[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Robust Parameter Selection



PureBytes Links

Trading Reference Links

Hello Joseph,

A subject dear to my heart...

My approach is somewhat agricultural, and goes like this:

I am assuming that one parm is an entry stop and one probably relates
to a trailing stop, leaving 4 to fool with. (Which is really too many,
but often only one or two parms really affect testing significantly
anyway).

Run a few tests to find starting points for each parm, say $1250 for
entry stop, whatever for trailing, a nominal length (if there is one)
and so on, divide your test data into suitable overlapping time
slices, then laboriously test each parm, one at a time, over each time
slice and chart the results in your favorite spreadsheet.

Be meticulous about this, and save each chart. I would suggest
charting something like Expectancy, Profit Factor and maybe Average
Trade $, whatever you prefer (mainly because I never really understood
Sharpe anyway). 

What you are looking for is a stable curve of results that also
remains relatively stable over time (hence the overlapping time
slices) which I believe represents a robust parm.

By stable I mean that there is no dramatic spiking of results, just a
steady curve.

This I believe is referred to as walk forward testing.

Done correctly this is a time-consuming endeavor, particularly if
done over many markets. With a little imagination much can be
automated.

Hope this helps.

-- 
Best regards,
 Ross                       mail to: Ross.Bond@xxxxxxxxxx

Friday, September 12, 2003, 5:07:59 AM, you wrote:

SJE> What would be a practical procedure to find the most robust parameter set
SJE> for a system that has 6 parameters?  For arguments sake, lets assume the
SJE> only result of each optimization run is the Sharp Ratio.  

SJE> Any thoughts on the subject would be greatly appreciated.

SJE> Thank you,

SJE> Joe