New Experimental Evidence on Expectations Formation

We measure belief formation in an experiment where participants are asked to provide forecasts of a stable and simple statistical process. We then estimate an empirical model that allows for under- and over-reaction, as well as stickiness. Our findings are threefold. First, the rational expectations hypothesis is strongly rejected in our setting, and we find little evidence of learning. Second, both extrapolation and stickiness patterns are statistically discernible in the data, but extrapolation quantitatively dominates. Third, our model coefficients are robust across different settings, including to changes in experimental setting: they do not depend on process parameters, individual characteristics, framing.