Traders' Roundtable

Just downloaded version 2.10.7.116 csi
Still getting discreoancies between this and my previous version - all other thngs being equal - system on one computer taking data from csi version 2.10.7 is giving different results to on another computer - all else equal in TB AND csi
I notice howver in the new version (csi) there is a box ticked called use advanced backadjuster - this is not on previous editions and cannot reference what this is via CSI manuals etc - would this perhaps be the difference
I know someone is going to say try it with /without new box ticked - I have 150 markets on the go here and before I manually go through everymarket and uncheck box just wondedred if anyone else had come across the same problem / issue or knew what differences are likely to be with this new functionality in CSI
Thanks in advance
C

Ok just realised you dont have to do this manually - you can do it once for portfolio as a whole (i.e. not use advanced back adjuster)
As per usual different results still remain same systems / same port admin file / same settings across teh board - always the same when installing on a new computer - Bugger !

Chris,

I am not familiar with the full-blown CSI UA package, I am currently testing using the "free" data made available here. I know you have posted on this before but I can't remember the details of the entire saga, so ignore me if I am asking the obvious ...

Where are you seeing the discrepency manifest? Is it in the contract data files that CSI constructs (i.e. the continuous contract) and the trading results or just in the trading results?

Obviously, if the differences are in the contract data, it is not possible to say if anything odd is happening in TBB because it is downstream of the data. I have found that the smallest difference in a data file that affects one trade early in a simulation can create a butterfly effect. Also obvious, if the data files are identical, CSI UA is not the source of the problem. So first task (I am sure you have done this) is to compare the continuous contract files on one pc to those on the other - they should be identical. Are they? Some scripting code such as R would help you do this quickly.

Have you tried copying the continuous contract files from the one computer to the other and running the simulations with data you know is identical. If identical data gives different results the issue is in the simulation. Do they give different results?

If the data is identical the issue is with the simulation. Each time you run TBB a trades file is generated. I assume you have compared the two trade files generated from identical data to find discrepencies in the trades. How significant are they? Does it start right at the beginning (butterfly effect - what I would expect is slightly different trade sizes, the differences growing until entire trades are missed or new trades are taken) or are there missed / extra trades or is the history unrecognizable? Once you have found the first erroneous trade, all others likely result from that.

Is the same version of TBB in use on both machines? What happens if you export the system from one machine and import it to the other? Same problem?

Can't think of anything more at this time. Let us know how you make out!

Thanks for the repsonse
the problem / issues are with CSI not TB

The TB versions are exported so are identical
The problemprobably arises with CSI data
I notice with the newcsi vesrion I amusing there is the enhanced / new back adjuster - with this there isnt even an option to tick a generate forward box-there are bound to be discrepancies somewhere ? that will explain the small differences
I will run some comparisonson files and see what I can come up with

Have you refreshed the database in both versions? Perhaps your older version is not in sync with the current update for some reason.

copying the portfile.adm from archives in the old version to the new one will sync the portfolio settings.

A colleague and I run identical systems and portfile.adm files. Identical versions of the software are used in each case. We need to keep exactly in sync. We download CSI at slightly different times from different geographical locations. Tiny differences arise in our ASCI files over time giving rise to the "butterfly effect" noted above.

At that stage it is necessary to run Windif on each set of ASCII files to find out in which symbols the difference lies. And then to refresh those symbols from the CSI data base.

Given the care we take to keep in sync, it is easy to appreciate how minute differences in data files can cause dramatic differences in test results.

There is no real answer other than extreme care and meticulous record keeping so that one can turn back the clock and see how the differences have arisen and how they can be reconciled.

I updated to the csi 2.10.7.xxx build from 2.9.3 last month and had no problems at all.
I ran both versions on the same computer for 3 weeks in january before I was satisfied there were no problems or differences in my data.

redrock's suggestion of copying the portfile.adm file is your best bet. transfer that file again and you will end up with the same settings.

I just tried .116 and concur with Chris that major differences are occuring - same portfile.adm on both computers and using the standard TB Donchian system with no frills.

Let the fun begin I suppose. Did they fix something that was previously broken or introduce a new back adjusting twist...

Yep Thers a new back adjuster - looks to have upgraded the C++ from earlier versions
No major complaints here - If I get a system thats within 0.25% of a previous system when using a new version of teh software its probably preety dam good i.e. 99.75% the same
On a 25 year backtest over 150 markets I'm 10 trades out ? I could either spend 57 hours trying to work out the cause or say - I guess thats pretty much identical give or take the odd qwerk
C

Using the TB standard Donchian system, about 25+ years of data on about 80 markets or so, I get bigger differences than you seem to be getting Chris.

Have you done all the usual refreshes etc etc ?

Also I'm noting a few missing days on Forex files I'm in touch with CSI at present about it

Also until we note what exactly the differences are in teh advanced back adjsuter its difficult to envisage what else may be different
As I say mine is pretty dam close but differences exist and the obvious butterfly problems then exude

It would have been nice... if CSI would have....highlighted the changes and possible consequences. Surly they would have been aware of the discrepancies between versions. sigh Now to discover what the changes are, and then which way is more 'correct' or accurate. (as roll your own, young man...echos through my head)

I see there was a V.115 which made a few changes. One in particular "Reseting of Days before expiration has been fixed." could be part of it... Or is it the new back builder...