Dear professor Pfeifer
I currently working on the project using business cycle accounting method and experiencing non-positive definite hessian matrix problem when model is estimated using MLE.
Sample codes for CKM 2007 experiences similar problem, as you mentioned in footnote in modfile.
‘CKM use a penalty function that punishes eigenvalues of the VAR processes for the wedges bigger than 0.995 (see footnote a) to Table I, p. 800). Because this cannot be easily done in Dynare the penalty is omitted, leading to a somewhat bigger maximum eigenvalue. However, this pushes the VAR coefficients closer to the boundary of the stability region. While hardly affecting the point estimates and conclusion of the paper, it makes the Hessian at the mode not positive definite and renders the standard errors invalid and the standard errors’
Is this problem can be solved using options developed in current version of Dynare or using Bayesian technique?
If not, could this problem can be solved by another data construction method (e.g. another detrending method different from original CKM)?