Economics experiments have produced widespread evidence that people suffer from a range of cognitive biases. However, unlike experiments, real-world institutions often allow decision makers to self-select out, potentially filtering (or amplifying) the impact of biases on economic aggregates. We study the economic impacts of such self-selection and how they depend on people's meta-cognitive awareness of their own errors. In a series of online experiments that cover a wide range of classical decision biases, we document large heterogeneity in how cognitive biases are related to the intensity of bets in speculative markets, bids for property rights in auctions and contributions to collective decisions. In some tasks, rational subjects are more confident than their biased counterparts and bet, bid and vote more aggressively. As a result, self-selection tends to filter the effect of irrationalities on aggregate quantities. However, in other tasks, confidence and performance are unrelated or even negatively correlated, so that experimental institutions do not filter errors and sometimes even magnify them. As a methodological blueprint, we show that a simple measure of the relative calibration of confidence strongly predicts the degree to which institutional self-selection filters the effect of irrationalities.
Join at http://imt.lu/seminar