Past is Prologue: Inference from the Cross Section of Returns Around an Event
Description
This package replicates the results in the paper, "Past is Prologue: Inference from the Cross Section of Returns Around an Event," by Jonathan Cohn, Travis Johnson, Zack Liu, and Malcolm Wardlaw, which studies how confounding contemporaneous news can generate misleading statistical significance in cross-sectional event studies of stock returns. The core hypothesis is that standard cross-sectional regression tests frequently over-reject the null of no differential event effect because firm characteristics are correlated with routine daily news unrelated to the focal event. Using firm-level daily returns from CRSP and accounting data from Compustat for U.S. public firms from 1991–2021, the data show that return–characteristic relationships are statistically significant on a large fraction of ordinary trading days, even in the absence of any event. The package contains cleaned input data, code to construct firm characteristics, and scripts that implement both conventional cross-sectional methods and the paper’s proposed time-series benchmarking approaches, including a novel GLS procedure that uses principal components of past returns to estimate daily covariance matrices. The main finding is that comparing event-day relationships to their empirical distribution on pre-event days sharply reduces false positives, while the GLS-based implementation substantially improves statistical power relative to OLS. Together, the data and code allow users to reproduce all tables and figures, assess the prevalence of spurious significance in standard event-study designs, and apply the proposed methods to other quasi-experimental settings in corporate finance.