Epstein-Zin preferences introduce big nonlinearities into the model that sometimes lead to numerical problems. Your U1sigma is about 21 in steady state. It enters as
exp(U1sigma)which is huge and can result in overflow. Your are lucky in that the residual is still within Dynare’s accepted tolerance. Thus, there is not too much to worry. But best practice in these cases is to introduce a normalizing constant into the utility function that makes U equal to 1 in steady state. That avoids most problems of this type.