Richard H. Thaler et les limites de la rationalité (with Gwen-Jiro Clochard and Guillaume Hollard) : Revue d'économie politique 2018/4 (Vol. 128), p. 535-548 (in French)
How serious is the Measurement Error problem in a popular risk-aversion task? (with Guillaume Hollard, Radu Vranceanu) Journal of Risk and Uncertainty
Self-Selection Filters Irrationality in One-shot Games (with Guillaume Hollard)
If economic agents engaging in strategic interactions have a sense of how well they are doing, they should avoid situations in which they may make embarrassing mistakes by opting out. We here propose to test that proposition and to explore its consequences regarding self-selection into strategic interactions. A crucial possible effect is that strategic interactions among self-selected agents (which are common in the field) may be better described by standard theory than situations in which agents did not have the opportunity to self-select (like standard experiments in the lab). We here investigate the effect of self-selection on the outcome of games.
We propose an experiment in which subjects can (1) report their subjective feeling of how well they are doing using a confidence measure that is new in this context and (2) self-select into strategic interactions or choose a sure payoff.
We find that self-selection is indeed driven by our measure of confidence and risk-aversion: the least confident and more risk-averse subjects are less likely to self-select into strategic interactions. Self-selection dramatically reduces the gap between theoretical predictions and actual behavior. We find that economic agents do have a sense of how well they are doing in games. Perhaps more surprisingly, this sense varies significantly from game to game at the individual level.
We complement our analysis by reviewing field evidence on self-selection, which confirms that self-selection promotes rationality.
Better Beliefs in Normal Form Games (with Guillaume Hollard)
Deviations away from equilibrium prediction in one-shot games are commonly observed in experimental games. Models were designed to capture the behavior of bounded-rational players in games. While the Quantal Response Equilibrium model allows players to better respond to correct beliefs, in the model we introduce, strategic players best respond to better beliefs, beliefs that are not perfect. Unlike Level-k models, strategic players hold plausible beliefs that are (almost) correct on average.
Specifically, our model assumes the existence of not forming beliefs (NFB) players that are either random or naive and the existence of strategic players. Strategic players best respond to their beliefs regarding the relative proportions of strategic and NFB players. Strategic players hold heterogeneous beliefs and fail to anticipate that they disagree on the relative proportion of NFB players.
We show that our model has interesting properties and can be brought to the data by adding no more than one free parameter.
Should we trust measures of trust? (with Héloise Cloléry, Guillaume Hollard and Inès Picard)
Trust is an important economic variable that may however be subject to measurement error, leading to econometric issues such as attenuation bias or spurious correlations. We use a test-retest protocol to assess the measurement error in the two main tasks that are used to elicit trust, namely survey questions and experimental games. We find that trust measures based on the trust game entail substantial measurement error (with up to 15\% of noise), while there is virtually no noise in stated trust measures. Given the specificity of our subject pool (students in a top Engineering school) and the short period of time between the test and the retest, we consider this percentage of noise as a lower bound. We also provide a subgroup analysis based on measures of cognitive ability and effort. We find substantial heterogeneity across sub-groups in trust-game behavior, but none for the survey questions. We finally discuss which measure of trust should be used, and the estimation strategies that can be applied to limit the effect of measurement error.
Best experimental design presented during a poster session during the ASFEE Conference in Rennes.