Elena PANOVA will defend her HDR on Thursday 20 November at 09:30, by ZOOM
"Five Applications of Game Theory to Problems Involving Differentiated Agents"
To attend the conference, please contact the secretariat Christelle Fotso Tatchum.
Memberships are:
- Thomas Mariotti: Senior Researcher, CNRS/TSE-R
- Andrew Rhodes: Professor of Economics, GE-TSE, Université Toulouse Capitole
- Enriquetta Aragones: Professor of Economics, Institut d'Anàlisi Econòmica (IAE) Rapporteure
- Annick Laruelle: Professor of Economics, Universidad del Pais Vasco Rapporteure
- Daniel F Garrett: Professor of Economics, University of Essex Examinateur
Abstract
This HDR application is based on five research papers that use game theory to analyze applied issues across different domains, ranging from political economy (continuing the research line initiated during my PhD studies) to network formation (which entered my research agenda at a later stage), and to the regulation of network operators, including cost sharing and incentive regulation (a topic that emerged from collaborations with private partners of the Toulouse School of Economics during 2019-2023). The common thread unifying these papers lies in their applied nature and, from a technical perspective, in the presence of heterogeneity among the participating agents. These papers are outlined below, in reverse chronological order.
A. Regulation of network operators.
1.1. Regulating investments when both costs and need are private
A critical issue for the regulation of public utilities is incentive provision for large investments. Part of the difficulty for regulation is that the eventual usefulness of these investments can be highly uncertain. An important example of this kind of scenario is the regulation of investments by an electricity transmission network driven by the anticipated transition to clean energies. In our research paper “Regulating investments when both costs and need are private” joint with Daniel F. Garret (revised and resubmitted for publication in Games and Economic Behavior) we describe the optimal regulation of large-scale infrastructure investments carried out in settings where their eventual usefulness or importance is difficult to predict. The agent undertaking the investment has superior information on two dimensions: the cost of investment and the likelihood it is useful or beneficial to the principal. Out of concern for the firm's ability to finance the investments, the regulation aims at ensuring non-negative profits in any period. We characterize the optimal screening mechanism and show that it depends heavily on the degree of asymmetric information about usefulness. As long as the variation in probability of usefulness is relatively low, the optimal regulation qualitatively resembles that in a setting with uncertainty only about cost. When the asymmetry of information about usefulness is severe, the optimal mechanism can feature upward distortions in investment and rent for all agent types. While our findings are arguably pessimistic about the nature of optimal regulation of investment projects with high uncertainty about their eventual usefulness or importance, we see them as a benchmark against which regulatory policy might aim to improve.
1.2. Sharing cost of network among heterogenous users
The problem of sharing common cost is relevant in many economic settings. Axiomatic approach to cost-sharing problems proposes cost-sharing rules characterised by a set of desirable (problem-specific) features. In my paper “Sharing cost of network among users with differentiated willingness to pay” published in Games and Economic Behavior (2023), I take axiomatic approach to propose a rule sharing the cost of fixed uncongested distribution network among users with differentiated willingness to pay. This problem is relevant, for example, in the gas distribution industry. The cost of gas distribution network is supposed to be paid from the regulated market revenues. Some consumers possess relatively high willingness to pay, such as industrial consumers or large buildings that heat with gas; other consumers, such as households, possess only a modest willingness to pay. How shall differences in consumers' locations and willingness to pay be reflected in their bills? In order to address this question, I analyse a model in which agents differentiated by their locations and their willingness to pay for the good, are connected to the source by a fixed uncongested tree-network. I show that the associated value-sharing problem is convex, which motivates my first axiom requiring that the solution lie in the core. This requirement, together with two additional axioms (the “classic” split- and merge-proofness, and a normative axiom specific to my game, which I term no spatial discrimination) characterize a computationally simple solution based on the idea of proportionality, which may serve a basis for elaboration of regulated gas tariffs.
B. Clustering in social networks with different minded participants
Social networks contribute to diffusion of information and behaviours. Growing evidence suggests that the speed and distance of diffusion through the network depend on its features. One key characteristic of social networks is clustering: the tendency of people's contacts to also be connected to each other, forming tightly knit groups. While previous studies have shown that this aspect of networks helps explain various phenomena, the reason behind its prevalence and impact remain largely unexplored. In our joint paper with Thibault Laurent entitled “Clustering in communication networks with different minded participants published in Social Choice and Welfare (2025), we examine how network structure in communication networks influences learning and social welfare when participants possess different prior opinions and face uncertainty about the external state. We consider a game in which the players have subjective imperfectly correlated priors on some relevant state of nature. They build network links, receive private signals on the state and truthfully announce their posterior expectations to their network neighbours in two successive rounds. A player’s disutility is measured by subjective posterior variance (his remaining uncertainty). We achieve tractability through two simplifying assumptions (imperfect knowledge of a distant network and an arbitrarily small correlation of priors). We provide a closed-form expression relating a player’s payoff to the network architecture, characterize the efficient network and show that it constitutes an equilibrium. We show numerically that our results may hold when both assumptions are relaxed, that is, when the players know their local and distant network and the correlation of their priors takes higher (but not too high) values. Furthermore, Rawlsian efficiency criterion may be replaced with (more common) utilitarian efficiency criterion. Our findings are likely to be relevant in contexts such as learning about the relative merits of different public policies or the adoption of innovations (e.g., technology, drugs, or products). In these scenarios, differences in prior beliefs can create barriers to learning, and clustering helps mitigate these barriers. This effect offers a potential explanation for the widespread occurrence of clustering in real-world social networks. world social networks.
C. Political campaigns and voting
During my PhD studies and the following years my research has focused on the field of political economy. Two papers outlined below represent this research line. The first paper seeks to provide a better understanding of voting behavior and electoral outcomes. A key argument for voting is its potential to choose public policies that are aligned with common interests on the basis of information gathered from different voters. The information-aggregation property of voting is established in settings where each voter behaves so as to improve the outcome, conditional on being pivotal. These objectives are called instrumental. A voter with instrumental objectives votes if he has relevant private (and not public) information and abstains otherwise (because he is more likely to spoil, rather than improve, the voting decision by better-informed voters). While there is some experimental evidence that voter information increases the turnout, voting with inferior information is quite common. Furthermore, the vote choices are sensitive to public information, such as campaign information and results of public opinion polls. These observations suggest that voter motivation is not purely instrumental. Several studies propose that voters derive utility from the very act of voting, termed expressive motivation, formalizing it in different ways (either as a benefit from voting in a certain way or as a payoff depending on the vote choices by the other voters). These studies have been criticized for the lack of robust predictions regarding the effect of expressive motivation on voting behaviour and outcomes. My paper entitled “A Passion for Voting” published in Games and Economic Behavior (2015) proposes a framework to analyse this effect. My modelling approach is motivated by the observed effect of habitual voting which has not been previously modeled by either instrumental- or expressive-voting theories: participation in one election increases the propensity to vote in the next election. I consider a voting game with successive elections. A continuum of voters chooses public policy using a simple majority rule with common objective to match their policy choice with a hidden state of nature. A minority of voters receive informative signals on the state; others remain ignorant receiving signals uncorrelated with the state. A voter pays some (time) cost for voting and receives the benefit which is equal to his confidence in his vote choice. He can abstain if he wishes. He has selective memory for actions: while initially he knows the quality of his private signal, by the next election he remembers only his past voting behaviour and the majority outcome. If the game was static or if voter memory was perfect, the ignorant voters would abstain, letting the informed voters select the outcome matching the state. However, in the repeated voting game an ignorant voter would like to deviate and vote: if by chance he pools with the winners, he gains confidence in his future vote choice allowing him to receive an expressive benefit from voting. I characterize the unique symmetric perfect Bayesian equilibrium and show that it accommodates relevant patterns: voting with inferior information, impact of public information on vote choices (with bandwagon and underdog effects) and habitual voting. I show that when the public signal favoring a given policy or electoral alternative is sufficiently strong, expressive voting may distort electoral outcomes, which is a possible reason for the observed persistence of public policies and high re-election rates. The second paper entitled “Partially Revealing Campaign Promises” published in the Journal of Public Economic Theory (2017) asks why incumbent politicians tend to fulfil their electoral promises while there are no legal restraints preventing them from breaking these promises. It models campaign promises as pure cheap talk. A candidate’s campaign promises are a partially revealing signal of her policy preference type. The incumbent’s policy choice is yet another signal of her type. Policy choice is a costly signal (unlike campaign promises). The incumbent tends to keep her electoral promises in order to preserve ambiguity about her type, which is necessary to assemble a winning majority for re-election. She keeps her promises regardless of information about the efficiency of different public policies which she receives upon taking office. Therefore, campaign promises generate inefficiencies in public policy.
