Improving Tests of Abnormal Returns by Bootstrapping the Multivariate Regression Model with Event Paraments

37 Pages Posted: 7 Feb 2004

See all articles by Scott E. Hein

Scott E. Hein

Texas Tech University

Peter Westfall

Texas Tech University - Area of Finance

Date Written: January 2004

Abstract

Dummy variable-based tests for event studies using multivariate regression are common in finance, especially for banking-related studies. These tests are not robust to nonnormality of the residual, even for arbitrarily large sample sizes. Such methods typically overstate the significance of the event, although on some occasions the normality-assuming tests lead to an understatement of significance. Alternative bootstrap methods to rectify this deficiency are described, investigated, and compared. Not all procedures are robust in the face of cross-sectional correlation; we find that it is necessary to bootstrap the residual vectors of a cross-sectional time-series regression model to adequately control type I error rates. Such properly conducted bootstrapping methods greatly improve upon traditional normality-assuming methods and also allow development of new and powerful event study tests for which there is no parametric counterpart; thus their use is generally recommended for dummy variable-based tests of events in financial studies.

Keywords: Event study, event parameter estimation, bootstrap, cross-sectional correlation, resampling, simulation, significance level

JEL Classification: C3, C31, G1, G14, G21

Suggested Citation

Hein, Scott E. and Westfall, Peter, Improving Tests of Abnormal Returns by Bootstrapping the Multivariate Regression Model with Event Paraments (January 2004). Available at SSRN: https://ssrn.com/abstract=496325 or http://dx.doi.org/10.2139/ssrn.496325

Scott E. Hein (Contact Author)

Texas Tech University ( email )

PO Box 42101
Lubbock, TX TX 79409
United States

Peter Westfall

Texas Tech University - Area of Finance ( email )

Lubbock, TX 79409
United States