|
From: Ferdinando A. <na...@am...> - 2006-10-12 19:28:59
|
Hi Yiping, > Just submitted a patch adding HyperSphere salvage algorithm to > PseudoSqrt. Luigi added your patch to the CVS, but I noticed it only works for correlation matrices, it fails with covariance matrices. The Spectral alternative approach works in both cases. It's probably just a normalization issue, would you consider looking into it? Besides you might want to consider the lower diagonal parametrization of the pseudo square root. This would make the optimization algorithm much more efficient by reducing the parameter by a factor of 2. Details about this approach can be found in "Monte Carlo methods in finance" by P. Jackel, and/or the following papers: http://www.quarchome.org/correlationmatrix.pdf http://www.fabiomercurio.it/riskcorr.pdf#search=%22rapisarda%20correlation%20geometric%22 ciao -- Nando |