|
PureBytes Links
Trading Reference Links
|
> -----Message d'origine-----
> De : Gary Fritz [mailto:fritz@xxxxxxxx]
> Envoye : mardi 17 juillet 2001 15:49
> A : Omega-List
> Objet : RE: Special thread for Pierre Orphelin, the man who do not want
> to see
>
>
> > If your system is so sensitive to epsilon rounding effect, yous
> > system will ot be stable over unseen data, where the precision of
> > the considered data is far below the epsilon value.
>
> Pierre, listen to what Mats is saying.
>
> 1. He is not looking at his system, he is looking at Omega's DMI.
Yes, I understand, but I supppose that the DMI will be used in a system ( if
not, I seriously doubt that the pixel size of the screen will show any
noticeable difference).
>
> 2. Omega's DMI produces DIFFERENT RESULTS than DMI written in C++.
This is possible.
Any further discussion must be conducted by evaluatiing the difference
between both codes.
I also seriously doubt that the difference will bring noticeable
modification in the system performance summary report.
>
> 3. The errors accumulate, and the results are SIGNIFICANTLY DIFFERENT
> in Omega's DMI than in a C++ implementation of DMI.
How much ?
>
> If Mats is correct (I haven't spent the time he has to verify it),
> then it seems clear to me that Omega's code is in error, and the
> erroneous code can produce incorrect trading decisions even if you
> don't assume your price data is accurate to 10^-7.
>
This is not what I pointed out.
I only mean that price data use 5 digits. The DMI values beyond 5 digits
have no meaning at all.
If the error due to the Omega code acts on these extra digits, they are of
no importance at all.
P.
|