We study risk-shifting behavior in a laboratory experiment, a setup that overcomes methodological hurdles faced by empiricists in the past. The participants are high-level managers. We observe risk shifting in a simple setup, but less so in a setup with a continuation value. Reputation effects also reduce risk shifting. When combined, a continuation value and reputation effects eliminate risk shifting. Our findings shed light on environments in which risk shifting is unlikely to happen, and why earlier studies produced conflicting results. In particular, our findings show that managers' concerns with their own reputations are an important factor that mitigates risk shifting.