Hello,
I am currently performing an event study. I have a Sample of 84 european banks with daily returns calculated on closing Stock prices and a widely diversified Benchmark portfolio. I have Multiple events over a time of 2 years where an event always occurs for all bank simultanousy measuring a Bail-in of a european bank. The events are clustered in time as well. Let T=0 denote the event data,-1 the day prior and +1 the day past the event. I use an estimation Window of exactly 80 days with upper Bound -2. I regress each banks stock return according to the Market Model on the Benchmark portfolio and estimate the normal return as out of sample prediction. The abnormal return hence equals the observed stock return-predicted return. If an event date occurs in a later events' estimation window, i simply delete(dummy out) the obersvations of the earlier event period(-1,0,1) to estimate normal Performance correctly. I define three Different estimation windows, [-1;-1], [0;0] and [0;1] to check for anticipatory effects as if they would habe been the true eventday.[0:1] displaying the cumulative ar of the event date and the following one. My event identifying is really accurate and thus prevents Type 2 error. When testing i apply the generalized rank test adjusting for forecast errors, allowing event induced variance, and cross sectional correlation, autocorrelation. I further defined subsamples splitting up the banks into systematically important banks or not and giips state banks or not and then rerunned the estudy command for them. All my test statistic are good and seem to have low Type 1 error comparing with other empirical results. Only that the Grank tests differ in significance between CAAR and portfolio CAR for a single day in the eventwindow when they are supossed to be equal by defitnition suprized me. If somebody could explain the difference between portfolio car and caar in this case would be nice.
But Now to my real Problem: there is no Option to test the equality of average abnormal returns using estudy. I want to test if the differences in abnormal returns between the groups are nonzero. In that matter just aggregating cross sectional for the respective event window and testing average abnormal return on -1 against -1 and 0 against 0 and so on. Applying two sample ttest, results in Overrejection of the Null. E.g. the gsib sample consist only 16 Banks and the abnormal returns are surely not approximative normal distributed . Using wilcoxon rank sum doesnt seem to imrpove my results. Comparing with 68 non gsib banks i get hillarious low p values because the test statistic isnt adjusted At All.
My Question: I need a proper test statistic to test differences between groups. I thought about someting like a sign test with standardized abnormal returns and estination window adjusted Standard deviations. However the estudy command doesnt display the standardized abnormal returns or z statistic or allow for test of differences in two subsamples. I did the same stuff manually without estudy. However im new to stata and to stupid to correctly compute Sd(SAR) and SARs in the weird enviroment of stata. Maybe somebody can Help me i would appreciate it.
Thanks for Your time.
Best regards