common: they
all contain one or more parameters. A parameter can be set to one of several possible numeric values, the
value of which can greatly alter the behavior and performance of the indicator. Because parameters exist
in every indicator, the system designer must set each parameter to some value . For each system designer,
the dilemma is what value to assign each parameter. That is where optimization occurs. Optimization is
the process of choosing the best parameter values for the indicator. So what do we mean by "best
FINDING THE BEST VALUE
One approach to finding the best parameter value is to backtest the indicator with a range of parameter
values and simply choose which parameter value produced the most profitable trading signals . This
approach has several problems. First, how far back should the testing go? Should the parameters be
changed during the test run? If so, when and how? The way that the system designer handles these
questions will greatly affect the validity and utility of the optimization process.
However, those systems that are optimized without the benefit of hindsight experimentation because of
the fear of curve-fitting are not properly optimized because no effort has been made to verify the validity
of their parameters. This process is simply choosing parameters randomly. This is the dart-board method
to optimization, because money is risked based on parameters that could have been chosen by the toss of a dart.