A version of the herding prediction model with a rational-expectations flavor is reexamined in the light of incentive theory. The welfare loss at the market solution with respect to the incentive-efficient solution can be decomposed into an information externality term minus an incentive cost term. It is found that the inefficiency of herding at the market solution is low when the cost of providing incentives is high. When the cost of providing incentives is low (and this happens when prior information is diffuse), the incentive-efficient solution approaches the team solution that fully internalizes the information externality. Then the herding problem at the market solution is at its worst.