----- Original Message -----
Sent: Sunday, March 13, 2005 10:07
AM
Subject: Re: [Metastockusers] What to
Look for in a System Test
Martin,
How do I remove the zero trades without
deleting the negative trades? The cleanup utility removes both and I am only
left with the winners. I found that entering a percentage for Money Market
was really throwing off my overall returns for a given set of equities. I
have just been sorting the results by the Net Profit column. I then look at
the graph at the top of the results window. If the negative bars out number
the positive ones I assume I do not have a good system. I know the magnitude
of the losses are incorrect but the actual losses I assume is
accurate.
I look at the trades that did well. For most of
my trials I will usually only get 1 or 2 trades in a 250 bar test. I don't
think I can just remove the losers and then explore for signals on the
equities that did well. Just because they hit on the last 250 bars is not an
indication of future performance. This is why I feel I need to have a
positive win ratio over all securities in my portfolio in order to have
confidence in my system. When I look at different charts with the indicators
superimposed I have confidence I can trade them successfully. But when I run
the system test my confidence goes to 0. With so few trades a year I would
need a good set of stocks to pick from.
Thanks, Scott
----- Original Message -----
Sent: Sunday, March 13, 2005 9:36
AM
Subject: Re: [Metastockusers] What to
Look for in a System Test
Scott
Look at each trade one at a time. The results
shown are not accurate. You must calculate the percentage of gain or loss
yourself. Start by deleting all symbols with zero trades.
Martin Blain
Burlington Ontario
----- Original Message -----
Sent: Saturday, March 12, 2005 8:47
PM
Subject: [Metastockusers] What to
Look for in a System Test
O.k.
I have been building system test after
system test, trying several different combinations of indicators.
Since I use EOD data I have given up on getting accurate exit values. I
get a bunch of high losses which I
attribute to the EOD data not exiting until the close.
I assume the object of a system test
is to have more winning trades than loosing ones? I have resorted to
adding one optimization so I can look at the overall system performance
on a total $$ basis. The only problem with this is that with the exits
being what they are, my losses are humongous. I have been trying to find
a system that is better than 50/50 but have yet to stumble on anything
that yields more winners than losers.
How do others go about analyzing system
tester results?
Thanks,
Scott