|
PureBytes Links
Trading Reference Links
|
----- Original Message -----
From: "Gary Fritz" <fritz@xxxxxxxx>
To: <omega-list@xxxxxxxxxx>
Sent: Saturday, February 28, 2004 2:25 PM
Subject: Re: QuantStudio
> > 1) Take any system that exits at the end of day so that P/L for a
> > given day is not affected by overnight trades. Go back say, one day.
> > Go to View->performance Report->daily Tab. Note the P/L. Now, go and
> > bring up the symbol again. Change number of days to look back to 2.
> > Go back and look at the daily report. Look at the P/L for the day
> > that you looked at - they are different.
>
> I don't understand what you mean by "go back one day, go back two
> days."
>
> Are you saying you change the start date of your chart, and then
> you see different results? That's certainly possible if the
> system code is carelessly written. E.g. you can use xaverages
> that take a long time to stabilize, and that can cause different
> results depending on your start date. But that's not TS's fault.
> It's executing the code correctly.
Look at what I wrote. Write a system where that EXITS at the close of the
day and takes no positions overnight. It is IMPOSSIBLE for a system that
takes all of it's signals for a given day from within that day, to have a
different P/L based on what days are loaded or not. Of course, in the case
where you are using indicator where adding a prior day will affect the
entries of today would not be either careless coding or incorrect
calculation by TS. But in the case I am talking about, I am NOT using
indicators that cross daily boundaries.
> > 2) Take any system that gives signals in realtime. Watch it during
> > the day without ever closing TS down. When the markets close,
> > shutdown TS and come back in, do not save the workspace. Now, run
> > the system again. Different values.
>
> I don't see this in my systems. I've checked this very carefully
> over the years. (The xaverage example above could cause problems
> like this, but again that's because of bad code, not TS errors.)
I see it all the time.
> > 3) This is is MUCH harder because it would require me to give you
> > code that I cannot. But here is the gist of it. Make a system. Apply
> > it to say ES and YM. Look at the signals it generates. Take note
> > that sometimes, where a signal UNQUETIONABLY should have been
> > generated, one was not. You sit there and stare at the code, and
> > unfortunately, you do not have a debugger. Printf debugging does not
> > help here.
>
> Whenever I've had one of these situations, it has virtually
> always turned out to be pilot error.
Guarantee you, it is not.
> > 4) I have about 12 charts open, each with 1 of two systems applied to
> > the symbol. Try to look at the performance report during trading
> > hours. Evertime I scroll to the bottom of the trades tab, it keeps
> > going off, doing something, and taking me back to the top of the
> > page.
>
> Never saw it.
See it all the time.
> > 5) A variant of the above is, often a symbol, for whatever reason,
> > keeps needing to "get data." I have no idea why this happens, and it
> > does not happen all the time. The effect is that the entire history
> > for the symbol has to be reloaded every minute or so. This makes
> > following a system under these conditions intolerable.
>
> Never saw it.
See it all the time.
> I'm not saying these things don't happen to you. But I'm saying
> I've been running TS2k for several years, intensively, and I
> don't run into these problems. Assuming these problems aren't
> caused by poorly-written systems, there is possibly something
> unstable about your setup -- memory, hardware, software
> interactions, who knows -- that triggers these problems.
Guarantee you, it is not. I have eight computers, all with massive amounts
of CPU and RAM in them. What is funny is that a friend of mine told me about
this case, and I could not believe it. We loaded the system on my machine,
and presto!
> Of course, TS2k *should* be stable enough that things like this
> don't happen even if the system is stressed, and it's not.
> Nobody (except maybe Pierre :-) would claim TS2k is bug-free or
> even particularly solid. I have my share of problems with it
> too. But most of those problems are annoyances, and none of them
> are "wrong result" problems like you're seeing.
Oh, there is a growing list of problems. But if your systems test accurately
for you, what can I say? They do not form me. FWIW, I really push TS to
"extremes."
> If there was a better answer out there, I'd jump on it. But I
> haven't seen anything yet that does what I want as well as TS
> does. So I grit my teeth and put up with TS's foibles.
TS is an ok tool for what it is as I have mentioned before, charting, that
sort of thing. The key is knowing it's idiosynchronansies and to sit and
watch a system trade in realtime very very carefully for days on end, and to
plot the indicators that the system is using to base it's opens and closes
(I often send it to the debug window as well, but it gets really tiring.)
The problem is, a real programming environment would have a debugger, and
the programmer would be able to go in, roll up his sleeves, signle step
through the cod, and figure out what was happening when the unexpected
happens. No such tool in TS. So the "pilot" is forever left trying to rework
the code in the hope that he hits some combination that will not confuse TS.
Ivan
> Gary
>
|