/


Measuring and managing market, credit and Op risk [109]

Go to: Summary | Previous | Next   
Bullet points include: PCA focuses just on magnitude of contribution to variance The trace of the covariance matrix (i.e. the sum of the variances of each security in the universe) equals the sum of its eigenvalues So even the most important PCA components might just be (larger magnitude) random noise Usually when asked to explain how something works, we expect the answers (i.e. ‘drivers’) to be ‘causative’ or ‘informative’, like extracting radio signals from background noise Is it possible instead to focus on meaningfulness? See e.g. www.nematrian.com/BlendingPCAandICA.aspx or Kemp (2011)

NAVIGATION LINKS
Contents | Prev | Next | ERM Lecture Series


Desktop view | Switch to Mobile