Hp Filter or not?

Dear All,

I am currently writing a DSGE model in a log linear form. In the model, there are both productivity and monetary shocks. Each shock follows an AR(1) process with persistent coefficient below 1. The aim is to compare the standard deviations of the model with the ones observed in the data. The data are Hp-filtered with a coefficient lambda=1600.

My question, maybe trivial for some of you, is the following: should I include the option “hp_filter=1600” in my stoch_simul block ? At first sight I would say no, since there is no growth in this model. But I’ve seen some examples with similar frameworks that include such a filter.

Thanks in advance for your reply.


One of the rules is to treat your data and your model consistently. If you use first differences for your data, do so in the model as well. If you use the HP-filter to extract trend deviations in the data, do so as well in the model.

The underlying idea is twofold:

  • You make sure that that the things you study are comparable. Filtered data considers only a particular frequency band that is thought of as “business cycle frequencies”. You want your model to explain the business cycle, i.e. you want it to perform similar in that particular domain/frequency band.
  • By treating the data and the model similarly, you make sure that all distortions and biases that might apply to the data treatment are also present for the model. This is related to the Sims-Cogley-Nason-approach. See Christiano, /Eichenbaum/Vigfusson (2006). “Assessing structural VARs”. NBER Macroeconomics Annual. Chap. 1, 1–106

Thank you very much for this quick and helpful answer.

All the Best