This project is read-only.

Calibration and estimation

Topics: Database Management, Dates, FAVAR, IRIS in General, Kalman Filtering, Models, Neural Networks, Reporting, Times Series, VAR, SVAR, FAVAR
Jun 8, 2014 at 12:35 PM
Edited Jun 8, 2014 at 2:09 PM
I have some question to group:
  1. can I calibrate in same time all parameters in a file, and estimate some of them, which are not structural in other matlab file?
  2. How to run impolse responde function if my standard deviation is calibrated?
  3. If I have different specification of shocks, exemple:
-epsilon_pi, which value in steady state is zero, and it apear in this form in equation (.....+0) sum relation
-exp(epsilon_h) the value of this expresion shoul be 1 ( product relation( (x*exp(0))....
thus, I don't need the log specification of shocks, but in levels. Can I do it in IRIS?
  1. If I don't have trends in model how to write corect this secvence of code (Tutorial SPBC)
    filterOpt = { ...
    'outoflik=',{'Short_','Infl_','Growth_','Wage_'}, ...
    'relative=',true, ...
optimSet = { ...
'maxFunEvals=',10000, ...
'maxIter=',100, ...
[est,pos,C,H,mest,v,~,~,delta,Pdelta] = ...
estimate(m,d,startHist:endHist,E, ...

Great thanks in advance!
Jun 9, 2014 at 7:51 AM
  1. Yes, of course. Calibrate parameters by whatever method you want. When you call estimate you choose which parameters you want to estimate. If you want to leave the parameters calibrated, don't put them in the prior structure.
  2. You can create impulse responses using model/simulate() with the 'anticipate=' option set to false and non-zero shock(s) in some period in the input database, and you can specify the shock size in the input database.
  3. You need to modify the filter options. As you go through the tutorials refer either to the IRIS Manual or to the help pages for model/estimate() and model/filter() to understand the settings.
Marked as answer by michaeljohnston on 6/8/2014 at 11:51 PM
Jun 9, 2014 at 10:43 AM
And in regards to shock specification, shocks in IRIS are always in levels. IRIS only supports Gaussian shocks in state space models and there is no sensible log transformation.