Error density estimation in high-dimensional sparse linear model (Q2304251)

A high-dimensional sparse linear regression model is studied, where the covariates are random and the number \(p\) of covariates may be larger than the sample size \(n.\) The number of nonzero coefficients \(s\) satisfies \(s=O(n^\gamma)\) with \(0\le \gamma <1.\) An improved two-stage refitted cross-validation procedure by random splitting technique is used to obtain the residuals of the model, and then the kernel density method is applied to estimate the error density \(f\). For the estimator, the consistency, asymptotic normality and the law of the iterated logarithm are proven, as well as the asymptotic normality of the estimator for a linear integral functional of \(f.\) Moreover, the relationship is established between the sparsity coefficient \(\gamma\) and the convergence rate of the density estimator \(\hat{f}.\) Simulation results show a good performance of \(\hat{f}.\) A real data example illustrates the method.