PureBytes Links
Trading Reference Links
|
Hi MG Ferreira:
I need to take some time to test a lot of variables, such as PC amd
SM. Due to the limitations of Metastock, it limits 20 variables for
maximum. Will it be ok to determine each indicator's PC and SM in
following approach, such as
For the inital stage, the PC and SM for all indicators are
determined, which is PC=85 [0-100] and SM=10 [0-100] and
PC=240 [0-500] and SM=15 [0-200], then I pick PC=240 and SM=15 to be
the default value for all indicators, then
when I determine PC and SM for W1
STEP 1
PC=??? and SM=?? for W1 [after evaluation PC=200 and SM=10]
PC=240 and SM=15 for W2 [default value]
PC=240 and SM=15 for W3 [default value]
PC=240 and SM=15 for W4 [default value]
STEP 2
PC=200 and SM=10 for W1
PC=??? and SM=?? for W2 [after evaluation PC=250 and SM=20]
PC=240 and SM=15 for W3 [default value]
PC=240 and SM=15 for W4 [default value]
STEP 3
PC=200 and SM=10 for W1
PC=250 and SM=20 for W2
PC=??? and SM=?? for W3 [after evaluation PC=80 and SM=5]
PC=240 and SM=15 for W4 [default value]
STEP 4
PC=200 and SM=10 for W1
PC=250 and SM=20 for W2
PC=80 and SM=5 for W3
PC=??? and SM=?? for W4 [after evaluation PC=130 and SM=7]
RESULTS
PC=200 and SM=10 for W1
PC=250 and SM=20 for W2
PC=80 and SM=5 for W3
PC=130 and SM=7 for W4
Does it cause any problem if I determine each PC and SM in this
approach? I assume that each indicator has his own learning rate, so
the PC and SM for each indicator will not be influenced each other.
Am I right?
On the other hands, in term of managing indicators, divergence
indicator is hard to manage too, since it is a leading indicator and
has no clearly defined buy and sell signal. There is no problem on
normalizating it, but can divergerence indicator be added into the
pool too? such as, if RSI(14) divergence occurs, and I assume that
trend could be changed in the near future, but there is no way to
know how long, then the period between the first day for divergence
[first] and the first day for changing in trend [later] will affect
the weight's performance if I do the same approach as what I have
done. Do you have any suggestion on managing divergence indicator?
Thank you
Eric :>
--- In equismetastock@xxxxxxxxxxxxxxx, mgf_za_1999 <no_reply@xxxx>
wrote:
> Hi Eric,
>
> I am glad you are smiling again!
>
> In terms of the ratio between the weights. PC will not influence
it a
> lot and SM will only influence it in the shorter term. Since the
> weights adjust themselves, a weight of 3% means in effect that that
> indicator has been discarded from the system. So you do not really
> have to worry about physically taking it out - it is out in
practise!
>
> Also, I'd be hesitant to change a system too much based on
hindsight.
> I am sure superf can write many comments on such an approach! You
> can now think of adding more indicators to the system. Maybe also
> change the range on PC from 0-100 to 0-200 or even 0-500 and play
> around with some of the bigger numbers, say 200. But there is not
too
> much in there - the next step is to have individual PC and SM
values,
> but the complexity it adds I think outweigh the benefit it
delivers.
> By the way, I think the system now resembles a neural network to
some
> extent.
>
> Since weights are adjusted dynamically they will change over time.
> This is good and it fades different indicators in and out as they
> deliver or fail to deliver performance. The only thing that can
> happen, and probably will happen a lot, is that a particular
indicator
> be faded out completely (like 3% weight). It is also common to
have a
> lot of indicators fade out and just one or two dominant ones remain.
>
> We sometimes force indicators to remain present in a system even if
> they don't work. You sometimes want to retain an indicator to keep
> the system from being overoptimised even if it does not seem to add
> value. This depends on the optimisation process too, so I am not
sure
> how it would apply here. Again, we are *not* optimising since we
> adjust weights on observed performance and we are not iterating over
> the whole sample. To do this you need the individual PC and SM
> variables, so I am not sure we really want to go that far here.
>
> Anyhow, keep us informed of the system's developments.
>
> Regards
> MG Ferreira
> TsaTsa EOD Programmer and trading model builder
> http://www.ferra4models.com
> http://fun.ferra4models.com
>
>
>
>
>
>
> --- In equismetastock@xxxxxxxxxxxxxxx, chichungchoi <no_reply@xxxx>
wrote:
> > Hi MG Ferreira:
> > I have tested it, and select 85 for PC and 10 for SM.
> > In term of ratio between 3 indicators, what should I need to pay
> > attention? such as
> > CASE 1
> > W1 = 10%
> > W2 = 30%
> > W3 = 60%
> > TOTAL = 100%
> >
> > CASE 2
> > W1 = 33%
> > W2 = 33%
> > W3 = 34%
> > TOTAL = 100%
> >
> > CASE 3
> > W1 = 3%
> > W2 = 17%
> > W3 = 80%
> >
> > In above samples, do you have any idea on how to handle or adjust
it
> > when the above situation occur? such as replacing a better
indicator
> > for the worst one (3%), adjusting inital weights ... etc.
> > Do you have any suggestion please?
> > Thank you
> > Eric :>
> >
> >
> >
> > --- In equismetastock@xxxxxxxxxxxxxxx, mgf_za_1999
<no_reply@xxxx>
> > wrote:
> > > Hi Eric,
> > >
> > > The second version is correct, with the small numbers. It
looks
> > fine
> > > and it does not matter that the weights are small, since they
are
> > all
> > > small. It is the same thing to use W1 of say 1000 and W2 of
2000,
> > > since relatively we use twice as much of W2 as of W1. Now this
is 1
> > > and 2. It could also be say 0.001 and 0.002, as long as the
> > relative
> > > relationship is maintained. Also, we now increase it by a
> > percentage,
> > > so again the size does not matter. If you increase 0.001 or 1
or
> > 100
> > > or 1000 by 10%, it all will increase by the same relative
amount.
> > >
> > > The PC and SM values become more important now. What are you
> > > currently using for PC? I think it needs to be bigger (say
enter 90
> > > or 100) rather than smaller.
> > >
> > > Regards
> > > MG Ferreira
> > > TsaTsa EOD Programmer and trading model builder
> > > http://www.ferra4models.com
> > > http://fun.ferra4models.com
> > >
> > > --- In equismetastock@xxxxxxxxxxxxxxx, chichungchoi
<no_reply@xxxx>
> > wrote:
> > > > Hi MG Ferreira:
> > > >
> > > > DR:=(C/Ref(C,-1)-1)*C;
> > > > P1:=DR*((T1>0)-(T1<0));
> > > > Sample for P1:
> > > > 04-27-05 -19.9116
> > > > 04-28-05 70.1324
> > > > 04-29-05 -0.4502
> > > > 05-03-05 14.9729
> > > > 05-04-05 -51.2578
> > > > 05-05-05 -117.6266
> > > > 05-06-05 27.6853
> > > > 05-09-05 0.0000
> > > > 05-10-05 66.3936
> > > > 05-11-05 78.9304
> > > > P1 is the penalty function, if I redefine the daily return to
be
> > > > DR:=(C/Ref(C,-1)-1), then
> > > > Sample for P1
> > > > 04-29-05 -0.0000
> > > > 05-03-05 0.0011
> > > > 05-04-05 -0.0037
> > > > 05-05-05 -0.0084
> > > > 05-06-05 0.0020
> > > > 05-09-05 0.0000
> > > > 05-10-05 0.0047
> > > > 05-11-05 0.0057
> > > > which seem solve fit the range, and the W1 is more stable,
> > > > Sample for W1
> > > > Loading 2000 days for data, current W1 = 1.156153
> > > > Loading 1000 days for data, current W1 = 1.055910
> > > > Loading 500 days for data, current W1 = 1.023015
> > > > Does it what you expect for the result?
> > > >
> > > > Am I on the right track to solve this problem by redefining
the
> > daily
> > > > return? Learning seems a little bit too slow, isn't it? since
the
> > > > weight took a long time to adjust a little bit.
> > > >
> > > > Thank you
> > > > Eric :>
------------------------ Yahoo! Groups Sponsor --------------------~-->
Has someone you know been affected by illness or disease?
Network for Good is THE place to support health awareness efforts!
http://us.click.yahoo.com/Rcy2bD/UOnJAA/cosFAA/BefplB/TM
--------------------------------------------------------------------~->
Yahoo! Groups Links
<*> To visit your group on the web, go to:
http://groups.yahoo.com/group/equismetastock/
<*> To unsubscribe from this group, send an email to:
equismetastock-unsubscribe@xxxxxxxxxxxxxxx
<*> Your use of Yahoo! Groups is subject to:
http://docs.yahoo.com/info/terms/
|