Econometric Analysis of Strategic-Interaction Models under Flexible Behavioral Assumptions

本项目致力于经济主体(个人、企业、工会、政府等)模型的计量经济学研究。在不完整的信息环境中相互作用。在这些环境中，每个代理人的观察到的行为至少部分由她对其他人的预期选择的主观和未观察到的(对研究人员)的信念决定。现有的关于这些模型的计量经济学工作绝大多数假设未观察到的信念与均衡行为一致，这对代理人的理性施加了极端的假设，并危及估计结果的稳健性和有效性。这个项目削弱了这一限制性假设，取而代之的是假设代理人的未观察到的信念只有在某种意义上是“可合理化的”，即他们不会为严格控制的对手的行动分配正的概率。可理性化的概念已经被博弈论者彻底研究过了，但到目前为止还没有被正式纳入计量经济学分析。它包括均衡信念作为特例，但允许广泛的替代行为。在方法论上，本研究项目是双管齐下的。首先，它描述了在可合理化信念的唯一假设下，可以确定给定模型的哪些统计特征。其次，它开发了用于估计和推断已识别特征的统计方法。使用观察到的行为和代理的其他可观察特征(如果有的话)，这些方法阐明了这样的问题：如果代理的信念只被假设满足最低理性要求，那么代理的行为在多大程度上受到其他人的预期选择的影响？理性的上限能被确定吗？纳什均衡能被排除吗？代理人使用哪些可用信息来构建他们的信念？在灵活的行为假设下进行反事实练习(例如，假设政策分析的效果)也是可能的。这些方法可以应用于各种各样的真实世界环境，包括：拍卖、市场进入/退出、公司之间的互动、公司或个人的选址模型、具有同行效应的个人选择模型等。它们还特别适合于实验数据集，这些数据集的可用性在最近几年一直在稳步增加。研究项目的更广泛的影响：如果该项目不仅为现有文献提供新的方法，并吸引该学科研究专家的注意，那么将产生更广泛的影响。主要目标是设计课程、研讨会和研讨会，不仅面向高级经济学博士生，而且面向对基于交互作用的模型中的估计和推理感兴趣的其他社会科学学科的学生和研究人员。这些课程不仅要学习理论，还要学习将这些方法应用到实际数据中所涉及的方法和计算问题。与之相辅相成的是由该领域知名专家举办的讲习班、研讨会和讲座。该项目旨在将博弈论者、计量经济学家、应用经济学家和决策者聚集在一起，在灵活的行为假设下，在复杂的互动环境中需要进行政策分析。本项目所产生的所有计算机数据程序、课堂讲稿和研究论文将广泛传播。

This project is devoted to the econometric study of models where economic agents (individuals, firms, unions, governments, etc.) interact with each other in an incomplete information environment. In these settings, each agent's observed behavior is at least partially determined by her subjective and unobserved (to the researcher) beliefs about the expected choices of others. The vast majority of existing econometric work for these models assumes that unobserved beliefs are consistent with equilibrium behavior, which imposes extreme assumptions about agents' rationality and jeopardizes the robustness and validity of the estimation results. This project weakens this restrictive assumption and assumes instead that agents' unobserved beliefs are only "rationalizable" in the sense that they do not assign positive probability to opponents' actions that are strictly dominated. The notion of rationalizability has been thoroughly studied by game theorists but has not ?until now- been formally incorporated into econometric analysis. It includes equilibrium beliefs as a special case, but allows for a wide range of alternative behavior.Methodologically, this research project is two-pronged. First, it characterizes which statistical features of a given model can be identified under the sole assumption of rationalizable beliefs. Second, it develops statistical methods for estimation and inference of the identified features. Using observed behavior and other observable features of agents (if any), these methods shed light on questions such as: How much are agents' actions affected by the expected choices of others if their beliefs are only assumed to satisfy a minimal rationality requirement? Can an upper bound for rationality be identified? Can Nash equilibrium be ruled out? What pieces of available information do agents use to construct their beliefs? Counterfactual exercises (e.g, the effect of hypothetical policy analysis) under flexible behavioral assumptions are also made possible. The methods can be applied to a wide variety of real-world settings, including: Auctions, market entry/exit, interaction among firms, location models for firms or individuals, models of individual choices with peer-effects, etc. They are also especially amenable to experimental data sets, whose availability has been steadily increasing in the recent past. Broader Impacts of Research Project:The broader impact if the project goes beyond contributing new methods to the existing literature and attracting the attention of research experts in the subject. A major goal is the design of courses, workshops and seminars aimed not only to advanced economics doctoral students, but to students and researchers of other social science disciplines interested in estimation and inference in interactions-based models. These courses would study not only the theory, but also the methodological and computational issues involved in the implementation of these methods to real data. They will be complemented by workshops, seminars and lectures by prominent experts in the subject. This project is designed to bring together game theorists, econometricians, applied economists and decision-makers in need of policy analysis in complicated interactions settings under flexible behavioral assumptions. All computer data programs, lecture notes and research papers which will result from this project will be widely disseminated.