Event studies as regression discontinuities in time

Event history analysis has a long history in the business literature where it is used to measure the effects on firm stock market value of various information shocks, such as acquisition announcements. The basic idea is to use the change in firm value in a short (e.g. three day) window around the announcement as an estimate of the market's valuation of the information. Careful implementation requires close attention to when information actually reaches the market, which may not be the time of the official announcement, and to the possibility of confounding announcements of other information by the same firm within the same window. Such confounding announcements could well be endogenous if firms try, for example, to release all their bad news at once, though my sense is that this endogeneity issue is not usually addressed. The general approach to confounding announcements, whether endogenous or exogeneous, is apparently to simply discard observations that have them, which potentially changes the nature of the parameter being estimated.

It struck me that it would be interesting to approach this older literature from the viewpoint of recent developments in regression discontinuity methods in labor economics. For example, one could try and come up with something other than a three day rule of thumb for the window width used to calculate the change in stock market value, using the same sort of strategy that has been used in papers that look generally at bandwidth choice for RD estimators in the labor literature.

Hat tip: David Benson, whose dissertation defense today, in particular his chapter on event history methodology, sparked these thoughts