A0495
Title: Estimating High-Dimensional Conditional Dependence Coefficients via Synthetic Control
Authors: Tsung-Chih Lai - National Chung Cheng University (Taiwan) [presenting]
Jia-Han Shih - National Sun Yat-sen University (Taiwan)
Abstract: The curse of dimensionality is addressed in estimating the dependent coefficient introduced by a past study, which measures the dependence between a scalar response Y and a potentially high-dimensional covariate vector X. The approach begins by observing that the problem of multivariate dependence between (Y, X) can be reformulated as a bivariate problem involving (Y, Y'), where Y and Y' share the same conditional distribution given X. Viewing (Y, Y') as potential outcomes, a jackknife synthetic control procedure is developed that constructs Y' for each observation through a weighted average of the remaining sample, with weights chosen to closely match on X. The dependence coefficient, which corresponds to Spearman's foot rule between Y and Y' in the continuous case, is then estimated using the sample counterpart. Simulation studies show that the proposed method performs exceptionally well in high-dimensional and even low-dimensional settings.
