Page 1 of 1
Thresholds
Posted:
Tue Jan 20, 2015 10:22 amby eta
Hi, I'm running a deterministic model which has different versions and generally solves correctly, meaning that it is analytically correct and produces sensible results. However, it happens that for certain values of the shock or of some parameters, the model produces flat dynamics. Does it mean that there are thresholds in the model ? Thanks
Re: Thresholds
Posted:
Tue Jan 20, 2015 1:09 pmby jpfeifer
Are you using Dynare 4.4.3? In older versions completely flat lines were sometimes given as output when the model solution completely failed. Are the diagnostics Dynare provides in this case sensible (e.g. residuals should not be NaN).
Re: Thresholds
Posted:
Tue Jan 20, 2015 1:31 pmby eta
Actually, none of the residuals of the static equations is NaN. However, in these cases the model solution fails, as you said. Probably this is due to large shocks (when I reduce the size it solves) and a large wage asymmetry. This means I have to set a relatively low value for this parameter if I want a such that reproduces my targets.
Re: Thresholds
Posted:
Tue Jan 20, 2015 1:41 pmby eta
By the way, I checked residuals (resid(1)) of another model that I have attached and that generally works. I get some big values. Is it OK ?
Re: Thresholds
Posted:
Tue Jan 20, 2015 5:38 pmby jpfeifer
You cannot use resid(1) to check convergence of deterministic simulations. But you can set
- Code: Select all
options_.debug=1;
before simul and will see
-----------------------------------------------------
-----------------------------------------------------
MODEL SIMULATION:
Largest absolute residual at iteration 1: 0.020
Largest absolute residual at iteration 2: 0.016
Largest absolute residual at iteration 3: 0.050
Largest absolute residual at iteration 4: 0.002
Largest absolute residual at iteration 5: 0.000
Simulation concluded successfully after 5 iterations.
Total time of simulation : 0.094
Convergency obtained.
-----------------------------------------------------
So everything is fine for the posted file.