hi!
I am learning the sw2007 code in your homepage. I had noticed that the parameters value are set in the mode. why is that? and when we do bayesian estimation ,we just set the prior of the parameters. do we need to calibrate parameters before estimation?
Hi, I guess this is just a shortcut to get results more easily, ie the optimization is initialized with values for the parameters found in a previous estimation. So if the provided values are the posterior mode, when you run again the optimization (triggered by the estimation command) the estimate of the posterior mode should not improve. It is not mandatory to use these values.
Best,
Stéphane.
1 Like
Adding to the answer: it is recommend to always fully calibrate your model before estimation. See Remark 2 (Using stoch_simul before Estimation) from Pfeifer(2013):“A Guide to Specifying Observation Equations for the Estimation of DSGE Models”
1 Like
thanks for your reply!
i still a little confused about the code.in the //estimated parameters initialisation block. the values are fully calibrated right? and in the estimated_params block , i see INITVAL which means initial value . so what is the reason to set these value?which values will be initialized to find the mode? thanks!
1 Like
Which file are you talking about?
hi! thanks for your reply
I am talking about the Smets_Wouters_2007 code in your github .
my question is how you set the parameters value in the following blocks in dynare.
// fixed parameters
// estimated parameters initialisation
i also noticed the estimated parameters block ,
estimated_params;
the parameters under these blocks are very accurate ,are they calibrated?
Parameters resulting from calibration are the ones that stay fixed during estimation. You need to set those before estimation. That is why the parameters under
// fixed parameters
are set according to what Smets/Wouters (2007) wrote in their paper. The parameters under
// estimated parameters initialisation
are set according to the posterior estimates in the paper. I set those to allow working with the model without estimating it (it was a common user request). In practice, if you are only interested in estimation, you would not need those as the parameters are subsequently estimated.
The initial values in the
estimated_params;
are again set to the posterior mode reported in the paper. The idea is the one @stepan-a discussed in his post above. Again, if you are interested in estimating the model, you do not have these values and usually set the initial value to your best guess. This often is the prior mean, which is what Dynare uses when you do not provide any values here.
1 Like
Yes ! I have the same confuse. What is the “INTVAL” block means? And If I am just getting ready to use Bayesian estimates, how can I make an initial estimate without a calibration value?
Please see the Dynare manual located in the doc-folder of your installation.