Brownian Distance Variance: private API
PortfolioOptimisers.BrownianDistanceVarianceFormulation — Type
abstract type BrownianDistanceVarianceFormulation <: AbstractAlgorithmAbstract supertype for the rows that tie the distance matrix of a BrownianDistanceVariance model to the portfolio returns.
The model holds the distances $|x_i - x_j|$ in a symmetric matrix variable $\mathbf{D}$, and a member of this family selects the rows that bound each entry from below, $D_{ij} \geq |x_i - x_j|$. Both members state the same feasible set, so both give the same optimum and the same risk. They differ in the type and the number of rows that the solver reads. The risk expression increases in each entry of $\mathbf{D}$, so the rows are tight when the objective minimises the risk. set_brownian_distance_variance_constraints! states the rows and the relaxation.
All concrete types implementing a formulation of these rows should subtype BrownianDistanceVarianceFormulation.
Related
PortfolioOptimisers.BDVarRkFormulations — Type
const BDVarRkFormulations = Union{RSOCRiskExpr, QuadRiskExpr}Groups the two encodings that BrownianDistanceVariance accepts for its risk expression.
Both report the bound itself, not a root. QuadRiskExpr states the risk as a quadratic form in the distance matrix. RSOCRiskExpr holds each of the two squares of the bound with a rotated second-order cone, so the risk expression is affine and a bound on the risk is a conic row. A bound on the quadratic form needs a bridge to a cone. The type bound of the constructor refuses SOCRiskExpr and SquaredSOCRiskExpr. set_brownian_distance_risk_constraint! states the expressions.
Related