摘要
We propose a Sparse-Group Independent Component Analysis (SG-ICA) method to extract independent factors from high dimensional multivariate data. The method provides a unified and flexible framework that automatically identifies the number of factors and simultaneously estimates a sparse loading matrix, enables us to discover important features and offers improved interpretability of the estimators. We establish the consistency and asymptotic normality of the loading matrix estimator, demonstrate its finite sample performance with simulation studies, and illustrate its application using the daily US Overnight Index Swap rates from Oct 2011 to Mar 2015 with 15 maturities ranging from 1 week to 30 years. With higher efficiency of extracting factors, the forecasting performance of the SG-ICA is remarkably better than the popular parametric DNS model in an era of quantitative easing with short-term interest rate being close to zero.