|
Abstract:
|
Modern advancement of technology permits us to accumulate more complicated data than before. Unlike the traditional time series where only a scalar or a vector is observed at each time point, in functional time series, a curve is observed at each time point. Correlation exists among the curves observed at different time points. In this paper, we consider a functional autoregressive (FAR) model with general order which is a generalization of the traditional AR model. To fit the FAR model and obtain the estimate of coefficient functions, we propose a signal compression procedure. To determine the optimal tuning parameters and optimal order of FAR model, we propose a window-shifting cross-validation procedure. We compare our proposed procedure to recently developed one using both simulated data and real data, which illustrate the good predictive performance of our method.
|
