Pareto optimality in spatial voting models (Q1091239)

This paper studies the Pareto optimality properties of the policy proposals that k candidates (k\(\geq 2)\) will make when the candidates have some uncertainty about voters' choice. Specifically, although each candidate c knows voter i's preference relation \(R_ i\) over the set of feasible policies, S, c is still uncertain about how i will actually vote. Candidates, however, hold a common subjective probability distribution \(P^ c_ i(\psi)\in [0,1]\), for the event ''voter i will vote for candidate c if the vector of candidate platforms is \(\psi\) ''. Three possible payoff functions are considered. These are (i) Expected vote, (ii) Expected plurality vote, (iii) Expected proportional vote. A crucial role is played by local responsiveness conditions which are restrictions on the degree to which the probability that a voter will vote for a particular candidate is strictly increasing in the voter's utility of proposals in a neighbourhood of that candidate's proposal. Let \(\psi_{-c}\in S^{k-1}\) denote the vector of candidate strategies excluding c's strategy, while \((x,\psi_{-c})\) is the strategy vector when x replaces c's strategy in \(\psi\). \(P(R_ i)\) is the asymmetric part of \(R_ i.\) Definition: Voting is locally responsive for c at \(\psi_{-c}\) in \(Z\subseteq S\) iff \(\forall x\in Z\), there is some voter i such that \(P^ c_ i(w,\psi_{-c})\in P^ c_ i(x,\psi_{-c})\) whenever \((w,x)\in P(R_ i)\) and \(w\in N_{\delta}(x)=\{s\in S|\) \(d(s,x)<\delta \}.\) The first result shows that under quite general assumptions, if each candidate expects locally responsive voting ''close'' to the Pareto set, then every pure Nash equilibrium strategy vector must be Pareto optimal. The second result shows that under a stronger local responsiveness condition any proposal outside the Pareto set is strictly dominated by a proposal inside the Pareto set. Hence, in any mixed strategy Nash equilibrium each candidate will assign zero probability to the set of points outside the Pareto set.