EcoSta 2021: Start Registration
View Submission - EcoSta2021
A0493
Title: Inference using nuclear-norm penalized estimator and its applications Authors:  Jungjun Choi - Rutgers University (United States) [presenting]
Hyukjun Kwon - Rutgers University (United States)
Abstract: The inference theory of the (debiased) nuclear norm penalized estimator of the latent approximate low-rank matrix is studied when the observation matrix is subject to missingness. The alpha test in empirical asset pricing and the average treatment effect estimator are provided as applications. Although the nuclear norm penalization causes shrinkage bias which makes inference infeasible, our debiasing procedure successfully removes it, and the resulting debiased estimator attains the asymptotic normality. Unlike other debiasing schemes for the inference using the nuclear norm penalized estimator, our debiasing method does not resort to sampling splitting. So our estimation step is simple, and we can avoid some undesirable properties of sample splitting. In addition, the heterogeneous missing probability is allowed, and the inverse probability weighting is used, which improves the estimation performance by treating units with different missing probabilities in an equal manner.