|
PureBytes Links
Trading Reference Links
|
Hi Jim,
Tuesday, February 11, 2003, 3:59:28 AM, you wrote:
JH> Your system will get a high return on account because there is no
JH> trade delay set “SetTradeDelays(0, 0, 0, 0);”. You are buying or
JH> selling at the close on the same day the trading signal is
JH> generated. Which would not be possible in real life. You can not
JH> trade at the close after the close.
You can, and I do sometimes, but it certainly has pitfalls that would
need to be addressed by any system that you would test this way.
The main problem is that some signals will be ambiguous just prior to
the close. Maybe the bid-ask will actually be straddling the signal
price, or jumping around on both sides of it. You would have to make
a guess to take such a trade (the back tester would not be guessing
of course), and sometimes you will get caught making the wrong guess.
When you compound this inherent problem with a large basket of
stocks, such a system really becomes unworkable. You simply cannot
monitor and execute just prior to the close on a large basket with
many ambiguous signals.
At times, however, and especially with very small baskets of stocks,
there is going to be no ambiguity. You are going to know there is a
signal prior to the close, with almost zero probability of that
signal coming off by the close. My experience with short-term
systems and trading suggests that these signals should be taken,
because they are more profitable and less risky then waiting for the
subsequent open.
I think that to test such a system, however, and to have the test
anything close to reality, you need at least 2 things:
1) a rather small basket of stocks -- small enough to execute
properly just prior to or on the close (would have to be only a small
handful, I would suggest)
2) back test only for those signals that would be unambiguous, i.e.,
your back test would only take signals x percent 'deep' into the
signal area, signals that would have an extremely low probability of
canceling out before the close
Depending on how deep (unambiguous) you require the signal to be for
purposes of back testing, you could end up with a system that might
actually trade *better* in practice than it back tests (wouldn't that
be a refreshing change?). ^_^ The reason is that you could require
a very deep signal for back testing, deeper than you think you would
actually need in real trading. So your back tested system would
likely take fewer trades than you would take in reality if you set it
up this way. The caveat of course is whether there would be a marked
performance difference between the 'deeper' signals and the ones you
might take in practice that you would be programming your back test
to ignore. Further back testing might shed some light on the
probabilities there, however.
Such a system is a problem, but I don't think we can dismiss these
systems out of hand. I sure don't think you can test them
realistically on even 50 stocks, however, or on even half that many.
Best,
Yuki
mailto:yukitaga@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/
|