The full insurance hypothesis states that shocks to the firm's performance do not affect workers' compensation. In principal-agent models withmoral hazard, firms trade off insurance and incentives to induce workers to supply the optimal level of effort. We use a long panel of matched employer-employee data to test the theoretical predictions of principal-agent models of wage determination in a general context where all types of workers, not only CEOs, are present. We allow for both transitory and permanent shocks to firm performance and find that firms are willing to fully absorb transitory fluctuations in productivity but insure workers only partially against permanent shocks. Risk-sharing considerations can account for about 10 percent of overall earnings variability, the remainder originating in idiosyncratic shocks. Finally, we show that the amount of insurance varies by type of worker and firm in ways that are consistent with principal-agent models but are hard to reconcile with competitive labor market models, with or without frictions.