[R-SIG-Finance] Genetic Algorithms & Portfolio Optimization

[Brian G. Peterson]

On Saturday, January 22, 2011 04:33:09 pm Lui ## wrote:
> Dear group,
> 
> I was just wondering whether some of you have some experience with the
> package "rgenoud" which does provide genetic algorithms for complex
> optimization problems. 

<...>

> What is your general experience? Did you ever try solving the
> Markowitz portfolio with the rgenoud package?
> I know that there are good solvers around for the qudratic programming
> problem of the markowitz portfolio, but I want to go into a different
> direction which translates into a quadratic problem with quadratic
> constraints (and I havent found a good solver for that...).
> 
> I am interested in your replies! Have a good weekend!

As others have already said, for a quadtratic problem with quadratic 
constraints, there is an exact analytical solution.

In these cases, you will be much better off both from a performance and 
accuracy perspective in using a quadratic solver (quadprog is most often 
applied in R, see list archives and many packages for examples).

Other portfolio problems may be stated in terms of linear solvers, which will 
likewise be faster than a global optimizer for finding an exact analytical 
solution.

If, however, your portfolio problem is non-convex and non-smooth, then a 
genetic algorithm, a migration algorithm, grid search, or random portfolios 
may be a good option for finding a near-optimal portfolio.  If this is your 
true goal, perhaps you can say a little more about your actual constraints and 
objectives (and use assets that are outside of your true area of interest, 
such as the S&P sector indices).

Regards,

  - Brian 

-- 
Brian G. Peterson
http://braverock.com/brian/
Ph: 773-459-4973
IM: bgpbraverock