Problems with DifEq and Estimate when using Gaussian

[jimarin]

Hi Dani and all the rest,

This e-mail perhaps involves several bugs (or even no bugs at all). I will try to
make myself understood.

I have tried to produce an artificial time series to test the estimator.

I used the routine Gaussian to produce a random series.

This poses, apparently, some problems as:

• each time the series is used it seems to be re-evaluated and produces new

values.

• the quality of the random series, given by its statistics, seems poor, a great deal

of correlation existing.

• behaviour of routine DifEq seems affected by the use of Gaussian. Normally it is

expected to return a time series that starts at least at the same instant of the
series being processed, at least when solving the system:

Z_t - Z_{t-1} = E_{t}

and providing initial value Z_{t0} is given.

• At the moment of using the Estimate function we have found that when building

the synthetic time series to be estimated with a MA parameter of 0.8 the Estimate
function returned in several ocassions a value of 0.9999 for the MA parameter.
This seems odd at least.

I send all used files to danirus in a separate mail.

Thanks !

Jose Ignacio

[Jorge]

Hi, Jima, could you make an attachment (​http://bugs.tol-project.org/attachment.cgi?bugid=170&action=enter), into this bug report, of the mentioned files,

thanks,

Jorge

[Víctor de Buen Remiro]

Hello Jima.

Sorry but if you doesn't create an attachment, we can't try to solve this bug.

Thanks to be patient

[Víctor de Buen Remiro]

We are making a cleaning in bugzilla and this bug has no enougth information.
So it will pass to REMIND status.