Journal Article

PIVOTAL ESTIMATION VIA SQUARE-ROOT LASSO IN NONPARAMETRIC REGRESSION

Belloni, Alexandre, Wang, Lie, Chernozhukov, Victor

2014

Overview

We propose a self-tuning $\\sqrt {Lasso} $ Lasso method that simultaneously resolves three important practical problems in high-dimensional regression analysis, namely it handles the unknown scale, heteroscedasticity and (drastic) non-Gaussianity of the noise. In addition, our analysis allows for badly behaved designs, for example, perfectly collinear regressors, and generates sharp bounds even in extreme cases, such as the infinite variance case and the noiseless case, in contrast to Lasso. We establish various nonasymptotic bounds for $\\sqrt {Lasso} $ including prediction norm rate and sparsity. Our analysis is based on new impact factors that are tailored for bounding prediction norm. In order to cover heteroscedastic non-Gaussian noise, we rely on moderate deviation theory for self-normalized sums to achieve Gaussian-like results under weak conditions. Moreover, we derive bounds on the performance of ordinary least square (ols) applied to the model selected by $\\sqrt {Lasso} $ accounting for possible misspecification of the selected model. Under mild conditions, the rate of convergence of ols post $\\sqrt {Lasso} $ is as good as $\\sqrt {Lasso's} $ rate. As an application, we consider the use of $\\sqrt {Lasso} $ and ols post $\\sqrt {Lasso} $ as estimators of nuisance parameters in a generic semiparametric problem (nonlinear moment condition or Z-problem), resulting in a construction of $\\sqrt n - consistent$ and asymptotically normal estimators of the main parameters.

• Publisher: Institute of Mathematical Statistics,The Institute of Mathematical Statistics
• Subject: 62G05 / 62G08 / 62G35 / Approximation / Eigenvalues / Estimators / generic semiparametric problem / Impact factor / Least squares / Linear regression / Mathematical models / Mathematical problems / model selection / Non Gaussianity / non-Gaussian heteroscedastic / nonlinear instrumental variable / Normal distribution / Oracles / Parameter estimation / Pivotal / Regression analysis / sqrt{n}-consistency and asymptotic normality after model selection / square-root Lasso / Statistical variance / Studies / Z-estimation problem