This paper compares the performance of four expected return estimation models-the MM, CAPM, 3FM, and 4FM-separately using OLS estimator, GARCH model, and mix of OLS and GARCH-for finding AR in traditional short-horizon event study, using simulation experiments. Moreover, Jensen's performance index is employed to evaluate the last three models. The conclusions are that under a traditional framework, OLS is not inferior to the other complex estimators. 3FM and 4FM somewhat dominate the MM and CAPM because of lower estimation bias and slightly smaller Type I error, though the last two models' performance is already satisfactory. Under Jensen's performance index, CAPM is superior to 3FM and 4FM due to lower Type I error and higher power. Both event-study frameworks have weak power in experiencing uncertain event day and/or tiny AR.