Question(s) on Cogley-Primiceri-Sargent (2010) replication

Hi all,

I’m currently working on a replication using the model presented by Cogley et al. 2010. The model code comes from the US_CPS10 mod file on the MMB-rep github repo.

I’m quite new to Dynare, and as such I was hoping to get some clarity on some questions.

  1. The parameters fp and fy (corresponding to taylor rule params for inflation and output respectively) appear inverted. That is, the prior mean of fp matches closely to the posterior mean of fy and vice versa, I’m wondering if this is perhaps a feature of my dataset or whether there is an error in my model code. I have gone through the equations but nothing stands out to me as an obvious error that might be causing this.

  2. Log data density: My estimated model has a positive log data density (=138), I understand that this is not necessarily an outright issue, but my understanding is that this is somewhat of a red flag. Are there any “usual suspect” areas that I should be looking at to investigate this?

  3. Unit roots: In the original CPS paper, there is no mention that I can see of filtering non-stationary variables. Specifically, the interest rate variable, R, in my dataset is non-stationary. Is it generally recommended that I apply a transformation to this series to get R as an I(0) variable?

Appreciate any assistance/direction you can provide.
US_CPS10.mod (5.0 KB)

Can you provide more details? Is the problem with your own data? Or the original one?

I am using data on the New Zealand economy to estimate the model (attached).
DSGE_DATA_CPS.xlsx (13.5 KB)

y is the difference in log gdp per capita, p is inflation calculated as quarterly difference in gdp deflator, R is the interest rate. I use the 90-day-Tbill rate as a proxy for the official cash rate.

Here is the log file I get after estimating the model.
US_CPS10.log (8.0 KB)
The parameters I am concerned about are fp and fy in particular as they are significantly different to those in CPS. I would expect some difference due to my data being different, but given most other parameters are close to those in CPS these two stood out to me as outliers. Appreciate any advice/insights you can offer me.

The mod-file is ill-suited for estimation. The parameter dependence is not correctly handled and there is no proper observation equation taking care of the non-zero means in the data. Please refer to
See Pfeifer (2013): “A Guide to Specifying Observation Equations for the Estimation of DSGE Models".