Prior Scenarios

A prior result carries its scenarios in X. A prior that reweights observations, such as an entropy pooling prior, states its prior probabilities on the scenarios of the prior that it wraps. keeps_observations tells it when the wrapped prior answers one scenario for each observation, so that it reads the count of the scenarios with no fit. A new prior implements a method of it when no configuration and no data makes its result drop a row. The other names of this topic are private, and the private page documents them.

PortfolioOptimisers.keeps_observations — Method

keeps_observations(pe::AbstractPriorEstimator) -> false
keeps_observations(pe::EmpiricalPrior) -> Bool
keeps_observations(pe::Union{<:FactorPrior, <:FactorBlackLittermanPrior,
                             <:BlackLittermanPrior, <:BayesianBlackLittermanPrior,
                             <:EntropyPoolingPrior, <:MeucciEntropyPoolingPrior,
                             <:HighOrderPriorEstimator,
                             <:HighOrderFactorPriorEstimator}) -> Bool
keeps_observations(pe::AugmentedBlackLittermanPrior) -> Bool
keeps_observations(pe::OpinionPoolingPrior) -> Bool
keeps_observations(pe::Online{<:AbstractPriorEstimator}) -> false

Return true when the result of a prior estimator carries one scenario for each observation that it receives.

A prior that reweights observations, such as EntropyPoolingPrior, states its prior probabilities on the scenarios of the prior that it wraps. A wrapped prior can carry fewer scenarios than it receives, so the reweighting prior reads the count with scenario_count before it fits the wrapped prior under the probabilities. A wrapped prior that keeps the observations carries the count that it receives, so the count needs no fit.

The fallback returns false, which makes no promise, and scenario_count then fits a prior that has no method of its own to read the count. A wrong true makes the reweighting prior check the probabilities against the wrong count. So a prior returns true only when no configuration and no data makes its result carry fewer scenarios than the observations that it receives. A CrossSectionalFactorPrior keeps the fallback, because it drops the observations that its Descriptors warm up over and the observations that its exposure lag uses. The answer reads the configuration alone. A Scenario Cap at or above the number of observations drops no row, but its prior still returns false.

Algorithm

The method that Julia selects is the algorithm.

Any other prior: return false.
An EmpiricalPrior: return true when it has no Scenario Cap, that is when pe.max_scenarios is nothing.
A prior whose result carries the scenarios of the prior in its pe field: apply this function to pe.pe. A factor prior cuts its asset rows to the rows of its factor prior, and the other priors forward the rows of their wrapped prior.
An AugmentedBlackLittermanPrior: return true when this function returns true for both pe.a_pe and pe.f_pe, because the result keeps the rows that both nested priors keep.
An OpinionPoolingPrior: return true when pe.pe1 is nothing, or when this function returns true for pe.pe1. The pool works on the rows of pe.pe1, and it refuses an opinion or a pe.pe2 that answers on fewer rows.
An Online wrapper of a prior: return false. A batch fit does not fit the wrapper, and the online flow replaces it with the prior that it wraps before the fit, so the wrapper makes no promise.

Arguments

pe: Prior estimator.

Returns

flag::Bool: true when the result carries one scenario for each observation, in order, and false when the prior makes no such promise.

Examples

julia> PortfolioOptimisers.keeps_observations(EmpiricalPrior())truejulia> PortfolioOptimisers.keeps_observations(EmpiricalPrior(; max_scenarios = 50))falsejulia> PortfolioOptimisers.keeps_observations(EntropyPoolingPrior(; pe = FactorPrior()))true

Related

source