Limits on Predictability of Risky Choice Behavior

AbstractResearch in decision-making has recently begun to emphasize predictive accuracy as the dominant principle for designing and evaluating choice models. This emphasis has led to development of increasingly more precise models of humans' risk preferences, as measured in experimental paradigms built upon certainty equivalence testing.Here, we argue that the level of precision attained by recent choice models is unexpected, because human preferences are irreducibly noisy. We support this argument by conducting experiments to measure intra-observer consistency in choice behavior in two common risk preference paradigms: decisions from description and experience. We find that while current choice models of decisions from experience align fairly well with the upper limits of choice consistency seen in our experimental data, choice models for decisions from description are significantly more consistent with humans' choices than the humans themselves are consistent with their own choices. We discuss some theoretical and practical implications of our results.

Return to previous page