Hi Prof. Pfeifer ,

I have a linear model with stationary variables. I evaluate it using the Bayesian method. I use the same HP filter for the data and for the model. The reviewer of my article says that the HP filter has not been used for a long time and refers me to the article “Estimation of DSGE models when the data are persistent”(2010), (Yuriy Gorodnichenko, Serena Ng ).

But as far as I understand, this article talks about filtering variables in non-linear models.

Help me please

Thank you,

Leonid.

Can you please explain how exactly you proceeded in your paper?

Hi Prof. Pfeifer ,

I worked in the standard way

For the time series of data, the trend components were removed using a one-sided Hodrick-Prescott filter. Then the linear model with stationary variables was estimated in a standard way by the Bayesian method using the obtained detrended data.

What did I do wrong?

Help me please

Thank you,

Leonid.

You did not say “one-sided” before. Using a two-sided HP filter would clearly be problematic. The one-sided one is a more defensible choice. It has for example been used in the context of Bayesian estimation in

See also

That being said, pretty much any filtering device will be controversial as you need to (implicitly or explicitly) take a stand on the underlying DGP, where your prior may be different from the referee’s. See also

Dear, Pfeifer,

Can i use filter FD (first differences) to filter data and model variables of linear model with stationary variables. Then compare the received moments with the one-sided Hodrick-Prescott filter moments and choose the best one.

Thank you,

Leonid.