Whether high-stakes exams such as the SAT or College Board AP exams should penalize incorrect answers is a controversial question. In this paper, we document that penalty functions can have differential effects depending on a student's risk tolerance. Moreover, literature shows that risk aversion tends to vary along other areas of concern such as race, gender, nationality, and socioeconomic status. In this article, we simulate Item Response Theory (IRT) data with and without a wrong answer penalty. In the presence of mild risk aversion, we find that students omit 12% more items than risk neutral individuals with identical ability. This translates into a nearly 2% difference in sum scores between the risk neutral and risk averse groups. We also find that penalty functions result in noisier estimates of student ability. These findings suggest that random guessing penalties should not be used in most circumstances, particularly for learning platforms. [This paper was published in: "Companion Proceedings 9th International Conference on Learning Analytics & Knowledge (LAK19)."]