Diffusion Bootstrap for High-Dimensional Linear Models
Abstract
Classical bootstrap methods can behave poorly in high-dimensional linear models: the pairs bootstrap often yields overly conservative inference, whereas the residual bootstrap can be anti-conservative, reflecting systematic failures in variance calibration. We propose a diffusion-based pairs bootstrap that replaces the empirical joint distribution with a learned generative law. We establish variance consistency under a score approximation assumption, using complementary SDE and PDE arguments. Counterexamples show that terminal convergence alone is insufficient for variance consistency. Experiments indicate that diffusion pairs bootstrap improves variance calibration and generally improves Type~I error calibration, including in settings not covered by our theory.
Cite
@article{arxiv.2607.24324,
title = {Diffusion Bootstrap for High-Dimensional Linear Models},
author = {Ce Liang and Wei Ma},
journal= {arXiv preprint arXiv:2607.24324},
year = {2026}
}