Do People Listen to Cassandra? Persuasion and Accuracy in Geopolitical Forecasts

In a world marked by persistent uncertainty, accurately predicting the future is vital. While certain people ("superforecasters") and practices (aggregation) can produce more accurate forecasts, the utility of these forecasts depends on whether they are persuasive enough to be heeded. This study investigates the relationship between persuasion and accuracy in geopolitical forecasts. In a forecasting competition, 153 forecasters predicted future global events and explained their rationales. Next, 474 evaluators rated these rationales for persuasiveness. Using a neural network-based language model, we found that although forecasting rationales contained valid cues for accuracy, evaluators were no more persuaded by accurate forecasts. Further analyses using large language models revealed that psychological attributes indicating expertise, trustworthiness, and emotional composure were associated with both persuasion and accuracy, while eloquence and confidence, though persuasive, were unrelated to accuracy. We conclude that people are sensitive to valid cues, but are misled by invalid ones, impairing overall calibration.