Current-price independence repairs exploration noise moments
ProvedBFWFluidLimit.conditionalIndependence_repairs_momentsLet F be the past information and G = F ∨ σ(P) include the full current price vector. Assume that every integrable measurable function of the current shock vector has the same conditional expectation given G and F, and each shock coordinate has conditional mean zero given F. If prices are bounded, then each shock coordinate has conditional mean zero given G, and each product P_i ε_i has conditional mean zero given F. This is the additional independence condition and its required consequence for the corrected version of BFW Theorem 3.1. No IID assumption on full observations or sub-Gaussian noise is used.
import Mathlib.MeasureTheory.Function.ConditionalExpectation.PullOut import Mathlib.Probability.ConditionalExpectation import Mathlib.MeasureTheory.Constructions.Pi open MeasureTheory ProbabilityTheory
namespace BFWFluidLimit
theorem conditionalIndependence_repairs_moments
{Ω : Type*} [mΩ : MeasurableSpace Ω] {μ : Measure Ω} [IsFiniteMeasure μ]
{N : ℕ} (F : MeasurableSpace Ω) (P E : Ω → Fin N → ℝ)
(hG : F ⊔ MeasurableSpace.comap P inferInstance ≤ mΩ)
(hind : ∀ φ : (Fin N → ℝ) → ℝ, Measurable φ → Integrable (φ ∘ E) μ →
μ[φ ∘ E | F ⊔ MeasurableSpace.comap P inferInstance] =ᵐ[μ] μ[φ ∘ E | F])
(hE : ∀ i, Integrable (fun ω => E ω i) μ)
(hzero : ∀ i, μ[(fun ω => E ω i) | F] =ᵐ[μ] 0)
(C : ℝ) (hbound : ∀ᵐ ω ∂μ, ∀ i, ‖P ω i‖ ≤ C) :
∀ i, (μ[(fun ω => E ω i) | F ⊔ MeasurableSpace.comap P inferInstance] =ᵐ[μ] 0) ∧
(μ[(fun ω => P ω i * E ω i) | F] =ᵐ[μ] 0) := by sorry
end BFWFluidLimit
Read-back
What the Lean code literally says, in plain math · gpt-6
Conditional expectations of every integrable measurable function of the full shock vector are unchanged by adjoining current prices. Together with integrable coordinates and past-conditioned zero means, this implies coordinate meanzero with current prices and past-conditioned meanzero of own-price times own-shock. Prices are bounded, and enlarged information lies below the ambient sigma-algebra. No convergence conclusion or IID full observations is assumed.