[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: Bootstrap with time series models

From   "Austin Nichols" <>
Subject   Re: st: Bootstrap with time series models
Date   Wed, 3 Oct 2007 17:14:20 -0400

Bob Yaffee refers to
which code would take quite a lot of adapting for your use.

I haven't read:
Politis, D.N., Romano, J.P. and Wolf, M. (1999). Subsampling, Springer, NY.
Lahiri, S.N. (2003). Resampling Methods for Dependent data, Springer, NY.
Politis, D.N. and Romano, J.P. (1994). The Stationary Bootstrap. JASA,
I suspect both subsampling and the block bootstrap are easy to
program: for the block bootstrap you just select _N/L observations
(from the first _N-L+1 obs) and the L-1 obs that follow them in each
iteration, then stitch them together, right (where the block length L
for each iteration is random and generated from a geometric
distribution)?  Perhaps an enhancement of the -gsample- program could
do it?  That would be up to its author, Ben Jann, of course...

The non-overlapping block bootstrap is already possible with the
cluster option.  Just define the cluster variable as blocks of time...

On 10/3/07, Robert A Yaffee <> wrote:
> Time series models require a block or wild bootstrap.
> Stata does not have either of these yet, though I think that Austin Nichols indicated that
> there is a wild bootstrap in workshop somewhere.  I think he said that it needs documentation.
> ----- Original Message -----
> >
> > Is there any way to use the Stata bootstrapping facility with time series
> > estimation functions, such as var or arima?
*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index