|
PureBytes Links
Trading Reference Links
|
I think that the maxbarsback setting is the problem. Remember that this is
really the number of bars used as a basis for your system and indicators and
will give you faulty values if it not set correctly. In theory it should
only give bad values for the beginning of the chart, but, Murphy's Law
always rules. Use at least 50 and remember that this just means that the
first 50 bars of the chart are used as the basis for your indicators and
systems. I used to think that it meant that the program was fetching data
before the beginning of the chart. Pro has the name changed now for
clarity's sake.
Jim Bronke
Phoenix, AZ
----- Original Message -----
From: "Jack Griffin" <jack_2231@xxxxxxxxx>
To: "Jim Mann" <mannjt@xxxxxxxxxxxxxx>; <omega-list@xxxxxxxxxx>
Sent: Saturday, August 11, 2001 8:30 AM
Subject: RE: Why bad realtime results? The plot thickens . . .
> Ok, thanks to the people on this list, I have had
> about 15 different suggestions on possible causes for
> the realtime problem I am having. To remind everyone,
> the problem is that my indicators and strategies are
> totally different if I let them eat realtime data for
> a few hours verses what I get if I load a fresh chart.
> More specifically, my indicator gives wrong results
> after about one hour of getting realtime data. The
> most important thing I have learned thus far is that
> MANY people are having this same problem.
>
> This is a very tough problem to debug, since I can
> only debug is while the market is open, and I have to
> wait about 2 hours before I can notice the differences
> in the lines. What I was doing on Friday was trying
> various suggestions, waiting 2 hours, taking a
> snapshot picture of the indicators using Hypersnap.
> Closing tradestation (but not the global server), and
> restarting it using the same chart.
>
> I have been able to rule out about 70% of the possible
> suggested causes and have posted the reasons to the
> list. The other 30% I am still working on. Here is
> one possible cause of the problem suggested
> independently by both Multitrack and Jim Mann that I
> had already ruled out, but am now reconsidering (which
> I am reposting for the benefit of the list):
>
> --- multitrak@xxxxxxxxxxxxxxxx wrote:
> > How many days of data are you loading into the
> > chart? You might be using
> > data with a different starting point than the
> > original real time data when
> > you reloaded the chart presumedly EOD or after RTH.
> > This might be enough of
> > a change in the data to affect the system results.
> >
> > MT
>
>
> While Jim Mann suggested:
>
>
> --- Jim Mann <mannjt@xxxxxxxxxxxxxx> wrote:
> > Your system is initializing differently depending on
> > the number of bars present before the signal.
> >
> > Jim
>
>
> My original response:
>
> > This is a very good point. If you are loading only
> 2
> > days, but your indicators use a very slow
> exponential
> > moving average (say 10-days), I can understand why
> > results would be different every two days or so if I
> restart.
>
> > As it turns out, the chart was set to load 180 days
> of
> > data, so even if I used moving averages of say 60
> > days, this is still no good reason why the
> indicators
> > would change substantially with 3 days of new data.
>
> Now the reason I am reconsidereing this idea is
> because I am using a maxbars back setting of only 1.
> This is because it decreases the memory requirements
> --- optimizations run dramatically faster. All
> necessary historical price is stored in variables or
> arrays.
>
> Is this the problem? Does the realtime engine in
> Tradestation 2000i assume that since maxbarsback=1
> that it can get by with only recalculating the
> realtime chart with the new data by using a few bars
> of fresh data?
>
> Jack
>
|