RI: Small: Modeling Platform Competition: A Multi-Agent Systems Approach

在许多重要的现实环境中，多个代理的交互由将它们聚集在一起的平台来管理。即使我们假设代理人的行为是理性的，这种互动的结果在很大程度上取决于规则、条例和平台的规范。当多个这样的平台在动态环境中竞争时，代理在知识和计算能力方面受到限制，仅根据传统经济理论预测社会结果变得非常困难。本项目的目标是利用计算方法研究三个重要领域的平台竞争：(1)金融市场；(2)肾脏交换；(3)在线情景劳动力和匹配市场。我们将详细探讨这些领域，首先关注以下问题：(1)在金融市场中，高频交易（HFT）对社会是有益的还是昂贵的？如果这是昂贵的，我们如何激励代理商转向市场平台，取消对高频交易的激励？(2)当多个活体捐赠者配对肾脏交易所在运行匹配算法的频率上竞争时，这是否对社会有害，因为它阻止了一个可能允许更多匹配的更厚的市场？这会不成比例地影响某些类型的患者吗？如果是这样，我们如何激励交易所或患者等待更长时间？(3)在允许用户搜索有形商品/耐用品（如汽车）或无形火柴的平台中，向参与者提供的信息的数量和类型如何影响社会福利？平台竞争的动态如何影响用户可获得的信息？在这个项目中，PI和他的学生将使用多种方法，包括计算博弈论、多智能体模拟和经验博弈分析，来模拟平台竞争的动态。应用计算视角来研究平台竞争的动态是特别重要的，因为这些环境有三个特征：(1)参与者之间发生的复杂互动（例如，交易者从市场价格中学习，平台决定向参与者提供多少信息）；(2)金融市场的做市商或肾脏交易所的清算算法等大参与者（平台本身或为平台提供重要服务的代理）所面临的决策问题的固有复杂性；(3)关注平台在竞争中如何演变的动态。这个研究项目将有助于阐明关于具体应用的重要问题，以及探索新的方法问题，因为它们出现在经济和社会系统的计算建模如何与经济学和运筹学中常见的建模类型结合使用时，可以促进更深层次的理解。

In many important real-world settings, the interactions of multiple agents are governed by the platform that brings them together. Even if we assume rational behavior on the parts of agents, the outcomes of such interactions depend heavily on the rules, regulations, and norms of the platform. When multiple such platforms compete in dynamic environments, and agents are constrained in terms of their knowledge and computational capabilities, it becomes very difficult to predict societal outcomes based only on traditional economic theory. The goal of this project is to employ computational methods to study platform competition in three important domains: (1) financial markets; (2) kidney exchange; (3) online episodic labor and matching markets. We will explore these domains in detail, focusing initially on the following questions: (1) In financial markets, is high-frequency trading (HFT) beneficial or costly to society? If it is costly, how can we incentivize agents to switch to market platforms that do away with the incentives for HFT? (2) When multiple living-donor paired kidney exchanges compete on how frequently they run their matching algorithms, is this socially harmful in preventing a thicker market that would potentially allow more matches? Does this disproportionately affect certain types of patients? If so, how can we incentivize exchanges or patients to wait longer? (3) In platforms that allow users to search for tangible goods/durables (e.g. cars) or intangible matches, how does the amount and type of information provided to participants affect social welfare? How do the dynamics of platform competition affect the information available to users?In this project, the PI and his students will use multiple approaches, including computational game theory, multi-agent simulation, and empirical game analysis, to model the dynamics of platform competition. Applying the computational lens to studying the dynamics of platform competition is particularly important because of three features of these environments: (1) The complex interactions that occur between participants (e.g. traders learning from market prices, platforms deciding how much information to provide to participants); (2) The inherent complexity of the decision-problems faced by the big players (the platform itself, or agents that perform an important service for the platform) like market-makers in financial markets or the clearing algorithm for kidney exchanges; (3) A focus on the dynamics of how platforms evolve under competition. This research program will serve to both elucidate important questions about the specific applications, as well as explore new methodological questions as they arise in how computational modeling of economic and social systems can contribute deeper understanding when used in conjunction with the types of modeling common to economics and operations research.