We study the functional ARMA(p, q) and a corresponding approximating vector model, based on functional PCA. We investigate the structure of the multivariate vector process and derive conditions for the existence of a stationary solution to both the functional and the vector model equation. We then use the stationary vector process to predict the functional process, and compare the resulting predictor to the functional best linear predictor proposed. We derive bounds for the error due to dimension reduction. We conclude by applying functional ARMA processes for the modelling and prediction of highway traffic data.      «

We study the functional ARMA(p, q) and a corresponding approximating vector model, based on functional PCA. We investigate the structure of the multivariate vector process and derive conditions for the existence of a stationary solution to both the functional and the vector model equation. We then use the stationary vector process to predict the functional process, and compare the resulting predictor to the functional best linear predictor proposed. We derive bounds for the error due to dimensio...     »