[R-SIG-Finance] clustering

BBands bbands at gmail.com
Wed May 7 23:37:12 CEST 2014


Hello,

Has anyone had any experience with clustering involving large numbers
of series? For example kmeans() or hclust(). We maintain a database of
5000 stocks, 200 industry groups and 15 market sectors. The groups and
sectors are calced by us and are equal weighted. We require both a
fundamental fit and a solid correlation for inclusion and maintain
non-correlated buckets, one in each sector, for stocks that don't fit
in the group they are 'supposed to' belong to. The idea of doing the
cluster work is to provide a rational check and balance. Are new
groups forming, old groups deteriorating, have we missed existing
groups and so forth? (If kmeans is chosen an estimate of the number of
groups and sectors is needed and it is not clear how best to get to
that estimate.) So: Have you had any relevant experience? Especially
with feeding large numbers and mostly positively correlated price
series to cluster analysis?

Thanks in advance,

     John
-- 
John Bollinger
www.BollingerBands.com



More information about the R-SIG-Finance mailing list