Time Series Forecasting by Principal Component Methods
Mariano J. Valderrama, Ana Maria Aguilera, Juan Carlos Ruiz-Molina · COMPSTAT · 1998
On the basis of Functional Principal Component Analysis (FPCA), two forecasting approaches for time series are developed in this paper. The first one uses weighted multiple linear regression among principal components whereas the second one applies Kalman filtering on approximate state-space models. The forecasting performance of both methods is discussed on a real financial time-series.