PureBytes Links
Trading Reference Links
|
Dave,
I sent a similar bug report in a few days ago. In my case, the
entries and exits wre market signals, so all trades should have been
exiting at the same time, but once in every 5 to 10 exits, one trade
would not exit and would remain "open" until the end of the test
period - even if the end was several years and 50 market exit
signals later. In one test a portfolio of 5 was down to only 1 stock
being traded after 10 years. The other 4 slots in the portfolio were
occupied by the trades that failed to exit.
As you say, it may be a tough bug for TJ to track down because, at
least in my case, it was only happening 4 times in several hundred
trades.
However, the "possible" bug my previous post referred to was of a
different nature. I have to make some time to explore it a bit more
before sending in a bug report -- I want to make sure the problem is
not something in my AFL code or my data.
But I will tell you what it looks like, in case you have already
seen somthing like it. It appears that in some conditions the
settradedelay does not take effect. Either that or I have discovered
the mother of all Holy Grails: 800%/year with only 18% drawdowns.
Have you noticed anything like this?
b
--- In amibroker@xxxxxxxxxxxxxxx, "Dave Merrill" <dmerrill@xxxx>
wrote:
> have you checked if there are important differences in the actual
trades? I
> assume, since you're comparing the results, that you're not using
any
> portfolio features that you *want* to make it behave differently
in that
> mode, right?
>
> one reason I ask about the actual trades is that while working on
> portfolio-based systems recently, I've noticed what may be some
strange sell
> behavior, or rather lack of sells. it's come up in a pretty complex
> environment, and I'm not 100% certain my code isn't causing it in
some way,
> but I haven't seen how it is, yet. I sent in a possible bug report
about it
> and haven't heard back, probably because it's not a simple, easy
to isolate
> situation that would have been easy to confirm or explain away.
>
> for example, one system switching on fairly fast indicators showed
some
> holdings lasting multiple years. I can see on the charts that the
relevant
> indicators crossed their thresholds many times, without generating
a sell.
> also, the system had a 30% stop loss in place, that likewise
didn't sell. I
> discovered this while investigating some trades with
uncharacteristically
> large losses, like 87%. sometimes things like that do happen if a
stock
> breaks really quickly through your stop, but here, the losses
occurred over
> a very long period. the expected sells never occurred, neither
from my
> indicator-based code nor from the stop loss.
>
> as I said, I wrote bugs@xxxx about this; hopefully I'll hear back
> at some point.
>
> dave
> I'm playing with some simple ideas right now, and I find that I
am
> getting dramatically different test results from the regular back
> tester and the new portfolio back tester. I realize the latter
is a
> beta while the former is not, however the former is showing a
nice
> profit on the report summary, while the latter is showing a loss.
> Anyone else getting conflicting results like that? Shouldn't the
> money totals be the same?
>
> Yuki
------------------------ Yahoo! Groups Sponsor ---------------------~-->
Rent DVDs from home.
Over 14,500 titles. Free Shipping
& No Late Fees. Try Netflix for FREE!
http://us.click.yahoo.com/ArdFIC/hP.FAA/ySSFAA/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/
|