An Empirical study of Risk Aversion using Natural Experiment

Abstract

This paper investigates the performance of the expected utility model versus a non expected utility model incorporating both loss aversion and narrow framing when describing the risky choice behavior in a natural experiment. It uses data from the Tunisian version of the TV game show ＂Deal or No Deal.＂ Based on the choices of 90 contestants, we first calculate the mean of the lower and the upper bounds of the contestants' relative risk aversion. We find that the expected utility model fail to describe all the possible scenarios of the risky choice behavior. Notably, we find when the contestant accepts an offer while he rejected better, the expected utility model produces an upper bound of risk below the lower bound. The model of loss aversion and narrow framing has better performance as it introduces a correction in the contestant utility via the loss aversion term-a greater sensitivity to losses than gains-witch solves the problem of inconsistency between upper and lower bounds. We then estimate a stochastic preference model via the maximum of likelihood approach. Results show important contestants' errors which reflect their difficulties to make choices in the game and confirm the superiority of the loss aversion and narrow framing model on the expected utility model in describing contestants' preferences. The model also provides an average estimate of the relative risk aversion of 0.1962 for initial wealth between 0 and 12,000 Tunisian National dinars (TND).