一种估计连续策略博弈的两阶段方法  

A Two-step Estimation for Incomplete Game with Continuous Actions

作　　者：马键[1] 王美今[1] 

机构地区：[1]中山大学岭南学院

出　　处：《统计研究》2010年第6期87-94,共8页Statistical Research

基　　金：广东省普通高校人文社会科学研究重点项目(项目批准号:06JDXM790002)资助

摘　　要：最近实证博弈研究的迅速发展为市场中的策略互动、政策分析和反事实实验提供了有效的工具。本文提出一种估计不完全信息连续策略博弈的两阶段方法,它可以处理私有信息的影响。第一阶段通过非参数分位数回归估计局中人的策略与期望支付函数;第二阶段利用贝叶斯—纳什均衡不等式构造模拟最小距离估计量,最终获得结构参数的估计。数值模拟显示本方法有良好的小样本表现。与现有文献的嵌套固定点方法相比,本方法不需计算均衡,极大降低了计算量,并减轻了多重均衡的干扰。本方法既可以用于估计离散状态博弈,也适用于连续状态博弈。Recent developments of empirical game research provide useful tools for analysis of strategic interactions in market,policy recommendations and counterfactual experiments. This article proposes a tow-step estimation for incomplete game with continuous actions,which copes with the influence of private information. In the first step,the strategy function and expected payoff function are estimated by nonparametric quantile regression. In the second step,the structure parameter is estimated by simulated minimum distance estimator. Numerical simulations are used to demonstrate the finite sample properties of the two-step estimator. This algorithm does not need to compute the equilibrium,thus alleviates computing burden. It could be used in both discrete and continuous state games.

关 键 词：连续策略博弈 两阶段估计方法 非参数分位数回归 模拟最小距离估计 

分 类 号：O212[理学—概率论与数理统计]