ICES Brown Bag for August 30

Events, Seminars | August 27, 2018

Join us for the first ICES Brown Bag Lecture of the Fall 2018 semester, featuring Paul Feldman.

Mr. Feldman, of University of California San Diego, will discuss his paper (co-authored with John Rehbeck) Revealing a Preference for Mixing: An Experimental Test on Risk (Abstract). The talk will take place on Thursday, August 30th, from 12:00 to 1:00pm, in room 5075 of the Vernon Smith Hall (formerly Metropolitan Building), Arlington campus.

Coffee and dessert will be provided.

Please visit the Brown Bag Schedule to learn more about the Brown Bag series.


There are now a cornucopia of theories of risk preferences that compete with expected utility. To test these competing theories, we use a novel elicitation procedure to acquire decisions from convex budgets in probability space. The elicitation procedure is analogous to the standard consumer problem and allows us to use the tools developed therein. Looking at aggregate data, we see that many individuals ($\approx$99 \%) randomize in a way that implies two indifference lines have different slopes, so individual preferences are often inconsistent with expected utility. Moreover, we find that a similar number of individuals have choices that would imply crossing linear indifference curves and are therefore inconsistent with the Betweenness class ($\approx$97 \%). In addition, we find that aggregate data satisfies a “law of demand” for lotteries. We next examine individual-level data using revealed preference methods as a pre-test for convex preferences. While no individual is consistent with a convex preference specification, we show that most individuals are “close” to a well defined convex preference relation. To further sort preferences, we perform cross-validation to select among popular models of convex risk preferences. We find that that the model of stochastic reference dependence with a choice acclimating personal equilibrium, a quadratic and referent dependent model, has minimal cross validation error for $\approx$57\% of individuals. Further, we provide additional non-parametric tests that support and clarify these results.

<<< Back to News Page