Consider a repeated principal-agent setting with verifiable effort and an extra profit that can materialize only if the agent is talented. The agent is overconfident and updates beliefs using Bayes' rule. The agent's principal-expected compensation decreases over time until high talent is revealed; thus he may be employed only if beliefs are sufficiently low. We apply these results to a firm's promotion policy, which may be based on success in a previous job even if jobs are uncorrelated. This provides an explanation for the “Peter Principle” in a setting with verifiable performance and highly confident workers (Benson et al., 2019).

The outcome of a foregone alternative is not always learnt. We incorporate this observation into the classic decision theoretic model of regret (Bell, 1982, Loomes and Sugden, 1982), allowing for the ex-post information available to a regret averse decision maker to vary with choice. We show that more ex-post information is never desirable for a regret averse individual. We then suppose that there are multiple regret averse decision makers and that the ex-post information available to each depends both on own choice and on the choices of others. Thus, what appeared to be a series of isolated single individual decision problems is in fact a rich multi player behavioural game that we term regret game, where the psychological payoffs that depend on ex post information are interconnected. For an open set of parameters, the regret game is a coordination game with multiple equilibria, and this despite the fact that all individuals share a uniquely optimal choice when the same decision problem is faced in isolation. We experimentally test the predictions and find strong support that regret averse individuals behave as predicted by our theory.

Choices under risk often violate canonical axioms, like transitivity and stochastic dominance. These violations are in line with context-dependent theories such as regret aversion and salience theory, which differ in whether they regard such violations as mistakes (e.g. due to blurred perception of probabilities) or as a manifestation of true preferences (e.g. regret avoidance). By implementing a display format of lottery choices with a transparent correlation structure, we investigate whether violations of axioms are mistakes or an expression of true preferences in a setting that, according to existing evidence, increases the explanatory power of context-dependent theories. We build on the experimental paradigm by Nielsen and Rehbeck (2022) and allow subjects to reconcile previously elicited axiom preferences and previously made lottery choices. Our experimental results show that significantly more subjects — among those who initially violated an axiom — keep their lottery choices and rather unselect the axiom. About a quarter of our subjects can be explained by a theory of context-dependent preferences. Of those observations, a majority of almost three quarters can be categorized as having intentionally violated the axiom and only roughly a quarter as having made a mistake.