By modeling the market for IPOs as a repeated game with imperfect monitoring, we establish that collusion among underwriters explains the concentration of spreads at 7%, along with other characters of the data on spreads. Furthermore, the structure of optimal spreads in the model explains the...