SVD as a pricing model?

theRedBaron · February 11th, 2015, 2:59 am

To be used to find mispricings in a basket of highly correlated assets. Is this idea viable? Any references? How would it work in practice?

daveangel · February 11th, 2015, 10:16 am

I don't think so - how can you find a mispricing in a basket as the basket will be the sum of its parts ?

theRedBaron · February 12th, 2015, 1:51 am

Not of the basket, but of individual contracts inside it. You know what I mean,

daveangel · February 26th, 2015, 5:59 am

no I am sorry I don't....

davidhigh · March 5th, 2015, 9:24 am

This isn't really an idea, it is just a statement. Can you please give more information on the form of the data on which you want to apply the SVD?Basically, an SVD -- or call it better Principal Component Analysis in this connection -- can be used to find the directions of maximum variance. From the first, it isn't made for outlier detection ... which of course doesn't mean it couldn't work somehow for that. Rather, as a first basic step, one usually looks at quartiles of the data and whether some points lie outside these quartiles.

Traden4Alpha · March 5th, 2015, 1:07 pm

One could use SVD/PCA as an extension of pairs trading in multiple dimensions but only if one assumes that the variance modes of the system are unchanging over time.