Multi-period decisions are decisions which determine an individual's payoffs in several periods in the future. This paper examines the theoretical foundations of the prevalent weighted average assumption. More specifically, we use a multi-period interpretation of the famous Ellsberg paradox in decision under uncertainty to show that in many cases of interest additively-separable functionals (in general) and weighted average ones (in particular) do not seem appropriate for the representation of the decision maker's preferences. We then suggest replacing the sure-thing principle, which may be used to axiomatize a weighted average functional, by a weaker version of it. Using the weakened axiom in Schmeidler's nonadditive measure model (reinterpreted for the multi-period context) yields an axiomatization of a larger class of decision rules which are representable by a weighted average of the utility in each period and the utility variation between each two consecutive periods. The weighted average assumption is a special case of the generalized model, a case in which the decision maker is variation neutral. Similarly, we define and characterize variation aversion and variation liking, and show an example of the economic implications of these properties.