We propose a computationally efficient method to construct nonparametric, heteroskedastic prediction bands for uncertainty quantification, with or without any user-specified predictive model. The data-adaptive prediction band is universally applicable with minimal distributional assumptions, with strong non-asymptotic coverage properties, and easy to implement using standard convex programs. Our approach can be viewed as a novel variance interpolation with confidence and further leverages techniques from semi-definite programming and sum-of-squares optimization. Theoretical and numerical performances for the proposed approach for uncertainty quantification are analyzed.

More on this topic

BFI Working Paper·Jun 18, 2025

Innovator Networks Within the Firm and the Quality of Innovation

Michael Gibbs, Friederike Mengel, and Christoph Siemroth
Topics: Uncategorized
BFI Working Paper·Jun 17, 2025

The Social Desirability Atlas

Leonardo Bursztyn, Ingar K. Haaland, Nicolas Röver, and Christopher Roth
Topics: Uncategorized
BFI Working Paper·Jun 10, 2025

Measuring Markets for Network Goods

Leonardo Bursztyn, Matthew Gentzkow, Rafael Jiménez-Durán, Aaron Leonard, Filip Milojević, and Christopher Roth
Topics: Uncategorized