Extreme events: blending PCA and ICA [24]

Go to: Summary | Previous | Next   
Bullet points include: PCA concentrates on magnitude (maximise aggregate contribution to variance). ICA concentrates on meaningfulness (and thus comes in more flavours, but often seeks to maximise kurtosis). In either case, most important component can be extracted by projection pursuit maximising a particular importance criterion. So we can blend the two together, using a blended importance criterion. But further refinements needed to cater for time-varying volatility and other behaviour linked to distributional mixtures. Such mixtures are important sources of fat-tailed behaviour in practice

Contents | Prev | Next | Library

Desktop view | Switch to Mobile