This set of codes is based on IMF's Quarterly Projection Model
- Credits to IRIS Solutions team (IRIS Toolbox)
- The codes are obtained from IMFx MPAFx Monetary Policy Analysis and Forecasting on EdX.
Please go to the site and enroll if you are interested.
The data is likely for the Czech Republic.
In the original codes provided by IMF, the model used is a calibrated one. Calibrating a large system such as the QPM can be challenging and requires a lot of tedious trial and errors. In a lot of cases, we also want to be able to estimate the parameters so that the output is as close to the data as possible.
In this repo, I added three scripts that do exactly this, with 3 methods (maximum likelihood, Bayesian, and System Priors).
The approach is based on A Cook-Book of IRIS by Jaromir Benes & Martin Fukac.
I adapted the codes there to produce a07_estimate_mle and a07_estimate_bayesian that perform the relevant estimation and then save the posterior mean into pE.
The script readmodel_est.m will replace readmodel.m in this case.
Benes later added a System-priors approach (here), which can only be implemented with IRIS 2021 onwards.
If you want to try this approach, run a07_estimate_system in the repo and remember to use IRIS 2021 instead of IRIS 2018.
(Note: In the Bayesian estimation, the old 6-element format {init, lb, ub, 'beta', mean, std} worked in IRIS 2018 but will fail in the 2021 version. The 2021 version only accepts distribution.Distribution object, a function handle, or a numeric penalty as the 4th element. If we follow the old syntax, a plain string like 'beta' doesn't match any of those checks and is ignored, leaving all priors as "Flat", and the estimation will default to MLE)
Based on this repo, I wrote a dynare implementation of the current QPM model. You can run qpm_dynare.mod via the command dynare qpm_dynare.mod (make sure to add DYNARE path first). Note that the end result is not very good because the Log-lik is about -4128, while other methods produce the Log-lik of about -622. I think there is semething wrong with my dynare code.
The calibrated model
The estimated model
- The calibrated model (IMF's EdX version)
a02_makedata -> a03_kalmanfilter -> a05_forecast
- The estimated model
a02_makedata -> (any) a07... -> a03_kalmanfilter_est
Make sure to enable the IRIS-Toolbox-Legacy-20211206 in start.m instead of IRIS_Tbx_20181028.
- Try dynare
copy history.csv from results folder, change the variable name of the observed variables to match varobs in the dynare file. Then run dynare qpm_dynare.mod
- This repository is for educational and research purposes only. This archieved version of the IRIS Toolbox here is obsolete and no longer in development.
- The author of this repository does not claim ownership of the original IMF QPM framework, IRIS Toolbox, or the IMFx/MPAFx materials.
- The materials may be subject to copyright and intellectual property rights of the IMF, IRIS Solutions Team, and/or their respective contributors.
- No official endorsement by the IMF, IRIS Solutions Team, or EdX is implied.
- Users are responsible for ensuring compliance with all applicable terms of use, licenses, and citations when using or redistributing these materials.