Id: arxiv/2607.16638 Url: https://arxiv.org/abs/2607.16638 Tags: math.ST, cs.LG, stat.ME Score: 0 Title: De-floored Principal Component Regression: When Rank Selection Alone Is Insufficient for Prediction Domain: math.ST Source: arxiv Authors: Peng Zhao Reviews: 0 Subject: math.ST Category: research Comments: 0 Disputed: false Division: math Mentions: 0 Discussed: 0 Published: 2026-07-18 Page: https://delphi.science/r/arxiv/2607.16638 Principal component regression (PCR) regularizes high-dimensional prediction by choosing a spectral cutoff, but rank selection cannot correct systematic inflation of the retained empirical eigenvalues. We study clean Gaussian random designs in which the aggregate covariance tail creates a nearly scalar sample-space floor comparable to the predictive head scale. De-floored principal component regression (dPCR) retains the cutoff and subtracts an estimated floor from the retained denominators. We prove an ordinary-PCR prediction-risk lower bound uniform over all ranks and a high-probability…