Essays in Decision Theory

When a choice model fails, the standard economics exercise is to weaken one assumption at a time to study what has changed. This is often accompanied by the understanding that future work will relax multiple assumptions simultaneously in order to explain actual behavior. This dissertation does exactly that, and by studying seemingly independent behavioral anomalies as related to one another we obtain new insights about why behavior departs from standard models. Chapter 1 studies how violations of structural assumptions like expected utility and exponential discounting can be connected to reference dependent preferences with set-dependent reference points, even if behavior conforms with these assumptions when the reference is fixed. This is done with the introduction of a unified framework under which both general rationality (WARP) and domain-specific structural postulates (e.g., Independence for risk preference, Stationarity for time preference) are jointly relaxed using a systematic reference dependence approach. The framework allows us to study risk, time, and social preferences collectively, where behavioral departures from WARP and structural postulates are explained by a common source—changing preferences due to reference dependence. In our setting, reference points are given by a linear order that captures the relevance of each alternative in becoming the reference point and affecting preferences. In turn, they determine the domain-specific preference parameters for the underlying choice problem (e.g., utility functions for risk, discount factors for time). Chapter 2, a joint work with Silvio Ravaioli, conducts an empirical test for one of the models in Chapter 1. It studies how the introduction of a very safe or very risky option affects risk attitude. In a laboratory experiment, we find that adding safer options increases displayed risk aversion, and it does so even when the added options are not chosen. This finding is robust across participants and treatments (e.g., degenerate and non-degenerate safe options). By contrast, we find that the addition of risky options does not result in a detectable change in risk attitude. Our results are in line with Chapter 1’s Avoidable Risk Expected Utility model. Chapter 3 studies choices over time, which allows us to study anomalies “at a given time” and “across time” as related to one another. This is achieved by studying how past choices affect future choices in the framework of attention. Limited attention has been proposed as an explanation for the failure of “rationality”, where better options are not chosen because the decision maker has failed to consider them. We investigate this idea in a setting where (1) the observable are sequences of choices and (2) the decision makers are aware of the alternatives they chose in the past when they face future choice sets. This provides a link between two kinds of rationality violations: those that occur in a cross section of one-shot decisions and those that occur within a sequence of realized choices. Unlike the former, the frequency of the latter is naturally bounded, and their occurrence helps pin down preferences whenever a standard model of limited attention cannot