- You need at least second order
- You need to correctly specify the auxiliary variables arising in EZ (see github.com/JohannesPfeifer/DSGE_mod/blob/master/Caldara_et_al_2012/Caldara_et_al_2012.mod)
- To avoid numerical issues, you should introduce a normalizing constant, see [Endogenous Growth Model)

How do you define “accurate”? We verify that the third-order approximation of Dynare is correct by cross-checking it with the Mathematica results of Fernandez-Villaverde et al (2012) “Risk matters”. If you have Euler errors in mind, then the answer depends on the respective model and you need to check. A general guide on accuracy is the Caldara et al (2012)-paper.