/

Extreme events: blending PCA and ICA [11]

Go to: Summary | Previous | Next   
Bullet points include: A common way of deriving factors that describe observed market behaviour. Typically introduced via eigenvalues and (normalised) eigenvectors of the return covariance matrix, V. i.e. solutions to Vx = lambda x; the lambda are the eigenvalues, the x are the eigenvectors. Any instrument’s behaviour then expressible as a linear combination of ‘signals’ associated with these eigenvectors. i.e. r(j, t) = a(j,1) .S1(t) + a(j,2).S2(t) + ... + a(j,n).Sn(t)   for instrument j. Eigenvectors are orthogonal, deemed to be ‘different ’ drivers of behaviour. Usually limit merely to ‘significant’ factors, and add back idiosyncratic risk

NAVIGATION LINKS
Contents | Prev | Next | Library


Desktop view | Switch to Mobile