|
PureBytes Links
Trading Reference Links
|
Hi Hal,
Instead of 4851 tries for nearly 1000 days of data you can
make two runs of 100-150 steps.
I begin with the crude
/*FIRST APPROXIMATION*/
X1=Optimize("X1",14,10,20,2);
Dbuy=Optimize("Dbuy",15,0,20,5);
Dsell=Optimize("Dsell",10,0,20,5);
s1=StochD(X1);
AVST=MA(s1,100);
Buy = Cross( s1,AVST-DBuy);
Sell = Cross( AVST+Dsell,s1);
for the last 200 days (I do not think you need more)
For 400 stocks it took less than 2.5 min with a
P3/800MHz. The optimum of this crude approximation
was
X1=12, DBUY=15, DSELL=0
Then I run the next, more detailed search close to the first
solution, ie
X1 in [10,14], Dbuy in [13,17] and Dsell in [0,4]
/*DETAILED APPROXIMATION*/
X1=Optimize("X1",14,10,14,1);
Dbuy=Optimize("Dbuy",15,13,17,1);
Dsell=Optimize("Dsell",3,0,4,1);
s1=StochD(X1);
AVST=MA(s1,100);
Buy = Cross( s1,AVST-DBuy);
Sell = Cross( AVST+Dsell,s1);
In less than 2.5 min I found the
X1=12, DBUY=14, DSELL=1
Take a look now at the interesting part of the search :
The last 200 days in my Market the majority is between
-25% up to -50% lower .
The crude approximation had an optimum Net % profit -4.5%
The detailed approximation improved to -3.3%
The conclusion was obvious: There is not a good "universal"
combination of X1, Dbuy, Dsell for the whole Market.
There are stocks with Net % profit more than 100% and this is
the target.
I do not know a general method to filter out these stocks, so I keep
on searching.
Dimitris Tsokakis
--- In amibroker@xxxx, Hal Brehe <infoads@xxxx> wrote:
> Hi Demitris,
>
> Thanks for your interest. Here is the "works".
>
> /*To Optomize*//*
> Dbuy=Optimize("Dbuy",14,0,20,1);
> Dsell=Optimize("Dsell",8,0,20,1);
> X1=Optimize("X1",13,10,20,1);
> s1=StochD(X1);
> AVST=MA(s1,100);
> Buy = Cross( s1,AVST-DBuy);
> Sell = Cross( AVST+Dsell,s1);*/
>
> Regards,
>
> Hal
>
> At 05:43 AM 10/26/01 +0000, you wrote:
> >Hi Hal,
> >Is it possible to give us the parameters and the formula
> >of your Optimization ?
> >Best Regards
> >Dimitris Tsokakis
> >--- In amibroker@xxxx, Hal Brehe <infoads@xxxx> wrote:
> > > Hi Dimitris,
> > >
> > > I'm using your Relative Slope system, and I thank you for this
> >wonderful
> > > gift to the group.
> > >
> > > Everything is working well with the system, except that my NYSE
> >database is
> > > too large to be practical to use with the Optimization feature.
> >This is
> > > because the total continuous optimization time is approximately
72
> >hours or
> > > 3, 24 hour days. With the Optimization system requiring 4851
> >tries, at the
> > > approximate half way mark, the actual optimization time is 90
> > > seconds/ticker, and the time between starts at 8 seconds &
getting
> >longer
> > > (this latter time is the time between the end and beginning of
the
> >next
> > > run). I assume it is AB finding the next "try" information).
This
> >timing
> > > was done on a 900MHz P4, a subset of the total record set
> >accomplished by
> > > making a "scan" for upper and lower price limits which were
stored
> >as a
> > > WATCH LIST (around 2300 tickers), and maximum data record
length of
> >3-1/4
> > > years.
> > >
> > > It is obvious to me that the total time requirement for the
> >Optimization
> > > run is around 72 hours, or three continuous days.
> > >
> > > In thinking over what would help, I believe that a ticker
reduction
> >scheme
> > > would be an asset. That is to say, make a preliminary "SCAN" to
> >reduce the
> > > number of tickers. Make a "Watch List" of the result, and make
the
> > > Optimization run on the Watch List. It might be some scheme to
> >eliminate "1
> > > out of X" tickers or possibly be made "random" rather than
> >sequential. I donno!
> > >
> > > Could you, or anyone else interested, get me started in
developing
> >this
> > > type of routine?
> > >
> > > Any assistance would be greatly appreciated.
> > >
> > > Regards,
> > >
> > > Hal
> >
> >
> >
> >
> >
> >Your use of Yahoo! Groups is subject to
http://docs.yahoo.com/info/terms/
|