Intermediate preferences and Rawlsian arbitration rules

Given the preferences of two agents over a finite set of alternatives, an arbitration rule selects some fair compromise. We study the idea that more consensus should not be harmful: the closer your preferences are to mine (in the sense of Grandmont's (1978) intermediate preferences), the better I like the selected alternative. We describe several Pareto optimal rules satisfying this principle. If, in addition, a condition akin to Suppes' (1966) grading principle is imposed, the rule must always choose an alternative maximizing the welfare of the worst-off agent, measured by the number of alternatives that he finds worse than the chosen one.Stimulating discussions with H. Moulin and helpful comments from J. Crémer are gratefully acknowledged. The author also wishes to thank a referee and an associate editor for challenging remarks. This research was partly supported by a CAFIR grant from the Université de Montréal.