[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Evaluating MS Trading Systems



PureBytes Links

Trading Reference Links

Dave, you can't test different time periods in the Explorer unless they all
end in the last (most current) date, can you?

By the way, being limited to 6 columns in the Explorer sure is a pain for
system testing!  Sure limits you to what you can report.

Tom
----- Original Message -----
From: "Dave Nadeau" <dave_nadeau@xxxxxxxxx>
To: <metastock@xxxxxxxxxxxxx>
Sent: Monday, June 25, 2001 12:00 PM
Subject: Re: Evaluating MS Trading Systems


> Herman,
>
> I don't use the MSBT, but do the same sort of thing using the Explorer.  I
will code many of the
> metrics that you are discussing and compare them across multiple
securities in a portfolio.  You
> have some very good ideas and I would like to experiment with some of them
in my future tests.
>
> Another idea that may be worth trying is this:  split your data into three
date ranges (assuming
> sufficient number of data points for a statistically valid result).  My
theory is that when I
> create a system that is successful, it exploits a market behavior that is
consistent and
> persistent.  An example is a trend following system.  Some markets trend
better than others,
> especially certain commodities.    PREMISE:  Generally speaking, i.e.
across a group of these,
> that market which trends will be more likely to trend in the future than a
market which is choppy.
>
> So taking the best performing issues in the first date range, should be
the better performing set
> in the next data range, and then, the third.  There will be variation
among some of the issues, of
> course, but in general, I'm looking for this to be true.  If so, then I
get a good feeling for the
> robustness of my system as well as the expectation of its performance in
the future.  When I find
> systems that do not do this, my sense is that they are not much better
than random, and are more
> of a curve fit rather than one I'd trust trading real dollars.
>
> This approach tends to be more of a securities fit, rather than a curve
fit.  It's just another
> way to slice and evaluate a system.
>
> Dave Nadeau
> Fort Collins, CO
>
> --- Herman van den Bergen <psytek@xxxxxxxx> wrote:
> > At 08:00 PM 6/24/01 -0500, you wrote:
> > >What is MSBT?
> > >Lionel Issen
> >
> > MSBT (Multiple Security Back Testing) is an add-in for Metastock. I use
it
> > to test my trading systems on various stock selections (50 - 8000
stocks).
> > It takes about ten minutes to process 8000 stocks and produces a Excel
> > compatible (.csv) report with about 50 stats (like in the MS system
tester)
> > for each stock tested.
> >
> > I like this add-in because it gives me an immediate idea of how robust
my
> > system is. Testing my system on a large universe of stocks and knowing
for
> > how many stocks the ROI actually improved, gives me a good (well,
sometimes
> > not so good...) indicator of the "robustness" of my system.
> >
> > The ratio (number of stocks with improved ROI)/(Total number of stocks)
> > makes a nice Robustness Index. It is also interesting to run tests on
> > different sectors or other categories of stocks. It would be interesting
to
> > learn what kind of RB-Index other developers manage to obtain - anybody
> > care to share? I typically get 35% however I have have a sorting problem
> > related to negative B/H indices - so I think my ratio is actually much
> > better. Anybody calculated this ratio?
> >
> > My systems tend to be too specific and finding stocks that trade well
with
> > my system was difficult. Using the MSBT you run your system on all 8000
> > stocks (or less) and see immediately which stocks performed well with
your
> > system. I had some interesting surprises :-)
> >
> > For more info visit http://www.holygrailsoftware.com/msbt.shtml , the
> > add-in costs $45.
> >
> > Happy trading!
> > Herman.
>
>
> __________________________________________________
> Do You Yahoo!?
> Get personalized email addresses from Yahoo! Mail
> http://personal.mail.yahoo.com/
>