AN EXAMINATION OF EVENT STUDY METHODOLOGY (NONPARAMETRIC, T-TESTS, CORRELATION, SKEWNESS, SAMPLE SIZE

This study seeks to increase our understanding of the econometric properties of alternative measures of excess returns and testing procedures. Recent methodological papers involving capital market event studies have provided some contradictory and erroneous results. This study identifies some weaknesses in those papers and resolves some of the conflicts. In addition, it develops some new empirical tests.;This study consists of two major parts. The first examines four questions relating to (i) the choice of a return generating model, (ii) the choice of the appropriate t-test for detecting a mean change, (iii) the suitability of including all affected securities in a test sample, and (iv) the advantages and disadvantages of accounting for cross-sectional dependencies among abnormal returns. The results show (i) that when all of the assumptions underlying the different return generating models are valid, the market model provides the best test, (ii) that without knowledge of the appropriate abnormal returns model the choice among t-tests is arbitrary but that a cross-sectional t-test is inefficient and biased, (iii) that indiscriminate inclusion of all available securities (even when they all are equally affected) may reduce the power of the test, and (iv) that there is no advantage in using a test which ignores cross-sectional dependencies.;The second part identifies a generalized abnormal returns model that leads to nine different t-tests, four of which have been used in event studies. Simulation is then used to examine the sensitivity of the t-tests, the sign test, and Wilcoxon's test. The results show (i) that skewness adversely affects the nonparametric tests and the tests cannot account for cross-sectional correlations, (ii) that for small samples, skewness affects the t-tests also, and (iii) the generalized least squares t-tests is highly sensitive to a parameter of the abnormal returns model. To overcome these problems, four empirical tests are proposed. Even when all the assumptions of the t-test are satisfied, these empirical tests are shown to compete well with the t-tests. But importantly, the empirical tests remain well specified even with skewed and correlated data.