On Prediction with Functional Autoregressive Processes

Functional autoregressive (FAR) processes have become an important tool in functional data analysis for modeling and forecasting data represented as continuous curves or functions over time. In contrast to classical autoregressive models, which deal with scalar or vector-valued observations, FAR models consider each observation as a function belonging to an infinite-dimensional Hilbert space. This representation allows the model to capture the temporal dependence and smooth structure of complex functional data more effectively.

This study focuses on the theory and application of FAR processes for prediction problems. The proposed methodology combines functional data analysis techniques with autoregressive modeling to estimate the relationship between successive functional observations. To address the high dimensionality of functional data, functional principal component analysis (FPCA) is employed to extract the main modes of variation and reduce computational complexity. The autoregressive operator is then estimated using these principal components to construct prediction models.

The performance of the FAR model is evaluated through simulated and real-world datasets from areas such as climate studies, electricity consumption, finance, and biomedical signal analysis. Experimental results indicate that FAR processes provide accurate and reliable forecasts by effectively capturing smooth temporal patterns and complex dependencies within the data. The study concludes that functional autoregressive processes constitute a flexible, robust, and efficient framework for forecasting functional time-series data in modern statistical applications.