VAR methods PCA vs FA

[mkheidzedavid]

Hello everyone!

I have a question regarding the choice between the Principal Component Analysis and Factor Model methods. PCA's main focus is the diagonal terms in covariance matrix and FA model concentrates on off-diagonal terms. Everything in calculation of both model is clear and understandable but I know not which one to choose in multi-factor model.
If single index model is not sufficient in explaining return and volatility we have a choice to more factors which is done through these 2 main methods(PCA,FA). So my question is: which one is preferable over another and why?

Thank you

 

[alain]

PCA works well if your system is highly correlated. If not, you will end up using a lot of principal components.

Which type of Factor Analysis are you talking about?

 

[mkheidzedavid]

I mean the explanatory factor analysis... That gives the latent factors given the set of some observed ones...