ZLB in getting_started.ipynb

[NicoGreggio]

Hello everyone,
first of all thank you Gregor for this fatastic library. I have a doubt about the zero-lower-bound ZLB in the example getting_started.ipynb. I do not understand why after running the Transposed Ensemble Kalman Filter (TEnKF) I get (as in the example) R values that are less than zero when the model should not allow it (also because the historical data never goes below zero) .
Thanks in advance

[gboehl]

Hi Niko,
Thanks for reaching out. As everything in the linearized model, the interest rate is expressed in terms of (percentage) deviations from steady state. The example model is calibrated to a steady state interest rate of 4% annualy, thus the steady state is ~1% quarterly in the model. The zero lower bound then is located around -1% in terms of deviations from st.st..

Have a look at my econpizza package if you are interested in fully nonlinear simulations in python.

Cheers,
Gregor

[NicoGreggio]

Hi Gregor,
thank you very much for your speed and accuracy.
Wow! In the coming weeks I will also study the econpizza library then 🥇.
Back to the question, if instead of being interested in the percentage deviations from steady state I want to see the values in levels of interest rate and GDP, should I use the mod.obs function?
I am interested to see if the model performs well in historical values and then see if it produces appropriate forecasts.

[gboehl]

Yes, exactly. mod.obs is what you want.

The model is just a small skale example NK model and will not perform well empirically. For that purpose you need at least a Smets-Wouters-type of model. This repo can get you started: https://github.com/gboehl/projectlib/tree/master/yamls

You should have a look at my ZLB estimation paper with Felix on ZLB estimations with this type of model: https://gregorboehl.com/live/bczlb_BS.pdf

[NicoGreggio]

Thank you very much Gregor,
if you pass through Italy I owe you at least a beer!