tests for statistical significance

[lomfy]

Hey there!

I'm about to conduct an event study using EViews, and thus constructed a time series containing event-induced or "abnormal" returns.
my goal is to test each and every one of this time series points for statistical significance (whether the respectie value is significantly different from zero).
i intend to apply the following tests (some of them have been designed especially for event study frameworks):

Parametric Tests
- Simple t-Test
- t-Test according to Patell (1976)
- t-Test according to Boehmer/Musimeci/Poulsen (1991)

Parametric Tests
- Sign test
- Wilcoxon rank sign test
- Rank sign test according to Corrado

Are there any code libraries or similar pre-arranged frameworks for calculating any of those test statistics? After all, they are all simple formulae, so I guess I could program them, however it would probably take a lot of (unnecessary) time..! :)

Thanks a lot for any hints!!

[Macbeth]

I am in the same situation as you were. Did you program the statistic all by yourself or did you find any solution ?

thank you

[lomfy]

Hey there, the best tip I can give here is: Do it yourself! Have a close look at the formula of the test statistic (actually, a VERY close look, as even one missing or wrong parenthesis can obviously screw things up completely...!), and then program it step by step (it helped me a lot to program interim variables, say for parts of the numerator / denominator of the test statistic, otherwise its to hard to see clearly what's going on in each step of the calculation). Then, play around with the outputs in an Excel file to get a feeling for whether your specification might be correct (I guess it would be best to find some program, that does calculate the test statistics in order to compare the outputs, but I didn't find anyone)...this is really important, as minor mistakes can have dramatic effects. But with hindsight, all of this stuff can be implemented, although it's sometimes necessary to ask in the forum for specific operations needed ;-) In my view, diy is optimal, as then (and only then) you really develop an understanding of what these test statistics actually do! This is extremely helpful when explaining/discussing the output!