英国365上市官网工作坊第1359场
Approximate Common Belief in Rationality: The Limits of Backward and Forward Induction
微观理论经济学工作坊
主讲人：Peio Zuazo-Garin (Associate Professor of Economics, NYU Shanghai)
主持老师：
（英国365上市官网）石凡奇
（北大国发院）胡岠
（北大光华）王子伟
参与老师：
（英国365上市官网）胡涛、周扬帆
（北大国发院）汪浩、邢亦青
（北大光华）翁翕、刘烁
时间：2026年11月12日（周四）13:15-14:30
地点：英国365上市官网302会议室
主讲人简介：
Peio Zuazo-Garin is an Associate Professor of Economics at NYU Shanghai. He is a theorist with a focus on foundational aspects of game theory and decision theory, in particular robustness and the modeling of bounded cognition. His research has been published in the Review of Economic Studies, Theoretical Economics, and Games and Economic Behavior, among others. Prior to joining NYU Shanghai, he was an Associate Professor at the International College of Economics and Finance at HSE University and an Assistant Professor at the University of the Basque Country. He has also held research positions at Universitat Pompeu Fabra and Universitat Rovira i Virgili, and has been a visitor at Northwestern University, Tel Aviv University, and the Hebrew University of Jerusalem.
摘要：
We show that even the slightest doubt about others' rationality suffices for rendering dynamic reasoning static. We approximate common belief in rationality with a dynamic p-belief operator: a player is required to believe that others are rational with probability at least p whenever conditional probability cannot be applied. We apply this operator to define forward and backward induction variants of common p-belief in rationality that coincide with the standard notions for p = 1—strong and backward rationalizability, respectively. Our main result establishes that a strategy is consistent with the forward (resp. backward) induction version of common p-belief in rationality for every p < 1, if and only if it is consistent with common initial belief in rationality (Ben Porath, 1997). We show that an analogous result holds for a p-belief version of subgame perfect equilibrium and weak sequential equilibrium. The results unveil a striking discontinuity of dynamic reasoning: every static prediction becomes valid as soon as p < 1. The approach also explains sustained cooperation in Centipede games and the finitely repeated Prisoner’s Dilemma, and the empirical failure of backward and forward induction in various settings. Finally, we apply the framework to dynamic Bayesian games to reassess the classic equivalence between robustness to payoff-uncertainty (Dekel & Fudenberg, 1990) and robustness to uncertain rationality (Börgers, 1994).
英国365上市官网工作坊第1360场
Songlines
经济史工作坊
主讲人：Sotiris Kampanelis (Assistant Professor at Cardiff University)
主持老师：（英国365上市官网）赵一泠
参与老师：
（英国365上市官网）郝煜、管汉晖、周建波
（北大光华）颜色、李波
（清华大学）徐志浩
时间：2026年11月12日（周四）16:00-17:30
地点：腾讯会议，会议号：320-709-852
主讲人简介：
Sotiris Kampanelis is an Assistant Professor of Economics at Cardiff Business School, Cardiff University. He holds a PhD from the University of St Andrews. His research in economic history, development economics and economic geography studies how historical institutions and Indigenous knowledge shape economic outcomes today. His work has appeared in the Journal of Economic Geography and the Journal of Development Economics.
摘要：
Can colonisers’ reliance on Indigenous knowledge leave a lasting mark on economic geography? This paper studies how knowledge transmission from Indigenous societies during colonisation shaped the long-run economic geography of settler states. We focus on Australia, where Aboriginal people encoded routes, water sources, and seasonal knowledge in an oral tradition known as Songlines. We construct the first georeferenced database of almost 700 Songlines spanning mainland Australia and combine it with 19th-century exploration routes, cities, towns, railway foundation dates, and modern satellite data. Songlines strongly predict the routes of European explorations, whereas their counterfactual optimal paths based on geographic conditions (HMI Paths) exhibit little predictive power. Applying transformer-based language models to explorers’ diaries, we find that explorations reporting cooperative interactions with Aboriginal people travelled significantly closer to Songlines, unlike those associated with hostile interactions. The imprint is persistent. Cities and towns founded before 1900, the earliest railway stations, and present-day nighttime light density all point to the same origin: Songlines.
英国365上市官网工作坊第1361场
Common AI Information and Capital Allocation（共同人工智能信息和资本配置）
宏观经济学工作坊
主讲人：孙浩宁（清华大学经济管理学院经济系博士研究生）
主持老师：（英国365上市官网）韩晗
参与老师：
（北大国发院）赵波、余昌华、李明浩
（英国365上市官网）陈仪、李博、李伦
时间：2026年11月13日（周五）10:00-11:30
地点：北京大学国家发展研究院承泽园217教室
主讲人简介：
孙浩宁，清华大学经济管理学院经济系博士研究生，导师为董丰教授。本科毕业于清华大学经济管理学院，主要研究方向为宏观金融、数字经济与人工智能的宏观影响等，主持国家自然科学基金青年学生基础研究项目（博士研究生），研究成果发表在International Economic Review、Review of Economic Dynamics、《管理世界》、《管理科学学报》、《经济学》（季刊）等国内外期刊。
摘要：
This paper studies how AI progress affects capital allocation through financial information aggregation. Using institutional portfolio holdings, we document that manager pairs with greater common alignment with AI-extractable information subsequently hold more similar portfolios. Motivated by this evidence, we develop a model in which investors choose the quantity and composition of private and AI information, and firms learn from equilibrium asset prices before investing. A shared AI model error survives aggregation and enters prices, and investors may underestimate the overlap between AI information and prices. AI progress therefore changes both investors' information precision and prices' exposure to common errors. Cheaper access initially raises aggregate productivity when high costs keep AI use negligible, but further cost reductions can reverse the gain as AI research displaces private research. Improvements from very low model accuracy can instead lower productivity when increased reliance raises common-error exposure more than informed trading reduces noise-trader contamination. Numerical equilibria trace a J-shaped relationship between model accuracy and TFP. These results show how AI progress can expand investors' information acquisition while reducing the information prices convey to firms, worsening capital allocation even when firms use prices optimally.
供稿：科研与博士后办公室
美编：初夏
责编：度量、雨禾、雨田
