Home / Summaries / Event Studies (based on ppt slides) / variance-event-events

The testing procedure

4 important questions on The testing procedure

3rd issue: equal weights - if we use aggregate stdev (as in the second test) we don't have equal weight to each stock return. The con is that then the variance of one stock if its out of proportion it will blow up the aggregate variance

The solution is we take the AR and divide it by its own std dev. So unweighted variance.

Fourth aspect: normal distributions, based on asymptotic theory, we need a high J so either few firm with many events or many firms with few firms. Otherwise also we'd have outliers have a large effect on small samples:

Use nonparametric tests
alternative: use wilcoxon signed rank test

5th aspect: role of sampling interval, date uncertainty

a. If you know exact time event, shorter date you have after the event the better (increases ability to statistically identify the event) moreover using shorter intervals will reduce the variances of the abnormal return, without changing the mean.

b. Inference with data uncertainty/ newspaper events: for instance if wondering whether event took place before or after market close,/ due events after the newspaper.

take one day after in the event window. So day 0 and day +1

Clip 6.5: Excel Clip 2: Application (con'd): Robustness checks

So remember one of the things that we can criticise is that the variance of the abnormal return might rise during the event window. So we said to compute the crossectional variance of all ar at this particular time period. (so variance of each firm AR /20), didn't work out variance decreased so its not because of that you can say.

For Car Same way, var of CAR of days/20 is the new variance.

if equally weighting; sqr of n * average of all variances

The question on the page originate from the summary of the following study material:

Event Studies (based on ppt slides)

View summary