Publications
Grabbing the Forbidden Fruit: Restriction-Sensitive Choice (with A. Ghersengorin) — Journal of Economic Behavior & Organization, 248, 107653, 2026
[Author version: March 2026]Revealed Deliberate Preference Changes (with S. Gleyze and A. Ghersengorin) — Games and Economic Behavior, 142, 357–367, 2023
Working Papers
Does Party Formation Solve Arrow’s Problem? Aggregation over an Endogenous Political Offer
[This version: August 2026] [Supplementary Appendix]Abstract
In Arrow’s theorem the options are given. In an election they are not: parties form, and citizens then rank them. Can party formation make preferences over parties consistent? I take a continuum of citizens in the plane and a mechanism that produces a finite offer of parties from each distribution. The answer turns on second choices: each duel between two parties is decided by the members of the others, according to which of the two they prefer. When parties sit at the mean of their members, these choices are free, and no anonymous, independent, non-imposed and non-null rule exists. For three parties whose producible vote shares form a convex set, a rule that ranks each pair strictly both ways on a range of shares exists if and only if a rule with one quota per pair does, so consistency alone yields a monotone rule. With four parties it does not: on shares that a fixed offer produces, such a rule exists but no monotone one. A mechanism escapes when it offers two parties, aligns them, or keeps them far apart, with exact ties broken by a suitable arbiter. Letting a party survive only when its existence changes the election screens the cycles under plurality without noise and three births, but mostly leaves one party. With three births, once every party has a chance of winning or a weight in the outcome, the impossibility returns. Parties solve Arrow’s problem only when they constrain the second choices of their members.
Latent Relevance and the Identification of Deliberate Preference Change — Submitted
[This version: November 2026]Abstract
A decision maker changes her preferences deliberately when she chooses which values to treat as relevant by maximizing a stable metapreference. Choices, however, identify relevant values only partially: a value can be relevant at a period without affecting any choice then, in which case it is latent. For data that can be rationalized with no value latent, we characterize those that leave no room for latent values: at every period, every value must be revealed relevant by that period’s choices, or revealed irrelevant by the requirement that a single way of weighing values fit every period. When the data leave no room for latent values, which is the case in particular when every value makes a difference on its own at some period, testing deliberation as if none were latent is without loss. Otherwise, latent values can turn an apparent reversal into a deliberate change. In a population, individuals can share a metapreference exactly when some relevance paths that fit their choices are each deliberate and form no cycle of conflicting moves, and two observations per individual can suffice to reject such a shared view.
Fear of Missing Out: Pragmatic Beliefs and the Reception of New Technologies (with F. Rocciolo)
[This version: October 2026]Abstract
When a new technology becomes available, people who use an older one often change their view of the future without having learned anything. We model this, with what people choose as the only data. A decision maker entertains several scenarios about the world and holds a baseline confidence in each. Facing the technologies on the table, one of which she uses, she gives more weight to the scenarios under which she would miss the most by standing pat: she fears missing out. She then maximizes expected utility. Three axioms on her choices, added to expected utility, characterize the model. The weak axiom of revealed preference holds across menus whose best payoffs, scenario by scenario, differ by a constant; translating a whole situation changes nothing; and a newcomer can move her choice among the old options only toward those that do well where it shines. Choices reveal her weights exactly, and an option that nobody chooses can reverse a choice. Evidence and stakes then enter her beliefs as two likelihoods that do not interact. A new technology is received through how far it pushes the frontier of what is already available, and radical technologies are favored. In a market, others’ adoption raises what each firm would miss by staying behind, and fear of missing out turns this payoff externality into a belief externality: adoption can tip, and rest on a confidence that no information supports; when the value of the technology lies in not falling behind, more adoption is worse for everyone, and a useless technology can be adopted by all. After the launch, holding the new technology turns the fear around: adopters fear having missed the scenario under which they should have stayed. Markets then churn and reverse their waves, unless switching costs make adopters loyal.
The Sound of Silence: Turnout Neutrality and the Choice of Groups (with N. Colin-Jaeger)
[This version: November 2026] [Earlier version, Generational Equity in Democracy: November 2025]Abstract
Abstainers are silent, and every voting rule must say what becomes of their share of the decision. Democracies promise one person, one vote, but they count one voter, one vote. Counting only the ballots is not neutral: it gives the same outcome as counting every abstainer like the average voter, so that a group that votes less weighs less. We study the alternative default, which hands each abstainer’s share to the voters of her own group, so that each group weighs its share of the population whatever its turnout. With a blank ballot that means let all voters decide, the two rules agree on what every ballot means and differ only on what silence means: one voter, one vote is the case in which every non-voter casts a blank ballot. The group default is also as far as a correction can go: when groups are united against each other, no rule that counts ballots equally when everyone votes, rewards nobody for staying home and ignores abstention spread evenly over all groups does more for the groups that vote less. We argue that the default is owed when turnout gaps are produced by the costs of voting and by a circle of neglect and abstention, and that, besides territory, only age meets the conditions for applying it today. In German federal elections, where turnout is high, it would have changed no party’s share by more than a quarter of a point; in British general elections, where the age gap in turnout is large, it would have moved one to two points from the Conservatives to Labour.
Fathers and Kings: Who Gains from the Family Metaphor?
[This version: October 2026]Abstract
Rulers have long called themselves fathers of their people. Who gains from the metaphor, and what does it do to families? In our model people judge husbands and rulers by one expectation about those in charge, right on average and wrong case by case, as in analogy-based expectation equilibrium. Any such analogy transfers trust from the authorities who behave better than their class to those who behave worse, and creates none. A ruler worse than fathers therefore gains from the family metaphor, and gains most where patriarchy is common, since fathers are then most of the authority people see. A ruler who chooses his analogy, which authority to resemble and how much, pools himself only with authorities better than he is, and they lose trust. Hence the metaphor he cultivates can discredit fathers but never launder them. It follows patriarchy where fathers deserve trust, and never sustains it where they do not. The metaphor sustains patriarchy in two cases only: when the ruler values patriarchy for itself, or when a tradition he cannot erase carries the analogy. In both, fathers do not deserve trust on their own, but the order around them does.
Impartial Resolution of Random Conflicts (with Y. Breitmoser)
[This version: July 2026] [Online Appendix]Abstract
A planner repeatedly chooses between two randomly drafted allocations, and whenever people disagree she must side with someone. We test welfare measures by a minimal requirement of impartiality: no one should win almost every conflict. Standard indices fail it. Since they ignore which proposals are drafted, a drafting process tailored to them can make one person a near-dictator. More strikingly, under a given process that treats people identically, the egalitarian-equivalent index lets the person who needs goods together more win almost every conflict under leximin once goods are many, and other rules only change who wins; under the same process, scoring each person by how her bundle ranks among the drafts splits the conflicts evenly. The freedom behind this is general: for any anonymous Paretian rule, some rescaling of unchanged preferences makes it side with a designated person almost always. We then derive the rankings that impartiality requirements single out. If a randomly drafted bundle, given to one person alone, must beat any benchmark equally often whoever receives it, a Paretian ranking that uses only frequencies is an anonymous rule applied to percentiles, each person’s share of drafts that she ranks no higher. Hammond equity selects their leximin, and consistency across drafting processes selects their sum, a continuous Borda rule that reward models trained on pairwise comparisons already compute. These rankings treat people alike in expectation under every rule; in actual decisions they do so only under a condition on preferences and proposals that no scale can create.
Consensus and Interpersonal Welfare Comparisons (with Y. Breitmoser)
[This version: November 2026] [Online Appendix] [Earlier version, Consensus on the Lesser Evil: November 2025]Abstract
What welfare comparisons follow when people with different preferences agree about physical resources? We study a rule restricted to bundles their recipients could receive in efficient divisions of a specified resource pool. If both people regard one bundle as at least as good as the other, its recipient must count as no worse off, and as better off if either preference is strict. We characterize all ways of assigning comparable welfare levels that preserve individual preferences and satisfy this rule. The two people’s indifference judgments give exact bounds on which bundles can represent equal welfare across them. Every continuous, order-preserving pairing of reference bundles within those bounds is possible. Although some comparisons remain open, the shared comparisons can identify a best facility assignment under symmetric, welfare-increasing social evaluation, without comparing welfare gains. Equal welfare at an efficient division where the bundles can be exchanged also excludes envy.
The Welfare Representation of Envy-Free Choice (with Y. Breitmoser)
[This version: September 2026] [Earlier version, Envy-free welfare maximization under risk and uncertainty: March 2026]Abstract
How can a planner compare people’s welfare so that the comparison respects envy freedom, the requirement that neither person prefer the other’s allocation? Conventional equivalents preserve individual preferences, but their reference assignments may give the wrong fairness judgment. We characterize fairness-compatible equivalence-based normalizations for ordered two-person efficient allocation paths with adequate reference coverage. The appropriate reference for assessing another person’s package is that recipient’s efficient assignment path; the resulting Contract Curve Equivalents reproduce both directions of envy. Different envy cutoffs generally require two directed pairs of indices, while relative-cost variation restricts their common scaling and, under further invariance conditions, identifies the physical references used. We also show why fixed own-welfare reports cannot generally support consistent efficient no-envy choice across problems. A compensated benefit reform may preserve all attainable own-utility profiles while changing or eliminating efficient envy-free allocations. Thus policy evaluation under no-envy requires information about how each person values the other’s package.
Description-dependent Choices (with D. Borie and D. Jullien) — Submitted
[This version: August 2025]Abstract
We propose a model of framing effects in which a choice reversal is due to differences in the descriptions of the same alternatives. Our contribution is to show under which conditions these framing effects can be seen as rational or not. Our results provide explanations that are aligned with existing empirical evidence not accounted for by other models and suggest experiments for further work.
Work in progress
Distributive Policy Choice with Incomplete Personal Linkage (with Y. Breitmoser) — draft, November 2026
Fair Insurance under Heterogeneous Moral Hazard (with Y. Breitmoser) — draft, November 2026
Efficiency and Envy under Payment Constraints (with Y. Breitmoser) — draft, November 2026
How Large Are Majority Cycles among Parties? — companion paper to Does Party Formation Solve Arrow’s Problem?
Political Metaphors in Parliamentary Debates — text-as-data project
Summary
Which images do members of parliament use when they talk about politics, from which source domains (body, war, journey, family, building, disease), to speak about which targets (the nation, the state, the economy, public services, opponents, the European Union), and how does this vary with the party, the position in government or opposition, the topic and the period? The corpus is the House of Commons 2015–2022 from ParlaMint (2.9 million sentences). The method is fixed in advance: metaphors are identified sentence by sentence by open-weight language models following the Metaphor Identification Procedure, sorted by five dictionaries into conventional and non-conventional ones, annotated by hand on a stratified sample by two annotators, and the classifier is validated against this human reference before it is applied to the whole corpus. Hypotheses on the images that government and opposition use are written before the corpus is classified. The project is the empirical counterpart of Fathers and Kings; other ParlaMint parliaments come next.
Cognitive congestion and the Rally Round the Flag effect (joint work with A. Pietri)
Abstract
We propose a cognitive micro-foundation for the “Rally Round the Flag”. Departing from traditional models of patriotic utility, we argue that international crises generate Cognitive Congestion: high-stakes security narratives capture the electorate’s limited attention, mechanically crowding out monitoring of the domestic economy. We show that the Rally effect is not an unconditional surge in support, but a conditional phenomenon of insulation: conflict boosts incumbent popularity only during economic downturns by obfuscating failure, whereas it penalizes incumbents during booms by dampening the credit for success. We characterize the equilibrium strategies of a “Spin Doctor” who optimizes narrative complexity to trap voters in a high-congestion equilibrium, and demonstrate that the ability to manipulate information costs increases the incentive for diversionary war. Our framework rationalizes recent cross-national evidence showing that the political returns to conflict are strictly context-dependent.
Old projects
Les cinq affirmations de la délibération partielle (French version)
[This version: December 2018]Abstract
Economists have been less and less reluctant to include endogenous preferences in their models. However, there is a lack of general models that describe the psychological process behind preference changes. This paper constitutes a first step toward a theory that fulfills this gap. It presents a conceptual analysis of a specific mechanism, in which preferences are awareness driven, that is, they are formed through a process of partial deliberation on the values that induce them. The paper is composed of five sections, each of them presenting and justifying the hypothesis required for this mechanism. According to the first hypothesis preferences are induced by values. According to the second, these values can be logically connected so that they form systems, that can be hierarchically ordered by a consistency criterion. The third hypothesis is that agents can have a partial representation of the values that induce their preferences. In other words, the agent is aware of only a limited set of the values that induce his preferences; values in this set are called reasons to act. With the fourth hypothesis the agent can change the reasons that induce his behavior. With the fifth hypothesis, we argue that agents do change their reasons to act by using the consistency criterion of the second hypothesis. To conclude I argue that the process of partial deliberation allows to extend the concept of rationality to the concept of autonomy.
A theory of preferences manipulation
Abstract
A sender often tries to figure out how a receiver’s preferences change in order to propose him a project that best fits both of their preferences. In parallel she may try to change the preference of the receiver. This paper models such situations by using an euclidean approach of partial deliberation, a mechanism in which preference changes are yielded by changes in the awareness of the receiver. Thus, the sender must choose a project and a disclosure strategy. I show how the set theoretic approach of partial deliberation can be adapted to an euclidean one, in which the receiver modifies the intensity she assigns to values. In this paper, I show that the axiological criterion used by the sender to change her value system can be represented by a function Φ that maps each vector v into ℝ. Then I apply this new conceptual framework to formalize imperfect empathy, i.e., a situation in which the sender is uncertain about what determine the receiver’s preferences. I also study situations in which the sender tries to use sequential disclosure strategy rather than a simultaneous one.
The dynamics of cultural mixing
Abstract
When looking at cultural evolution as a transmission mechanism the question of the preservation of heterogeneity in cultural attributes (for subsequent generations) becomes of paramount importance. Models that incorporate such heterogeneity currently focus on a preference-based opposition between cultural traits (Bisin and Verdier, 2001; Benabou and Tirole, 2006). The parameters of the individuals’ utility functions used in these models and the shape of the functions themselves clearly reflect this opposition between cultural traits. Furthermore, it is the modeler who decides which is the relevant opposition to focus on, and the cultural dynamics are usually constrained by the extreme points of this opposition. The formation of new cultural traits is not considered. In this paper, we propose a complementary approach where the dynamics of cultural transmission are based on a principle of compatibility between cultures: cultural transmission is possible only for “mixable cultures”, while cultures that contradict each other cannot be mixed. This particular strategy allows for the building of a dynamic model in which new cultural traits can emerge. We characterize a given cultural trait by a two-valued function or “meaning function” that represents the relation that individuals make between cultural symbols like words, sentences, gestures or signs. This formulation of the problem makes it possible to define the opposition of the meaning function through the product of two functions: the first function would negatively relate the two cultural symbols, while the second function would relate them positively. Two cultural traits are (said to be) “opposed” if they are globally opposed in their meaning function. The concept of a “mixable cultural trait” allows us to produce a graph in which the nodes represent the individuals. The neighbors of each individual represent other individuals with whom a node can be mixed. By considering only the neighborhood that individuals (maximally) choose to modify their meaning function and therefore to modify their cultural trait. We show the convergence of this algorithmic process and study applied situations.