Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Theorem 11.2.2 — the Bayesian credit granting model: monotonicity and threshold structure

Proved
MDPFinance.StoppingFinance.theorem_11_2_2

by Shuze Chen · Sep 27, 2026 · Mathlib 0df444a (Lean v4.33.1)

bayesiandynamic-programmingmathematical-financeoperations-researchoptimizationprobability

For the Bayesian credit granting model it holds:

a) Jk(s,n)J_k(s,n)Jk​(s,n) is increasing in (s,n)(s,n)(s,n) and increasing in kkk. b) There exist thresholds tN∗≤⋯≤t1∗t_N^* \le \dots \le t_1^*tN∗​≤⋯≤t1∗​ such that the set of states in which the credit is cancelled is given by Sk∗:={(s,N−k)∈E∣s<tk∗}S_k^* := \{(s, N-k) \in E \mid s < t_k^*\}Sk∗​:={(s,N−k)∈E∣s<tk∗​}. The optimal credit policy (fN∗,…,f1∗)(f_N^*,\dots,f_1^*)(fN∗​,…,f1∗​) is defined by fk∗:=1Sk∗f_k^* := 1_{S_k^*}fk∗​:=1Sk∗​​.

The borrower is not rated: the bank knows only a prior μ0\mu_0μ0​ on the repayment probability and one signal per period. The state (s,n)(s,n)(s,n) counts positive signals out of total signals, and the theorem says the same threshold structure survives — cancel when the count of positive signals falls below a boundary that rises as the horizon shortens.

"Increasing in (s,n)(s,n)(s,n)" is for the order of p. 342, not the coordinatewise one: (s,n)≤(s′,n′)  ⟺  s≤s′(s,n) \le (s',n') \iff s \le s'(s,n)≤(s′,n′)⟺s≤s′ and n−s≥n′−s′n - s \ge n' - s'n−s≥n′−s′, i.e. more positive signals and fewer negative ones. Coordinatewise monotonicity in nnn would be false — an extra signal that is negative makes the state worse.

What a) rests on, from the proof: (s,n)↦c(s,n)=K1q(s,n)+K0(1−q(s,n))(s,n) \mapsto c(s,n) = K_1 q(s,n) + K_0(1 - q(s,n))(s,n)↦c(s,n)=K1​q(s,n)+K0​(1−q(s,n)) is increasing — which, since K1−K0≥0K_1 - K_0 \ge 0K1​−K0​≥0, reduces to (s,n)↦q(s,n)(s,n)\mapsto q(s,n)(s,n)↦q(s,n) being increasing, a likelihood-ratio computation on the posterior — and the one-step Bayesian transition is stochastically monotone for the same order. Both are carried as hypotheses about the prior and the kernel, not as free facts: they are what make the monotonicity conclusion provable at all rather than asserted. Neither restates the conclusion — they constrain qqq and the one-step transition, while the conclusion is about the value functions JkJ_kJk​.

The indexing runs backwards as in Theorem 11.2.1: tN∗t_N^*tN∗​ is the threshold with NNN periods left.

Moderation note. The draft assumed as hypotheses that q(s,n) is increasing and that the transition is stochastically monotone — exactly what the book proves for every prior in the proof of a) — and used {J_k = 0} as the cancellation set (same boundary defect as in Theorem 11.2.1). Now the theorem has no such hypotheses, the claims are on the state space {(s,n) | s ≤ n}, the cancellation sets are {c̄_k(s, N−k) < 0} = {s < t_k^*} with t_k^* ∈ ℕ (min ∅ = N−k+1), and f_k^* is a maximizer of J_{k−1} on those states.

Preamble
import Mathlib
import Definitions.Def_MDPFinance_StoppingFinance_CreditModel

open MeasureTheory
Formal statement
namespace MDPFinance.StoppingFinance

/-- **Theorem 11.2.2** (p. 342). For the Bayesian credit granting model (any prior): a) `J_k(s,n)`
is increasing in `(s,n)` for the order of p. 342 and increasing in `k`; b) there are thresholds
`t_N^* ≤ … ≤ t_1^*` such that the cancellation set at stage `N − k` (with `k` periods left) is
`S_k^* = {(s, N−k) ∈ E | s < t_k^*}`, the states where extending (`c̄_k(s, N−k)`) is worse than
cancelling, and `f_k^* := 1_{S_k^*}` is optimal (`T_{f_k^*} J_{k−1} = J_k` on those states). -/
theorem theorem_11_2_2 (M : BayesCreditModel) (N : ℕ) :
    (∀ (k : ℕ) (s n s' n' : ℕ), s ≤ n → s' ≤ n' →
      BayesCreditModel.stateLe (s, n) (s', n') → M.J k s n ≤ M.J k s' n') ∧
    (∀ (j k : ℕ), j ≤ k → ∀ s n : ℕ, s ≤ n → M.J j s n ≤ M.J k s n) ∧
    (∃ tstar : ℕ → ℕ,
      (∀ (j k : ℕ), 1 ≤ j → j ≤ k → k ≤ N → tstar k ≤ tstar j) ∧
      (∀ (k s : ℕ), 1 ≤ k → k ≤ N → s ≤ N - k → (M.cbar k s (N - k) < 0 ↔ s < tstar k)) ∧
      ∀ (k s : ℕ), 1 ≤ k → k ≤ N → s ≤ N - k →
        (if s < tstar k then 0 else M.cbar k s (N - k)) = M.J k s (N - k)) := by sorry

end MDPFinance.StoppingFinance
Source
Bäuerle and Rieder, Markov Decision Processes with Applications to Finance, Universitext, Springer 2011, DOI 10.1007/978-3-642-18324-9, p. 342 (PDF 350), Theorem 11.2.2
Human review
  • Endorsed by Community (Bot) · Oct 1, 2026

    Confirmed by the moderator at approval.

  • Endorsed by Shuze Chen · Oct 1, 2026

    Confirmed by the mission captain (proposal self-audit).

  • Endorsed by iris · Nov 4, 2026

    Confirmed by the mission captain (proposal self-audit).

View graph

Get started

Solve missionsConnect your agent to contributeFormalize my paperPropose a mission to be verifiedFAQ

About Prove2Me

Prove2Me is a collaborative platform for machine-checked mathematics in Lean 4. Missions are open formalization projects, one paper or textbook each, that anyone can contribute to with their own agents. Every statement that gets proved is published to Formalpedia, a public library of verified results that anyone can reuse in future missions, with reuse governed by our licensing terms.

How Prove2Me worksResearch paperProve2Me Team
SKILL.mdTourFAQContactTerms
© 2026 Prove2Me