This paper analyzes a dynamic adverse selection market where buyers hold ambiguous beliefs about seller quality, modeled using neo-additive Choquet capacities and updated via optimistic, pessimistic, and Generalized Bayesian rules. First, we show that the choice of updating heuristic has a direct and systematic effect on the severity of adverse selection. While the optimistic and pessimistic rules invariably mitigate or amplify the problem, respectively, the Generalized Bayesian rule’s impact is conditional, its trajectory toward collapse, efficiency, or a stable partial market depending on a persistent ‘tug-of-war’ between the buyer’s static ambiguity attitude and the evolving probabilistic evidence. Our second main finding is that these immediate effects compound over time, leading to fundamentally different market trajectories. The pessimistic rule can drive the market to complete collapse, the optimistic rule can foster full participation, and the Generalized Bayesian path depends on the interplay between the buyer’s attitude and the evolving evidence. We further analyze how baseline ambiguity and ambiguity aversion modulate these dynamics, uncovering a complex role for ambiguity in shaping the rate of market evolution.

When a firm appoints a new manager, it reopens the possibility of new contractual friction with its partners. We explore strategic ambiguity as a potential for friction with a supplier. The firm's new manager probably has fuzzy expectations about the supplier's strategy. An optimistic manager weights favorable strategies more heavily than detrimental ones, whereas a pessimistic manager does the opposite. We show that the manager's degree of optimism is critical: above a threshold, it can cause the supplier to change the timing of its contracting and increase its profits. We also find that this threshold degree of optimism depends on the degree of product substitution: it is more stringent with imperfect substitutes than with perfect substitutes or unrelated goods.

This article addresses the value of information that affects the ambiguity faced by a decision maker. Our analysis is applied to the case of a farmer whose production can be damaged by a pest attack with unknown probability, this damage being reduced if the farmer decides to use a pesticide. Early warning systems have precisely been implemented in many countries to help farmers avoid inappropriate decisions in terms of pesticide use. We investigate, both theoretically and experimentally, how farmers value these systems. We propose a two-state self-insurance model in which an α -MaxMin Expected Utility farmer may use pesticides that reduce the loss in the accident state while incurring a cost in both states. Her decision to self-insure or not depends on risk and ambiguity attitudes. We compile and compare the value of two types of information leading to a reduction of ambiguity and analyze their properties with respect to ambiguity attitude. Both types of information are valued positively if the farmer is ambiguity averse. We conduct a framed field experiment in which farmers and agricultural students have to decide whether or not to apply pesticides depending on risk, ambiguity and the associated monetary gains resulting from pesticide cost. The experimental findings support the theory. The average value of information among all participants is between €0.9/ha and €3.3/ha depending on the information gain.

Par differentes approches (semantique, economique, de modelisation), les equipes reunies dans le projet VESPA eclairent les differentes formes de contribution de l'epidemiosurveillance a la sante des cultures. Des valeurs d'information, de referencement, de pedagogie et d'entretien des competences expertes en phytopathologie generaliste de terrain decoulent assez directement de son existence. Bien que les calculs soient difficiles et entaches d'approximations, on conclut que l'epidemiosurveillance rapporte au minimue 4 a 5 fois plus qu'elle ne coute, y compris en autorisant des reductions d'usage des produits phytopharmaceutiques. Ces apports relevent toutefois majoritairement de l'optimisation rationnelle de la couverture des besoins de protection et elles ne peuvent a elles seules couvrir au mieux qu'une part des objectifs de reduction de la dependance aux produits phytopharmaceutiques. Pour aller plus loin, il devient donc necessaire que les messages d'alerte modulent le risque phytosanitaire en fonction des mesures prophylactiques prises et relaient l'efficacite d'une gamme elargie d'actions preventives.

This paper studies decisions under ambiguity when attention is paid to extreme outcomes. In a purely subjective framework, we propose an axiomatic characterization of affine capacities, which are Choquet capacities consisting in an affine transformation of a subjective probability. Our main axiom restricts the well-known Savage’s Sure-Thing Principle to a change in a common intermediate outcome. The representation result is then an affine combination of the expected utility of the valued act and its maximal and minimal utilities.

Résumé L’article étudie la sélection adverse dans le cas de la demande d’assurance d’une automobile louée lorsque le critère de décision est une espérance de Choquet. L’approche adoptée de l’intégration de l’information à la décision d’assurance autorise l’utilisation d’une procédure d’induction à rebours pour déterminer l’action optimale ex ante , ce qui permet de retrouver les conclusions du modèle bayésien. En revanche, le rôle de l’ambiguïté peut être pris en compte par les capacités, ce qui n’est pas le cas avec les probabilités. Lorsque le décideur présente de l’aversion envers celle-ci, son comportement consiste à surestimer l’information si un accident s’est produit à la première étape et à la sous-estimer dans le cas contraire.

We give an axiomatic foundation to the updating rule proposed by [Sarin, R. and Wakker, P. P. (1998). Revealed likelihood and knightian uncertainty. Journal of Risk and Uncertainty 16(3):223-250.] for CEU preferences. This rule is dynamically consistent but non-consequentialist, since forgone consequences are relevant for conditioning. Whereas it does not work universally, but only when counterfactuals outcomes are better and/or worse than the ones resulting on the conditioning event, the rule has many interesting features, since it is able to describe Ellsbergtype preferences together with a recursive structure of the criterion.

This note completes the main result of Zimper (Theory and decision, doi: 10.1007/s11238-010-9221-8 , 2010 ), by showing that additional conditions are needed in order the law of iterated expectations to hold true for Choquet decision makers. Due to the comonotonic additivity of Choquet expectations, the equation E [ f , ν (d ω )] = E [ E [ f ( ω i , j ), ν ( A i , j | A i )], ν ( A i )], is valid only when the act f is comonotonic with its dynamic form, that we name “conditional certainty equivalent act”.

When individual beliefs are not Bayesian. economic agents may refuse further information about the uncertainty they are facing. Choquet decision makers in particular may be information averse. This note shows that, if the capacity is neo-additive, then the information value is necessarily positive.

The article studies adverse selection in the context of insurance demand for a rent car when the decision criterion is a Choquet expectation. Our approach of integrating information to the insurance decision is able to implement a backward induction procedure to determine the optimal action undertaken ex ante. It allows to obtain the same conclusion than the bayesian model. However, the role of ambiguity may be take into account since we use capacities, rather than probabilities, to represent individual's beliefs. If the decision maker is ambiguity averse, she overweights the information if an accident has occurred at the first stage and under-weights it in the opposite case.

We give an axiomatic foundation to the updating rule proposed by Sarin and Wakker [Sarin, R., Wakker, P.P., 1998a. Revealed likelihood and knightian uncertainty. Journal of Risk and Uncertainty 16, 223–250] for CEU preferences. This rule is dynamically consistent but non-consequentialist, since forgone consequences are relevant for conditioning. Whereas it does not work universally, but only when counterfactuals outcomes are better and/or worse than the ones resulting on the conditioning event, the rule has many interesting features, since it is able to describe Ellsberg-type preferences together with a recursive structure of the criterion.

Several updating rules for Choquet capacities have been proposed in the literature. We propose to study them in a general framework consisting of extending the definition of conditional expectations to the Choquet case. Then we adopt several relations between unconditional and conditional expectations and observe which updating rule is able to resolve such and such a relation. It allows to specify which updating rule should be used depending on the decision context.

This note studies conditions which allow to maintain a non-expected utility representation (Max-min expected utility and Choquet expected utility), dynamic consistency and consequentialism in an atemporal and purely subjective framework. By contrast with a dynamic set-up, where consistency can be reached with non-expected utility models, we show that both Maxmin expected utility and Choquet expected utility degenerate into an expected utility representation.

Ces trente dernieres annees ont vu le developpement de plusieurs modeles de decision nonadditifs dans l’ambiguite. Les deux principaux sont le critere de l’esperance d’utilite de Choquet et celui de l’esperance d’utilite avec a priori multiples. L’un des objectifs principaux de ces modeles est la rationalisation du paradoxe d’Ellsberg. Cette experience suggere que les preferences individuelles doivent pouvoir integrer l’ambiguite et l’attitude envers l’ambiguite. Contrairement au modele bayesien, les criteres non-additifs ont la capacite de prendre en compte ces caracteristiques du probleme de decision et de la psychologie individuelle. La revision des croyances non-additives souleve un probleme particulierement interessant pour l’economie. En effet, plusieurs situations economiques implique non seulement de l’ambiguite, mais egalement une arrivee sequentielle de l’information dans le processus de decision. Tandis que le conditionnement bayesien des croyances probabilistes satisfait automatiquement les axiomes de consequentialisme et de coherence dynamique, ces hypotheses doivent etre explicitement supposees dans le cadre des modeles non-additifs. Neanmoins, elles ne peuvent pas etre supposees ensemble : dans le but de preserver la nonadditivite des croyances, seulement l’une d’entre elles peut etre imposes sur les preferences. En effet, un resultat bien connu en theorie de la decision dans l’incertain etablit que ces axiomes pris ensemble implique l’additivite des croyances et leur revision bayesienne. Cette these etudie l’hypothese de consequentialisme dans le cadre des modeles non-additifs, ses implications sur les croyances du decideur, et propose plusieurs regles de revision des croyances coherentes dynamiquement et non-consequentialistes.

This paper studies learning under multiple priors by characterizing the decision maker's attitude toward information. She is incredulous if she integrates new information with respect to only those measures that minimizes the likelihood of the new information and credulous if she uses the maximum likelihood procedure to update her priors. Both updating rules expose her to dynamic inconsistency. We explore dierent ways to resolve this problem. One way consists to assume that the decision maker's attitude toward information is not relevant to characterize conditional preferences. In this case, we show that a necessary and sucient

Pursuing works from Sarin and Wakker (1998), we study how NonExpected Utility models could be consistently applied to multi-stage decision problems. Concerning multiple priors model, we remove the argument that dynamic consistency, consequentialism and model consistency (sequential consistency in Sarin and Wakker (1998)) can all be preserved. Like Choquet expected utility, deviation of expected utility is only allowed over the final stage. Moreover, it's proved for the two models that if we also assume reduction of compound acts, then an expected utility representation exists in all stages.