constrained_objective: calculate a numeric return value for a portfolio based on a set of constraints and objectives

Description

Function to calculate a numeric return value for a portfolio based on a set of constraints and objectives. We'll try to make as few assumptions as possible and only run objectives that are enabled by the user.

Usage

constrained_objective_v1(w, R, constraints, ..., trace = FALSE, normalize = TRUE, storage = FALSE)
constrained_objective_v2(w, R, portfolio, ..., trace = FALSE, normalize = TRUE, storage = FALSE, env = NULL)
constrained_objective(w, R, portfolio, ..., trace = FALSE, normalize = TRUE, storage = FALSE, env = NULL)

Arguments

an xts, vector, matrix, data frame, timeSeries or zoo object of asset returns.

a vector of weights to test.

portfolio

an object of class portfolio specifying the constraints and objectives for the optimization, see portfolio.

...

any other passthru parameters.

trace

TRUE/FALSE whether to include debugging and additional detail in the output list. The default is FALSE. Several charting functions require that trace=TRUE.

normalize

TRUE/FALSE whether to normalize results to min/max sum (TRUE), or let the optimizer penalize portfolios that do not conform (FALSE)

storage

TRUE/FALSE default TRUE for DEoptim with trace, otherwise FALSE. not typically user-called.

constraints

a v1_constraint object for backwards compatibility with constrained_objective_v1.

env

environment of moments calculated in optimize.portfolio

Details

If the user has passed in either min_sum or max_sum constraints for the portfolio, or both, and are using a numerical optimization method like DEoptim, and normalize=TRUE, we'll normalize the weights passed in to whichever boundary condition has been violated. If using random portfolios, all the portfolios generated will meet the constraints by construction. NOTE: this means that the weights produced by a numeric optimization algorithm like DEoptim, pso, or GenSA might violate constraints, and will need to be renormalized after optimizing. We apply the same normalization in optimize.portfolio so that the weights you see have been normalized to min_sum if the generated portfolio is smaller than min_sum or max_sum if the generated portfolio is larger than max_sum. This normalization increases the speed of optimization and convergence by several orders of magnitude in many cases.

You may find that for some portfolios, normalization is not desirable, if the algorithm cannot find a direction in which to move to head towards an optimal portfolio. In these cases, it may be best to set normalize=FALSE, and penalize the portfolios if the sum of the weighting vector lies outside the min_sum and/or max_sum.

Whether or not we normalize the weights using min_sum and max_sum, and are using a numerical optimization engine like DEoptim, we will penalize portfolios that violate weight constraints in much the same way we penalize other constraints. If a min_sum/max_sum normalization has not occurred, convergence can take a very long time. We currently do not allow for a non-normalized full investment constraint. Future version of this function could include this additional constraint penalty.

When you are optimizing a return objective, you must specify a negative multiplier for the return objective so that the function will maximize return. If you specify a target return, any return that deviates from your target will be penalized. If you do not specify a target return, you may need to specify a negative VTR (value to reach) , or the function will not converge. Try the maximum expected return times the multiplier (e.g. -1 or -10). Adding a return objective defaults the multiplier to -1.

Additional parameters for other solvers (e.g. random portfolios or DEoptim.control or pso or GenSA may be passed in via ...

Description

Usage

Arguments

Details

See Also