Kurtosis 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,
    i,
    r::Kurtosis{<:Any, <:Any, <:Integer},
    opt::Union{AbstractProgrammeAllocationSet, RiskJuMPOptimisationEstimator},
    pr::AbstractPriorResult,
    args...;
    prefix,
    kwargs...
) -> Any

Add the approximate kurtosis of a Kurtosis to the model, and return its risk expression.

This method applies when r.N is an integer. cokurtosis_spectrum writes the fourth moment as a sum of $N^2$ squared quadratic forms, and the method keeps the $N_f$ forms of the largest eigenvalues. A larger r.N is more accurate and more expensive. The rank of $\mathbf{K}$ is at most $N(N+1)/2$, so every $N_f \geq N(N+1)/2$ keeps every form whose eigenvalue is not zero.

The method reads the measure's kt when the measure states one, and the prior's otherwise. It accepts any prior result, and assert_high_order_quantity refuses the measure when neither holds a cokurtosis matrix. Every approximate measure of the model that reads the prior's matrix reads one decomposition, get_kt_Akt_pm.

Mathematical definition

\[\begin{align} \widetilde{m}_4(\boldsymbol{w}) &= \sum_{i=1}^{N_f} \left(\boldsymbol{w}^\intercal \mathbf{B}_i \boldsymbol{w}\right)^2 \leq m_4(\boldsymbol{w})\,. \end{align}\]

Where:

  • $\widetilde{m}_4(\boldsymbol{w})$: Approximate fourth central moment of the portfolio returns.
  • $N_f$: Number of forms that the method keeps, r.N clamped to the range from $1$ to $N$, times $N$.
  • $m_4(\boldsymbol{w}) = (\boldsymbol{w} \otimes \boldsymbol{w})^\intercal \mathbf{K} (\boldsymbol{w} \otimes \boldsymbol{w})$: Fourth central moment of the portfolio returns, with $\otimes$ the Kronecker product.
  • $\mathbf{B}_i$: $N \times N$ matrix of eigenpair $i$ of the block vectorisation of $\mathbf{K}$, the reshaped $\sqrt{\lambda_i}\, \boldsymbol{q}_i$. cokurtosis_spectrum gives the eigenvalue $\lambda_i$, clamped at zero, and the unit eigenvector $\boldsymbol{q}_i$. The index $i$ counts down from the largest eigenvalue.
  • $\mathbf{K}$: Cokurtosis matrix of the risk measure, $N^2 \times N^2$. It is the measure's own kt when the measure states one, and the prior's otherwise.
  • $\boldsymbol{w}$: Portfolio weights vector $N \times 1$.

The model replaces $\boldsymbol{w} \boldsymbol{w}^\intercal$ with the lift $\mathbf{W}$, so each quadratic form becomes $\mathrm{Tr}(\mathbf{B}_i \mathbf{W})$, which is affine in the variables. A second-order cone bounds the norm of the vector of the $N_f$ forms.

Algorithm

  1. Check that r.kt or pr holds a cokurtosis matrix, with assert_high_order_quantity.
  2. Read sc, the constraint scale, and W, the lift of set_sdp_constraints!.
  3. Compute Nf, which is r.N clamped to 1:N, times N.
  4. Create sqrt_kurtosis_risk and x_kurt, and register them under their keys.
  5. Read vals_A and vecs_A, the spectrum of r.kt from cokurtosis_spectrum, else the spectrum of the prior's matrix from get_kt_Akt_pm.
  6. Build Bi, the matrices of the Nf largest eigenvalues.
  7. Register the two rows below under their keys.
  8. Bound sqrt_kurtosis_risk and add the risk expression with set_kurtosis_risk!, and return the risk expression.

JuMP formulation

Variables

  • W: read, $\mathbf{W}$, the lift that set_sdp_constraints! builds once for the owner of the weights.
  • kurtosis_risk_: created, the scalar $g$.
  • x_kurt_: created, the vector $\boldsymbol{x}_{kt}$ of $N_f$ entries.

Constraints

  • capprox_kurt_soc_: $(s_c g,\, s_c \boldsymbol{x}_{kt}) \in \mathcal{K}_{2}$, which is $g \geq \lVert \boldsymbol{x}_{kt} \rVert_2$.
  • capprox_kurt_: $s_c \left(x_{kt,j} - \mathrm{Tr}(\mathbf{B}_j \mathbf{W})\right) = 0$, $\forall\, j = 1,\ldots,N_f$.

Each name carries the index i of the measure, and the namespace prefix.

Where:

  • $g$: Epigraph variable of the square root of the kurtosis, $g \geq \lVert \boldsymbol{x}_{kt} \rVert_{2}$.
  • $\boldsymbol{x}_{kt}$: Vector whose squared norm is the kurtosis expression. The exact formulation states $\mathbf{G}_{kt} \mathbf{L}_2 \mathrm{vec}(\mathbf{W})$, and the approximate formulation states entry $j$ as $\mathrm{Tr}(\mathbf{B}_j \mathbf{W})$.
  • $\mathbf{W}$: Lifted weight matrix, symmetric $N \times N$. set_sdp_constraints! bounds it by $\mathbf{W} \succeq \boldsymbol{w} \boldsymbol{w}^\intercal / k$.
  • $k$: Variable that scales the budget in the homogenised problem.
  • $\mathcal{K}_{q} = \{(t, \boldsymbol{x}) : t \geq \lVert \boldsymbol{x} \rVert_{q}\}$: Norm cone of order $q$. $\mathcal{K}_{2}$ is the second-order cone.
  • $s_c$: Constraint scale. It multiplies both sides of a row, so a positive value leaves the feasible set unchanged.

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.

  • At $\mathbf{W} = \boldsymbol{w} \boldsymbol{w}^\intercal / k$, $\lVert \boldsymbol{x}_{kt} \rVert_2^2 = \widetilde{m}_4(\boldsymbol{w}) / k^2$, which lies at or below $m_4(\boldsymbol{w}) / k^2$. The two are equal when $N_f \geq N(N+1)/2$.
  • With $N_f \geq N(N+1)/2$, $\lVert \boldsymbol{x}_{kt} \rVert_2^2$ is the squared norm of the exact formulation in another basis, and the relaxation of the exact formulation applies. With a smaller $N_f$ the expression bounds $m_4$ in neither direction away from rank one.
  • kurtosis_risk_ lies at or above $\lVert \boldsymbol{x}_{kt} \rVert_2$. It is tight when the objective minimises the risk.
  • A bound on kurtosis_risk_ holds the weights to $\widetilde{m}_4$, not to $m_4$. With $N_f < N(N+1)/2$ the returned weights can have a fourth moment above the bound.

Arguments

  • model::JuMP.Model: The JuMP optimisation model.
  • i: Constraint index for unique variable and constraint naming.
  • r::Kurtosis: The measure, read for kt, N, alg and settings.
  • opt::RiskConstraintOwner: The owner of the risk constraint, a risk-based JuMP optimisation estimator or a programme Allocation Set.
  • pr::AbstractPriorResult: The prior result. It supplies kt when the measure states none.

Keyword arguments

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

Validation

Returns

Related

References

  • [12] D. Cajas. Advanced Portfolio Optimization: A Cutting-edge Quantitative Approach (Springer Nature Switzerland, 2025). Section 7.2.5.2, Equations 7.110 and 7.111. The original source of this formulation was not found. The formulation follows this work and is not checked against the original.
  • [195] D. Cajas. Approximation of Portfolio Kurtosis through Sum of Squared Quadratic Forms. Available at SSRN 4472793 (2023).
source
PortfolioOptimisers.set_risk_constraints! — Method
set_risk_constraints!(
    model::Model,
    i,
    r::Kurtosis{<:Any, <:Any, Nothing},
    opt::Union{AbstractProgrammeAllocationSet, RiskJuMPOptimisationEstimator},
    pr::AbstractPriorResult,
    args...;
    prefix,
    kwargs...
) -> Any

Add the kurtosis of a Kurtosis to the model in the exact semidefinite formulation, and return its risk expression.

This method applies when r.N is nothing. It is problem 13 of [66], with the budget homogenised by $k$. Equation 6 of that work states the fourth moment as a quadratic form in the entries on and below the diagonal of $\boldsymbol{w} \boldsymbol{w}^\intercal$, and the model replaces $\boldsymbol{w} \boldsymbol{w}^\intercal$ with the lift $\mathbf{W}$.

The method reads the measure's kt when the measure states one, and the prior's otherwise. Every exact measure of the model that reads the prior's matrix reads one factor, get_chol_or_Gkt_pm. The elimination and summation matrices come from dup_elim_sum_selector. They depend on the number of assets alone, so the method builds under a LowOrderPrior when the measure holds its own cokurtosis matrix. assert_high_order_quantity refuses the measure when neither holds one.

Mathematical definition

\[\begin{align} m_4(\boldsymbol{w}) &= \left[\mathbf{L}_2 \mathrm{vec}(\boldsymbol{w} \boldsymbol{w}^\intercal)\right]^\intercal \mathbf{S}_2 \mathbf{K} \mathbf{S}_2^\intercal \left[\mathbf{L}_2 \mathrm{vec}(\boldsymbol{w} \boldsymbol{w}^\intercal)\right] = \left\lVert \mathbf{G}_{kt} \mathbf{L}_2 \mathrm{vec}(\boldsymbol{w} \boldsymbol{w}^\intercal) \right\rVert_2^2\,. \end{align}\]

Where:

  • $m_4(\boldsymbol{w}) = (\boldsymbol{w} \otimes \boldsymbol{w})^\intercal \mathbf{K} (\boldsymbol{w} \otimes \boldsymbol{w})$: Fourth central moment of the portfolio returns, with $\otimes$ the Kronecker product.
  • $\mathbf{L}_2$: Elimination matrix, $N(N+1)/2 \times N^2$, which dup_elim_sum_matrices builds. $\mathbf{L}_2 \mathrm{vec}(\mathbf{A})$ holds the entries on and below the diagonal of $\mathbf{A}$, column by column.
  • $\mathbf{S}_2$: Summation matrix, $N(N+1)/2 \times N^2$, which dup_elim_sum_matrices builds. For a symmetric $\mathbf{A}$, $\mathbf{S}_2 \mathrm{vec}(\mathbf{A})$ holds each entry on and below the diagonal of $\mathbf{A}$, times the number of times it occurs in $\mathbf{A}$.
  • $\mathbf{K}$: Cokurtosis matrix of the risk measure, $N^2 \times N^2$. It is the measure's own kt when the measure states one, and the prior's otherwise.
  • $\mathbf{G}_{kt}$: Factor of the projected cokurtosis matrix, $N(N+1)/2 \times N(N+1)/2$, with $\mathbf{G}_{kt}^\intercal \mathbf{G}_{kt} = \mathbf{S}_2 \mathbf{K} \mathbf{S}_2^\intercal$. It is the transpose of the square root that matrix_square_root takes under the mtx_sqrt of the measure.
  • $\boldsymbol{w}$: Portfolio weights vector $N \times 1$.

Algorithm

  1. Check that r.kt or pr holds a cokurtosis matrix, with assert_high_order_quantity.
  2. Read sc, the constraint scale, and W, the lift of set_sdp_constraints!.
  3. Read L2 and S2 for the size of W, with dup_elim_sum_selector.
  4. Read G, the factor of r.kt projected by S2 under r.mtx_sqrt, else the factor of the prior's matrix from get_chol_or_Gkt_pm.
  5. Create sqrt_kurtosis_risk, and register it under its key.
  6. Build L2W once for the namespace, and build x_kurt.
  7. Register the cone row below under its key.
  8. Bound sqrt_kurtosis_risk and add the risk expression with set_kurtosis_risk!, and return the risk expression.

JuMP formulation

Variables

  • W: read, $\mathbf{W}$, the lift that set_sdp_constraints! builds once for the owner of the weights.
  • kurtosis_risk_: created, the scalar $g$.

Expressions

  • L2W: $\mathbf{L}_2 \mathrm{vec}(\mathbf{W})$, registered once for each namespace.
  • x_kurt_: $\boldsymbol{x}_{kt} = \mathbf{G}_{kt} \mathbf{L}_2 \mathrm{vec}(\mathbf{W})$.

Constraints

  • ckurt_soc_: $(s_c g,\, s_c \boldsymbol{x}_{kt}) \in \mathcal{K}_{2}$, which is $g \geq \lVert \boldsymbol{x}_{kt} \rVert_2$.

Each name except L2W carries the index i of the measure, and every name carries the namespace prefix.

Where:

  • $g$: Epigraph variable of the square root of the kurtosis, $g \geq \lVert \boldsymbol{x}_{kt} \rVert_{2}$.
  • $\boldsymbol{x}_{kt}$: Vector whose squared norm is the kurtosis expression. The exact formulation states $\mathbf{G}_{kt} \mathbf{L}_2 \mathrm{vec}(\mathbf{W})$, and the approximate formulation states entry $j$ as $\mathrm{Tr}(\mathbf{B}_j \mathbf{W})$.
  • $\mathbf{W}$: Lifted weight matrix, symmetric $N \times N$. set_sdp_constraints! bounds it by $\mathbf{W} \succeq \boldsymbol{w} \boldsymbol{w}^\intercal / k$.
  • $k$: Variable that scales the budget in the homogenised problem.
  • $\mathcal{K}_{q} = \{(t, \boldsymbol{x}) : t \geq \lVert \boldsymbol{x} \rVert_{q}\}$: Norm cone of order $q$. $\mathcal{K}_{2}$ is the second-order cone.
  • $s_c$: Constraint scale. It multiplies both sides of a row, so a positive value leaves the feasible set unchanged.

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.

  • $\lVert \boldsymbol{x}_{kt} \rVert_2^2 = \mathrm{vec}(\mathbf{W})^\intercal \mathbf{K} \mathrm{vec}(\mathbf{W})$. A sample cokurtosis is a mean of $(\boldsymbol{y}_t \otimes \boldsymbol{y}_t)(\boldsymbol{y}_t \otimes \boldsymbol{y}_t)^\intercal$ over centred observations $\boldsymbol{y}_t$, with non-negative weights. For such a matrix the expression is the mean of $(\boldsymbol{y}_t^\intercal \mathbf{W} \boldsymbol{y}_t)^2$, and $\mathbf{W} \succeq \boldsymbol{w} \boldsymbol{w}^\intercal / k$ gives $\boldsymbol{y}_t^\intercal \mathbf{W} \boldsymbol{y}_t \geq (\boldsymbol{y}_t^\intercal \boldsymbol{w})^2 / k$. So the expression lies at or above $m_4(\boldsymbol{w}) / k^2$.
  • kurtosis_risk_ lies at or above $\lVert \boldsymbol{x}_{kt} \rVert_2$, so at or above $\sqrt{m_4(\boldsymbol{w})} / k$. It is tight at $\mathbf{W} = \boldsymbol{w} \boldsymbol{w}^\intercal / k$, which the objective reaches when it minimises the risk and no other row acts on $\mathbf{W}$.
  • A bound on kurtosis_risk_ is exact, because the weights satisfy it when $\sqrt{m_4(\boldsymbol{w})}$ is at most the bound.

Arguments

  • model::JuMP.Model: The JuMP optimisation model.
  • i: Constraint index for unique variable and constraint naming.
  • r::Kurtosis: The measure, read for kt, mtx_sqrt, alg and settings.
  • opt::RiskConstraintOwner: The owner of the risk constraint, a risk-based JuMP optimisation estimator or a programme Allocation Set.
  • pr::AbstractPriorResult: The prior result. It supplies kt when the measure states none.

Keyword arguments

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

Validation

  • r.kt or pr holds a cokurtosis matrix. assert_high_order_quantity raises an ArgumentError otherwise.
  • The projected matrix has a square root under r.mtx_sqrt, as matrix_square_root states. Raises a LinearAlgebra.PosDefException.

Returns

Related

References

  • [66] D. Cajas. Convex Optimization of Portfolio Kurtosis. Available at SSRN 4202967 (2022). Section 2, Equation 6, and Section 3.1, Equation 13.
source

References

[12]
D. Cajas. Advanced Portfolio Optimization: A Cutting-edge Quantitative Approach (Springer Nature Switzerland, 2025).
[66]
D. Cajas. Convex Optimization of Portfolio Kurtosis. Available at SSRN 4202967 (2022).
[195]
D. Cajas. Approximation of Portfolio Kurtosis through Sum of Squared Quadratic Forms. Available at SSRN 4472793 (2023).