Could you help me?
I’ve tried to estimate model, but results are strange. I’ve used maximum likelihood estimation with errors of measurement.
Some of standard deviations of parameters estimations are extremely small (T-stats more then 10000). As I understand, it means problems with convergence. But each of algorithms that I’ve used can’t improve likelihood function. I’ve tried your default algorithm (mode_compute=4), fminunc, fminsearch and my stochastic search algorithm (for that algorithms I’ve changed dynare code).
Do you have any ideas why it could happen?
Thank you for help
So small standard errors are often the sign that optimization stopped agains a boundary. It may be that your model wants parameter values that are outside the range you thought was possible or hit against the boundary of unicity of a statble trajectory.
You may try to estimate it without some observable variable to try to detect which observed variable creates the problem.
Note also that it is much more difficult to estimate with ML rather than with Bayesian methods.
Thank you for ideas. I’ll try to use Bayesian method (I thought that it’s calculation have the same difficulty as ML).
As I understand, after detecting of variable which led some of parameters to bounder, I have to make some changes at model which prevent this. I’ll try to use this way.
Could you help me with Bayesian method?
I’ve tried to use it, but matlab display a error. After calculation of density I see the error. I’ve make most of parameters equal to constant, but it doesn’t help.
Last row of correct message + error message:
Log data density [Laplace approximation] is -892293.025081.
??? Error: File: C:\MATLAB6p5\work\dynare_v3\matlab\metropolis.m Line: 1464 Column: 22
Expected a variable, function, or constant, found “)”.
Ошибка в ==> C:\MATLAB6p5\work\dynare_v3\matlab\dynare_estimation.m
В строке 776 ==> metropolis(xparam1,invhess,gend,data,rawdata,bounds);
Ошибка в ==> C:\MATLAB6p5\work\dynare_v3\modeli\modelka2.m
В строке 172 ==> dynare_estimation(var_list_);
Ошибка в ==> C:\MATLAB6p5\work\dynare_v3\matlab\dynare.m
В строке 26 ==> evalin(‘base’,fname) ;