Re: st: bootstrap with overlapping clusters

[Steven Samuels <ssamuels@albany.edu>]

Nick, bootstrapping time series is a specialized topic not much  
related to panel surveys-and it is a topic in which I am not expert.   
In the book by Efron and Tibshirani (An Introduction to the  
Bootstrap, 1993, Chapman and Hall), two methods are discussed: the  
"moving blocks" method and bootstrapping residuals from the fit of a  
time-series model. I know that the topic has come up before in  
Statalist, without much of a useful answer. A Google search on  
"bootstrap time-series" will show you some recent approaches.  I'm  
sorry I can't help you more.

Steve

On May 23, 2007, at 3:01 PM, Nicholas.Kraninger@bos.frb.org wrote:


> I believe the goal of this bootstrapping exercise is to approximate  
> the
> distribution of a certain statistic estimated from time series  
> financial
> data (e.g. returns of stock XYZ).  I don't actually know all of the
> particulars; I am a research assistant.  A cluster is defined as the k
> previous observations, the current observation, and the k following
> observations.  So as we move through time, there will be substantial
> overlap in the clusters.  I know how to use the bootstrap cluster  
> options
> to do this exercise with non-overlapping clusters, and I didn't  
> know if it
> was possible to use Stata's built-in bootstrap to do the same thing  
> for
> the overlapping case as well.  I think the situation is in many ways
> analogous to the panel study of households that you mention.  I  
> will post
> more details if I find out more specifics of the project.
*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/