A0689
Title: Estimation of single index models in moderately high dimension
Authors: Kazuma Sawaya - The University of Tokyo (Japan) [presenting]
Yoshimasa Uematsu - Hitotsubashi University (Japan)
Masaaki Imaizumi - The University of Tokyo (Japan)
Abstract: A new estimator is proposed for semiparametric single index models in the moderately high-dimensional regime, where the number of covariates $p$ grows proportionally with the sample size $n$. Asymptotic unbiasedness and asymptotic normality of the estimator without the sparsity condition of the true parameter are also guaranteed. The estimation is based on the deconvolution method and the generalized approximate message-passing algorithm. Numerical simulations show the validity of our theory.
