The Ratchet Effect Re-examined: A Learning Perspective

We study dynamic moral hazard where principal and agent are symmetrically uncertain about job difficulty. Since effort is unobserved, shirking leads the principal to believe that the job is hard, increasing the agent's continuation value. So deterring shirking requires steeper incentives, which induce the agent to over-work today, since he can quit if the principal believes that the job is easy. With continuous effort choices, no interior effort is implementable in the first period. The agent's continuation value function is non-differentiable and convex, since the principal makes the agent indifferent between his discrete (participation) choices in the second period. The problem can be solved if the agent's participation decision is made continuous, or if there are long-term commitments, and we provide conditions for the first order approach to work. However, the impossibility result recurs in other agency models that combine discrete and continuous choices.