Abstract
Eliciting all the components of prospect theory – curvature of the utility function, weighting function and loss aversion – remains an open empirical challenge. We develop a semi-parametric method that keeps the tractability of parametric methods while providing more precise estimates. Applying the new method to the datasets of Tversky and Kahneman (1992) and Bruhin et al. (2010), we reject the convexity of the utility function in the loss domain and show that the probability weighting function does not exhibit duality and equality across domains, in line with cumulative prospect theory and in contrast with original prospect and rank dependent utility theories. Furthermore, our method highlights that the overweighting of tail probabilities is more pronounced in the gain domain than in the loss domain. Overall, our results show that the utility function varies little across domains, thus suggesting that probability distortions are key to capture differences in risk attitudes in the gain and loss domains.