文摘
Functional principal component regression for modeling and forecasting a sliced functional time series is considered. When new data points in the most recent curve are observed, dynamic updating methods are introduced to improve point and interval forecast accuracies. Via simulation and a data set, we found the vector autoregressive model outperforms the autoregressive integrated moving average model for forecasting principal component scores. Via simulation and a data set, a robust functional principal component analysis ought to be used when functional curves contain outliers.