[Libreoffice-bugs] [Bug 121396] FORECAST.ETS.PI functions produces different results with identical data.

[bugzilla-daemon at bugs.documentfoundation.org]

https://bugs.documentfoundation.org/show_bug.cgi?id=121396

Winfried Donkers <winfrieddonkers at libreoffice.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|critical                    |enhancement
           Priority|high                        |medium
            Summary|Function                    |FORECAST.ETS.PI functions
                   |FORECAST.ETS.PI.ADD()       |produces different results
                   |produces different results  |with identical data.
                   |with identical data.        |
            Version|5.2.7.2 release             |5.2.0.4 release

--- Comment #3 from Winfried Donkers <winfrieddonkers at libreoffice.org> ---
The description of the behaviour is correct. The behaviour is intentional.
The Prediction Interval (or better accuracy indicator) is calculated with
random numbers and multiple (I think it is 2000) samples.

I fully agree that the current behaviour is not perfect and can be misleading.
(On the other hand it underlines that the accuracy indicator is an
approximation.)

I can also inform you that I have been working on a calculation that produces
consistent results that are a good approximation of the accuracy ever since the
FORECAST.ETS functions were introduced early in 2016. However, it is a very
difficult task. I have consulted several mathematicians at a number of
universities, but so far no practical (any solution must be suitable to be
implemented without severe penalties with respect to performance of Calc)
solution has come up.
Currently, I am conferring with a Statistics Section of the Mathematics Faculty
that I hadn't contacted before. I keep on looking and hope to succeed
eventually.

I have changed the importance of this bug report to enhancement as there is no
bug. I will use this bug report/enhancement request to submit improvements to
the FORECAST.ETS.PI functions (the Additive and Multiplicative share the same
principle).

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/libreoffice-bugs/attachments/20181113/a1e7ef71/attachment.html>