Thanks a lot Allin.
After one full day of thinking I eventually reached the same
conclusion ;-)
I was indeed working on a code like yours but for the fact that, given
that I am actually doing the bootstrap on a panel, as proposed by:
Silvia Gonzalves (2009), "The moving blocks bootstrap for panel linear
regression models with individual fixed effects"
after having created the matrix from the dataset:
matrix X = { dataset }
I reshape the matrix to have all the cross-sectional observations for a
given period on the same row:
matrix XW = {}
c = cols(X)
loop for i = 1..n --quiet
XW = XW~X[$i*t-t+1:$i*t,1:c]
endloop
Bye
Giuseppe Vittucci
On Wed, 2011-02-02 at 21:39 -0500, Allin Cottrell wrote:
> On Tue, 1 Feb 2011, Giuseppe Vittucci wrote:
>
> > Is there a simple way to do block PAIRS bootstrap?
>
> Create a matrix, X0, holding the data pairs, and an appropriately
> sized matrix, X1, for the result.
>
> matrix X0 = {your_list}
> matrix X1 = zeros(T, cols(X0))
>
> Create a random vector, R, to pick the starting indices of the
> blocks to be selected (you can use the muniform() function,
> suitably scaled).
>
> Loop across the elements of R, picking the selected row-blocks
> from X0 and inserting them progressively into X1.
>
> row = 1
> loop i=1..n
> r = R[i]
> matrix tmp = X0[r:r+b-1,]
> X1[row:row+b-1,] = tmp
> row += b
> endloop
>
> If you want the columns of X1 as series, then use "nulldata" with
> a suitable number of observations plus the --preserve option, and
> do
>
> series y = X1[,1]
> series x = X1[,2]
>
> or similar. Use "setobs" if you want to impose a time series
> interpretation on the resampled data.
>
> Allin Cottrell
>
>
>
>
> _______________________________________________
> Gretl-users mailing list
> Gretl-users(a)lists.wfu.edu
> http://lists.wfu.edu/mailman/listinfo/gretl-users
