|
PureBytes Links
Trading Reference Links
|
Hello,
Thanks, TJ, for the info
With optimization capability of Fred's PT, it is possible to simulate
the random Watchlist effect TJ proposed by using AB's random() and
there is no need to use Jscript or VB at all.
By incorporating random() as part of the scoring process, we can turn
on/off a candidate easily but randomly. For example...
- random(1) >= 0.5, then 50% chance
- random(1) >= 0.3, then 70% chance
- random(1) > random(1), then unpredictable...
Depending on the outcome of the random(1), randomly, a candidate
MIGHT be disqualified which might force PT to select another one.
The BEST part is that PT automatically accumulates optimization
results in a separate CSV file which make post processing extremely
easy.
This is NOT exactly what TradeSim has done, but I bet with enough
iterations and some post processing DIY, it should give similar info
and it saves more than $700.
So, as soon as TJ adds the optimization to portfolio testing, it
becomes native, but it can be achieved today without using OLE or
TradeSim.
Thomas
--- In amibroker@xxxxxxxxxxxxxxx, "Tomasz Janeczko" <amibroker@xxxx>
wrote:
> Hello,
>
> As for
> > portfolio tester gives you very specific
> > candidates regardless
> > how many candidates are out there because there is not
> > enough equity to
> > trade them all. You get the same candidates regardless how
> > many times you test
> > it. AB choose the candidates using its internal rules.
>
> The rules for selecting candidates for trading if
> there are more signals than can be traded (because
> of limited funds or limited 'number of open positions') are:
>
> - absolute value of score (positionscore variable - ranking mode
only)
> - dollar position size
> - alphabetic order
> - order type (longs take precedence over short)
>
> so when there is no ranking or score is the same
> then the orders with higher position size take precedence,
> if equal then alphabetic order decides
> if the same symbol - then long is preferred
>
> One possibility of 'robustness' testing is setting
> up random watch lists containing subset of your
> universe and test on them and compare results.
>
> With OLE automation interface it is possible
> to setup watch lists and perform automated backtests.
> (actually only 'old' backtester can be launched by automation
> right now, but automation is planned also for portfolio
> backtester).
>
>
>
> Best regards,
> Tomasz Janeczko
> amibroker.com
> ----- Original Message -----
> From: "tchan95014" <tchan95014@xxxx>
> To: <amibroker@xxxxxxxxxxxxxxx>
> Sent: Wednesday, October 01, 2003 9:03 PM
> Subject: [amibroker] robust system development
>
>
> > Hello,
> >
> > I want to throw some random thoughts to the list to see if
some
> > discussions can be stimulated.
> >
> > 1) With the wonderful Excel 3D macro available in file area,
it
> > is easy to eyeball
> > the 'most robust or stable' area when 2 parametes are used.
> > How do we do if there are more parameters out there?
> > Is it enough just use 2 at a time and repeat?
> > How if the robust parameter pair we get changes each time?
> > How many repeat is enough?
> > It is very labor intensive and we get ONLY the net$
optimal,
> > nothing else.
> > The result is not quantifiable, you eye ball it, you feel
> > right and that is all.
> >
> > Mark Johnson propsed an idea years ago, (go to Yahoo AAFT
> > group, you might still
> > be able to find it) using what he called 'hyper-sphere' to
> > include ALL the
> > parameter sets within a defined radius and then compare
the
> > chosen statistics
> > amongst all the 'spheres' to find the optimal parameter
set.
> > The idea is
> > same as the 3D macro, just to extend it to N parameters
and
> > many more statistics
> > can ge generated and compared.
> >
> > Theorerically, there can be as many hyper-spheres as your
> > parameter set candidates,
> > say after an optimization run, you have 10,000 entries,
then
> > each entry can have
> > its own hyper-sphere with itself as the center point,
> > potentially with different
> > radius.
> >
> > Within each sphere, all sorts of statistics can be
calculated,
> > including net$,
> > and then can be compared against all other spheres for you
to
> > decide which one
> > is the best.
> >
> > I think it is a good idea, better still, if this idea can
be
> > included into AB.
> > I would like to know if there are other opinions or ideas
out
> > there.
> >
> > 2) In portfolio testing, Chuck R. brought up this question
> > and proposed a solution
> > very early on, but there are other questions.
> >
> > Since resource is always limited, how if there are more
> > candidates than one
> > can act for each signal?
> >
> > Current backtester gives you ALL the candidates giving
> > each candidate all your
> > equity, while portfolio tester gives you very specific
> > candidates regardless
> > how many candidates are out there because there is not
> > enough equity to
> > trade them all. You get the same candidates regardless how
> > many times you test
> > it. AB choose the candidates using its internal rules.
> >
> > Even using S&R, you just artificially limit your own
> > candidates per the rules
> > defined by yourself instead of AB.
> >
> > Chuck proposed using TradeSim, which uses Monte Carlo
> > simulation, to see more
> > realistic test results. I think it is a good idea to
> > confirm the robustness
> > of a system.
> >
> > Even so, how about in real time trading? Do we just choose
> > whatever the candidates
> > AB or S&R chooses for us and trade them?
> >
> > 3) Following the idea in (2), for each system we develop,
> > regardless how many data
> > points we use, they are all history.
> > We tried to use as many as possible data points hoping to
> > cover all the
> > possible scenarios that might happen in the future, we use
> > this process
> > to develop 'robust' systems. Of course, there are many do
> > not agree with
> > the idea of 'using as many as possible data points', but
> > that is beyond
> > the argument here.
> >
> > But, regardless, it is always single path and it is always
> > histoy.
> > Is this approach good enough?
> > Is this the best we can do?
> >
> > I heard that synthesized data set are used to make the
> > data universe even larger.
> > I also learned that Monte Carlo simulation might have some
> > uses here.
> >
> > Any ideas?
> >
> >
> > Thomas
> >
> >
> >
> >
> > Send BUG REPORTS to bugs@xxxx
> > Send SUGGESTIONS to suggest@xxxx
> > -----------------------------------------
> > Post AmiQuote-related messages ONLY to: amiquote@xxxxxxxxxxxxxxx
> > (Web page: http://groups.yahoo.com/group/amiquote/messages/)
> > --------------------------------------------
> > Check group FAQ at:
http://groups.yahoo.com/group/amibroker/files/groupfaq.html
> >
> > Your use of Yahoo! Groups is subject to
http://docs.yahoo.com/info/terms/
> >
> >
> >
------------------------ Yahoo! Groups Sponsor ---------------------~-->
Buy Ink Cartridges or Refill Kits for your HP, Epson, Canon or Lexmark
Printer at MyInks.com. Free s/h on orders $50 or more to the US & Canada.
http://www.c1tracking.com/l.asp?cid=5511
http://us.click.yahoo.com/mOAaAA/3exGAA/qnsNAA/GHeqlB/TM
---------------------------------------------------------------------~->
Send BUG REPORTS to bugs@xxxxxxxxxxxxx
Send SUGGESTIONS to suggest@xxxxxxxxxxxxx
-----------------------------------------
Post AmiQuote-related messages ONLY to: amiquote@xxxxxxxxxxxxxxx
(Web page: http://groups.yahoo.com/group/amiquote/messages/)
--------------------------------------------
Check group FAQ at: http://groups.yahoo.com/group/amibroker/files/groupfaq.html
Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/
|