Locally robust implementation and its limits (Q694745): Difference between revisions

From MaRDI portal
Import240304020342 (talk | contribs)
Set profile property.
Set OpenAlex properties.
Property / full work available at URL
 
Property / full work available at URL: https://doi.org/10.1016/j.jet.2012.05.012 / rank
 
Normal rank
Property / OpenAlex ID
 
Property / OpenAlex ID: W2132105654 / rank
 
Normal rank

Revision as of 02:09, 20 March 2024

scientific article
Language Label Description Also known as
English
Locally robust implementation and its limits
scientific article

    Statements

    Locally robust implementation and its limits (English)
    0 references
    0 references
    0 references
    0 references
    13 December 2012
    0 references
    The authors study a notion of locally robust implementation that captures the idea that the planner may know agents' beliefs well, but not perfectly. The concept of locally robust implementation introduced is weaker than ex-post implementation, yet the authors show that no regular allocation function is locally robust implementable in generic settings with quasi-linear utility, interdependent and bilinear values, and multi-dimensional payoff types. This result is an extension of a result of \textit{P.\ Jehiel} et al. [Econometrica 74, No. 3, 585--610 (2006; Zbl 1127.91046)]. The authors work with the following setup. There are two social allocations \(x \in \{0,1\}\) and there are two agents \(i\in \{1,-1\}\) with payoff types \(\theta_{i}\) drawn from \(d_i\)-dimensional cubes \(\Theta_i = [0,1]^{d_i}\) . Agents have quasi-linear Bernoulli utility functions of the form \(u_i = x v_{i} (\theta_i,\theta_{-i})-p_i\), where \(p_i\) is a monetary payment by agent \(i\) and \(v_i\) is \(i\)'s smooth interdependent value function. Let \(\Delta(X)\) denote the space of probability measures on \(X\). Baseline beliefs are given by continuous functions \(\pi_{i}^{\star} : \Theta_i \rightarrow \Delta(\Theta_{-i})\), where \(\Theta_i\) is equipped with the sup-norm and \(\Delta(\Theta_{-i})\) with the metric of absolute variation. The planner's local uncertainty about agents' beliefs is modeled by assuming that there is \(\varepsilon > 0\) such that agent \(i\)'s type space \(T_i \subset \Theta_i \times \Delta(\Theta_{-i})\) satisfies \(\theta_i \times B_{\varepsilon} (\pi_{i}^{\star} (\theta_i)) \subset T_i\) for all \(\theta_i\), i.e.\ \(T_{i}\) includes all \(\varepsilon\)-perturbed beliefs. A \(\pi_{i} \in \Delta(\Theta_{-i})\) is interpreted as a belief over \(T_{-i}\) with marginal \(\pi_{i}\) over \(\Theta_{-i}\) such that \[ \pi_{i}\{(\theta_{-i},\pi_{-i}^{\star}(\theta_{-i})): \theta_{-i} \in \Theta_{-i}\} = 1; \] i.e., agent \(i\) could have different beliefs about \(-i\)'s payoff types but \(i\) believes with probability one that \(-i\)'s beliefs are specified by \(\pi_{-i}^{\star}\). Set \(\Theta = \Theta_i \times \Theta_{-i}\). A (possibly stochastic) allocation function \(q : \Theta \rightarrow [0,1]\) is locally robust implementable if there exist type spaces \(T_i\) as above and a (possibly belief-dependent) payment function \((p_{i},p_{-i}) : T_i \times T_{-i} \rightarrow {\mathbb R}^2\) such that the direct revelation mechanism \((q, (p_{i},p_{-i}))\) is incentive compatible on \(T_i \times T_{-i}\), i.e.\ \[ \mathbb{E}_{\pi_{i}} [v_i(\theta) q (\theta) - p_i (t)] \geq \mathbb{E}_{\pi_{i}} [v_i(\theta) q (\theta') - p_i (t')] \] for all \(t_i = (\theta_i,\pi_i)\) and \(t_i' =(\theta_i',\pi_i')\), where \(\theta = (\theta_i, \theta_{-i})\), \(\theta' = (\theta_i', \theta_{-i})\), \(t = (\theta_i,\pi_i,\theta_{-i},\pi_{-i})\), \(t' = (\theta_i',\pi_i',\theta_{-i},\pi_{-i})\). A stochastic allocation function \(q : \Theta \rightarrow [0,1]\) is regular if it is smooth with \(\nabla_{i}q(\theta) \neq 0\) for all \(i\) and \(\theta\). A deterministic allocation function \(q : \Theta \rightarrow \{0,1\}\) is regular if it maximizes a smooth, non-satiated objective function, i.e.\ there exists \(\psi : \Theta \rightarrow \mathbb{R}\) satisfying \(\nabla_{i}\psi(\theta) \neq 0\) for all \(i\) and \(\theta\) and having an interior \(\theta \in \Theta\) such that \(\psi(\theta) = 0\), such that \(q(\theta) \in \arg \max_{x \in \{0,1\}}x \psi(\theta)\). The main result of the paper is that if \(d_{i} \geq 2\), for generic bilinear value functions, no regular allocation function is locally robust implementable; being generic means that in a canonical representation of the bilinear value functions \(v_{i}\) certain matrices satisfy some linear independence assumptions. The authors interpret this negative result as in many payoff environments even local robustness is too demanding when applied to social choice functions. The authors also construct a regular stochastic allocation function \(q\) which is locally robust implementable but not ex-post implementable.
    0 references
    locally robust implementation
    0 references
    ex-post implementation
    0 references
    regular allocation function
    0 references
    quasi-linear Bernoulli utility function
    0 references
    bilinear value function
    0 references

    Identifiers

    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references