Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

(A.2) for a subunit j≠ij \ne ij=i: expected payoff under β∗/βi\beta^*/\beta_iβ∗/βi​ equals the expected conditional expectation

Proved
IncentivesInTeams.Conglomerate.appendix_A2_subunit

by mikedeng1 · Sep 26, 2026 · Mathlib 0df444a (Lean v4.33.1)

conditional-expectationindependencep2o-batch-p100ap2o-gran-per-chapterp2o-plan-paperp2o-v1team-theory

Consider a conglomerate model whose component weights are probability weights, a joint strategy β∗\beta^*β∗, a subunit iii, a strategy βi∈Bi\beta_i \in B_iβi​∈Bi​, and a subunit j≠ij \ne ij=i. Let y^\hat yy^​ denote the information functions when β∗/βi\beta^*/\beta_iβ∗/βi​ is played. Then

E[vj[δj∗(y^j(s)),δ0∗(y^0(s));sj]]=E[hj(y^0(s))],E\big[v_j[\delta_j^*(\hat y_j(s)), \delta_0^*(\hat y_0(s)); s_j]\big] = E\big[h_j(\hat y_0(s))\big],E[vj​[δj∗​(y^​j​(s)),δ0∗​(y^​0​(s));sj​]]=E[hj​(y^​0​(s))],

where hj(y0)h_j(y_0)hj​(y0​) is subunit jjj's factor of the head's conditional expectation under β∗\beta^*β∗ (model file): the average of vj[δj∗(ζj∗(t),γ0j∗(z0)),δ0∗(y0);t]v_j[\delta_j^*(\zeta_j^*(t), \gamma_0^{j*}(z_0)), \delta_0^*(y_0); t]vj​[δj∗​(ζj∗​(t),γ0j∗​(z0​)),δ0∗​(y0​);t] over the states ttt of SjS_jSj​ whose message under β∗\beta^*β∗ is the one recorded in y0=(z0,m)y_0 = (z_0, m)y0​=(z0​,m).

This is equation (A.2) of the paper for a subunit j≠ij \ne ij=i, with its right-hand side E[E{vj[δj∗(yj∗),δ0∗(y0∗);sj]∣y0∗(s)=y^0}]E[E\{v_j[\delta_j^*(y_j^*), \delta_0^*(y_0^*); s_j] \mid y_0^*(s) = \hat y_0\}]E[E{vj​[δj∗​(yj∗​),δ0∗​(y0∗​);sj​]∣y0∗​(s)=y^​0​}] in the factorized form. Together with the head's analogue it gives (A.1).

Formalization Note. Under β∗/βi\beta^*/\beta_iβ∗/βi​ the strategy of subunit j≠ij \ne ij=i and the head's strategy are those of β∗\beta^*β∗, so the left-hand side is the jjj-th payoff component under β∗/βi\beta^*/\beta_iβ∗/βi​. Information follows the one-exchange protocol of the model file.

Preamble
import Mathlib
import Definitions.Def_IncentivesInTeams_Conglomerate_Model
Formal statement
namespace IncentivesInTeams.Conglomerate

theorem appendix_A2_subunit {ι : Type*} [Fintype ι] [DecidableEq ι] {S₀ : Type*} [Fintype S₀] {S : ι → Type*}
    [∀ i, Fintype (S i)] {Z₀ : Type*} {Z M₀ M : ι → Type*} {D₀ : Type*} {D : ι → Type*}
    (T : Model S₀ S Z₀ Z M₀ M D₀ D) (hT : T.WeightsOK)
    (βs : JointStrategy S₀ S Z₀ Z M₀ M D₀ D) (i : ι)
    (b : SubStrategy (S i) (Z i) (M₀ i) (M i) (D i)) (hb : b ∈ T.B i) (j : ι) (hji : j ≠ i) :
    T.expect (T.subPayoff (βs.update i b) j) =
      T.expect (fun s => T.subFactor βs j ((βs.update i b).headInfo s)) := by sorry

end IncentivesInTeams.Conglomerate
Source
Groves, Incentives in Teams, Econometrica 41(4), 1973, p. 630, Appendix, display (A.2) and the l.h.s. (A.2) chain
Read-back

What the Lean code literally says, in plain math · claude-opus-5-5

Setting. ι\iotaι is a finite set of subunits. S0S_0S0​ and each SkS_kSk​ are finite. The model's weights P0P_0P0​ and PkP_kPk​ are assumed to be probability weights, with product P(s)=P0(s0)∏kPk(sk)P(s)=P_0(s_0)\prod_kP_k(s_k)P(s)=P0​(s0​)∏k​Pk​(sk​) and E[X]=∑sP(s)X(s)\mathbb E[X]=\sum_sP(s)X(s)E[X]=∑s​P(s)X(s). The payoffs are v0:D0×S0→Rv_0:D_0\times S_0\to\mathbb Rv0​:D0​×S0​→R and vk:Dk×D0×Sk→Rv_k:D_k\times D_0\times S_k\to\mathbb Rvk​:Dk​×D0​×Sk​→R.

A joint strategy β\betaβ has head functions (ζ0,(γ0k)k,δ0)(\zeta_0,(\gamma_0^k)_k,\delta_0)(ζ0​,(γ0k​)k​,δ0​) and subunit functions (ζk,γk,δk)(\zeta_k,\gamma_k,\delta_k)(ζk​,γk​,δk​), as follows:

  • ζ0:S0→Z0\zeta_0:S_0\to Z_0ζ0​:S0​→Z0​, γ0k:Z0→M0k\gamma_0^k:Z_0\to M_{0k}γ0k​:Z0​→M0k​, δ0:Z0×∏M→D0\delta_0:Z_0\times\prod M\to D_0δ0​:Z0​×∏M→D0​;
  • ζk:Sk→Zk\zeta_k:S_k\to Z_kζk​:Sk​→Zk​;
  • γk\gamma_kγk​ and δk\delta_kδk​ act on Zk×M0kZ_k\times M_{0k}Zk​×M0k​.

Its information maps are ykβ(s)=(ζk(sk),γ0k(ζ0(s0)))y_k^\beta(s)=(\zeta_k(s_k),\gamma_0^k(\zeta_0(s_0)))ykβ​(s)=(ζk​(sk​),γ0k​(ζ0​(s0​))) and y0β(s)=(ζ0(s0),(γk(ykβ(s)))k)y_0^\beta(s)=(\zeta_0(s_0),(\gamma_k(y_k^\beta(s)))_k)y0β​(s)=(ζ0​(s0​),(γk​(ykβ​(s)))k​). Subunit jjj's payoff under β\betaβ is

ujβ(s)=vj(δj(yjβ(s)),δ0(y0β(s)),sj).u_j^\beta(s)=v_j(\delta_j(y_j^\beta(s)),\delta_0(y_0^\beta(s)),s_j).ujβ​(s)=vj​(δj​(yjβ​(s)),δ0​(y0β​(s)),sj​).

For fixed β∗\beta^*β∗ and yˉ=(z0,m)\bar y=(z_0,m)yˉ​=(z0​,m), the factor Fj(yˉ)F_j(\bar y)Fj​(yˉ​) is

Fj(yˉ)=cavg⁡Pj({t∈Sj:γj∗(ζj∗(t),γ0j∗(z0))=mj}, t↦vj(δj∗(ζj∗(t),γ0j∗(z0)),δ0∗(yˉ),t)).F_j(\bar y)=\operatorname{cavg}_{P_j}\Big(\{t\in S_j:\gamma^*_j(\zeta^*_j(t),\gamma_0^{j*}(z_0))=m_j\},\ t\mapsto v_j\big(\delta^*_j(\zeta^*_j(t),\gamma_0^{j*}(z_0)),\delta^*_0(\bar y),t\big)\Big).Fj​(yˉ​)=cavgPj​​({t∈Sj​:γj∗​(ζj∗​(t),γ0j∗​(z0​))=mj​}, t↦vj​(δj∗​(ζj∗​(t),γ0j∗​(z0​)),δ0∗​(yˉ​),t)).

Here cavg⁡w(E,X)=∑EwX/∑Ew\operatorname{cavg}_w(E,X)=\sum_EwX/\sum_Ewcavgw​(E,X)=∑E​wX/∑E​w, with value 000 when the denominator is 000.

Theorem. Take a joint strategy β∗\beta^*β∗, a subunit iii, a subunit-iii strategy b∈Bib\in B_ib∈Bi​, and a subunit j≠ij\ne ij=i. Let β∗/b\beta^*/bβ∗/b be β∗\beta^*β∗ with subunit iii's strategy replaced by bbb. Then

E[ujβ∗/b]=E[s↦Fj(y0β∗/b(s))].\mathbb E\big[u_j^{\beta^*/b}\big]=\mathbb E\Big[s\mapsto F_j\big(y_0^{\beta^*/b}(s)\big)\Big].E[ujβ∗/b​]=E[s↦Fj​(y0β∗/b​(s))].

In words: FjF_jFj​ is built from β∗\beta^*β∗ but evaluated at the head information produced under β∗/b\beta^*/bβ∗/b.

β∗\beta^*β∗ is not required to lie in any strategy set, and no positivity of the conditioning events is assumed. Where such an event has weight 000, FjF_jFj​ takes the value 000.

Degenerate cases. The weight hypothesis forces all state sets to be nonempty. If ι\iotaι has exactly one element, no j≠ij\ne ij=i exists and the statement is vacuous.

Human review
  • Endorsed by Shuze Chen · Sep 27, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Sep 27, 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