Theorem 9.4.2 — the optimal execution rate is monotone and unit-Lipschitz in the inventory
ProvedMDPFinance.JumpMarkets.theorem_9_4_2Theorem 9.4.2 (p. 296).
a) The value function is the unique fixed point of in . b) There exists an optimal Markov strategy such that and satisfies , and is the corresponding number of share process.
The structural answer to the execution problem: sell more when you hold more, but never more than one extra share per extra share held. Both inequalities are the theorem.
alone would be a generic monotonicity claim. It is the upper bound that carries the content — the optimal block size is unit-Lipschitz in the inventory, so the execution schedule cannot jump. Dropping it loses half the theorem, and it is the half that takes the work: the book proves it by contradiction from the strict convexity (9.19) of .
, not . This is a minimisation with a convex cost and a convex value function, the opposite curvature from §9.3's terminal wealth problem; the two sets are not interchangeable.
is the smallest minimizer of over (9.22), the book's own choice; with an arbitrary selection of minimizers the unit-Lipschitz bound need not hold.
Optimality is stated at the level of the discrete-time model, where lives:
minimizes the operator's argument at the value function, so the stationary policy is optimal for
the discrete-time model, and — by Theorem 8.3.2, which p. 295 invokes and which belongs to chunk
08 — that determines an optimal control process for the continuous-time problem. The inner
optimisation is over a finite set, so no sSup/sInf junk-value device is needed here,
unlike in §9.3.
Moderation note. The draft assumed V to be some fixed point of 𝒯 in IM_cx and concluded its uniqueness, never connecting it to the value function; b)'s optimal strategy was absent (only the minimizer property of f^*). Now a) is: the value function V of the embedded model lies in IM_cx, is a fixed point of 𝒯, and is the only one there; b) f^* = f^*_V is monotone and unit-Lipschitz in x, and there is an admissible measurable policy f(t,x)(s) = f^*(t+s,x) attaining V (the strategy π^*_t = f^*(t, X_{t-})).
import Mathlib import Definitions.Def_MDPFinance_JumpMarkets_TradeExecution open MeasureTheory open scoped ENNReal
namespace MDPFinance.JumpMarkets
/-- **Theorem 9.4.2** (p. 296). a) The value function `V(t,x)` is the unique fixed point of
`𝒯` in `IM_cx`. b) There is an optimal Markov strategy `π^*_t = f^*(t, X_{t-})` — the policy
`f(t,x)(s) := f^*(t+s, x)` of the embedded model attains `V` — where `f^*` is the smallest
minimizer **(9.22)** at `V`, and `f^*(t,x) ≤ f^*(t,x+1) ≤ f^*(t,x) + 1`. -/
theorem theorem_9_4_2 (M : TradeExecution) :
∃ V : ℝ × ℕ → ℝ, M.IMcx V ∧
(∀ p ∈ M.E, M.Jinf p = ENNReal.ofReal (V p)) ∧
(∀ p ∈ M.E, M.T_op V p = V p) ∧
(∀ v : ℝ × ℕ → ℝ, M.IMcx v → (∀ p ∈ M.E, M.T_op v p = v p) → ∀ p ∈ M.E, v p = V p) ∧
(∀ t ∈ Set.Icc (0 : ℝ) M.T, ∀ x : ℕ,
M.fstar V (t, x) ≤ M.fstar V (t, x + 1) ∧
M.fstar V (t, x + 1) ≤ M.fstar V (t, x) + 1) ∧
(∃ f : ℕ → ℝ × ℕ → Offer, TradeExecution.IsPolicy f ∧
(∀ n, ∀ p ∈ M.E, ∀ s : ℝ, 0 ≤ s → (f n p).val s = M.fstar V (p.1 + s, p.2)) ∧
∀ p ∈ M.E, M.Jinfpi f p = M.Jinf p) := by sorry
end MDPFinance.JumpMarkets
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.