A1108
Title: Dynamic subgroup analysis on heterogeneous regression model
Authors: Haowen Zhou - University of Virginia (United States) [presenting]
Abstract: In recent years, the heterogeneous effect model, rather than a conventional homogeneous effect model, has become prevalent in various areas such as precision medicine and market segmentation. Yet it remains challenging to deal with such heterogeneity changing over time. To fill this gap, we propose a dynamic subgrouping framework on a heterogeneous regression model, which can capture the temporal pattern on heterogeneous covariates-effects. We impose the novel multidirectional separation penalty on the individualized covariates-effects to pursue subgroups of individuals dynamically while leveraging the temporal pattern of subpopulations by modeling the subgroup centers with smoothing splines. In contrast to all existing approaches, we allow the individuals to change their underlying subgroup memberships over time. We lay out the theoretical framework for the proposed model and estimates. An efficient ADMM algorithm with computational scalability is developed for model estimation. The outperformance of the proposed model has been validated by simulation studies and empirical data analysis in the stock market.
