Tuesday 28th March at 15:00 in Building D6/Room Bracco,
John Hey (University of York) will give a seminar on:
Learning under Ambiguity when Information Acquisition is Costly: an Experiment
Martin Forster, Konstantinos Georgalos, John Hey
Abstract
Consider a decision-maker (DM) confronted with two bags each containing Red and Blue balls. In one bag, Bag A, which we term the Risky Bag, exactly half the balls are Red and half the balls are Blue. In the other bag, Bag B, which we term the Ambiguous Bag, the proportion of red balls is unknown, but which could be either ½+α or ½-α (where the magnitude but not the sign of α is known). The DM must choose a bag and a colour, one ball is randomly drawn from the bag and, if it is the colour he or she chose, the DM is rewarded with a reasonable amount of money, otherwise gets nothing. Before making a choice, the DM can buy information about the composition of Bag B, in the form of a Brownian motion that depends upon the true composition of Bag B. Epstein and Ji (2020) provide a solution to the optimal strategy of a DM with MaxMin preferences. This depends upon his or her aversion to ambiguity. If this is sufficiently large, the DM should not buy information and should just choose a bag and a colour at random; if, however, ambiguity aversion is sufficiently low, the DM should keep on buying information until a threshold of the Brownian motion is crossed. We report on an experiment designed to test to see if subjects follow the optimal strategy. We find that many deviate from this optimal strategy. There could be other explanations of their behaviour; one of these is the assumption that the DM does not have MaxMin (ambiguity) preferences, but instead is a risk-averse EU maximiser. We explore these possibilities.