Process Components for Multivariate Time Series Analysis
John G. Crookes, K. Croston, Panayiotis Sypsas · Journal of the Operational Research Society · 1980
A technique for analysing multiple time series has been developed. Its efficiency on a number of types of series has been tested. The method involves seeking time dependent components analogous to principal components but using criteria based on time dependent functions of the resulting components. The functions are those used in the Box-Jenkins forecasting system and results in components ideally suited to forecasting using that system. The initial results are extremely encouraging and the paper consists of a description of the method together with a summary of the results for a wide range of series types. The system is offered as a diagnostic tool useful for those who need to identify the underlying sources of variation present in large volumes of data.