Learning in Dynamic Incentive Contracts

We derive the optimal dynamic contract in a continuous-time principal-agent setting, in which both investors and the agent learn about the firm’s profitability over time. Because investors learn about the firm’s future profitability from current output, which also depends upon the agent’s actions, deviations by the agent distort investors’ beliefs. We characterize the optimal contract, and show that the performance sensitivity of the agent’s payoff is set to account for both moral hazard and asymmetric information. We show that the optimal contract can be implemented by compensating the agent with equity and allowing him to manage the firm’s cash reserves by setting the its payout policy. Under this contract, the firm accumulates cash until it reaches a target balance that depends on the agent’s perceived productivity. Once this target balance is reached, the firm starts paying dividends equal to its expected future earnings, while any temporary shocks to earnings either add to or deplete the firm’s cash reserves. The firm is liquidated if it exhausts its cash reserves. We also show that once the firm initiates dividends, these dividends are smooth relative to earnings, and that liquidation is first-best, despite the agency problem. * Stanford University and UC Berkeley/NYU. This research is based on work supported in part by the NBER and the National Science Foundation under grant No. 0452686.