This paper proposes a new decomposition of welfare effects simulated by CGE models. To date, welfare decompositions are based on first order approximations of the CGE specifications. These locally-based decompositions may have a poor explanatory power from an empirical standpoint or may be path dependent. Our approach overcomes these issues and is based on Taylor series approximations of CGE specifications. Then it is a generalization of current ones which still allows to attribute changes in welfare to sources corresponding to the alleviation, or exacerbation, of existing market imperfections and distortions. Our decomposition approach is also attractive because that i) it can be applied to any globally regular representation of preferences, ii) it can be implemented in both level and linearized CGE models and iii) it eases the comparisons of welfare effects across individuals. We implement our approach to a widely used CGE model and show empirically that it performs well in most cases.
