Development and testing of a multivariate, seasonal ARMA(1,1) model

Abstract

A time series model of the ARMA class for seasonal, multisite applications is presented. Methods of estimating model parameters are developed and two methods (maximum likelihood and the method of moments) are compared. For model verification, a data generation experiment provides diagnostic checks at three levels: (1) properties of the model residuals; (2) preservation of short-term statistics (serial and cross-correlation); and (3) preservation of long-term characteristics (drought and reservoir storage properties). The model is then used to model bivariate, monthly river flow on the Yampa and White Rivers in northwest Colorado.