dear professor Pfeifer
I have checked my file with a fully calibration version and it works, I also use command " identification", it says “All parameters are identified”.
but when I begin to estimate the model, I get:
POSTERIOR KERNEL OPTIMIZATION PROBLEM!
(minus) the hessian matrix at the “mode” is not positive definite!
=> posterior variance of the estimated parameters are not positive.
You should try to change the initial values of the parameters using
the estimated_params_init block, or use another optimization routine.
Warning: The results below are most likely wrong!
what seems to be the proble?
Without the codes it is impossible to tell. Check your observation equations and the
mercury.mod (11.9 KB)
mycmrobs.mat (4.2 KB)
thanks professor , here is my mod.file and data~
You posted the wrong codes. They return
ERROR: There are 40 equations but 41 endogenous variables!
mycmrobs.m (262 Bytes)
data.mat (4.8 KB)
mercury.mod (11.4 KB)
sorry, this one is correct~
Using the mod-file, I ran
estimation(datafile = 'mycmrobs.m',
mh_replic = 20000, mh_nblocks = 2, mode_check,
mode_compute =0,bayesian_irf,irf=20,moments_varendo,mh_jscale = 0.0003,mcmc_jumping_covariance='identity_matrix') k cxf omegabar sigma net bf_obs gdp_obs invs_obs q zetai b_obs,Z_obs ;
i.e. used an identity matrix instead of the Hessian. The resulting posterior shows a lot of persistence related to net exports. I am not sure your model is able to capture its dynamics.
net export NX appears in my model twice: GDP=C+I+G+NX, and an AR(1) SHOCK PROCESS：ln(NX)=pho_nx*NX(-1)+e_nx;
will it have a big impact on model behavior, and after I change the data source of NX, whics seems to be less persistent, it still does not work.
If mode-finding is such a problem, try running the MCMC with an identity matrix as the proposal density (as I did above) and see what happens.