List of Abstracts


MONDAY, the 18TH OF DECEMBER 2023


«We consider a family of exchange economies with complete markets where consumers have multiprior preferences representing their ambiguity aversion. Under a linear independence assumption, we prove that regular economies are generic. Regular economies exhibit enjoyable properties: odd finite number ofequilibrium prices, local constancy of this number, local differentiable selections of the equilibrium prices. Thus, even if ambiguity aversion is represented by non-differentiable multiprior preferences, economies retain generically the properties of the differentiable approach.»


Parimutuel betting is commonly known as pool betting. It works differently from a traditional betting market in that, instead of betting against the bookmaker, all bets are cumulated into a pool and then payoffs are awarded based on the number of bets taken and the amount the player has wagered. A general model of betting markets will be proposed to encompass the various models of (i) Parimutuel betting, (ii) Fixed-odds bettings in which bettors are betting against bookmakers, and (iii) Online betting in which gamblers can be hybrids and can both make bets and take bets.  In Parimutuel betting, what happens when states collectively believed to be impossible happen?  In a model without any bookmakers, how do we model the Parimutuel House without letting it have any overround leverage as it is the case when betting against « standard » bookmakers? For the first issue, we will propose several solutions when agents have Von Neumann-Morgestern utility functions.  For the latter one, we will model the of Parimutuel House as a « neutral » bookmaker or a non-profit firm, whose role is only to collect  the bets, transport and allocate the pool to the winning bettors.  

.


Rational Inattention holds the promise of being a canonical model of costly information acquisition. Due to its tractability, the workhorse model of Rational Inattention, which bases the cost of information on Shannon entropy, has spurred a first wave of applications. However, it has likely reached its capacity, as it cannot capture the wide heterogeneity of attention problems that economic agents encounter. In this paper, we develop tools to analyze Rational Inattention problems for generalized entropy functions. We demonstrate that Rational Inattention is better understood through the lens of the Fenchel conjugate of the entropy function, rather than through the entropy function itself. Leveraging this dual perspective, we propose entropy functions that generalize the Shannon model, preserving its tractability. We also highlight new properties of information costs and explain their behavioral implications.


We study a broad class of preferences over menus (or opportunity sets) that obey the independence of common alternatives axiom and admit a representation that suits the peculiarity criteria, that is, there exists a utility function V over the collection of all menus of alternatives in which the menu A is at least as good a B if and only if the value V(A\B) is greater or equal than V(B\A). Further, in dealing with some special cases, we discuss the well-known case of an additive representation and characterize a novel representation called the maximal peculiar rule.


We consider a reasoner who selects a set of distributions given a database of observations. A likelihood region is monotonic with respect to the likelihood function. We provide axiomatic foundations for such a selection rule. Starting with an abstract set of theories, we propose conditions on choice functions (across di§erent databases) for which there exists a statistical model such that the choice function is a likelihood region relative to that model, for the general case and for the case of a Öxed likelihood-ratio threshold. We interpret the results as supporting the notion of likelihood regions for the selection of theories.  


This article considers Savage's theorem in a configuration relaxing his technical axioms P6 and P7 that ensure a continuum nature on the set of states. With the only enrichment on fundamentals being the connectivity of outcomes set, we show that a weakened version of Independence property is sufficient for establishing a utility function, a subjective probability, and an expected utility behavior. The proof does not require the existence of a pair event, an idea that was initiated by Ramsey (1926) and was in application after by Gul (1992).


This paper decision-theoretically investigates belief aggregation method which allows consistently updating the aggregate belief. After confirming that the Pareto axiom and dynamic consistency require decision power of an individual to evolve proportionally to how much his/her prior has been successful, we propose a weaker Pareto axiom which applies only to one-step-ahead uncertainties and puts no restriction on how decision powers should evolve. We show in the binary tree domain that taking the median belief works and even satisfies full ex-ante efficiency under a single-crossing condition and dynamic voting implements it. Finally, we investigate the role for ambiguity aversion.


We provide a unified approach to stochastic dynamic programming with recursive utility based on an elementary application of Tarski's Fixed Point Theorem. We establish that the exclusive source of multiple values is the presence of multiple recursive utilities consistent with the given aggregator, each yielding a legitimate value to the recursive program. Unbounded returns are encompassed by means of an approximation method. We also present a spectral radius condition ensuring a unique value to the recursive program in some circumstances. We finally apply our theory to Epstein-Zin preferences. Overall, acknowledging the unavoidable failure of uniqueness in general, we argue that, contrary to a certain practice in the literature, the greatest fixed point of the Bellman operator should have a privileged position. 


We provide a new foundation of risk aversion by showing that the propension to exploit insurance opportunities fully describes this attitude. Our foundation, which applies to any probabilistically sophisticated preference, well accords with the commonly held prudential interpretation of risk aversion that dates back to the seminal works of Arrow (1963) and Pratt (1964). In our main results, we first characterize the Arrow-Pratt risk aversion in terms of propension to full insurance and the stronger notion of risk aversion of Rothschild and Stiglitz (1970) in terms of propension to partial insurance. We then extend the analysis to comparative risk aversion by showing that the notion of Yaari (1969) corresponds to comparative propension to full insurance, while the stronger notion of Ross (1981) corresponds to comparative propension to partial insurance.  


Studies show that people’s beliefs about randomness are systematically misspecified – for instance they do not expect streaks to persist. In a canonical coin-tossing environment, this paper models such “streak aversion” in terms of a belief in Mean Reversion along the sequence of outcomes. Beliefs that exhibit Mean Reversion can be represented as if the bias of the coin is path-dependent and self-correcting. Consistent with other findings, such beliefs may fail the Law of Large Numbers. In the setting of Bayesian inference, Mean Reversion ensures that the agent never rules out the true parameter. In an evolutionary setting, Mean Reversion agents are never pushed out of the evolutionary race by standard agents who correctly understand randomness. This paper suggests several directions for both theoretical and empirical investigation of beliefs about randomness.


Suppose that a decision maker (DM) uses features (namely, measurable characteristics) to describe alternatives. An analyst cannot observe directly which features the DM considers, how she evaluates them, and which procedure she uses when choosing. Can we identify which features matter? We propose a “Pareto dominance” approach: the only assumption we make is that the DM does not make choices that are Pareto dominated in the set of relevant features. We characterise exactly which (collections of) pairs of a choice observation and a feasible set are informative about the features that the DM uses in her internal representation of alternatives. Our results can be leveraged for the study of AI algorithms and the design of experiments.


This paper models an agent that ranks actions that have uncertain payoffs after observing a signal that could have been generated by multiple information structures. Under the assumption that the agent's preferences conform to the multiple priors model, we show that a simple axiom characterizes an updating rule in which the priors and information structures considered are subjective. Our axiom only requires that whenever all the possible sources of information agree that it is more "likely" that an action that has uncertain payoffs is better than one that has certain payoffs, the agent prefers the former. We also characterize several special cases of generalized Bayesian updating, including Full Bayesian updating, Maximum Likelihood updating, and some more general rules. Finally, we consider the situation where the informational content of a signal is purely subjective. We characterize the existence of a subjective set of information structures under full Bayesian updating for two extreme cases: (i) there is no ex-ante state ambiguity, and (ii) there is no signal ambiguity. 


In many economic models an agent has sub-utility functions defined over different categories of goods, which are then aggregated into an overall utility function. We develop a new revealed preference method which could determine if choice behavior arises from maximizing a weakly separable utility function. We use our approach to test for weak separability in the data collected in an experimental study on distributional preferences by Fisman, Kariv, and Markovits (2007).  We find that a substantial proportion of participants have social preferences (preferences over money received by anonymous others) which are independent of the amount of money the participants themselves receive; in other words, they have weakly separable utility functions.  We also provide evidence based on our non-parametric approach that, when deciding how much to give to other people, most participants value gains in the efficiency of payouts over gains in equity.


We demonstrate how to extend well-known comparative results under expected utility to smooth preferences which fail to satisfy the Independence axiom. We provide three major sets of results. First, in order to motivate our approach, we show that well known equivalences under EU between partial derivatives of the local (i.e., Bernoulli) utility functions and comparative statics break down when we consider preferences that do not satisfy the Independence Axiom. In contrast, the comparative statics rely on total derivatives of the local utility functions. We then formulate sets of sufficient conditions in order for monotone comparative statics to hold so long as preferences are sufficiently smooth. In particular, we consider cases where actions are arguments in payoffs, the local utility functions and the distribution over outcomes. We develop new tools to link specific conditions such as log-supermodularity and single-crossing of the local utility functions to conditions on utility over lotteries. Last, we apply our results to a variety of well-known economic problems, including portfolio choice problems where preferences, returns or wealth might change, as well as to precautionary savings.


We illustrate the strong implications of recursivity, a standard assumption in dynamic environments, on attitudes toward uncertainty. In intertemporal consumption choice problems, recursivity always implies constant absolute ambiguity aversion (CAAA) when applying the standard dynamic extension of monotonicity. Our analysis also yields a functional equation called “generalized rectangularity”, as it generalizes the standard notion of rectangularity for recursive maxmin preferences to general certainty equivalents. Our results highlight that if uncertainty aversion is modeled as a form of convexity of preferences, recursivity limits us to only recursive variational preferences.


We study efficient allocations when consumers have heterogeneous smooth ambiguity preferences with a common, point-identified, set of relevant probability measures. Furthermore, aggregate endowment is ambiguous. We characterize economies where the representative consumer is also of the smooth ambiguity type and find efficient shar- ing rules in these economies. With heterogeneous ambiguity aversion, sharing rules exhibit systematic departures from those that obtain in vNM-economies. The representative consumer’s nature departs from the typical single-agent assumption, making for more compelling asset- pricing predictions. The insights extend to the case where models are only set-identified.es.


We study dynamic choice under uncertainty with growing awareness of states. First, we introduce a preference-based notion of a decision maker's awareness level. Second, we axiomatically characterize a family of conditional preferences which adheres to an Awareness-Dependent Bayesian representation with respect to a conditional probability system a la Myerson (1986), defined on the revealed awareness-unawareness structure. Finally, we derive a consequentialist (i.e., information preserving) evolution of conditional probability systems in the spirit of reverse Bayesianism in Karni and Vierø (2013).


Recursive utilities which represent preferences as solutions to non- linear difference equations are used extensively in macroeconomics and asset pric- ing. They offer great flexibility in modeling time preferences, intertemporal risk aversion and attitudes to timing of resolution of uncertainty. We revisit the issues of existence and properties of recursive utilities in regard to discounting, risk and timing attitudes in a Markov setting. As these properties are of ordinal nature, we focus on ordinal representations of recursive utilities which allow for simulta- neous transformations of the aggregator and the certainty equivalent. Taking into account ordinal representations delivers much weaker conditions for existence, risk aversion and preference for early resolution of recursive utilities than the conditions previously known in the literature. We examine Epstein-Zin and risk-sensitive recursive utilities that are often used in applications. Further, we introduce a novel class of Koopmans recursive utili-ties that are generated by an additively separable aggregator with non-linear dis-count function and the expectation operator as a certainty equivalent. Koopmans recursive utilities feature non-exponential discounting and a clear-cut separation between timing and risk attitudes, and offer great flexibility in these dimensions. 


One often has to aggregate over two or more components: risk & time & persons as in this conference, or over commodities, locations, etc. In classical models (e.g., discounted expected utility) the order of aggregation then does not matter. This leads to our first result: the prettiest axiomatization of discounted expected utility you ever saw. Modern behavioral generalizations relax classical complete separability but maintain weak separability (= Pareto optimality etc.). A paradox: for two or more components such weakening is just impossible. One then crucially has to choose the order of aggregation. Many ongoing, seemingly unrelated, debates in different fields amount to the same underlying million-$ question: ex-ante vs. ex-post fairness, equity in Harsanyi’s veil of ignorance, monotonicity in Anscombe-Aumann’s ambiguity framework, incentive compatibility of random incentives, hedging in ambiguity measurements, etc. We show that a century-old forgotten theorem from macro-economics, Nataf (1948), can resolve all aforementioned paradoxes and debates in their various fields.