Brownian Distance Variance Constraints

The methods below are public. The other names of this topic are private, and the private page documents them.

PortfolioOptimisers.set_risk_constraints! — Method
set_risk_constraints!(
    model::Model,
    ,
    r::BrownianDistanceVariance,
    opt::Union{AbstractProgrammeAllocationSet, RiskJuMPOptimisationEstimator},
    pr::AbstractPriorResult,
    args...;
    prefix,
    kwargs...
) -> Any

Add the BrownianDistanceVariance bound of the net portfolio returns to model, and return its risk expression.

r.alg1 selects the risk expression, and r.alg2 selects the rows that tie the distance matrix to the returns. When the objective minimises the risk, bdvariance_risk equals the functor of the measure at the solved weights.

The build is keyed on prefix alone, with no measure index. A second call with the same prefix uses the registered expression and builds no rows. The settings of the first BrownianDistanceVariance take effect, and a second measure logs a warning, because its bound and its scale have no effect.

Algorithm

  1. When bdvariance_risk is registered under prefix, go to step 6.
  2. Read the net portfolio return series net_X with set_net_portfolio_returns!, and its length T.
  3. Create Dt, and register Dx, the matrix of the pairwise differences of net_X.
  4. Register the risk expression of r.alg1 with set_brownian_distance_risk_constraint!.
  5. Add the rows of r.alg2 with set_brownian_distance_variance_constraints!.
  6. Convert the bound in r.settings to the units of a deviation with variance_risk_bounds_val, and take the expression that it acts on from set_brownian_distance_bounds_expr!. The risk expression has degree two in the weights, and the row of set_risk_upper_bound! multiplies the bound by $k$, so under a ratio objective the bound holds only on an expression of degree one.
  7. Add the bound on the deviation and the risk expression with set_unindexed_risk_bounds_and_expression!, and return the risk expression.

JuMP formulation

Variables

  • Dt: created, the symmetric $T \times T$ matrix $\mathbf{D}$.
  • bdvariance_dev: created by set_brownian_distance_bounds_expr! when r.settings holds a bound, the square root of the risk expression.

Expressions

  • Dx: $\boldsymbol{\Delta} = \hat{\boldsymbol{r}} \boldsymbol{1}^\intercal - \boldsymbol{1} \hat{\boldsymbol{r}}^\intercal$.

The two calls register bdvariance_risk and the rows, and their docstrings state them. Each name carries the namespace prefix.

Where:

  • $\mathbf{D}$: Symmetric matrix variable of the distances between the portfolio returns.
  • $\boldsymbol{\Delta}$: Matrix of the differences $\Delta_{ij} = \hat{r}_i - \hat{r}_j$.
  • $\hat{\boldsymbol{r}}$: The net portfolio returns net_X, $T \times 1$.
  • $\boldsymbol{1}$: Column vector of ones, $T \times 1$.
  • $T$: Number of observations.

Relaxation

The encoding is not exact: the entries below bound the quantity instead of reproducing it, and the bound is tight only under the condition stated here.

  • bdvariance_risk lies at or above $\mathrm{BDVar}(\hat{\boldsymbol{r}})$, and is tight when the objective minimises the risk. The two calls state the reasons.
  • $\mathrm{BDVar}(\hat{\boldsymbol{r}})$ lies at or above the distance variance of $\hat{\boldsymbol{r}}$. It is tight only for a constant series.

Arguments

  • model::JuMP.Model: The JuMP optimisation model.
  • ::Any: The index of the measure in the risk vector. The method does not read it.
  • r::BrownianDistanceVariance: The measure, read for alg1, alg2 and settings.
  • opt::RiskConstraintOwner: The owner of the risk constraint, a risk-based JuMP optimisation estimator or a programme Allocation Set.
  • pr::AbstractPriorResult: Prior result containing the returns matrix X.

Keyword arguments

  • prefix::Symbol: Model State namespace of the build (default: empty, i.e. the bare key).

Returns

  • bdvariance_risk: The risk expression.

Related

source