|
PureBytes Links
Trading Reference Links
|
My opinion:
Best to employe a set of artificial data comprised of sections of actual
history which has then been filtered thru a series of range-limited
random-number generators to change both the volatility and the trend. To
reverse the secular trend, simply invert the from-to time period.
A goal of reduced input parameters is simply a function of developing
correlation coefficients for them based on multi-variant analysis of the
inputs and the related output (net profit, win ratio, etc.).
Many times, the initial large set of parameters can be reduced significantly
by converting them to formulas internally.
> -----Original Message-----
> From: Mel [mailto:melsmail@xxxxxxxxxxx]
> Sent: Monday, July 10, 2000 5:35 AM
> To: OmegaList
> Subject: Bias in Testing
>
>
>
> Testing on back adjusted data (going back 18 years) it is not too hard to
> come up with systems that appear to do well particularly in a long term
> rising scenario. Question is how can I have faith in results
> where the early
> back adjusted prices are significantly higher with relation to the real
> prices at the time. Obviously any optimization of any system will have
> significantly different weighting with relation to the real prices at the
> time. How can this weighting be equalized - would detrending of the back
> adjusted data reduce/eliminate any bias or should one incorporate a data2
> (real contract) price in the system testing? My goal is to reduce
> the number
> optimization inputs to a minimum so that as a robust system I can
> trust the results. Any thoughts?.
>
>
> Mel Fox
>
>
>
|