This paper studies the provision of incentives in a universal bank. This is regarded as a (common) agent serving different clients with potentially conflicting interests: for example, it may buy assets on behalf of investors and sell assets on behalf of issuing firms. The clients offer incentive schemes to the bank and they behave non-cooperatively. The bank decides a level of effort and, when firewalls are absent, a level of collusion, modelled as a costly and unproductive redistribution of wealth among the clients. The main conclusion is that in the absence of firewalls the equilibrium incentive schemes are steeper. This means that the level of effort is higher and may compensate the (ex post) inefficiency of collusion. Moreover, this is shown not to hold in the presence of one naive player who does not recognize the existence of the conflict of interest. The model allows to draw conclusions about the desirability of firewalls or of softer measures like the imposition of transparency requirements.
