A regression model that uses more than one explanatory variable. The canonical weights provide an index for the dependent variables but without a theory. Despite a number of attempts, it seems to have no value for forecasting (e.g., Fralicx and Raju, 1982, tried but failed).