Lo and MacKinlay’s variance ratio test

[babu]

i am a new user of eviews i have a problem i want to perform Lo and MacKinlay’s variance ratio test and multiple variance ratio (MVR) Chow and Denning test

is it possible to perform this test by using this software .if yes would you please help me in this regard how can i perform these tests.

thanks with kind regard

arshad

[EViews Glenn]

At the moment, EViews doesn't have built-in procedures for variance ratio tests. (I will note that this is a feature that is on our list of things to look at for upcoming versions of EViews.)

For now, however, it is not particularly difficult to compute the moments for the test statistic using the @mean and @varp or @sumsq functions. Under the heteroskedastic null, the distribution of the test statistic is normally distributed, so finding the p-value is straightforward.

Here's a snippet of code that I just whipped up for this case. You can put in a program and then run it. Here %X is the name of the series of interest, and we set the comparison lag !q to be 4. I note that you say you are a new user of EViews. This snippet uses some of the programming constructs that are described in the manual. It's probably worth a quick perusal to get a sense for what we are doing here...

Code: Select all

' Lo and MacKinlay 1989, JEMA, p. 203-238.

' get name of series to test and the lag
%x = "x1"
!q = 4

' (3) and (4)
series xdiff_1 = {%x} - {%x}(-1)
series xdiff_1_demean = xdiff_1 - @mean(xdiff_1)
scalar var_1 = @varp(xdiff_1_demean)
!nq = @obs(xdiff_1_demean)

' (5)
series xdiff_!q = {%x}  - {%x}(-!q)
series xdiff_!q_demean = xdiff_!q - !q*@mean(xdiff_1)
scalar var_!q = @sumsq(xdiff_!q_demean) / (!q*!nq)

' (6b)
scalar Mr_!q = (var_!q / var_1) - 1
scalar Mr_!q_se = @sqrt(2*(2*!q - 1)*(!q-1)) / (@sqrt(!nq) * @sqrt(3*!q))
scalar Mr_!q_adj = Mr_!q  / Mr_!q_se
scalar Mr_!q_pval = @cnorm(-@abs(Mr_!q_adj))


The results will be in the scalars "Mr_4" and "Mr_4_pval". Note that I'm computing the centered ratio in Mr_4.

You can add the unbiased adjustments described in Lo and MacKinlay (7a) and (7b) without too much difficulty. If, however, you want to allow for a heteroskedastic or weakly dependent null, it will be more difficult, since you'll need to compute a weighted sum of the estimated autocorrelation variances.

Adding Chow and Denning to the mix adds an additional layer of complexity.

In short, computing the variance ratio test is pretty straightforward for the heteroskedastic, single comparison case, but is a bit tricky for more general calculations. Perhaps someone else has already done some work on this or cares to expand on what I've done above.

[babu]

Thanks Dear you really did hard work for me.i am very thankful to you