| This dissertation explores models of learning and utility using two experimental designs. The first is a supply-chain experiment, the newsvendor game, which generates consistent patterns of under-ordering when profitability is high, and over-ordering when profitability is low. The experience-weighted-attraction learning model provides a good fit to these data. The model parameters suggest that this "pull-to-the-center" result occurs because subjects have some forgetfulness and a very strong tendency to ignore alternative choices, even if they are very profitable. There is also some evidence of an S-shaped utility function, suggesting that risk attitudes may change for gains and losses. A treatment in which decision and feedback frequency were restricted did not improve orders and simply led to slower learning. The second experiment involves an asset market with a risky and a riskless asset, which generates pricing bubbles that grow with the amount of wealth in the market. Only a few subjects in the market actually participate in the bubble, and these usually are worse off at the end of the experiment than those who did not speculate. The market data are fit to a behavioral portfolio-choice model that contains flexible formulations for risk attitudes and price forecasts based on past events. The results suggest the presence of increasing relative risk aversion and decreasing absolute risk aversion, and an extrapolative formulation of subjective price beliefs based on past prices and order-book effects. |
