If the initial parameters did not solve, the code for displaying results and saving was not reached. The also led to later crashes with saving as the -append option was used and no file existed. If no valid draw can be found, the function exits in any case, so no additional check is needed.
(1) Added more checks on the content of the provided mode file compared the the list of declared parameters (the condition on the number of parameters is not strong enough).
(2) Added a mechanism to adapt the content of the mode file if possible. For instance, if the estimated parameters are a subset of the parameters in the mode file, we only need to discard some of the parameters in the mode file.
(3) Added output argument in dynare_estimation_init, which returns the hessian matrix (hh) with the estimated mode.
2) fixed bug for the ML base (bayestopt_.pshape==0);
3) make more transparent the error when the model does not solve at prior_mode or prior_mean;
4) when prior_mean or prior_mode fail, try 50 times with randomly generated samples from the prior before interrupting identification.
This commit fixes issues related to fabiac forum request and subsequent discussion with Johannes.
http://www.dynare.org/phpBB3/viewtopic.php?f=1&t=4402&p=10771&hilit=identification&sid=64c6f9d987a2641e79dd5722137eb483#p10771
1) allow to compute derivatives starting from NUMERICAL derivatives of jacobian and steady state: this has a minor cost in accuracy and allow apply without errors identification and estimation with numerical derivatives;
2) added trap in dynare_estimation_init: if steadystate changes param values, automaticly shifts to numerical derivs of jacoban and steady state + analytic derivatives of all the rest;
3) bug fixes for 2nd order derivatives w.r.t. model parameters;
-) never use normalization for SVD and brute force covariance checks;
-) Jacobians are normalized for checking the rank condition: this should allow for more uniformity in results across models.
-) removed obsolete commented lines or unused variables;
-) when Asymptotic Hessian is available, take SVD of that for identification patterns;
-) when computing analytic Hessian, scores are not necessary (no_DLIK=1);
-) fixes in output info when Hessian is available.
9747cb78940f41ba3d6c262b2e77a0aa01576d3d
730100282f17974d86510a9ebbaad735f27c46ba
* 1) bug fixes: non-initialized ide_strength when no param is identified; normaliz1 set to one when prior std is inf; HIGHEST SVD bar fixed (wrong singluar values were plotted).
2) Simplified advanced output (no more multicollinearity and pair-wise correlations!).
3) Beautified output of collinear groups of variables.
4) use of sensitivities normalized with std errors.