事件研究法的实现、问题和拓展

Questions,Applications and Extensions in Event Study Approach

事件研究法是一种广泛应用于社会科学各领域的因果推断和政策评估工具。近年来事件研究法通常与双重差分法搭配使用,用于检验平行趋势假设和政策动态因果效应。本文结合前沿理论计量研究成果,试图对事件研究法进行系统性的归纳梳理,以厘清事件研究法在实践中需要注意的一些问题。本文梳理了事件研究法的历史发展脉络,详细展示事件研究法的实现方式、识别假设以及其与双重差分法之间的紧密联系。本文归纳分析了事件研究法实践中可能会面临的一些常见问题,包括基期选择、平行趋势检验、数据归并和截断、控制组间时间趋势差异等问题的影响和应对方法。本文还详细分析了异质性处理效应情形下的事件研究法估计偏误来源、修正思路、平行趋势检验等问题。

Event study is an empirical research method used to study the impact of specific events or policy shocks on individual behavior.In recent years,event study has gained increasing importance in empirical research and has been widely applied by researchers in various social science fields.The popularity of event study in recent years is largely due to its close connection with the canonical difference-in-differences(DID)method.Using event study to test pre-parallel trends has become one of the standard steps in the current DID approach.The role of event study goes far beyond testing pre-parallel trends.Compared with the DID method,event study has many attractive features.First,it can visually demonstrate the dynamic changes in policy effects,which is the most appealing characteristic of the method.Second,recent advances in theoretical econometrics have demonstrated that the estimated results of the DID method may be biased if there are heterogeneous treatment effects at the group or time level.An event study is more robust than the DID method in this scenario.This study attempts to provide a comprehensive analysis and summary of the implementation,identification assumptions,and practice suggestions of the event study method to guide researchers in various social science fields.Based on the cutting-edge econometric literature,we found that 1)the estimated coefficient of the TWFE model in the event study approach is essentially a weighted average of a bunch of canonical 2×2 DID estimates.2)Heterogeneity in treatment effects at the group and time(or both)levels can lead to estimation bias in the DID approach based on the two-way fixed effects model.If the dynamic paths of treatment effects remain consistent across groups(homogeneous treatment effect paths assumption),the event study approach can consistently estimate the average treatment effects for each relative period of the treatment.3)Researchers can assess the existence of heterogeneous treatment effects by comparing the differences between the estimation results of the TWFE event study and new HTE-robust estimators and then identify potential factors that may cause heterogeneous treatment effects.This evaluation can help assess whether endogeneity issues may arise in the timing of treatment.This study provides several recommendations for using the event study method.1)We suggest that a pre-event period should be used as the base period,e.g.,-1 period.2)When testing pre-parallel trends,we suggest that a joint test of the significance of pre-event coefficients should be conducted.3)Binning data after the treatment period may lead to estimation bias,so it is essential to avoid binning post-treatment data in practice.If pre-parallel trends are met,binning can be performed for the pre-treatment period.Trimming data requires consideration of potential sample selection issues and loss of estimation efficiency.4)Controlling group heterogeneity time trends directly in the TWFE model may cause estimation bias.Using the imputation method can effectively control potential differences in pre-trends between groups,thereby avoiding estimation bias.