This paper is concerned with the design of non-cooperative game forms for economic decision problems. A decision problem is presented which admits non-dictatorial game forms with the following properties: Nash equilibria exist and all Nash equilibrium outcomes are Pareto optimal; or dominant strategies exist and all dominant strategy equilibria are Pareto optimal; but not both. This is, any (non-dictatorial) game form whose Nash equilibria are well behaved does not have dominant strategies, and any game form with well behaved dominant strategy equilibria must have at least one non-optimal non-dominant strategy Nash equilibrium.

Economic theorists have interpreted the "efficient markets hypothesis" to assert that equilibrium asset prices reveal all decision-relevant information in the market. This paper establishes conditions on investors' utility functions of future wealth which are necessary for the efficient markets hypothesis to be satisfied and be robust to slight perturbations of endowments and the joint distribution of current information and future asset values. The main result states that over the relevant range of future wealth values, there are three possible cases: (i) all investors are risk-neutral; (ii) modulo a change in the wealth origin, each investor has constant relative risk aversion with the same constant for all investors; or (iii) all investors have constant absolute risk aversion.

Non-manipulable direct revelation social choice functions are characterized for societies where the space of alternatives is a euclidean space and all voters have separable star-shaped preferences with a global optimum. If a non-manipulable choice function satisfies a weak unanmity-respecting condition (which is equivalent to having an unrestricted range) then it will depend only on voters' ideal points. Further, such a choice function will decompose into a product of one-dimensional mechanisms in the sense that each coordinate of the chosen point depends only on the respective coordinate of the voters' ideal points. Each coordinate function will also be non-manipulable and respect unanimity. Such one-dimensional mechanisms are uncompromising in the sense that voters cannot take an extreme position to influence the choice to their advantage. Two characterizations of uncompromising choice functions are presented. One is in terms of a continuity condition, the other in terms of “phantom voters” i.e. those points which are chosen which are not any voter's ideal point. There are many such mechanisms which are not dictatorial. However, if differentiability is required of the choice function, this forces it to be either constant or dictatorial. In the multidimensional case, non-separability of preferences leads to dictatorship, even if preferences are restricted to be quadratic.