Parameter Selection

"Select parameters for your system
which give you the best results."

The method traders use to evaluate system rules is intended to explore their potential and suitability for trading.

This partly involves testing against historical prices for all the commodities on the trader's 'list' and over an extended time period, normally somewhere between 5 and 10 years.

Nearly all system rules work by using variable values called parameters that provide the detailed information for generating signals to commence or terminate trades.

For example, values may be required for moving averages that the system uses.

So how do you find sensible values for these parameters?

The logical thing would seem to be to locate the values that give the best results and select those as the ones you would use to trade the system.

But that is not the way.

Surely selecting what gives you the best results can't be wrong?

Yes it can be wrong and it is wrong. Within your price history data there are sure to be some exceptional movements. Any 'massive jumps' or unusual price behavior around these points is likely to show highly profitable trades.

Selecting these price 'spikes' when choosing parameter settings does not make a great deal of sense because they are caused by unexpected one-off situations that would be most unlikely to repeat.

Selection based on 'best results' is an example of what is known as curve fitting - fitting the results to the data.

Curve fitting produces misleading results and is a common mistake made by newcomers to commodity trading. Often, it is done accidentally, which is why traders need to guard against it at all times.

So what is the right way to search for parameter values?

The right way is to 'look at' all the possible values and select within ranges that demonstrate robust results i.e. that contain areas yielding similar results.

This can involve a very large number of runs (thousands or perhaps tens of thousands) and a very large quantity of results to look through!

Because of the amount of work involved in doing this job thoroughly, it is often neglected or skimped.

There are special techniques for doing this work that make it much easier and more reliable. Also, the quality and efficiency of the software used has a major bearing on the trader's ability to do the work properly.

Traders should not select parameters which give the best results but those that reveal robust results and are located among similar neighbors.

 

Modus Trading
Contact David

Copyright David Bromley 2006
All Rights Reserved.

 

 

 

 

[Close This Window] Sample content here.