学生汇报，汇报人：李星宇、江弘毅、张紫荆

时间：2025年5月16日（周五）14：00-15：30

地点：北大国发院承泽园131教室

汇报人1：李星宇 （导师：易君健、孙振庭）

题目：《Causal Inference with Social Interactions: A Structural Break Viewpoint》

摘要： 

This article studies causal inference with social interactions in a non-experimental setting with a non-staggered binary treatment. We characterise the potential outcomes by a factor model that allows for interference between any two units. Under this specification, the observed outcomes can be represented by a structural break model and the treatment effects are exactly the outcome changes induced by this break. Since the structural break literature has not yet provided any estimator for such estimands, we propose an innovative estimation procedure for treatment effects. Under standard assumptions, the estimator of every individual and time specific treatment effect is proved to be consistent and asymptotically normal as the numbers of units, pre-treatment and post-treatment times go to infinity. We find consistent estimators for the asymptotic variances, which enables asymptotically pivotal inference on treatment effects. As a by-product of causal inference, we contribute to the structural break literature by providing a valid approach to the estimation and inference of outcomes changes induced by a structural break. Furthermore, we extend our method to models with covariates. Finally, we investigate the performances of the proposed method in finite samples by Monte Carlo experiments.

汇报人2：江弘毅（导师：沈艳，黄卓）

题目：《Factor Approach for Program Evaluation with Spatial type Dependence: Estimation and Inference》

摘要：

We consider the inference on individual and time specific treatment effects on the treated under the framework of spatial panel data models with common shocks. Quasi-maximum likelihood estimator (QMLE) and imputation are used to estimate the parameters and a residual-based bootstrap resampling procedure is used to construct the prediction intervals. The proposed prediction intervals are proved to    have asymptotic validity as the number of pre-treatment times and control units go to infinity. Monte Carlo experiments illustrate that our method performs well in finite samples under a wide variety of data generating processes.

汇报人3：张紫荆 （导师：黄卓）

题目：《未预期央行货币政策沟通与股票市场响应——基于文本分析的实证研究》

摘要：

 央行货币政策沟通文本具有复杂性和多重含义，传统测度方法往往忽略部分信息维度。本文基于2009—2023年我国央行货币政策沟通事件，采用大语言模型（LLM）和词典法，从原始沟通文本和市场预期信息中提取超预期信息，系统分析其对股票市场的影响。结果表明，相较于单纯测度政策文本字面内容，衡量央行与市场预期偏差对股市响应的解释力更强。同时，相较于其他测度央行货币政策沟通事件中未预期信息的方法，文本分析方法能更全面、直观地展示沟通事件对市场的影响。本文研究发现：（1）超预期宽松沟通带来正向市场反应，超预期紧缩沟通则产生负向影响，且市场对紧缩性超预期沟通更敏感；（2）口头沟通侧重宏观预期引导，书面沟通主要通过具体政策操作影响市场；（3）LLM在口头沟通分析中优势更显著，词典法更适用于书面沟通；（4）在散户占比较高和信息透明度较低的公司中，股价对政策沟通的反应更强烈。