QuoteOriginally posted by: neuroguyIn practice:RegressionPCAFactor AnalysisNumerical OptimisationPortfolio theory:Optimisation (i.e. Langrange multipliers etc)Linear algebrabut these things dont work in unmodified form because asset prices are not smooth, so one might be also interested in methods to estimate covariance matriciesin what way are asset prices not smooth ?

You must be missing frenchX Dave !!!

QuoteOriginally posted by: spv205You must be missing frenchX Dave !!!haha

QuoteOriginally posted by: daveangelQuoteOriginally posted by: neuroguyIn practice:RegressionPCAFactor AnalysisNumerical OptimisationPortfolio theory:Optimisation (i.e. Langrange multipliers etc)Linear algebrabut these things dont work in unmodified form because asset prices are not smooth, so one might be also interested in methods to estimate covariance matriciesin what way are asset prices not smooth ?Fair enough. I wrote this in a bit of a hurry. Firstly the above applications are areas that apply to some buyside activities. Earlier the OP asked what asset managers might use.Secondly what I mean is that the expected returns in some alpha model may be random variables themselves. Hence straightforward optimisation can lead to overtuned results that are not robust. Hence one might wish to include uncertainty in those factors. Similarly mean-var optimisation applied using a covariance matrix that is derived from real data (again that is a random realisation from some set of matricies) may fail to be robust because one is optimising with noise. So I guess what I am saying is that because these things are randomly realised they might have spurious maxima. Of course something can be smooth and suffer from spurious maxima, so I might have been mis-speaking. But I have in mind a sort of 'spiky' surface. This is not smooth by the standards of standard differentiation which is where that choice of words originated.