How Better Information Can Reduce the Credibility of Experts' Advice

Two experts make yes/no predictions about whether an event, such as an economic crisis, will occur. They receive private signals about the likelihood of the event that are much more informative than the coarse recommendations they can ultimately convey. The experts may care about both absolute and relative performance. We show that when both experts receive strictly better information, the error rates of their all their unanimous predictions in the new equilibrium may be higher than before the information improvement. This unintuitive phenomenon raises the question of how society can optimally incentivize such experts. We find that, without the possibility of huge transfers, society's ability to extract good advice or even to interpret imperfect advice is severely limited