I have been trying to run an estimation from Alpanda and Kabaca (2020) on a subsample of the original data (only EA and USA) but have been unsuccessful so far.
I’ve instead decided to start from square one and simply try to replicate their results with the original data. The authors have provided the _mode file, so you can get the exact same estimates, but I cannot figure out how they obtained that specific _mode file. I have tried running both mode 5, 6, 7 and 8 but all have failed as none achieve convergence. Also, the estimation will not run without the ‘lik_init=2’ option.
I have attached the data, the data transformation code, the original code, and their _mode file.
If anyone can provide me with some hints to achieving the same _mode file, I will be very grateful!
The model has unit roots, so you need the diffuse_filter or lik_init=2.
mode-finding is often a tricky and time-intensive process that may involve a lot of trial and error. That is the reason one provides mode-files. Tracking the way how you get there may be complicated. It would be more of an issue you were able to find a better mode.
I have now simply begun to estimate the shock persistence and -std. deviation on Euro Area data. The estimation runs with mode_compute=6, but I am unsure whether the mode_check and posterior plots are okay. I have attached them.
Running mode_compute=5 produces an error “Error using chol. Matrix must be positive definite.”. I have attached the log.
Any suggestion on how to move forward ? Do I just try all modes or do I have to try them in a specific order?
These are two different issues: the mode and the Hessian at the mode. Judging from the posterior plots, the MCMC did not converge and not sample the posterior yet. When I run mode_compute=5, I get a well-defined mode: tidy_mode.mat (35.0 KB)
Thank you very much for your kind help, Professor Pfeifer I have tried to expand the estimation to also estimate some structural parameters. However, the mode_check plots look off for some of them and if I increase the number of iterations to 50,000 it crashes as the matrix is not positive definite. Everything is attached. Do you have any suggestions as to how I might improve the results? tidy_PriorsAndPosteriors.zip (334.3 KB) tidy_log mode and check.zip (320.5 KB) tidy.mod (26.1 KB) tidy_data_growth.m (1.0 KB) EA_data_updated.xlsx (492.7 KB)
I hope you have time to look All help is greatly appreciated!!
I have tried to expand the estimation to also estimate some structural parameters. However, the mode_check plots look off for some of them and if I increase the number of iterations to 50,000 it crashes as the matrix is not positive definite. Everything is attached. Do you have any suggestions as to how I might improve the results?