How do departures of individual beliefs from Bayesian rationality impact welfare and informational efficiency in financial markets? We allow agents' posterior beliefs to depart from the rational expectations hypothesis and to over/under-react to news (i.e.to changes in fundamentals and/or prices). We find that both in a financial market with sequential trading and unit demand and supply, as well as in a financial market with simultaneous trading, with volume and unbounded signals, over-reaction has a positive informational externality, while under-reaction exacerbates any informational friction. Over-reaction mitigates individual losses due to non-Bayesian beliefs. Because agents over-react to news, prices aggregate more efficiently private information, relative to the rational case. As a consequence, over-reaction decreases the likelihood of informational cascades, increases the informational content of prices, and it increases welfare.