Stochastic stability in best shot network games (Q423720)

The paper under review studies stochastical stability of best shot network games under perturbations. The authors show that the stochastically stable states are completely depending upon the definition of the perturbations. NEWLINENEWLINENEWLINE NEWLINELet us first recall the basic setup. Let \(I\) be a finite set of players with cardinality \(n\). A network is represented through an \(n \times n\) symmetric matrix \(G\) with \(G_{ij}=1\) if two different players \(i\) and \(j\) are linked together as neighbors and zero otherwise. The neighbors of player \(i\) are given by \(N_i = \{j\in N: G_{ij}=1\}\), a path between \(i\) and \(j\) is an ordered set of nodes \((i, h_1, \dots, h_l, j)\) such that \(G_{ih_1}=G_{h_1h_2}=\cdots = G_{h_lj}=1\). Each player \(i\) takes two action \(x_i=1\) (contributor) or \(x_i=0\) (defector). A state of the system is \(x= (x_i)_{1\leq i \leq n}\). The best reply function of \(i\) is a map \(b_i(x)\) which sends a state to a utility maximizer, then \(b_i (x) =1\) if \(x_j =0\) for all neighbors of \(i\), and \(b_i(x)=0\) otherwise. This game is called a best shot network game. NEWLINENEWLINENEWLINE NEWLINEA state \(x\) is a pure strategy Nash equilibrium if and only if the set of satisfied players is \(S(x) = \{i\in I: x_i = b_i(x)\} = I\) (equivalent to unsatisfied players empty). Let \({\mathcal N}\) be all the possible Nash equilibria in pure strategies. It is an NP-hard problem to enumerate all the elements of \({\mathcal N}\). A set \({\mathcal E}\) of states is ergodic (recurrent) if each state \(x\in {\mathcal E}\) can be moved finitely with positive probabilities to any other state in \({\mathcal E}\). If \(\{x\}\) is an ergodic set, then \(x\) is absorbing. Let \({\mathcal R}\) be the set of all recurrent states, and \({\mathcal A}\) the set of all absorbing states. NEWLINENEWLINENEWLINE NEWLINEThe authors show that the set of Nash equilibria coincides with the set of absorbing states (Proposition 1), and there are no other recurrent states (Proposition 2). If the perturbation is a regular perturbed Markov chain, then a state \(x\) is stochastically stable if and only \(x {\mathcal E}\) is a Nash equilibrium, where a stochastically stable \(x\) is such that the limit distribution \(\lim_{\epsilon \to 0}\mu^{\epsilon} (x) = \mu^* (x) > 0\) (Proposition 6). If the perturbation is an individual perturbation process, then a state \(x\) is stochastically stable if and only \(x\in \mathrm{arg}\max_{x' \in {\mathcal{N}}} \|C(x'\|\), where \(C(x) = \{i\in I: x_i =1\}\) (Proposition 9). This main result characterizes stochastically stable equilibria for best shot network games. The authors also identify the stochastically stable ones under perturbations affecting all agents in Proposition 10. NEWLINENEWLINENEWLINE NEWLINEThe work is to analyze \textit{H. P. Young}'s framework [Individual Strategy and Social Structure. Princeton University Press, Princeton (1998)]. Many examples are given to illustrate concepts and ideas. The model in this paper may not be that close to any specific real-world situation. The authors will look at more general models of network games and the stability selection in future research.