Finding hidden structure of sparse longitudinal data via functional Eigenfunctions

Abstract

ABSTRACT In this research, we are interested in finding the hidden dependence structure of sparse longitudinal data. Finding the hidden dependence structure of sparse longitudinal data is difficult due to the starting and end times being different. We propose that finding the directional dependence structure of the eigenfunctions by sparse functional principal component analysis (FPCA) may be a good alternative solution to find the hidden dependence structure of sparse longitudinal data. To verify this idea, we apply sparse FPCA to simulated data and two real datasets, wage sparse longitudinal data and Korea composite stock price index (KOSPI) high-frequency minute tick data and then apply vine copula and copula dynamic conditional correlation with asymmetric GARCH model to the functional eigenfunctions from FPCA.