The accuracy of welfare computations

With the development of DSGE modelling, welfare is often used as a criteria for comparison of alternative policies. As one must rely on numerical approximation to solve DSGE models, one should wonder abnout the accuracy of welfare measures used in the comparison. Kim and Kim had already shown a famous case where linear approximation can lead to welfare reversal. In this paper, I compare welfare measures obtained by different approximation methods, perturbations and projections on simple and less simple DSGE models. Both conditional and asymptotic welfare are examined as well as approximation of asymptotic measures obtained by pruning