AR(1) with autocorrelation 0.9998

I am estimating a model DSGE where I find mode of autocorrelation of one AR(1) process near to 1, especifically 0.9998 with standard deviation 0.0000. So my question is if it is a serious problem? and what could I do if I want to solve the possible problem?

Thanks a lot

Hi Aldo,

When your persistence parameter hits the upper bound then this could be due to a wrong observation equation. Have you checked whether the observation equation matches the detrended data? It seems your data wants have a unit root, which later materialises in a high autocorrelation, such that the model “matches” the data.



Thanks dear Robert,
I took first-log diference . Problem appears in parameter rho_g. I attach plots of my data and results prometheo_TeX_binder_onlyrhophitilde.pdf (862.7 KB)

Excuse me, any other suggestions please?

It’s hard to tell, but @Robert’s intuition is correct. If you look at the mode_check plot, you will see that the likelihood favors a unit root - and only the prior pushed the autocorrelation slightly below 1. This indicates that there is some persistence in the data that the model is unable to capture with stationary exogenous processes. The most common case is a forgotten constant in an observation equation. In this case, the only way to generate a permanent deviation from the steady state is to have permanent shocks. If you can exclude this is the case, then there may simply be too much persistence in your data for the model to handle, indicating model misspecification. For example, data_RstarU and PD_HHobs do hardly look mean-reverting in your sample.

1 Like