Task Complexity, Equilibrium Selection, and Learning: An Experimental Study
通过实验室实验，研究任务复杂性如何影响协调博弈中的均衡选择，并比较三种适应性学习模型（古诺、虚拟行动、收益强化）对数据的拟合效果。
We consider several coordination games with multiple equilibria each of which is a different division of a fixed pie. Laboratory experiments are conducted to address whether “task complexity” affects the selection of equilibrium by subjects. Three measures of task complexity—cardinality of choice space, level of iterative knowledge of rationality, and level of iterative knowledge of strategy—are manipulated and tested. Results suggest the three measures can predict choice behavior. Since strategically equivalent games can have different task complexity measures, our results imply that subjects are sensitive to game form presentation. We also fit data using three adaptive learning models: 1) Cournot, 2) Fictitious Play, and 3) Payoff Reinforcement, in increasing order of required cognitive effort. The Fictitious Play model, which tracks only cumulative frequencies of opponents' past behaviors fits the data best.