Stochastic Stability in Networks and Markets

在过去的25年里，博弈学习理论已经成为分析有限理性和经济行为的主要建模工具之一。大多数模型依赖于一种称为随机稳定性的技术，在这种技术中，代理被赋予有限理性的行为规则(例如模仿、短视的最佳回答或强化)，并且明确纳入了决策错误的可能性。从形式上讲，这导致了一系列扰动的、离散时间的马尔可夫链，这使得能够分析长期行为和经济结果的稳定性。在这篇文献中已经取得了大量的结果。应用包括抽象博弈中的均衡选择、寡头垄断理论、信号和保险市场、竞争理论和网络形成等等。然而，除了少数例外，文献没有确定一般原则。基本模型的改变是可能的，从一个研究文章到另一个研究文章，其结果可能会有惊人的变化。这些变化从实质性的(全局与本地交互；代理的内存长度)到纯粹的技术(修订机会；打破平局的假设)。基于首席调查员在这一领域的丰富经验，该项目提出了一项具体战略，以确定在游戏、行为规则和互动结构的联合空间中潜在的经济相关结果的一般原则。这些包括但不限于：(I)协调博弈中有效结果的选择，作为技术选择或经济努力水平的基础；(Ii)所谓的有限人口、进化稳定状态的稳定性，这些状态预测了在例如寡头垄断竞争中比纳什均衡更具竞争性的结果；(Iii)在社会经济背景下合作结果的生存。除了这一基本理论策略外，该项目还提出了两条进一步的路线。第一个集中在网络上的游戏，也就是明确考虑到经济互动本质上是局部的。其目的是获得导致每一类经济结果的网络特征的完整特征。这包括对互动和信息的区分，即信息溢出的可能性。第二种考虑了当交易者是有限理性时市场机构(例如B2B市场平台)的选择、演化和设计。后一条研究路线的议程包括在行为实验室进行市场实验。

In the last 25 years, the theory of learning in games has become one of the major modeling tools for the analysis of bounded rationality and economic behaviour. Most models rely on a technique known as stochastic stability, where agents are endowed with boundedly rational behavioural rules (for example imitation, myopic best reply, or reinforcement) and the possibility of decision mistakes is explicitly incorporated. Formally, this results in a family of perturbed, discrete-time Markov chains which enable the analysis of long-run behaviour and the stability of economic outcomes. A large number of results have been obtained in this literature. Applications include equilibrium selection in abstract games, oligopoly theory, signaling and insurance markets, contest theory, and network formation, among many others. However, with only a few exceptions, the literature has not identified general principles. Changes in the basic models, which are possible along a staggering number of possible dimensions, alter the results from research article to research article. Those changes range from the substantial (global vs. local interactions; length of agents' memory) to the purely technical (revision opportunities; tie-breaking assumptions). Building upon the Principal Investigator's extensive experience on this field, the project proposes a specific strategy to identify the general principles underlying economically relevant results in the joint space of games, behavioural rules, and interaction structures. Those include, but are not limited to: (i) the selection of efficient outcomes in coordination games as those underlying technology choice or economic effort levels; (ii) the stability of so-called finite-population, evolutionarily stable states, which predict more competitive outcomes than Nash equilibria in e.g. oligopolistic competition; (iii) the survival of cooperative outcomes in socioeconomic contexts. In addition to this basic theoretical strategy, the project proposes two further lines. The first one concentrates on games on networks, that is, takes explicitly into account that economic interactions are local in nature. The aim is to obtain full characterizations of the network characteristics leading to each class of economic outcomes. This includes the distinction between interaction and information, that is, the possibility of information spillovers. The second one considers the selection, evolution, and design of market institutions (as e.g. B2B market platforms) when traders are boundedly rational. The agenda for this later research line includes carrying out market experiments in the behavioural laboratory.