[R-SIG-Finance] Splitting time series into blocks/regimes?

[R. Michael Weylandt]

I find this idea methodologically iffy, but perhaps you could do a
little bit to make it a well posed question:

i) Are you picking these periods as relatively equal divisions or time
or as ex-post artifacts of "all that mess"

ii) What is "the clustering algorithm"? Aren't there quite a few?

iii) What does this mean: "the 3 big blocks/regimes need to be
contiguous within each block/regime itself..."

Perhaps you should look into reformulating your idea in terms of
regime switching models.

Michael Weylandt

On Tue, Feb 28, 2012 at 12:55 PM, Michael <comtech.usa at gmail.com> wrote:
> Hi all,
>
> I have a time series of historical SP500 returns.
> How do I split it into a few non-overlapping blocks(regimes) in time in an
> automated fashion?
> For example, given a prespecified number N which is the number of
> blocks/regimes, lets say 3.
> And given 5 year history of SP500 time series and returns.
> Manually, the period before Sept. 2008 is one regime. The period in-between
> Sept. 2008 and arguably May 2010 is the 2nd regime.
> Then the third regime is from May 2010 until today.
> So then we have three blocks/regimes splitted.
> Is there a way to split automatically?
> I have been thinking about using clustering algorithm on the returns.
> However, what is needed is a constrained version of the clustering
> algorithm, i.e. the 3 big blocks/regimes need to be contiguous within each
> block/regime itself...
> I am looking for a way to do this in R...
> Any thoughts are highly appreciated!
> Thanks a lot!
>
> FYI: I have asked a similar question before on this list but somehow I felt
> it's probably better to renew the question with my newer and more concrete
> examples. Also I've listed my question here so the answers to my question
> could be reference for fellow statisticians in the future...
> http://stats.stackexchange.com/questions/23817/splitting-time-series-data-into-blocks-regimes
>
>        [[alternative HTML version deleted]]
>
> _______________________________________________
> R-SIG-Finance at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-finance
> -- Subscriber-posting only. If you want to post, subscribe first.
> -- Also note that this is not the r-help list where general R questions should go.