The RATS Software Forum

Hi,

I'm trying to estimate potential output of industrial production in two versions of a univariat model. In the first version, the trend follows a RW with drift. The result looks reasonable ok though the cycle seems to be trending over time. I had to peg one variance in the transition equations to zero.

In the second version, potential output is modelled as a local trend. I peg two of the variances in the transition equations to zero. My problem is that the estimated potential output mimicks industrial production hence making the cycle almost zero.

I suppose there are other problems involved here as well but I can't figure them out now. Would greatly appreciate any suggestions on how to improve the estimations. I have attached the programme file and the data file.

best

Marcus

There's nothing wrong with the program. However, quoting from the State Space e-course:

3.3 Estimating the Local Trend Model
Before you do this, you should first ask whether you really want to. Attempts
to estimate the variances in local trend models often produce results which
look “wrong”. In particular, the common problem is that maximum likelihood
produces a trend which isn’t stiff enough to fit our understanding of what a
trend should be. In short, there’s a reason that the HP filter has pegged rather
than estimated ratios.