what date should input start in Estimate function?

[Chakib Faghloumi]

Estimating a daily model when all output and input start in y2000m01d02 whith
dif 7, AR1, and MA 7, I suppose that the first date should be y2000m01d10 but
Estimate ask for inputs to start the date y1999m12m12. Why? What should be the
theorical start date for the inputs ?
I will attach an example

[Chakib Faghloumi]

Compile estimation.tol

[Víctor de Buen Remiro]

Hello Chakib:

First date of inputs should be the output one forward brought by sum of

• Difference Deg(DIF)
• Maximum degree of OMEGA polynomials of input's transfer function

Stochastic polynomials AR and MA have no influence on forwardness since initial
values of ARMA subprocess are estimated by estatistical methods.

So, input begining cannot be after output begining as you are proposing. In your
case, if all OMEGA polynomials have null degree then input must start 7 days
before output does, never 8 days after. To explain that required date were
y1999m12m12, it's to say, 3 weeks before output starting day y2000m01d02, I
think there should be a B^{14 OMEGA polynomial in transfer function.}

Any way, in current version of Estimate function there are some errors that could
cause unexpected forwardness. If you want to send to me an stand-alone set of
TOL files in order to reproduce the model by them self, I could determine if this is
really a bug. If you don't know how to do this, you can contact with Fredy or Ivan
who will can help you.

This problem is related with non resolved bug 81
​http://bugs.tol-project.org/cgi-bin/bugzilla/show_bug.cgi?id=81

Victor de Buen

* This bug has been marked as a duplicate of 81 *

[imendez]

Hello, Víctor and Chakib.

I have two explanations:

• Chakib doesn't mean that input beginning date must be "m" days after the

output beginning, where "m" es the output difference polynomial degree plus the
maximum degree of OMEGA polynomials of input's transfer function.

He means that if your output first date is y2000m01d02 and "m" is 21, the

first date taken into account for the OUTPUT must be y2000m01d23, and Estimate
function must consider inputs first date as y2000m01d02, instead of y1999m12d12
(y2000m01d02 - 21), because some inputs could not exist at this day, and then
you would have to calculate the initial date for inputs and output.

• Víctor, you require to Chakib a "stand-alone set of TOL files in order to

reproduce the model by them self". He has attached this in a rar file but the
file hasn't got any extension, you only have to add the rar extension to his
file and unzip it.

Thank you. Regards

Iván Méndez

[Víctor de Buen Remiro]

Hello, Chakib and Ivan

Well, I'm sorry. I didn't undestand it nor I viewed the attachment but now I can
view it.

If I've understood you, then this bug is exactly the same as bug 81

​http://bugs.tol-project.org/cgi-bin/bugzilla/show_bug.cgi?id=81

It isn't? I expect to get some time to dedicate to it, though this is not exactly a bug
but an enhacement, because you can explicit the correct date as optional
argument of Estimate function.