Dear All,
I would like to perform a non-linear estimation exercise by setting the starting points of the parameters at the posterior mode from a complete linear estimation e.g mode compute plus MC exploration.
Would the following set of commands perform what I need?
STEP 1: set parameters at the mode
set_param_value(‘alpha’,oo_.posterior_mode.parameters.alpha);
%
M_.Sigma_e(1,1)=oo_.posterior_mode.shocks_std.epsM.^2;
M_.Sigma_e(2,2)=oo_.posterior_mode.shocks_std.epsZ.^2;
STEP 2: define priors starting point:
estimated_params;
alpha, alpha, , , uniform_pdf, , , 0.3, 1;
stderr epsM, 0.06, , , uniform_pdf, , , 0.00001, 100;
stderr epsZ, 0.07, , , uniform_pdf, , , 0.00001, 100;
ME
stderr C_obs, 0.004, uniform_pdf, 0.00001, 100;
end;
If this is fine, is there a way of calling elements from the shock standard deviation matrix M_.Sigma_e within the prior block?
In STEP 1, alternatively, I was considering using a string match on elements of matrix M_. after the command:
param_post_mode=get_posterior_parameters(‘mode’,M_,estim_params_,oo_,‘lik_’)
% INPUTS
% o type [char] = ‘mode’ or ‘mean’.
% o M_: [structure] Dynare structure describing the model.
% o estim_params_: [structure] Dynare structure describing the estimated parameters.
% o field_1 [char] optional field like ‘mle_’.
Many thanks in advance.
Best regards,
DB