Documentation

Copula.RandomVariable.Monotone

← Copula mathematical handbook

Monotone functional dependence and the Fréchet bounds #

Nelsen, An Introduction to Copulas, second edition, Theorem 2.5.4 and Corollary 2.5.5 (random-variable version, for laws on Fin 2 → ℝ with continuous marginal CDFs).

For such a law μ, the Sklar copula is the law of (F₀ x₀, F₁ x₁). It equals the comonotonic copula M exactly when F₀ x₀ = F₁ x₁ almost surely, and the countermonotonic copula W exactly when F₁ x₁ = 1 - F₀ x₀ almost surely. If the second coordinate is almost surely a strictly increasing (respectively decreasing) function of the first, these almost-sure identities hold, so the copula is M (respectively W).

The converse implication is proved as well, which gives the full Theorem 2.5.4: the copula is M if and only if x₁ = f x₀ almost surely for a function f that is monotone on a set carrying the law of x₀ (ofContinuousMarginals_eq_comonotonic_iff_exists_monotoneOn), and W if and only if the same holds with an antitone f (ofContinuousMarginals_eq_countermonotonic_iff_exists_antitoneOn). The function is explicit: f = G₁ ∘ F₀ (respectively f = G₁ ∘ (1 - F₀)), with G₁ = realQuantile the quantile of the second marginal, and the carrier set is {x | 0 < F₀ x < 1}.

Monotonicity cannot in general be required on all of ℝ: if x₀ is uniform on [0,1] and x₁ = Φ⁻¹(x₀) is standard normal, then the copula is M, but a monotone f : ℝ → ℝ with x₁ = f x₀ almost surely would have to be -∞ on (-∞, 0). Nor can strict monotonicity be required, since F₀ is constant on gaps in the support of x₀. For globally monotone f the sufficiency direction is eq_comonotonic_of_ae_eq_monotone.

The Sklar copula of a bivariate law with continuous marginals is M if and only if the probability integral transforms of the two coordinates agree almost surely.

The Sklar copula of a bivariate law with continuous marginals is W if and only if the second probability integral transform is almost surely one minus the first.

If the second coordinate is almost surely g of the first, its law is the image law.

Nelsen, Theorem 2.5.4 (sufficiency, increasing case). If the second coordinate is almost surely a strictly increasing function of the first, the Sklar copula is M.

Nelsen, Theorem 2.5.4 (sufficiency, decreasing case). If the second coordinate is almost surely a strictly decreasing function of the first, the Sklar copula is W.

theorem ProbabilityTheory.Copula.eq_comonotonic_of_isSklarCopula_of_ae_eq_strictMono {μ : MeasureTheory.ProbabilityMeasure (Fin 2 → ℝ)} (hc : ∀ (i : Fin 2), Continuous ↑(ProbabilityTheory.cdf (marginal μ i))) {C : Copula 2} (hC : IsSklarCopula μ C) {g : ℝ → ℝ} (hg : StrictMono g) (h : ∀ᵐ (x : Fin 2 → ℝ) ∂↑μ, x 1 = g (x 0)) :

Any Sklar copula of a law with continuous marginals whose second coordinate is almost surely a strictly increasing function of the first is M.

Any Sklar copula of a law with continuous marginals whose second coordinate is almost surely a strictly decreasing function of the first is W.

Monotone functions on a carrier set: the full Theorem 2.5.4 #

A coordinate of a law with continuous marginal CDF takes each fixed value with probability zero.

theorem ProbabilityTheory.Copula.cdf_marginal_one_of_ae_eq_monotoneOn (μ : MeasureTheory.ProbabilityMeasure (Fin 2 → ℝ)) (hc1 : Continuous ↑(ProbabilityTheory.cdf (marginal μ 1))) {f : ℝ → ℝ} {S : Set ℝ} (hf : MonotoneOn f S) (h : ∀ᵐ (x : Fin 2 → ℝ) ∂↑μ, x 0 ∈ S ∧ x 1 = f (x 0)) {y : ℝ} (hy : y ∈ S) :

If x₁ = f x₀ almost surely with f monotone on a set carrying x₀, the marginal CDFs are linked by F₁ (f y) = F₀ y on that set.

theorem ProbabilityTheory.Copula.cdf_marginal_one_of_ae_eq_antitoneOn (μ : MeasureTheory.ProbabilityMeasure (Fin 2 → ℝ)) (hc : ∀ (i : Fin 2), Continuous ↑(ProbabilityTheory.cdf (marginal μ i))) {f : ℝ → ℝ} {S : Set ℝ} (hf : AntitoneOn f S) (h : ∀ᵐ (x : Fin 2 → ℝ) ∂↑μ, x 0 ∈ S ∧ x 1 = f (x 0)) {y : ℝ} (hy : y ∈ S) :

If x₁ = f x₀ almost surely with f antitone on a set carrying x₀, the marginal CDFs are linked by F₁ (f y) = 1 - F₀ y on that set.

theorem ProbabilityTheory.Copula.eq_comonotonic_of_ae_eq_monotoneOn (μ : MeasureTheory.ProbabilityMeasure (Fin 2 → ℝ)) (hc : ∀ (i : Fin 2), Continuous ↑(ProbabilityTheory.cdf (marginal μ i))) {f : ℝ → ℝ} {S : Set ℝ} (hf : MonotoneOn f S) (h : ∀ᵐ (x : Fin 2 → ℝ) ∂↑μ, x 0 ∈ S ∧ x 1 = f (x 0)) :

Nelsen, Theorem 2.5.4 (sufficiency, increasing case, general form). If the second coordinate is almost surely f of the first, with f nondecreasing on a set carrying the first coordinate, the Sklar copula is M.

theorem ProbabilityTheory.Copula.eq_countermonotonic_of_ae_eq_antitoneOn (μ : MeasureTheory.ProbabilityMeasure (Fin 2 → ℝ)) (hc : ∀ (i : Fin 2), Continuous ↑(ProbabilityTheory.cdf (marginal μ i))) {f : ℝ → ℝ} {S : Set ℝ} (hf : AntitoneOn f S) (h : ∀ᵐ (x : Fin 2 → ℝ) ∂↑μ, x 0 ∈ S ∧ x 1 = f (x 0)) :

Nelsen, Theorem 2.5.4 (sufficiency, decreasing case, general form). If the second coordinate is almost surely f of the first, with f nonincreasing on a set carrying the first coordinate, the Sklar copula is W.

theorem ProbabilityTheory.Copula.eq_comonotonic_of_ae_eq_monotone (μ : MeasureTheory.ProbabilityMeasure (Fin 2 → ℝ)) (hc : ∀ (i : Fin 2), Continuous ↑(ProbabilityTheory.cdf (marginal μ i))) {f : ℝ → ℝ} (hf : Monotone f) (h : ∀ᵐ (x : Fin 2 → ℝ) ∂↑μ, x 1 = f (x 0)) :

Sufficiency with a globally nondecreasing function.

Sufficiency with a globally nonincreasing function.

Almost surely the first probability integral transform lies strictly inside (0, 1).

Almost surely the quantile of the second marginal inverts its CDF.

Nelsen, Theorem 2.5.4 (necessity, increasing case). If the Sklar copula is M, then x₁ = G₁ (F₀ x₀) almost surely, where G₁ is the quantile of the second marginal.

Nelsen, Theorem 2.5.4 (necessity, decreasing case). If the Sklar copula is W, then x₁ = G₁ (1 - F₀ x₀) almost surely, where G₁ is the quantile of the second marginal.

Nelsen, Theorem 2.5.4 (increasing case). For a law with continuous marginals, the Sklar copula is M if and only if the second coordinate is almost surely a nondecreasing function of the first, where the function need only be nondecreasing on a set carrying the first coordinate.

Nelsen, Theorem 2.5.4 (decreasing case). For a law with continuous marginals, the Sklar copula is W if and only if the second coordinate is almost surely a nonincreasing function of the first, where the function need only be nonincreasing on a set carrying the first coordinate.

theorem ProbabilityTheory.Copula.IsSklarCopula.eq_comonotonic_iff_exists_monotoneOn {μ : MeasureTheory.ProbabilityMeasure (Fin 2 → ℝ)} (hc : ∀ (i : Fin 2), Continuous ↑(ProbabilityTheory.cdf (marginal μ i))) {C : Copula 2} (hC : IsSklarCopula μ C) :
C = comonotonic 2 ↔ ∃ (f : ℝ → ℝ) (S : Set ℝ), MonotoneOn f S ∧ ∀ᵐ (x : Fin 2 → ℝ) ∂↑μ, x 0 ∈ S ∧ x 1 = f (x 0)

Theorem 2.5.4 for an arbitrary Sklar copula of a law with continuous marginals (increasing case).

Theorem 2.5.4 for an arbitrary Sklar copula of a law with continuous marginals (decreasing case).