Working Papers
Experimental Evidence on Misguided Learning
[pdf] [appendix] [instructions (German)]
This paper studies how people form beliefs in environments with multiple unknown parameters, some of which are relevant to agents' self-esteem. In particular, we examine how initial bias in beliefs about an ego-relevant characteristic affects learning about the state of the world. Using data from a laboratory experiment, we demonstrate that the learning process of an overconfident agent is self-defeating: the agent repeatedly takes suboptimal actions, misinterprets the output, and forms increasingly mistaken beliefs about the state. Therefore, we corroborate the theory of misguided learning formulated by Heidhues et al. (2018). We provide the first empirical evidence that allowing a biased agent to experiment and acquire new information is not only ineffective but in some cases counterproductive. Furthermore, we move beyond the theory as we examine how learning about multiple parameters evolves in ego-relevant and ego-neutral environments.
Belief-Based Utility and Signal Interpretation
Do people update their beliefs differently after positive versus negative feedback? The existing literature disagrees on the magnitude and direction of the bias. In this paper, I propose a new experiment guided by a simple model of belief choice. The experimental data reveal a strong asymmetry in updating after "good" versus "bad" news. Moreover, I design a control condition that allows a clear identification of belief manipulation and provides robust evidence on the underlying mechanism. Participants in the control group evaluated hypothetical signal realizations---they faced the same decision as subjects in the treatment group but without receiving a signal. I document no asymmetry in the control condition, as predicted by the model. The difference-in-difference analysis reveals that people tend to distort their perception of a positive signal only after its realization. The results point towards the role of affect (or utility from beliefs shifted by the signal) in asymmetric updating. The proposed method can be applied more broadly to study belief-based utility and its role in belief formation.
Hope for the best, prepare for the worst: signal anticipation and ex-ante belief manipulation
[Draft available upon request]
In this paper, I experimentally test a model of belief choice with reference-dependent utility. The basic idea is that people can "prepare themselves" for the arrival of new information by adopting overly pessimistic beliefs. By distorting her prior beliefs, an agent can 1) hedge against a painful downward shift in beliefs after a negative signal and 2) enhance a pleasant surprise from a positive signal. To test the model, I designed a lab experiment in which subjects solve an IQ test and subsequently report beliefs about their relative performance. I introduce an exogenous variation in subjects' expectations over the upcoming signal, which allows me to identify belief manipulation. The results confirm the main predictions of the model, substantiating the claim that utility from beliefs is reference-dependent. Furthermore, I examine a previously unexplored link between gain-loss attitudes and overconfidence, and confirm it in the data.
Work in Progress
Estimating Belief-Based Utility Using Experimental Data
Does the world get crazier or is it just me? Learning about ability and an external parameter