NuclearPhynance Archive

Hi,



We are contemplating buying MOSEK for portfolio optimisation. We are trading systematic equity strategies and would need SOCP.



What are people using on NP?

MOSEK and Gurobi are names that come back often. Would you rather one against the other? Any other software we should consider?



btw, the following is useful if anybody else is looking for an optimiser: http://plato.asu.edu/guide.html



Thanks!

Ilog cplex is very good, especially for mixed integer stuff... expensive though

We've been doing a review recently and decided on Gurobi after a review of a few weeks. Our main interface would be R so that and support were our primary concern.



I wasn't involved in the trials but the focus was on nonlinear systems of equations and semidefinite programming.

Thanks,



This is interesting.

I've found mosek to be great for convex opt however often time off the shelf software ends up being a huge constraining factor and you end up having to write stuff on your own.

COmputational INfrastructure for Operations Research (COIN) has a huge list of open-source projects.



In MATLAB, I usually use MATLAB's Optimization Toolbox and also CVX.



In Python, I have used CVXOPT for general convex optimization and PuLP for linear programming (including integer programming).



These are OK for prototyping and for relatively small problems.

Thanks for the links.



We went with Mosek in the end and we are quite happy with is so far.



I found CVXOPT very slow and I don't remember it being able to solve SOCP problems. As you say it is ok for small problems.

Argog,



There's also Lobo & Vandenberghe & Boyd's SOCP: an "implementation of a primal-dual potential reduction method for solving second-order cone programming problems. Package includes C implementation, Matlab interface functions, documentation and examples."



It's old stuff, back from 1995. It's also academic software, which is usually atrociously documented. Yet, it's free.



More recent software would be the Embedded Conic Solver (ECOS), "a numerical software package for computing solutions to (or certifying infeasibility of) of conic optimization problems".

Cheers!



ECOS was very slow when I tried it.

I will give Lobo & Vandenberghe & Boyd's a try.

second this. had exactly the same issue. 2000+ asset constrained opts in cvxopt took 10+ mins to solve. mosek is < 2 secs.