Overconfident forecasters and the impact of inflation information: evidence from a randomized survey experiment

We exploit a randomized information intervention in a quarterly survey of Italian firms to study how access to recent inflation data affects deviations from the full-information rational expectations (FIRE) benchmark in managers' inflation forecasts. Treated firms receive the latest inflation reading before submitting their forecasts and, relative to non-treated firms, display less underreaction at the consensus level and less overreaction at the individual level, moving forecasts closer to the FIRE benchmark. A model that combines noisy information with overconfidence in private information provides the best overall fit to the data, outperforming alternative frameworks featuring diagnostic expectations, internal cognitive constraints, or over-extrapolation. Intuitively, with noisy information, average forecasts react sluggishly to news, so an informative public signal speeds up aggregate updating and reduces consensus underreaction. With overconfidence, managers overweight private signals and the public signal shifts weight away from private information, attenuating individual overreaction.