Theorem 4 (Implied ambiguity) (corrected: requires for every ) — is minimized on at with value
ProvedImpliedAmbiguity.theorem_4ambiguity-aversionmean-variancep2o-correctedp2o-gran-per-chapterp2o-plan-paperp2o-v1portfolio-choice
Let be a symmetric positive definite matrix (the covariance matrix of the risky asset returns), the expected returns, the risk-free return and a portfolio of risky assets. Write and suppose
Let (15), (17) and (19). Assume moreover that for every , i.e. is not a positive multiple of the tangency portfolio . With the Sharpe ratio put
Then, as a function of on :
- ;
- for every ;
- is the only minimizer of on ;
- the minimum value, the implied ambiguity of , is
The implied ambiguity is the smallest coefficient of relative ambiguity aversion with which a smooth-ambiguity (KMM) investor with CARA utilities and jointly normal returns holds as optimal; it is an explicit, strictly decreasing function of the normalized Sharpe ratio .
Differences from the source:
- correction: no hypothesis on v^θ → ∀ θ > 0, v^θ ≠ 0 (Lemma 4's hypothesis, by which the paper proves the theorem); reason: the printed theorem is false when x is a positive multiple of x̄ — N = 1, Σ_R = 1, µ = 2, R_f = 1, x = 1 satisfies (14), θ̄ = 1, η^θ = 1/θ − 1 has no minimum on (0, 1), and (24) gives θ* = 1 ∉ (0, θ̄).
- reading: "η^θ is minimized when θ is equal to θ*" → θ* ∈ (0, θ̄), θ* minimizes η^θ on (0, θ̄), and it is the only minimizer there; reason: Lemma 4, which the paper's proof applies, asserts a unique minimizer in (0, θ̄).
- encoding: Σ_R ∈ S^N_++, µ − R_f1, θ̄, η^θ defined for v^θ ≠ 0, Sr on ℝᴺ∖{0}, (·)^{1/2}, "Let θ > 0 and suppose (16)" →
Matrix.PosDef,excessReturn μ Rf,thetaBar, totaletaThetaminimized over exactlySet.Ioo 0 θ̄, totalsharpe,Real.sqrt, θ bound insideIsMinOn; reason: Lean convention; on (0, θ̄) every quantity is the paper's.
Preamble
import Mathlib import Definitions.Def_ImpliedAmbiguity_excessReturn import Definitions.Def_ImpliedAmbiguity_thetaBar import Definitions.Def_ImpliedAmbiguity_vTheta import Definitions.Def_ImpliedAmbiguity_etaTheta import Definitions.Def_ImpliedAmbiguity_sharpe import Definitions.Def_ImpliedAmbiguity_tangency open Matrix
Formal statement
namespace ImpliedAmbiguity
/-- Hara & Honda, *Implied Ambiguity: Mean-Variance Efficiency and Pricing Errors*, working
paper (October 9, 2018; published version Management Science 2022, doi:10.1287/mnsc.2021.4097),
Theorem 4 (Implied ambiguity), pp.21–22: Let `Σ_R ∈ S^N_++`, `µ ∈ R^N`, `R_f ∈ R` and `x ∈ R^N`,
and suppose (14) `(µ − R_f1)ᵀx > 0`. Define `θ̄` as in (15) and `η^θ` as in (19). Then, as a
function of `θ` defined on `(0, θ̄)`, `η^θ` is minimized when `θ` is equal to
(24) `θ* ≡ θ̄ / (1 + (1 − (Sr(x)/Sr(x̄))²)^{1/2})`, and the minimized value is
(25) `η^{θ*} = 2(1 − (Sr(x)/Sr(x̄))²)^{1/2} / (1 − (1 − (Sr(x)/Sr(x̄))²)^{1/2})`.
**Correction.** The printed theorem is false when `x` is a positive multiple of the tangency
portfolio `x̄ = Σ_R⁻¹(µ − R_f1)` (e.g. `N = 1`, `Σ_R = 1`, `µ = 2`, `R_f = 1`, `x = 1`: `θ̄ = 1`,
`η^θ = 1/θ − 1` on `(0, 1)` has no minimum and (24) gives `θ* = 1 ∉ (0, θ̄)`). The hypothesis `hv`
is Lemma 4's hypothesis "there is no θ > 0 for which v^θ = 0" (p.41), by which the paper proves
Theorem 4.
"η^θ is minimized when θ = θ*" is read as: `θ* ∈ (0, θ̄)`, `θ*` minimizes `η^θ` on `(0, θ̄)`, and
it is the only minimizer there (Lemma 4: "a unique θ").
**Formalization Note.** `S^N_++` is `Matrix.PosDef`; `µ − R_f1` is `excessReturn μ Rf`; `θ̄`,
`v^θ`, `η^θ`, `Sr`, `x̄` are `thetaBar`, `vTheta`, `etaTheta`, `sharpe`, `tangency`; `(·)^{1/2}` is
`Real.sqrt`. `η^θ` is a total function of `θ` and is minimized over exactly `Set.Ioo 0 θ̄`. -/
theorem theorem_4 {N : ℕ} (SigmaR : Matrix (Fin N) (Fin N) ℝ) (hSigmaR : SigmaR.PosDef)
(μ : Fin N → ℝ) (Rf : ℝ) (x : Fin N → ℝ)
(h14 : 0 < excessReturn μ Rf ⬝ᵥ x)
(hv : ∀ θ : ℝ, 0 < θ → vTheta SigmaR (excessReturn μ Rf) x θ ≠ 0) :
let c := excessReturn μ Rf
let θbar := thetaBar SigmaR c x
let q := Real.sqrt (1 - (sharpe SigmaR c x / sharpe SigmaR c (tangency SigmaR c)) ^ 2)
let θstar := θbar / (1 + q)
θstar ∈ Set.Ioo 0 θbar ∧
IsMinOn (etaTheta SigmaR c x) (Set.Ioo 0 θbar) θstar ∧
(∀ θ ∈ Set.Ioo 0 θbar, etaTheta SigmaR c x θ = etaTheta SigmaR c x θstar → θ = θstar) ∧
etaTheta SigmaR c x θstar = 2 * q / (1 - q) := by sorry
end ImpliedAmbiguity
Source
Hara & Honda, Implied Ambiguity: Mean-Variance Efficiency and Pricing Errors, working paper (October 9, 2018; published version doi:10.1287/mnsc.2021.4097), Theorem 4, eqs. (24)–(25), pp.21–22; correction from Appendix C, Lemma 4, p.41
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.