📝 SummarXiv

Abstract

Ensemble forecasts have become a cornerstone of large-scale disease response, underpinning decision making at agencies such as the US Centers for Disease Control and Prevention (CDC). Their growing use reflects the goal of combining multiple models to improve accuracy and stability versus using a single model. However, recent experience shows these benefits are not guaranteed. During the COVID-19 pandemic, the CDC's multi-model forecasting ensemble outperformed the best single model by only 1%, and CDC flu forecasting ensembles have often ranked below multiple individual models. This raises a key question: why are ensembles underperforming? We posit that a central reason is that both model developers and ensemble builders typically focus on stand-alone accuracy. Models are fit to minimize their own forecasting error, and ensembles are often weighted according to those same scores. However, most epidemic forecasts are built from a small set of approaches and trained on the same surveillance data, leading to highly correlated errors. This redundancy limits the benefit of ensembling and may explain why large ensembles sometimes deliver only marginal gains. To realize the potential of ensembles, both modelers and ensemblers should prioritize models that contribute complementary information rather than replicating existing approaches. Ensembles built with this principle in mind move beyond size for its own sake toward true diversity, producing forecasts that are more robust and more valuable for epidemic preparedness and response.