Sunk Cost Fallacy Is Associated With Biased Predictions
Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Overly optimistic predictions may cause people to persist too long in investments, reflecting the sunk cost fallacy. Past literature indicates that people use recent and longer-term trends while making reward-related predictions. However, it remains unclear how trend shape (e.g., linear, exponential, asymptotic) influences the timing and accuracy of these predictions. This information may correspondingly interact with individual factors (e.g., socioeconomic status, age, gambling tendencies, geographic income inequality) to influence decision making. We conducted two experiments to understand how different trends impact predictions. In Experiment 1 (N=159), participants completed a novel stock market task in which they predicted future stock prices based on exponential trends. Participants could choose to sell their stock at any point over ten turns or hold it to earn a fixed final bonus. We found that participants had lower earnings and delayed selling poor stocks when trends initially started slowly but accelerated rapidly. Experiment 2 (N=360) extended this paradigm by comparing exponential and inverse exponential trends using a modified Gold Mine task. Participants decided whether a mine would yield above or below a threshold, paying for additional turns to gather more information. We observed a strong optimism bias for fast-starting trends, leading participants to persist longer and exhibit the sunk cost fallacy. Additionally, psychosocial factors moderated this behavior: older individuals with lower socioeconomic status, and those with greater gambling tendencies residing in areas with higher income inequality, exhibited greater optimism with faster-starting trends. These results indicate that biased predictions stemming from the perceived nature of trends contribute significantly to the sunk cost fallacy. Our findings clarify mechanisms underlying predictive behavior and may help address biases in predictions.