I recently applied PCA to a set of variables, which consisted of the percentage discount/premium associated with nine listed investment companies. Nine variables is not exactly swimming in data, I know, but PCA/FA can still be a very useful tool to help uncover patterns within a correlation matrix. Essentially, I wanted to know whether the discount/premium associated with these nine LICs could be accounted for by a single component. Based on the results of the analyses, the answer was a clear 'No'. There appears to be two nearly totally orthogonal (uncorrelated) components. I found this interesting for a variety of reasons. One of the most obvious is that despite the fact that two LICs may have very similar underlying portfolios, their share price may not necessarily correlate with each other very strongly. This implies that selecting a LIC from each of the two components may be considered a more diversified portfolio than selecting two LICs that load onto the same component.
Why it is that these two components exist remains an interesting question. I've got some speculative hypotheses. Ultimately, though, the results appear to be relatively robust and may help serve to generate some interesting trading strategies (i.e., for retail traders; these stocks are relatively too illiquid for any large scale trading).
http://www.youtube.com/watch?v=_uYASFVUNpQ
http://www.youtube.com/watch?v=_uYASFVUNpQ