Title: (Mis)perceptions of stability and learning

Authors: Samuel D. Hirshman and Alex Imas

Expectations and learning from new information are critical inputs to economic behavior. We study how people update their beliefs in stable vs. unstable information environments. We document two novel empirical facts using learning experiments with simple data generating processes. First, people in stable environments update their beliefs “as-if” the environment is unstable. They update their beliefs too much relative to Bayesian learning in response to signals, especially in later rounds of our task. Second, people in simple unstable environments underreact to the possibility of change.  When told about the possibility of change at a specific point in time, people do not update their beliefs enough. We rule out misperceptions of signal diagnosticity, memory constraints, and cognitive uncertainty as drivers of our effects. Finally, we provide convergent evidence that people have a fundamental misperception of stability. They perform substantially better when in an information environment calibrated to prior participants’ perceptions of stability. Our experiments shed light on forecasters’ tendency to both over and underreact to new information.