Ellsberg's famous objection to Savage has led to the development of many new theories of choice under uncertainty. These decision-theoretic alternatives typically generalize Savage's subjective expected utility theory (EUT) and fall within two classes --- utility models of multiple versus single prior(s). We provide the first nonparametric revealed preference analysis of such models using individual-level data collected from a portfolio choice experiment. We find that for many subjects there is considerable scope for multiple prior utility models (and little scope for single prior utility models) to explain behavior not accounted for by subjective EUT. These results stand in sharp contrast to earlier results under pure objective risk (Dembo et al., 2021) which provide little scope to canonical non-EUT alternatives in explaining behavior.

Individuals often attach a special meaning to obtaining a certain goal, and getting past a threshold marks the difference between what they consider a success or a failure. In this paper we take a standard von Neumann-Morgenstern Expected Utility setting with an exogenous reference point that separates success from failure, and define attitudes towards success and failure as features of preferences over lotteries. The distinctive feature of our definitions is that they all concern a local reversal of the decision maker's risk attitude between riskaversion and risk-lovingness across the reference point. Our findings provide a unified view of several well-known models of reference-dependent preferences in economics, finance and psychology, and also include novel representations. Moreover, we introduce orderings over the primitive space of preferences to define different attitudes with which each attitudes can be displayed, and characterize them in terms of the representation, with indices analogous to the well-known Arrow-Pratt index of risk aversion. Our findings shed new light on frequently used notions of reference-dependent preferences, and suggest that new comparative statics analyses be conducted in these settings. Finally, we argue that our framework may prove useful to incorporate, within a standard economic model, behavioral manifestations of personality traits that have received increasing attention within the empirical economics literature.

In an experimental study, we investigate individual preferences when objects of choice are two-outcome (objectively ambiguous) prospects with imprecise winning probabilities, i.e., the probability of a given outcome belongs to an interval [p* , p*]. We postulate a generalization of α-Maxmin expected utility (α -MEU; e.g., Olszewski, 2007; Ahn, 2008) that substitutes rank-dependent utility (Quiggin, 1982) for expected utility. Further, we allow the weighting function of lower bound probabilities, w*, to be different from that of upper bound probabilities, w*. Our observations show that w* differs from w*, hence falsifying the assumption that w* = w*; (e.g., Baillon et al., 2012). Specifically, we show that our model could disentangle the possibility effect, captured by a concave w*, and the certainty effect revealed through a convex w*. We also observe that inverse S-shaped probability weighting observed under both risk and ambiguity results from a mere convex combination of w* and w*.

An extension to original prospect theory is provided in which the gain-loss utility is formally identified. To this aim, we demand some standard properties, including monotonicity with respect to first order stochastic dominance, and combine them with a new property called gain-loss consistency. For simple prospects that give a gain and a loss (and else the reference point), these properties characterize an extension of original prospect theory with probability weighting that depends on whether probabilities are attached to gains or to losses. For general prospects with multiple gains and multiple losses, the probability weighting functions apply to the overall probability of gaining and of losing. Further, the gain-loss utility function determines the value of a prospect by transforming the expected utility of the gain part and the expected utility of the loss part of that prospect before those components are aggregated to obtain that prospect’s value. This way, the gain-loss utility can be decomposed into a basic utility for outcomes and a transformation thereof that captures loss aversion.