Here's an example by Marco Avellenada from NYU titled "Statistical Arbitrage in the U.S. Equities Market". The idea of this paper involves capturing mean reversion in the residual returns of a security after removing the principle components of return.
As another example, here is some research by Mark Kritzman showing how spikes in the "absorption rate" are associated with drawdowns in the US equity market.
I wonder if there are other strategies involving the eigenportfolios or behavior of principal components other than the dominant eigenvector (i.e. the market portfolio) and non-random eigenvectors.