ARC Talk: Filippo Massari, Università di Bologna: Good Biases in Misspecified Learning

We study a parsimonious learning rule that encompasses well-documented learning biases. We prove a "universal-advantage" theorem which provides conditions under which our learning rule will outperform Bayes. For example, agents who display an under-reaction bias will never be outperformed by a Bayesian and will outperform a Bayesian in some environments. The opposite is true for agents who over-react. Beyond under- and over-reaction, our learning rule can encompass under- and over-inference, frequency bias, and extreme belief aversion. Our results suggest that predominant learning biases in the population have emerged due to evolutionary forces, rather than cognitive failures.

Hybrid Talk: https://arc.mpib-berlin.mpg.de/public-with-recording/

Meeting-ID: 650 1166 5219

Passcode: 125701