Risk aversion and generalized Schur (QZ) decomposition

Dear Dynare team,

I am using Dynare (4.5.7 and also 4.5.3) to compute a model with super high risk aversion. I noticed that if I set the relative risk aversion to be more than 15, then Dynare complains that “generalized Schur (QZ) decomposition failed”.

When I run model_diagnostics(M_,options_,oo_), I got message about The presence of a singularity problem, which comes from the collinearity problem.

These errors only occur if I set risk aversion to be super high, e.g. higher than 15.

I am wondering does it mean that Dynare cannot handle large risk aversion due to the QZ decomposition?

Thank you very much in advance. Looking forward to your helps.


This is most probably not a problem with the QZ decomposition per se, but rather an numerical issue introduced by the high exponent. Say consumption in steady state is 0.3. Marginal utility will be 0.3^(1-15)=4.7830e-08
Note that you are already moving dangerously close to machine precision.