#257: Broken estimation of models with non-diagonal covariance matrix for shocks and calibrated correlation ------------------------+--------------------------------------------------- Reporter: sebastien | Owner: stepan Type: bug | Status: new Priority: critical | Milestone: 4.3 Component: Estimation | Version: 4.2.5 Keywords: | ------------------------+---------------------------------------------------

Comment(by stepan):

Fixed in git#e12748329be28ee1b960744de36fab0b4ff2c8b1. Note that if the user calibrate a covariance, it is implicitly assumed that the correlation is calibrated.

#257: Broken estimation of models with non-diagonal covariance matrix for shocks and calibrated correlation -------------------------+-------------------------------------------------- Reporter: sebastien | Owner: stepan Type: bug | Status: closed Priority: critical | Milestone: 4.3 Component: Estimation | Version: 4.2.5 Resolution: fixed | Keywords: -------------------------+--------------------------------------------------

Comment(by jpfeifer):

The issue is not solved as e.g. dsge_likelihood does not use set_all_parameters.m to update the covariance matrix in each step. The attached mod-file should be added as a unit check.

Currently the attached mod-file leads to a complex likelihood as the covariance matrix is not positive definite. This shows another issue related to ticket 59. As complex likelihoods only result in a penalty, but not in an error code different from 0, initial_estimation_checks.m does not display an error message with these starting values.

#257: Broken estimation of models with non-diagonal covariance matrix for shocks and calibrated correlation -------------------------+-------------------------------------------------- Reporter: sebastien | Owner: stepan Type: bug | Status: reopened Priority: critical | Milestone: 4.3 Component: Estimation | Version: 4.2.5 Resolution: | Keywords: -------------------------+--------------------------------------------------

Comment(by stepan):

Fixed in git#06ca265272d278d6f4ec9b746c33b5b92da62de9 (checked with fs2000+calibrated correlation between e_a and e_m). I did not add as2007_correlated_shocks.mod to the test suite because I was not able to find a mode (ie positive definite hessian at the mode) with one call to csminwell.