Creating and validating risk models [15]

Go to: Summary | Previous | Next   
Bullet points include: However many securities we have, we can only identify from past time series data at most n (or rather n-1) different factors influencing returns on entire market, where n is the number of return periods we have Result derivable from eigenvalue / principal components analysis or observe that all factor series are linear combinations of n such series, first one consisting of (1, 0, 0 ,…), next one (0, 1, 0 ,…) etc. Most of the less important principal components or equivalent turn out not to be statistically significant, c.f. random matrix theory So most of the fine structure of most risk models (except if based on market implied distributions and if derivatives market is deep enough) is ultimately based on ‘expert judgement’

Contents | Prev | Next | ERM Lecture Series

Desktop view | Switch to Mobile