Changes

The following model is essential a two principal (the "Interests"), single agent (the "Executive"), common knowledge agency model. It is a direct application of the Bernheim and Whinston (1986) model, as implemented by Grossman and Helpman (1994).

:<math>U_j=u_j(x)-c_j(x) \quad</math> where <math>u_j(x)=-\beta_j(x-z_j)^2\;\,\beta_j>0</math>

The executive choses a policy <math>x \in \mathbb{R}\,</math> and has an additively seperable utility function:

The status quo policy is taken to be <math>x>0\,</math> and the sequence of the game is a simultaneous move on support schedules by the principals followed by a choice of policy by the executive. The principals make menu offers, that is they state a resource contribution for each potential policy outcome, and these offers are binding. Once the executive has made the policy choice the contributions are transfered from the principals according the menu value of the chosen policy.

It should be noted that as a result of additive seperability in the utility functions with respect to the contributions, and that both principal(s) and agent value the seperable contribution identically, the contributions are transfers and the agent will maximize the joint surplus - this is proved below. Having two principals introduces competition which favors the agent in an enter/don't enter prisoner's dilemma game that occurs before this game and allows the agent to extract rents from both principals; however, we could correctly determing the outcome of this game by using a single representative principal whose utility function is the sum of the two principals, and then computing a standard principal-agent model.

Bernheim and Whinston (1984) provide four necessary and sufficient conditions for a sub-game perfect Nash equilibrium in this model. In the notion of the model, these are:

:c) <math>\forall j (i \ne j) x^*\,</math>maximizes <math> \quad \{u_e(x) + c_j^*(x) + c_i^*(x)\} + \{u_j(x) - c_j^*(x)\} = \{u_e(x) + u_j(x) + c_i^*(x)\} </math>

:d) <math>\forall j (i \ne j) \exists x_j\,</math>s.t.<math>\quad u_e(x)+c_j^*(x)+c_i^*(x)\;</math> where <math>c_J^*(x) = 0 \quad \therefore x_j\,</math> maxes <math>\quad u_e(x) + c_i^*(x)\;</math>

The first order conditions of (b) and (c) taken together imply:

:<math>c_J^*\prime(x^*) = u_J^*\prime(x^*)\quad</math>and therefore that the contribution schedules are locally truthful around <math>x^*</math>.

The contribution schedules are constructed as linear functions of the utility of the principals, specifically:

:<math>c_J^\tau(x,\tau_j) = max\{0,u_j(x) + \tau_j\}\quad</math> Note that the linear term is added not subtracted, but that it may be negative.

The principal's utility maximization and conditions (a) and (b) imply:

And so the we have both that (as <math>U_e^\hat(x) = u_e(x) + u_g(x) + u_h(x) \quad</math>) the principal's problem is to maximize the joint surplus and that the contribution schedules are truthful.

Using (d) we can note that if player <math>i</math> doesn't contribute then the agent choses: