Market Risk [37]

Go to: Summary | Previous | Next   
Bullet points include: Suppose we have N instruments and estimate the covariance matrix from T observations per instrument where T much less than N (e.g. as would normally be the case for a whole market model). Then at most T-1 non-zero eigenvectors and most of the smaller ones often indistinguishable from ones that would arise randomly. Places fundamental limits on reliability of factor analysis (or any other risk modelling derived from historic return series), see e.g. Kemp (2010)

Contents | Prev | Next | ERM Lecture Series

Desktop view | Switch to Mobile