Hi everyone,

I am estimating a DSGE model for two subsamples because observed GDP is not variance stationary over the entire dataset. The first subsample has higher GDP volatility, and the second subsample has lower GDP volatility (which started after a structural adjustment program was implemented).

My problem is that the correlation between `C`

and `I`

has different signs for the 2 subsamples. Does this mean that the nature of shocks over the business cycles for the 2 samples are different? And if so, can I use different sets of shocks in the 2 models for the 2 subsamples?

Thanks for taking the time to write a reply.

This is tricky. In general, there are two things that can be the case:

- The structure of the model and the stochastic processes have not changed at all, but the actual shock realizations in the two subsamples are very different, which can just happen by chance.
- There is indeed a structural break that causes the covariance structure to differ.

Now, in your case you allude to a structural adjustment program. This suggests that there was indeed a break. So you may want to estimate the model for the two periods separately, possibly using the first period results as the prior for the second one.

Hi Prof. Pfeifer, many thanks for your response.