TIME SERIES MODELS FOR COMPOSITIONAL DATA

Compositional time series are becoming widely applicable in many situations. Differing from traditional methods,, that compositional data are firstly removed the constraints, and then standard time series techniques are used on the transformed data, we consider the compositional data series models built on the original compositional data. In this paper, C-white noise and C-stationary processes, based on inner products of compositional data in Simplex space, are defined. The C-AR model, C-MA model, C-ARMA model and C-ARIMA model processes are then deduced.