Non-negative variable

[jpfeifer]

That’s the problem with linearizations:

Non-negativity constraints

First, Dynare does not impose non-negativity constraints. How should it do that and know for which variables such a constraint applies? Second, in a linearly approximated model, there are no bounds on the endogeneous variables as every policy rule is linear. Hence the curvature implied by the Inada conditions is always lost. Using higher order approximation may make the problem better but will not get rid of it (unless you go to order infinity)