Implementation of Nicholas J. Higham's algorithm "Computing the Nearest Correlation Matrix - A Problem from Finance", 2002.
I found this algorithm to be very slow and to not successfully converge for the problem I was working on. I found a number of useful implementations of this algorithm at https://nickhigham.wordpress.com/2013/02/13/the-nearest-correlation-matrix/, and found one there that worked better.
Hi Erlend, all,
first of all, many thanks for providing this useful tool. I used it to transform symmetric 80x80 correlation matrices (~1300 correlation matrices, one for each day during my observation period). I encountered 2 difficutlties: 1. The program took ages to complete (14 hours) (despite high-performance Xeon workstation) and the result was not satisfying because only 15 of the resulting transformed correlation matrices were positive-definite. Could you please comment on options to increase speed and how to ensure that the resulting matrices are positive-definite? Many thanks for your feedback
I have found its use rather straightforward. Just pass the (quasi)correlation matrix for which you'd like to find the nearest valid correlation matrix (described in the reference above) ... and it's done.
The only parameter that you can tweak it seems to be the number of iterations that originally is set to 100 times the squared dimension of your problem.
No (useful) help, no comments, no H1 line. How do you expect that anybody else than you is able to use this function properly ...
Updated to add a BSD License.
Create scripts with code, output, and formatted text in a single executable document.