Finance: best way of replicating a series using regressions


I have approximately 50 series of foreign currency rates, many of which are fairly highly correlated. For each one, I want to find the linear combination of say 5 of the other series, which best replicates it. How would I go about doing this? Multiple regression?

I could test every possible combination of 5 series amongst the 49 remaining ones, until I minimize Rsquared? I think this would be computationally extremely expensive. (fwiw, I'm planning to use Python and Scipy).

Help much appreciated for this rusty applied mathematician! :tup:

I think factor analysis is the appropriate analysis to solve your problem. I'm fairly new to statistics so I could be wrong. If you're more interested in structures, have a look at principal component analysis (PCA).

I have no idea on how to implement this in Python, but give a try.

Hope this helps,