What is it about?
Forecasting models are often ranked by mean squared error (MSE), but the lowest-error forecasts can be overly smooth. We characterize the trade-off between accuracy and realism, showing that a small increase in MSE can often produce a much larger improvement in how well forecasts reflect real future variability.
Featured Image
Photo by Adam Śmigielski on Unsplash
Why is it important?
Ranking models by MSE alone can favor oversmoothed forecasts even when another model has almost the same error and is much more realistic. Across nine datasets, allowing up to 5% higher MSE produced a median 17.3% improvement in marginal realism.
Perspectives
The most satisfying result for me was seeing forecasting strategies occupy clearly different parts of the accuracy–realism frontier. My PhD has focused on understanding these strategies, and it was exciting to find such a clear interpretation: they are not simply better or worse, but can favour different trade-offs between accuracy and realistic variability.
Riku Green
University of Bristol
Read the Original
This page is a summary of: Expectations vs. Realities: The Cost of MSE-Optimal Forecasting Under Conditional Uncertainty, August 2026, ACM (Association for Computing Machinery),
DOI: 10.1145/3770855.3818087.
You can read the full text:
Contributors
The following have contributed to this page
