Fisher Student Statistics

arnao · December 6th, 2004, 1:45 pm

I have been using Fisher and Student statistics to test wether a filter/forecast is significant...and I am very disapointed because it seems the fact filter/forecast is significant one week does not imply it will be significant for the weeks after ! (I have checked some basic filters on equity weekly historical data)What are better statistics (than Fisher and Student) to look for, considering stochastic series ?

AndreaClaudia · December 8th, 2004, 2:20 pm

Hi arnao!Why don't you use the bootstrapped innovations? A T-Test would imply that equity data is normally distributed, which is clearly not the case.Hope this helps!Cheers

arnao · December 9th, 2004, 1:07 pm

Thanks for the advice, sure non-normality doesnt help and I was probably wrong to suppose that Fisher and Student tests where robust with respect to non-normality of my data.any references ? (google only returned 3 links for "bootstrapped innovations")what about Mann-Whitney U-test ? ps : actually I just want to test whether means of several groups are equals or not.

AndreaClaudia · December 14th, 2004, 3:17 pm

Well, if you just want to test whether two means are equal, then you can either use the Standard Two-Sample T-test (assuming asymptotic normality) (and test the null whether the difference between the two means is different from zero) or the Wilcoxon signed or rank-sum test. But then I don't understand what you need the innovations for. Anyways, there are tons of literature on bootstrapping (also on Google

)Hope this helps.Cheers.