|
PureBytes Links
Trading Reference Links
|
Have done a number of tests to see if the releases differs in results. I
also tried to make a comparision in execution speed, but was fooled by the
fact that SP4 overwrites your settings on how many days to load of data.
Thus, the thing that struck me is that if you change number of days to
calculate from 3300 (my setting) to 500 (Omega setting), the speed of a run
with the workspace assistant is affected greatly. Going from 49 minutes to
17 minutes for running through my indicators.
This would be a nice enhancement to have in the workspace assistant: to get
to select between a number of preset settings of number of bars to load
during the run. For example: Systembacktest-setting loads a great number of
bars, EOD run-setting loads enough many of bars for indicators to be
correct, Chart study-setting loads many bars, but not as many as
systembacktest, and so on.
Second thing: All the indicators run were set to do a fileappend on every
alert. They were run first on SP2, then on SP4. The output was then compared
through a program. Not so many differences, but enough to be worth
mentioning, and for some indicators frightening many:
* There were alerts that are shown using SP2 on screen, but that were
not output to files. Did not know of this bug before, it is very infrequent
and might not be repeatable, but it did exist in my SP2. One of the
indicators I got this error on was the Prettygood oscillator distributed in
this forum earlier. I have rewritten it a bit (as have probably many more),
but it still could not be considered very start point sensitive, and it
gives these glitches.
* I got proof of what I have read about in this list about
Tradestation trying over and over again to decide how many maxbarsback to
use if this is not hard coded. The proof i got was from some triggers
triggering repeatedly at the first date where maxbarsback was enough (those
triggers were such that they happened to trigg at the first date for that
stock). Guess there is a lot of time to save by always setting a maxbarsback
number yourself, rather than using the default setting of "auto".
* In SP2 I had an occurence opposite the first error mentioned above,
a trigger that made it to the alertfile, but is not visible on screen. This
error might be connected to the autodetect problem mentioned above. Since
alerts seem to be output for every try of maxbarsback settings, you might
get an alert registered that is later in the run found to be illegal since
the maxbarsback has to be changed in order for the run to go through as a
whole. This might be a problem if you are feeding signals to other systems
using fileappend or the "print to omega database" utility. It seems not to
happen in SP4.
Most functions compared quite okey, with only a few differences, and where
SP4 seemed to be the one having the logic on itīs side when I hunted for the
reason. Three functions behaved very different. One of them was based on
references, and I would not trust neither SP2 or SP4 to do okey for this
one. I know of bugs in SP2, but I have seen similar bugs reported for SP4 as
well.
The other two indicators are more disturbing, I can not figure out what is
really going on. Either the function is illogical and very dependant on
starting points, or there is a bug in SP2/SP4, or the problem is lack of
significant digits which causes randomness in behaviour. I will have to
investigate these ones more before I know. What is special with one of the
functions is it is calling other functions at different times with different
parameters. One of the functionscalled are called from within the
plot-command itself. Do not know about why the problem occurs, there is a
lot of logic in the calculation, but the behaviour was very unsimilar
between SP2 and SP4. The other one was the pretty good oscillator we
received in this forum a while ago. There were numerous differences between
the SP2 output and the SP4 output. Where I could compare, the SP2 output
seemed to have problems, but there were too many differences in between the
two releases to check them all. I will rerun comparision on this later on,
when I do the update on another machine and remember that the update
overwrites how many bars to use for daily charts.
End conclusion: If you are running SP2, it is buggy, probably more buggy
than SP4. I have during this comparision found a number of bugs, bugs that
affect my backtesting. I have found a number of them that are dead certain
bugs in SP2. I can not say for sure there are bugs in SP4 except for the
reference thing, and I will now have to do the switch a lot faster than
planned, there are too many bugs in SP2 to go on using it.
--- Mats ---
|