ℓ[subscript 1]-penalized quantile regression in high-dimensional sparse models
Name
Belloni-2010-`1-PENALIZED QUANTILE REGRESSION IN HIGH-DIMENSIONAL.pdf
Size
532.69 KB
Format
Adobe PDF
Checksum (MD5)
7b17caf66e4f7130664d5a47a405716b
Author(s)
Belloni, Alexandre
Chernozhukov, Victor V.
Date Issued
February 2011
Journal
The Annals of Statistics
Publisher
Institute of Mathematical Statistics
Citation
Belloni, Alexandre, and Victor Chernozhukov. “ℓ 1 -penalized quantile regression in high-dimensional sparse models.” The Annals of Statistics 39, no. 1 (February 2011): 82-130. 2011 © Institute of Mathematical Statistics
Version
Final published version
Abstract
We consider median regression and, more generally, a possibly infinite collection of quantile regressions in high-dimensional sparse models. In these models, the number of regressors p is very large, possibly larger than the sample size n, but only at most s regressors have a nonzero impact on each conditional quantile of the response variable, where s grows more slowly than n. Since ordinary quantile regression is not consistent in this case, we consider ℓ[subscript 1]-penalized quantile regression (ℓ[subscript 1]-QR), which penalizes the ℓ[subscript 1]-norm of regression coefficients, as well as the post-penalized QR estimator (post-ℓ[subscript 1]-QR), which applies ordinary QR to the model selected by ℓ[subscript 1]-QR. First, we show that under general conditions ℓ[subscript 1]-QR is consistent at the near-oracle rate √s/n√log(p v n), uniformly in the compact set U C (0,1) of quantile indices. In deriving this result, we propose a partly pivotal, data-driven choice of the penalty level and show that it satisfies the requirements for achieving this rate. Second, we show that under similar conditions post-ℓ[subscript 1]-QR is consistent at the near-oracle rate √s/n√log(p v n), uniformly over U, even if the ℓ[subscript 1]-QR-selected models miss some components of the true models, and the rate could be even closer to the oracle rate otherwise. Third, we characterize conditions under which ℓ[subscript 1]-QR contains the true model as a submodel, and derive bounds on the dimension of the selected model, uniformly over U; we also provide conditions under which hard-thresholding selects the minimal true model, uniformly over U.
MIT Department
Massachusetts Institute of Technology. Department of Economics
Terms of Use
Article is made available in accordance with the publisher's policy and may be subject to US copyright law. Please refer to the publisher's site for terms of use.
Persistent DSpace Link
DOI of Published Version
https://doi.org/10.1214/10-AOS827
