|
PureBytes Links
Trading Reference Links
|
Bob - I agree, but wow, the amount of logic and complexity with this
approach is significant....
i.e., what measure to use to determine a change is required ?; how does that
measure correlate to changes in the parameters of the system ?; how often
should the system make a change ? How will it know exactly which parameters
to change ? Does the system need to maintain multiple sets of "books" in
order to be able to effectively revert back to it's original parameters ? Or
should the initial parameters be foresaken forever, never to be revisited ?
All good questions.....and this approach certainly flys against the
"simple-is-better" philosophy espoused many times on this list.
> -----Original Message-----
> From: Bob Fulks [mailto:bfulks@xxxxxxxxxxxx]
> Sent: Thursday, December 20, 2001 2:28 PM
> To: rascal2
> Cc: omega-list@xxxxxxxxxx
> Subject: Re: Optimization Question
>
>
> At 1:25 PM -0500 12/20/01, rascal2 wrote:
>
> >I know that theoretically this should not work, but in fact it
> seems to work
> >pretty well. His system consistently turns a better profit with smaller
> >drawdowns that it would when optimized over a longer period and then run
> >against the same data sets.
> >
> >I'd be interested in any thoughts anyone might have on this approach.
>
> Unfortunately, you do not give enough information about what he
> is doing to be able to comment intelligently.
>
> If his optimization runs show very erratic values vs. each input
> parameter and he happens to just be picking some combination of
> input values that give the best numbers on that data, he is
> simply curve-fitting and it will not work consistently.
>
> But if the plot of results vs. each input parameter is a nice
> smooth curve with broad maxima, then he may, in fact, be adapting
> his system for some gradually shifting characteristic of the
> market. That would tend to improve the performance week-by-week
> until some point where the market characteristic changes quickly
> and then he could lose a lot of money until it settles out again.
>
> If he is adapting manually, a better approach would be to figure
> out what characteristic he is "chasing" and provide code that
> will automatically measure that characteristic and adjust the
> parameters for him.
>
> Bob Fulks
>
>
|