Missing Values @logl

[CharlesLogan]

Greetings.

I am attempting to use MLE to estimate an equation, however I am getting "missing values in @LOGL series at current coefficients at observation NNNN". Is there a way Eviews can show which variable is causing the problem? I use OLS to compute starting values for the coefficients and the OLS runs properly, but the MLE does not. Why would this be? If the OLS runs properly, shouldn't the MLE?

Any advice will be much appreciated. Thanks.

[EViews Gareth]

An equation object will automatically drop observations for which there are NAs in any of the variables. A LogL object will no do this, which will cause the error message.

[CharlesLogan]

OK - Thanks. Is there a way to find out which variable is causing the problem? There should be an easier way to find the problem than sifting through all 50 regressors one-by-one.

[EViews Gareth]

Unfortunately not. The problem is that it might not be an NA in a variable at all - rather it could be part of the calculation involving multiple variables that gives an NA. For example you might have

log(x1-x2)

as part of the LogL object. Neither X1 nor X2 have NAs, but any observation for which X2 is bigger than X1 will yield an NA for that expression.

[EViews Glenn]

What you can do is to look at the contents of the series that you generate inside of the logl to see where you might have a problem. I'd start at the first expression in the logl and look at the contents of the series. If that looks okay, go to the next expression, and so on until you find the problem. If you need to, you can break apart a single expression into multiple lines. Once you find the series that looks wonky, that should give you some idea of how to fix the problem.

One other thing. If your MLE is based on an OLS type model, the problem is probably in your variance specification...