## F Test for 2 series of variances

For econometric discussions not necessarily related to EViews.

Moderators: EViews Gareth, EViews Moderator

patrix
Posts: 10
Joined: Thu Sep 09, 2010 11:00 am

### F Test for 2 series of variances

Eviews 7
I had a daily series and converted it to lower frequency (quarterly), contraction method: variance. Now I have a quarterly series of variances and need to compare with another quarterly variance series, which is the benchmark. I need to use a F test, but I cant find a way for that. Help, please!
Thank you!

random_access
Posts: 30
Joined: Thu Apr 30, 2009 1:39 am

### Re: F Test for 2 series of variances

hi there
u can select both groups and right click open as a group and then view t-test and choose variance......i guess this will give u F-test result and t-test result for mean difference.

patrix
Posts: 10
Joined: Thu Sep 09, 2010 11:00 am

### Re: F Test for 2 series of variances

Thank you, but this is not what I need.
I calculated the quarterly variances (for two series of exchange rates). I want to compare a certain quarter variance of series x with the corresponding quarter variance of series y: for example var (x) and var (y) for 2005Q2.
Well it's easy to generate the the ratio of the 2 series (samples), var(x)/var(y). Then I need to compare the ratio with a cuantile of the F distribution (5%). The variances were calculated with slightly different d.f.'s (because in each quarter there are different holidays for each series of exchange rates).
I could simply look in the F distribution table, but isn't there any solution for automating the calculation?

Group of variables i use: varqx, varqy, dfqx, dfqy,
where varq are the quarterly variances for the x and the y, and de dfq are the degrees of freedom for the x and the y of each quarter.
Thank you again!

patrix
Posts: 10
Joined: Thu Sep 09, 2010 11:00 am

### Re: F Test for 2 series of variances

Well, after a few days of thinking and struggeling with the Eviews documentation and other manuals, I found the solution. I hope will not be spending so much time for every problem I encounter in Eviews.
Here is the solution: I used functions and (simple) programing
@fdist (x,v1,v2), retuns the p-value for the value x of the test-statistic and the V1 (numerator d.f.) and V2 (denominator d.f.)
Then it's easy to compare the value obtained with the F statistic calculated and to conclude about rejecting or accepting H0.
N.B: there's also the @qfdist(p,v1,v2), that returns the quantile for p=0.05 (e.g.), and v1, v2 d.f.'s

The program:
vector (47) res_ftest
for !i = 1 to 47
res_ftest(!i)=@fdist(fs(!i),nobseu(!i)-1,nobser(!i)-1)
next

N.B: fs is calculated as the varx/vary