Page 1 of 1

Different test results with CSI 2.9.1

Posted: Thu Mar 02, 2006 5:54 pm
by jkk3131
I have searched for this topic but haven't found it.
I recently downloaded the 2.9.1 version of CSI after having used an older version. When I run my tests in TB now (for futures), I get very different results from those of the previous CSI version. I believe that I have kept the backadjusting algorithm the same as well as the roll criteria, and have kept all the parameters the same in TB.
Has anyone had a similar experience? I would appreciate any suggestions.

Jonathan

Posted: Sat Mar 18, 2006 6:01 pm
by Tim Arnold
What I would look for is new settings and preferences in the new version, that might default differently than you expect. The smallest change to any setting in the backadjusting algorythm used for your data will affect your test results in a snowball (snowball down a hill, not thrown) type manner.

Posted: Sun Mar 19, 2006 6:40 am
by Turtle40
jkk3131,

I had the same experience when I changed from 2.9.0 to 2.9.1. I run 2 identical setups of UA and TBB, and always compare test results when I change either installation. I found small differences between the 2 computers after I had put 2.9.1 on both machines. This showed up in the instument test results, and I discovered that the CU results were very different.
On closer inspection of the chart, the CU data had become corrupted on one machine, and the Dec. contract rolled straight to June, missing out Mar. altogether. I deleted and re-added CU_20, and that curred that problem.
What I did notice were the roll parameters which were still not the same, and within my portfolio some roll triggers were set differently, by the default setup of UA.
The confirmation signal in UA was set to "Roll on second consecutive trigger" for all futures, except for about 5 which were set to "Roll on third consecutive trigger". I have no idea why, but having gone through all the futures individually between both machines, I now get identical test results.
I don't really know if there is a "correct" way to set the roll parameters, as this obviously affects the test results. Maybe someone else could shed some light on this area.

-Turtle40

Posted: Sun Mar 19, 2006 3:10 pm
by jkk3131
Thanks, Turtle40 and Tim, for your replies. I will try to double check every aspect of the new CSI for differences from the prior version, and check each of the futures carefully as well. As I get to know the UA software better, perhaps I'll figure out what is causing the different test results.

Jonathan