Abstract:We propose a two-step procedure to model and predict high-dimensional functional time series, where the number of function-valued time series $p$ is large in relation to the length of time series $n$. Our first step performs an eigenanalysis of a positive definite matrix, which leads to a one-to-one linear transformation for the original high-dimensional functional time series, and the transformed curve series can be segmented into several groups such that any two subseries from any two different groups are uncorrelated both contemporaneously and serially. Consequently in our second step those groups are handled separately without the information loss on the overall linear dynamic structure. The second step is devoted to establishing a finite-dimensional dynamical structure for all the transformed functional time series within each group. Furthermore the finite-dimensional structure is represented by that of a vector time series. Modelling and forecasting for the original high-dimensional functional time series are realized via those for the vector time series in all the groups. We investigate the theoretical properties of our proposed methods, and illustrate the finite-sample performance through both extensive simulation and two real datasets.

What problem does this paper attempt to address?

This paper mainly discusses the modeling and prediction problems of high-dimensional functional time series. In the high-dimensional setting, when the number of function value time series p is relatively large compared to the length of the time series n, new challenges arise. The traditional approach is to first extract features for each component series, such as through functional principal component analysis (FPCA) or dynamic FPCA, and then use vector autoregressive models to handle these vector time series. However, this approach may not be very effective as it does not fully utilize the information between different component series. The paper proposes a two-step approach to address this problem. Firstly, the original high-dimensional functional time series is transformed into a segmented form through linear transformation, so that the sub-series within each segment are mutually independent at all time lags. This step is called segment transformation, which is achieved by solving the eigenvalues of a positive definite matrix, and the transformed sequences are rearranged to identify potential group structures. Secondly, finite-dimensional dynamic structures are established for each transformed sub-series, utilizing the non-zero lag autocovariance functions for dimension reduction to better utilize the sequence correlation information of the time series. The paper demonstrates the superiority of this approach on real-world datasets, such as global age-specific mortality data, showing that predicting the transformed sub-series is more accurate than directly predicting the original series. In addition, the paper proposes a new function thresholding procedure that ensures estimation consistency in high-dimensional settings and establishes theoretical properties. In summary, the main contributions of the paper include: 1. Introducing a segment transformation method that converts the correlation between sequences into autocorrelation within each segment, simplifying modeling and improving prediction performance. 2. Developing a variation-based multivariate FPCA method to identify and model the finite-dimensional dynamic structures of each transformed sub-series. 3. Providing theoretical analysis that demonstrates the excellent properties of the proposed method in high-dimensional cases and validating the effectiveness of the method through simulations and real-world data.

On the modelling and prediction of high-dimensional functional time series

Factor modelling for high-dimensional functional time series

Covariance Function Estimation for High-Dimensional Functional Time Series with Dual Factor Structures

High-dimensional functional time series forecasting: An application to age-specific mortality rates

Factor Models for High‐Dimensional Functional Time Series II: Estimation and Forecasting

Forecasting High-Dimensional Functional Time Series: Application to Sub-National Age-Specific Mortality

Factor Models for High‐Dimensional Functional Time Series I: Representation Results

A Non-linear Function-on-Function Model for Regression with Time Series Data

A robust functional time series forecasting method

Nonlinear prediction of functional time series

Deep Functional Factor Models: Forecasting High-Dimensional Functional Time Series via Bayesian Nonparametric Factorization

Graphical Principal Component Analysis of Multivariate Functional Time Series

A Functional approach for Two Way Dimension Reduction in Time Series

Modeling High-Dimensional Time Series: A Factor Model with Dynamically Dependent Factors and Diverging Eigenvalues

Variable Selection and Minimax Prediction in High-dimensional Functional Linear Model

Shape-Preserving Prediction for Stationary Functional Time Series

Parallel Extraction of Long-term Trends and Short-term Fluctuation Framework for Multivariate Time Series Forecasting

Partially Functional Linear Regression in High Dimensions

Functional volatility forecasting

Functional Relation Field: A Model-Agnostic Framework for Multivariate Time Series Forecasting