State space and NAIRU estimation

[Halfdan]

Hi,

I some questions regarding the fomulation of my state space for use in NAIRU estimation using the Kalman filter. First, I'd like to ask some more general type questions. I have no issue with setting up the state space model and estimating it, but my results are not believable so I'm sure I have to make some ammendments. Particularily I'm unsure of how to specify the errors in my signal- and state equation. Let's say I have one signal equation modeling inflation as a function of the unemployment gap and other factors. My state equation models the beaviour of the NAIRU (random walk for now). So the state space could look something like this:

@signal inflation=c(1)*inflation(-1)+c(2)*inflation(-2)+c(3)*(uenmployment-NAIRU)+c(4)*(supply shock)+error
@state NAIRU=NAIRU(-1)+error2

First, I would like to make sure that both my errors are defined as N(0,variance). What's the code for doing that? As far as I've understood, the results are very sensitive to the signal/noise ratio, i.e. var(error2)/var(error1). How can I set a particular value for the signal/noise ratio? How can I make sure that there is a long-run Phillips curve by limiting c(1)+c(2)=1?

Any suggestions for estimating the signal/noise ratio for my particular data?

Thanks for your help!

Halfdan

[trubador]

Have you seen the previous discussion on this issue at: http://forums.eviews.com/viewtopic.php?f=4&t=4366

[Halfdan]

Have you seen the previous discussion on this issue at: http://forums.eviews.com/viewtopic.php?f=4&t=4366

Yes, I have, but I can't find the answers to my specific questions regarding the specification of the errors and how to set the signal-to-noise ratio.

Halfdan

[trubador]

As you might have noticed, error specification is pretty straightforward:

Code: Select all

[var = exp(c(1))]

If you want to set a particular value for signal/noise ratio, then you can:

Code: Select all

[var = exp(c(1))] [var = 0.5*exp(c(1))]

However, if you do not have a priori information, then it would be wise to estimate it along with your model (i.e. simply define a seperate error structure).
Finally, as for the long-run Phillips curve restriction, try:

Code: Select all

c(1)*inflation(-1) + (1-c(1))*inflation(-2)

If you need more details, please go over the related chapter in the EViews' user manual.