Dear Professor Pfeifer,
I was interested in the IRF matching procedure you have uploaded on your GitHub. I have used your code and your RBC model applied to Italian Data (Consumption, Output, Investment and Government spending) from 2007-2020 and it works well.
However, I tried your code on a more complex model (Kirchner & Wijnbergen (2016)) downloaded on the Macro-model data base that I have pimped a bit (Usefull public spending and Non Ricardian household) in order to get a positive response in consumption.
When I use this new model the IRFs for output Public spending and consumption start oscillating although this is not the case for standard IRFs produced by dynare .
I do not understand why the same code works for one model but not for the other one?
my guess is that it is due to the estimation of the AR(2) 's parameters?
BestMatching_my_model.zip (24.2 KB)
Please find attached the different files I have used
It seems the reason is that your empirical IRFs are oscillating and the model now mirrors that.
Dear Prof Pfeifer,
I am sorry for jumping into the discussion. Is there a way to do IRF matching for 3rd order approximation (i.e uncertainty shock) ? If yes can you briefly illustrate how it can be done? Thank you so much.
The code at DSGE_mod/RBC_IRF_matching at master · JohannesPfeifer/DSGE_mod · GitHub should be straightforward to modify for third order. Within
DSGE_mod/IRF_matching_objective.m at master · JohannesPfeifer/DSGE_mod · GitHub
you would need to put the code for the IRFs at third order, which will depend on which type of IRFs you want. Do you want to use GIRFs at the stochastic steady state/EMAS?
Dear Prof Pfeifer,
Thank you for your prompt reply. Yes, indeed, I simulate the uncertainty shock using NLMA toolbox from Meyer Gohde and Lan (2013). This toolbox gives GIRFs at the stochastic steady-state and does not use the pruning method if I understand correctly. I award of Dynare do it with pruning and IRF from the ergodic mean (not sure about this point).
Is it easier to switch to Dynare 3rd order simulation for IRF matching?
Can it produce a huge difference from using NLMA toolbox ?
Thank you so much for help.
The basic framework should be able to accommodate the NLMA toolbox as well, but may require more work to integrate it.