EViews.com

Good morning,
first of all I'm sorry for my English...I'm Italian so I'm sure that I will do many mistakes, I'm sorry!

In my work I have two series: DJ and MSCI with the sample (01/01/1980 - 01/30/2009) .
First I have to calculate earnings so I do:
- series dldj = dlog(DJ)
- series dlmsci = dlog(MSCI)
and so I create the series Y=dldj-dlmsci .

Then I create a dummy variable in this way:

smpl @all
genr dummy = 0
smpl 20/01/1981 30/01/1981
genr dummy = 1
smpl 20/01/1989 30/01/1989
genr dummy = 1
smpl 20/01/1993 30/01/1993
genr dummy = 1
smpl 20/01/2001 30/01/2001
genr dummy = 1
smpl 20/01/2009 30/01/2009
genr dummy = 1
smpl @all


So now I've got a problem because I've to estimate this

@logl loglik

eps = y-c(1)-c(2)*dummy

sik = exp(c(3)+c(4)*dummy)
sika = @sqrt(sik)

ink = eps/sika

loglik = log( @dged(ink,2+c(5)) ) - log(sik)

but there is always an error, that is "Missing values in @LOGL series at current coefficients at observation 1/01/1980" if my sample begin with 01/01/1980, instead if my sample for example begin with 01/15/1980 there is the same problem, that is "Missing values in @LOGL series at current coefficients at observation 01/15/1980"...I don't understand why...can you help me please?

I hope to have an answer

Sincerely

It's hard to tell without looking at your data.

But you can do a quick little analysis which should give you some hints. When you run the likelihood estimation, we'll try to evaluate all of the expressions in your LOGL object. So you should start by looking in the series EPS, SIK, SIKA and INK (in that order) to see if there are any values there that look out-of-line or are missing.

I'm sorry but I didn't find the error because for example in eps there is only one number...
I attach my data

Thanks

Can you send the workfile as it exists after you perform the estimation to support@eviews.com. It's better than having us recreate the data using commands since we want to make certain that we have exactly the workfile as you are using it.

I've looked at the workfile and there are a couple of things going on.

First, there is a missing value in the Y series for the first observation. While EViews often removes observations with missing values for data before performing estimation, this is done judiciously with non-linear specifications and in particular for LOGL objects since missings could simply be a function of the current coefficient values. In this case, we don't adjust the sample, but put the onus on the user.

I'll note that the error message that we provided indicated exactly the problem, that we were unable to compute the LOGLIK value for this first observation. So, if you change your estimation sample to remove observations where Y is missing, then you'll be able to perform the estimation.

One additional issue that I noticed when looking at this is that you have a couple of parts of your specification where you have values that are required to be non-negative. As we note in our manual, you can sometimes get into trouble in iterative procedures with this since it is possible that the terms will wander into non-valid regions for some coefficients. You may wish to reparameterize your df and variance terms to avoid this possibility.

I know that there is a missing value in the Y series for the first observation, this is because Y derives from the difference between DLDJ and DLMSCI in which there is already a missing value. But I think that this is not the problem because as I told you the first time, even if I change the sample, there is always the same error, for example: if the sample begins with 02/01/1980, the error will be "Missing values in @LOGL series at current coefficients at observation 02/01/1980"; if the sample begins with 05/01/1980, the error will be "Missing values in @LOGL series at current coefficients at observation 05/01/1980", and so on...

I also try to take the absolute value of each series so that I avoid the problem of negative values...but there is always the same error, that is "Missing values in @LOGL series at current coefficients at observation ......"

I forgot to mention that your starting values were bad. Look in the C vector and change them to something more appropriate.