EViews.com

Hello everyone,

I am trying to estimate potential output using the univariate Kalman filter. I am following the Clark (1987) model which assumes that gdp is composed of a trend component and a cyclical component. The trend is modelled as a random walk process with a drift, the drift is assumed to be a pure random walk and the cycle is assumed to follow an AR(2) process.

When I run my model using the log of gdp, I get the error: ""WARNING: Singular covariance - coefficients are not unique". However, when I run my model with gdp then the estimation works.

My questions are the following:
1) Is it ok if I use gdp instead of log(gdp)?
2) If I want to obtain the estimated the trend and cycle, are those the "filtered state estimates" or the "smoothed state estimates"?


Many thanks in advance!
Best,
jana

The Clark model is hard to estimate for a variety of reasons.

It is not okay to do GDP instead of log(GDP), although it's hard to say how far off it is.

You probably want the smoothed estimates, although it depends what you're going to do with them.

Thank you for your answer startz.

I just want to estimate the potential output and output gap using a variety of methods (HP filter, Kalman filter, Production function approach). For those purposes, should I use the smoothed estimates from the Kalman?

Probably smoothed.

Thank you!