Standard errors in the state space

[timwatson]

Hi,

I am having trouble getting error bands for a NAIRU estimated using the kalman smoother in the Eviews State Space. My code and data is attached below:

pc bc nairu.prg

(11.59 KiB) Downloaded 275 times

Data:

USNAIRU_Data.xls

(166 KiB) Downloaded 224 times

As you can see the output basically says the NAIRU is estimated with a very small constant error which cannot be correct.

Would it be likely that bootstrapping would avoid this problem?

Cheers

Tim

[trubador]

Just change the following line:

Code: Select all

SS_all.append @state nairu=nairu(-1)+[var=!SIGMA8^2]

with this one:

Code: Select all

SS_all.append @state nairu=nairu(-1)+[var=SIGMA(8)^2]

[timwatson]

Thanks Trubador, much appreciated, but I still get the same problem with that coding error corrected.

The standard error is constant, and very small.

[EViews Glenn]

Running the model I get a Failure to Improve the likelihood with gradients that are decidedly far from zero. I wouldn't view this estimate as having converged.

There are a lot of coefficients 15+2 variance parameters with only 208 observations. Perhaps other starting values might help, but it wouldn't surprise me to find that the likelihood is quite flat in some regions. It appears that in the region you are in, it's quite flat and the estimates are getting stalled.

[timwatson]

Running the model I get a Failure to Improve the likelihood with gradients that are decidedly far from zero. I wouldn't view this estimate as having converged.

There are a lot of coefficients 15+2 variance parameters with only 208 observations. Perhaps other starting values might help, but it wouldn't surprise me to find that the likelihood is quite flat in some regions. It appears that in the region you are in, it's quite flat and the estimates are getting stalled.

Thanks Glen, somebody had mentioned to me that this may be the issue. I have played around with some different starting values, but this hasn't helped a great deal so far. I will keep trying.