Overreaction in Expectations under Signal Extraction: Experimental Evidence

We experimentally evaluate three behavioral models of expectation formation that predict overreaction to new information: overconfidence in private signals, misperceptions about the persistence of the data-generating process (DGP), and diagnostic expectations. In our experiment, participants repeatedly forecast the contemporaneous and one-step-ahead values of a random variable. They are incentivized for accuracy and are fully informed about the DGP and its past realizations. We consider a signal-extraction treatment in which participants receive noisy private signals about the unobserved contemporaneous value, as well as a treatment without signals, and we vary the persistence of the AR(1) process. At the individual level, we find systematic overreaction even when the DGP is not persistent, and regardless of whether a signalextraction problem is present. By contrast, consensus forecasts exhibit underreaction when agents observe different private signals, consistent with an information-friction explanation. We then study the same behavioral mechanisms in a simple macroeconomic consumption model and find that misperceptions about the persistence of productivity generate sizable distortions in consumption dynamics and provide a better fit to U.S. consumption and productivity data than overconfidence, diagnostic expectations, or rational Bayesian learning.