|
PureBytes Links
Trading Reference Links
|
Hello pastal,
If I had no hypothesis about parameters, 12 months of data (Jan thru
Dec), max of one
month at a time to "optimize" I'd approach it something like this:
--pick two months, say March and September, and optimize each.
--inspect the results to find a set a paramenters that perform well in
each time set. I'd be more confident is the ones I chose were
accompanied by "nearby" values. E.g. say lengths between 30 and 35
consistently tested well in both time sets. then maybe I'd pick 32 as
my hypothesized best parameter.
-- I'd run the same optimization (possibly tighter ranges) on five
other months interspersed thru the year and expect
to see 30-35 consistently be profitable.
--if so, you could "add" the system reports together to make a
composite performance report over the entire 12 months--if you need
it.
--if not profitable, I'd pick another parameter from my initial set,
and cross-validate with the last 5 months I've held out. If I find
nothing, I'm on the next idea.
Best regards,
Jim Johnson mailto:jejohn@xxxxxxxxxxx
--
Wednesday, August 14, 2002, 11:51:09 PM, you wrote:
AM> Paolo wrote:
>>does not load more than 75-days data, I am obliged to optimize
>>every month and in spite of this the system has good results. Is there
>>anything wrong in all that?
AM> No, there is nothing wrong with that. I imagine many trading systems
AM> would benefit from periodic re-optimization, because markets change
AM> their characteristics as time goes on.
AM> I have noticed this with Swinger2-EOD, for example. Post 9/11, it works
AM> consistently a lot better with different parameters.
AM> -Alex
|