Skip to content

A High Dimensional Wild Bootstrap Max-Test for Detecting the Presence of Significant Predictors

Home / Papers / A High Dimensional Wild Bootstrap Max-Test for Detecting the Presence of Significant Predictors

A High Dimensional Wild Bootstrap Max-Test for Detecting the Presence of Significant Predictors

Authors: Jonathan B. Hill Date: 2026-04-30 Paper ID: openalex:2604.27813

Summary

This paper introduces a block bootstrap max-test designed for detecting significant predictors in high-dimensional settings where the number of covariates p grows exponentially relative to sample size n. By employing a max-statistic over marginal regression parameters under a physical dependence framework, the method bypasses the need for complex covariance estimation and Bonferroni corrections. The approach provides robust size control and effective signal detection for weakly dependent, non-stationary time series data, as demonstrated through empirical analysis of the VIX volatility index.

Key Contributions

  • Proposes a block bootstrap max-test for high-dimensional correlation screening where the number of covariates p grows exponentially with n.
  • Eliminates the need for explicit covariance matrix estimation and post-estimation Bonferroni corrections by using a max-statistic approach over marginal regression parameters.
  • Demonstrates robust size control and performance against sparse signals under weak dependence and non-stationarity, validated via VIX index empirical application.

Open Questions & Future Work

Archivist Review

The paper proposes a high-dimensional bootstrap max-test for predictor screening. Since the method is primarily a statistical inference framework rather than a novel representation or forecasting architecture, no specific ‘concept’ stood out as a generally reusable temporal method worthy of a permanent vault entry. The open questions, however, highlight substantial theoretical bottlenecks in high-dimensional dependence analysis and were approved for tracking.

Approved Open Questions

  • Adaptive Covariate Selection Under Dependence: Developing such an index would bridge the gap between simple marginal screening and complex post-model-selection inference, potentially enhancing the detection of informative predictors in dependent data.
  • Weakening Moment Conditions for Screening: Weakening these moment requirements is technically important to ensure the theory covers a wider array of financial and economic models where extreme observations are common.

Metadata & Links

url
https://arxiv.org/abs/2604.27813
paper_id
2604.27813
paper_source
openalex
domain
time-series
tags
architectures
datasets
concept_slugs
dataset_slugs
skill
TimeSeriesSkill
created_at
2026-05-03T06:03:34Z