|
PureBytes Links
Trading Reference Links
|
Right on track and I understand searching for a different strategy when one
stops working. A few years ago while in the development stages I found that
I had hundreds of systems that did well during various time frames. So I
built a spreadsheet to compare them on a daily basis for 28 years of history
but could only do 9 at a time. It took about 3 months to make it do what I
wanted it to do to allow me to pick the 9 "best" systems out of what really
turned out to 65 acceptable systems none of which were good all the time but
all had their moments where they were very good. The spreadsheet logic was
based on the equity curve of each system and the idea was to switch to the
best system when the current one was failing. I was surprised to find that
that method did not work and in fact when it looked like the current system
had been doing well and had just about peaked out on a relative equity
curve, select the system with the worst recent performance produced better
overall results. I abandoned that spreadsheet and ended up modifying it to
take 9 system results day by day from tradestation and go thru time giving
each of the system results a different weight and build a composite buy/sell
signal.
That work fine but again it was just another gigantic curve fitting process
as it did not work well on future data. I was about to turn it into a
walkforward system but TS2000i came out so I start working on putting all of
my spreadsheet logic into TS2000i since the memory constraints were gone.
But I ran into trouble moving forward and backward in time with easylanguage
because I use xaverage and it relies on an accumulation of prior bars.
That's why I now use Winbatch.
This thread has been very informative about what seems to be some kind of
"footprint" which can be used to describe what could be called the current
environment. And the concept and being able to identify a changing footprint
is also very good. But I don't know how to do it. Hearing other people talk
about implying that it works and is possible helps me to keep on searching.
Sometimes I think that this mechanical trading stuff is impossible but I
continue to try because I am lazy and somehow I think that I can create a
system that will work now and in the future so that I can stop this search
and let the computer tell me when to buy and sell so that I can have a life
or something.
----- Original Message -----
From: "M. Simms" <prosys@xxxxxxxxxxxxxxxx>
To: "Jon Woods" <justus5@xxxxxxxxxxxxx>
Sent: Friday, May 04, 2001 7:20 PM
Subject: RE: Trading Systems
> Yes......tell me if I am on the right track with your thinking:
>
> .....2 levels of adaptation.....
> 1) internal...the code making minor adjustments in stoploss settings, and
> entries based on previous trades
> 2) external...more drastic adaptation based on different "sets" of the
> parameter values to accomodate for extreme changes in the
marketplace...from
> trendy to choppy....volatile to quiet, etc.
>
> I've often wondered whether #2 above should just be a whole set of
different
> strategies instead of one, huge adaptive one with several parameters.
> Thus, if one strat quits working, just switch to one which "fits" the
> current market better.
>
> > -----Original Message-----
> > From: Jon Woods [mailto:justus5@xxxxxxxxxxxxx]
> > Sent: Friday, May 04, 2001 5:17 PM
> > To: prosys@xxxxxxxxxxxxxxxx
> > Cc: omega-list@xxxxxxxxxx
> > Subject: Re: Trading Systems
> >
> >
> > yes, reasonably stable or understandably stable.
> > as a byproduct of the yeartoyear analysis I am trying to associate the
> > varying of the parameters with the trendiness and trendlessness and the
> > transitions among them as mentioned in prior posts in this thread. I am
> > searching for numeric patterns of my parameters to assist in what
> > could then
> > be used as a further level of adaptiveness.
> > In other words, when I get stopped out and switch to a new set of
> > parameters, maybe I can have the system chose from more than one set of
> > alternate parameters and have that choice be based on what has
> > been working
> > but has suddenly or slowly stopped. I think I would rather have a system
> > quit working quickly than to have one slowly deteriorate and cook me
like
> > the story of the frog in water that is slowly brought to a boil.
> >
> > ----- Original Message -----
> > From: "M. Simms" <prosys@xxxxxxxxxxxxxxxx>
> > To: "Jon Woods" <justus5@xxxxxxxxxxxxx>
> > Sent: Friday, May 04, 2001 4:07 PM
> > Subject: RE: Trading Systems
> >
> >
> > > Wow - pretty impressive use of macros.....
> > >
> > > So I guess your goal in this is to insure the starting
> > parameters that are
> > > optimal remain stable
> > > from year-to-year testing periods ?
> > >
> > > > -----Original Message-----
> > > > From: Jon Woods [mailto:justus5@xxxxxxxxxxxxx]
> > > > Sent: Friday, May 04, 2001 12:06 PM
> > > > To: prosys@xxxxxxxxxxxxxxxx
> > > > Subject: Re: Trading Systems
> > > >
> > > >
> > > > 1) I used the demo version of winbatch. then recorded keystroke
> > > > macro to set
> > > > the dates to from 3/18/1970 to 1/1/1972 (200 max barsback)
> > then I called
> > > > another keystroke macro to run the optimization. While the
> > optimization
> > > > runs, the main macro trys to open the view of the optimzation
results
> > and
> > > > wait 10 seconds and repeats until the optimization results window
> > > > opens and
> > > > that indicates the optimization is done. I then build a file name
with
> > the
> > > > "to" year in and save the workspace as like wf1972.orw then the from
> > year
> > > > and to year are incremented by 1 until the to year is 2001.
> > > > 2) It takes about 12 hours to do the run and 3 months would
> > > > therfore take 48
> > > > hours. Guess I need to be more patient. After the run, I go
> > through each
> > > > year and (not with winbatch yet) open each performace window and
> > > > save it as
> > > > an excel file and then look at all the parameters with excel. The
test
> > > > interval I don't think has much impact on the adaptivity other
> > > > than to find
> > > > out if the adapting is working. MY adaptiveness come from having
> > > > developed a
> > > > fairly go system that exited long and short positions when the
> > > > trade went a
> > > > certain percent away from the entryprice. The adaptiveness comes
into
> > play
> > > > by using 2 sets of parameters, one for normal and the other for when
I
> > get
> > > > stopped out.
> > > > 3) The premise of my system is the relationship of today's
> > high and low
> > > > versus historical closes ( I trade profunds ultrabull and ultrabear
at
> > > > almost end of day). And if you plot the following on dialdata spx
> > endofday
> > > > data xaverage(jdiv(c-l,h-l),150) it should show that from april 1985
> > thru
> > > > july 2000 that number was above .5.
> > > > Prior to 1985 it went above and below .5 then it stayed above .5 for
> > more
> > > > than 14 years. That's all I could find that seemed to be related
> > > > to my code.
> > > >
> > > > ----- Original Message -----
> > > > From: "M. Simms" <prosys@xxxxxxxxxxxxxxxx>
> > > > To: "Jon Woods" <justus5@xxxxxxxxxxxxx>
> > > > Cc: "Omega-List" <omega-list@xxxxxxxxxx>
> > > > Sent: Friday, May 04, 2001 10:32 AM
> > > > Subject: RE: Trading Systems
> > > >
> > > >
> > > > > 1) Exactly how did you employe Winbatch to assist you hear ? this
is
> > > > > interesting....
> > > > >
> > > > > 2) I vote "don't trade" until you discover the optimal
> > "out-of-sample"
> > > > test
> > > > > interval.....
> > > > > which you say is one year. Why not try 9,6, and then 3 months ?
> > > > >
> > > > > 3) most importantly: have you determined WHY it failed ?
> > > > >
> > > > > Note: if the system were truly adaptive, then the shorter
> > optimization
> > > > > periods should have been determined automatically.
> > > > >
> > > > > > -----Original Message-----
> > > > > > From: Jon Woods [mailto:justus5@xxxxxxxxxxxxx]
> > > > > > Sent: Friday, May 04, 2001 10:20 AM
> > > > > > To: omega-list@xxxxxxxxxx
> > > > > > Subject: Re: Trading Systems
> > > > > >
> > > > > >
> > > > > > I optimized my TS2000i adaptive "H L C only" SP500 strategy from
> > 1970
> > > > thru
> > > > > > June of 2000.
> > > > > > I traded it from June 2000 to Feb 2001.
> > > > > > I got killed. The system didn't work.
> > > > > > So then I used Winbatch to optimize each year individually.
> > > > > > Then I took the parameters for each year and used them for the
> > > > > > NEXT year and
> > > > > > used those results to evaluate the performance.
> > > > > > Results = No losing years and all years beat the SP500.
> > > > > > My conclusion is that my strategy seems to be adapting.
> > > > > > But I would like to hear comments as to whether I should trade
it
> > > > > > now or do
> > > > > > some more testing.
> > > > > > I vote to trade it now.
> > > > > >
> > > > > >
> > > > > >
> > > > > >
> > > > >
> > > > >
> > > >
> > >
> > >
> >
>
>
|