I'd like to run a variance ratio test in the form I saw in a paper. Here is what they do:
So I tried to replicate this in eviews:We follow Lo and MacKinlay (1988) and Campbell, Lo, and MacKinlay (1997) in the construction of variance ratio (VR) statistics and of the heteroscedasticity consistent significance tests. To improve the small sample properties of the tests, in all reported VR analysis, we use overlapping observations, also following Campbell, Lo, andMacKinlay (1997). We first estimate autocorrelation coefficients of order one, five, and ten for the daily and weekly return series both for individual stocks and size-ranked portfolio returns and then compute VR statistics. If returns are uncorrelated over time, their variance is a linear function of the time interval over which they are computed. So, for instance, the variance of five-week returns should be equal to five times the variance of one-week returns and a VR of order five is obtained by dividing the former by five times the latter.
Test spec: Orginal data
Ticked unbiased vairance, heteroskedastic robust S.E, and demean data.
Probs: asyomptoic normal.
Now here is the question: I set the "user-specified list or vector" to be 1 5 10 to try to match the above quote. However I get an error message when trying to run the test (along the lines of "must be a set of unique integer values greater than one").
Question 1: Why can't I set it to include one? e.g. 1 5 10
Question 2: Maybe 1 is included automatically? I notice if I run the VR test using 5 and 10 only that in the test details section of the output the period 1 is included but it is not in the "individuals tests" part... (screenshot attached).
Possibly a bit of a confused post, but given my username you gotta forgive me that!
CG