PortfolioOptimisers.jl

Appearance

18

Risk-measure ↔ optimiser compatibility ​

PortfolioOptimisers.supported_risk_measures Function
julia
supported_risk_measures(::Type{<:AbstractOptimisationEstimator})
supported_risk_measures(opt::AbstractOptimisationEstimator)

Return the abstract risk-measure category an optimiser accepts.

For a leaf optimiser the answer is derived entirely from its family supertype — the same abstract type its constraint builders already dispatch on — so it cannot drift from what the optimiser actually accepts (see ADR 0018):

  • risk-taking JuMP optimisers (RiskJuMPOptimisationEstimator: MeanRisk, NearOptimalCentering, FactorRiskContribution, RiskBudgeting) accept RiskMeasure;

  • clustering optimisers (ClusteringOptimisationEstimator: HierarchicalRiskParity, HierarchicalEqualRiskContribution, SchurComplementHierarchicalRiskParity) accept OptimisationRiskMeasure (i.e. RiskMeasure ∪ HierarchicalRiskMeasure);

  • every other leaf optimiser (naive, finite allocation) accepts no risk measure and returns Union{}.

Meta-optimisers (NestedClustered, Stacking, SubsetResampling) have no risk measure of their own — they delegate to their constituent optimisers and accept the intersection of what those optimisers accept, i.e. a measure only when every one of them accepts it:

  • NestedClustered — the inner (opti) and outer (opto) optimisers must both accept it;

  • Stacking — the outer (opto) and all inner (opti) optimisers must accept it (the inner optimisers may differ, so the category is intersected by broadcasting over the inner vector);

  • SubsetResampling — its single inner optimiser (opt) must accept it.

Called on an instance, the delegation reads the actual child optimisers, so it is exact even when a Stacking's inner optimisers are heterogeneous. Called on a type, it reads the child field types: exact for single-child metas and inner vectors whose element types are preserved, and conservatively Union{} where a field's concrete children are erased to an abstract type.

Related

source
PortfolioOptimisers.supports_risk_measure Function
julia
supports_risk_measure(::Type{<:AbstractOptimisationEstimator},
                      ::Type{<:AbstractBaseRiskMeasure}) -> Bool
supports_risk_measure(opt::AbstractOptimisationEstimator,
                      r::AbstractBaseRiskMeasure) -> Bool

Whether an optimiser accepts a given risk measure, as a Bool.

This is the programmatic form of the measure↔optimiser compatibility table. It is derived from supported_risk_measures — supports_risk_measure(O, R) === R <: supported_risk_measures(O) — so predicate, category, and dispatch share one source of truth and cannot drift. Pass both arguments as types, or both as instances (an instance query resolves meta-optimiser delegation against the actual child optimisers); see supported_risk_measures.

Examples

julia
julia> supports_risk_measure(MeanRisk, Variance)
true

julia> supports_risk_measure(HierarchicalRiskParity, EqualRisk)
true

julia> supports_risk_measure(MeanRisk, EqualRisk)
false

Related

source
PortfolioOptimisers._child_category Function
julia
_child_category(T::Type)

Return the risk-measure category accepted by a meta-optimiser child of type T, or Union{} when T is not an optimiser (e.g. an abstract or erased child field type). Used by the type-level meta-optimiser methods of supported_risk_measures to intersect the categories of their inner/outer optimisers.

Related

source