Error message while performing MLE

[Zappa_F]

Hello.

I am trying to estimate a simple GARCH(1,1) with maximum likelihood. My code is:

Code: Select all

%daily = "test" !obs = @val(%1) pageselect {%daily} genr y = {%0} genr x = dlog(y) '###################################### param c(1) 0 c(2) 0.1 c(3) 0.8 c(4) 10 logl ll1 ll1.append @logl loglf ll1.append res = x - c(1) ll1.append sig2 = @nan ( c(4) / (1 - c(1) - c(2)) + c(2) * res(-1)^2 + c(3) * sig2(-1) , 0.1) ll1.append loglf = -0.5 * log(2*3.14) - 0.5 * log(sig2) -0.5 * res^2 / sig2 ll1.ml(showopts) show ll1.output

As %0 I want to use a Dow Jones Industrial Average series starting from 01/04/1960 until 10/27/2011. Each time I want to run the program, though, Eviews tells me:

Missing values in @LOGL series at current coefficient at observation 01/04/1960 in “DO_LL1.ML(showopts)"

Well, as far as I believe the Problem is that for the first observation x = NA. Therefore I took the @nan command in front of the sig2 equation. But it still doesn't work.

What is wrong with my code?

Cheers,

Felix

[EViews Gareth]

More likely sig2 goes negative, then you take the log of it.

[Zappa_F]

Hi Gareth,

thank's for your thoughts on this so far. I took the absolute value of sig2

Code: Select all

ll1.append loglf = -0.5 * log(2*3.14) - 0.5 * log(@abs(sig2)) -0.5 * res^2 / @abs(sig2)

But still the same error message pops up...

If I wanted to take only the second until last observation of x, what command would I have to use? I'm asking because the first entry in x is a "NA". I still believe that this causes the problem...

Felix

[EViews Gareth]

Set the workfile sample prior to estimating.