Una nota sulla distanza tra modelli ARIMA per serie storiche correlate

M. Corduas
In this paper we consider a distance measure between ARIMA models which has been proposed as a measure of structural diversity between time series. Specifically, the metric is defined on the space of linear, convergent and Gaussian processes as the Euclidean distance between the weights of the corresponding AR(oo) representations. We discuss how the distribution properties of that measure are modified by the presence of correlation between time series.