On the equivalence between (quasi-)perfect and sequential equilibria (Q2453494): Difference between revisions

The authors begin their considerations with introducing an extensive form as a tuple \(\Gamma = ({\mathcal N}, X, C, P, \rho, H)\) described by the following six conditions: {\parindent= 0.5 cm \begin{itemize} \item[(1)] \({\mathcal N}=\{1,\dots, N\}\) is a finite set of players; \item [(2)] the pair \((X,C)\) is a tree with a finite set \(X\) of nodes containing some distinguished initial node \(\tilde{x}\) and a set \(Z\) of terminal nodes, and with \(C=\{C(x): x\in X\setminus Z\}\), where \(C(x)\) is a finite set of choices at \(x\); \item [(3)] \(P: X\setminus Z \rightarrow {\mathcal N}\cup \{0\}\) is a player function that assigns to each nonterminal node \(x\) the player \(P(x)\) who has to move at \(x\) (if \(P(x)=0\) then chance determines the choice taken at \(x\)); \item [(4)] the function \(\rho\) independently assigns to each node \(x\) with \(P(x)=0\) a completely mixed probability measure \(\rho(\cdot|x)\) on \(C(x)\); \item [(5)] \(H = (H_i)_{i\in {\mathcal N}},\) where every \(H_i\) is an information partition of player \(i\) with the property that \(h\in H_i\) implies \(h\subset \{x\in X: P(x)=i\}\), for every \(h\in H_i\) holds \(C(x)=C(x')\) whenever \(x,x'\in h\), and for any \(h, h'\in \bigcup_{j\in {\mathcal N}} H_j\) the sets \(\bigcup_{x\in h}C(x)\) and \(\bigcup_{x\in h'}C(x)\) are disjoint; \item [(6)] the extensive form \(\Gamma\) has perfect recall (in the sense of Kuhn). \end{itemize}} Next an extensive form game \(\Gamma(u)\) is obtained by specifying a payoff vector \(u=(u_1,\dots, u_N)\) with \(u_i =\{u_i(x): x\in Z\}\) describing player \(i\)'s payoff at any terminal node \(x\). The main result of the paper says that there is a closed, lower-dimensional, semi-algebraic subset \(U_0\subset U=(\mathbb{R}^Z)^N\) such that for every \(u\in U\setminus U_0\) the sets of perfect, quasi-perfect and sequential equilibria in the game \(\Gamma(u)\) coincide.