Further tail coefficients of families of Nelsen's Table 4.1 #
Applications of Nelsen, An Introduction to Copulas, second edition, Corollary 5.4.3:
- for a strict generator,
λ_L = lim_{x → ∞} ψ(2x)/ψ(x)(BivariateGenerator.hasLowerTailDependence_of_tendsto). Families 10, 13 and 17 are strict withψ(2x)/ψ(x) → 0, soλ_L = 0(hasLowerTailDependence_nelsen10,hasLowerTailDependence_nelsen13,hasLowerTailDependence_nelsen17). For family 13,ψ(2x)/ψ(x) = exp((1 + x)^{1/θ} − (1 + 2x)^{1/θ}); for family 10 it is((e^x + 1)/(e^{2x} + 1))^{1/θ} ≤ (2e^{−x})^{1/θ}; for family 17,ψ(s) = G(e^{−s})withG(0) = 0,G'(0) ≠ 0, soψ(2x)/ψ(x) ~ e^{−x}. λ_U = 2 − lim_{x → 0⁺} (1 − ψ(2x))/(1 − ψ(x))(BivariateGenerator.hasUpperTailDependence_of_tendsto). For family 21,1 − ψ(x) = (1 − (1 − x)^θ)^{1/θ}near0, so the limit is2^{1/θ}andλ_U = 2 − 2^{1/θ}(hasUpperTailDependence_nelsen21).
The helper tendsto_ratio_of_hasDerivAt states the elementary fact that if H(0) = 0 and
H'(0) = d ≠ 0 then H(2x)/H(x) → 2 as x → 0⁺.
theorem
ProbabilityTheory.Copula.tendsto_ratio_of_hasDerivAt
{H : ℝ → ℝ}
{d : ℝ}
(hH0 : H 0 = 0)
(hd : HasDerivAt H d 0)
(hd0 : d ≠ 0)
:
Filter.Tendsto (fun (x : ℝ) => H (2 * x) / H x) (nhdsWithin 0 (Set.Ioi 0)) (nhds 2)
If H(0) = 0 and H'(0) = d ≠ 0, then H(2x)/H(x) → 2 as x → 0⁺.
Lower tails of strict families #
theorem
ProbabilityTheory.Copula.hasLowerTailDependence_nelsen13
(θ : ℝ)
(hθ : 0 < θ)
:
(nelsen13 θ hθ).HasLowerTailDependence 0
Nelsen's family 13 has no lower tail dependence.
theorem
ProbabilityTheory.Copula.hasLowerTailDependence_nelsen10
(θ : ℝ)
(hθ : 0 < θ)
(h1 : θ ≤ 1)
:
(nelsen10 θ hθ h1).HasLowerTailDependence 0
Nelsen's family 10 has no lower tail dependence.
theorem
ProbabilityTheory.Copula.hasLowerTailDependence_nelsen17
(θ : ℝ)
(hθ : θ ≠ 0)
:
(nelsen17 θ hθ).HasLowerTailDependence 0
Nelsen's family 17 has no lower tail dependence.
Upper tail of family 21 #
theorem
ProbabilityTheory.Copula.hasUpperTailDependence_nelsen21
(θ : ℝ)
(hθ : 1 ≤ θ)
:
(nelsen21 θ hθ).HasUpperTailDependence (2 - 2 ^ θ⁻¹)
Nelsen's family 21 has upper tail-dependence coefficient λ_U = 2 − 2^{1/θ}.