Shape-preserving prediction for stationary functional time series (Q2233562)

This article presents a novel method for prediction of stationary functional time series, in particular for trajectories that share a similar pattern but display variable phases. More precisely, the authors develop a new prediction method for stationary functional time series that display a common pattern. The method is the first to incorporate functional registration into prediction of functional time series. The prediction algorithm jointly predicts the amplitude and phase components. These two predicted components are then combined to form the final prediction. To overcome the classical limitations, the authors develop a shape-preserving (SP) prediction method that incorporates both vertical and horizontal variations. The SP method has two main advantages. First, if the curves display a common pattern and significant phase variation, considering the vertical variation only would lead to the loss of main features. Comparatively, the new methodology separates amplitude and phase components first, thus the SP method can preserve the pattern better. Second, the SP method is ``natural'' because (1) it is based on a bijective transformation, thus no additional adjustments are needed to transform the data into a warping function, which avoids bias; (2) the method does not directly apply linear models to nonlinear objects, thus making the prediction natural and avoiding extremely small values. As an illustration of the effectiveness of the SP method the authors consider the analysis of the non-metanic hydrocarbons concentration. The analysis shows that the prediction by the SP method captures the common pattern better than other existing prediction methods along with a competitive prediction accuracy.