Demande d'assurance : faut-il abandonner le critère de l'espérance d'utilité ?

Karine Darjinoff, TEAM
François Pannequin, GRID, ENS Cachan

Résumé. Cet article présente une réflexion sur le modèle d'espérance d'utilité (EU). L'objectif est d'évaluer ce critère selon sa capacité à prédire les comportements de demande d'assurance, à l'aide des études expérimentales et des observations réelles. Quatre faits principaux sont retenus : la prédominance des contrats de franchise sur le marché, le développement économique et l'assurance, le choix entre assurance complète et rétention de risque, la perception des risques. Si les études expérimentales, issues de la littérature remettent généralement en cause la théorie normative, elles omettent d'évaluer le résultat fondamental de la théorie de la demande d'assurance : l'optimalité du contrat de franchise. Un test original permet de lever cette insuffisance ; les prédictions du modèle EU ne sont pas totalement vérifiées.
Mots clés : Assurance, risque, utilité espérée, effet de contexte.

Abstract. This paper presents a reflection on the expected utility model. The aim is to assess this criteria according to his ability to predict insurance behavior, with experiments and real observations. Four main facts are : the predominance of deductible insurance policy on the market, economic development and insurance, the choice between complete insurance and retention and the perception of risk. We present the predictions of expected utility theory and several results are presented. Thus, some experimental and real results do not respect the predictions. (1) Insurance seem to be a normal good. (2) Subjects have diverse attitudes : subjects who are risk-averse in one situtation become risk-seeking in another, and vice-versa (Slovic et al., 1979). (3) They prefer extreme insurance policies (Schoemaker and Kunreuther, 1979). (4) Transformation effects in the loss distribution are generally consistent with the theoretical work of Eeckhoudt, Gollier and Schlesinger (1991). However, if there is a mean-preserving transformation (an increase in risk) affecting only portions of the loss distribution above the deductible, then some people change their optimal deductible (Brandao, 1994). (5) Expected utility theory with a concave utility function implies that probalistic insurance must be preferred to regular insurance, but the majority of subjects reject probalistic insurance (Kahneman et Tversky, 1979). (6) Subjects are sensitive to te context of decision (distorted in their perceptions of risk and framing effects), subjects overweight low probabilities, and underweight high ones. Furthermore, subjects' pricing decisions are sensitive to ambiguity and aversion to ambiguity decreases as the probabilities of loss increases, there is a preference for ambiguity for high probabilities (Hogarth and Kunreuther, 1989). We show that belief, perceptions and judgements must be taken into account in individual-decision making and further research must determine which theory better explains observed behavior. If experimental studies, stemming from literature, generally call the normative theory into question, they omit to assess the main result of the theory of insurance demand : optimality of deductible insurance contracts. An original test let renove this insufficiency ; these experimental results are not completely consistent with the EU model's predictions.
Keywords : Insurance, risk, expected utility, context effect.

JEL Classification : D81.