In this paper, we match data on student performance in a multiple-choice exam with data on student risk preferences that are extracted from a classroom experiment. We find that more-loss-averse students leave more questions unanswered and perform worse in the multiple-choice exam when giving an...