Employs weights based on relative errors produced by two or more models to formulate a consensus forecast. This method better captures changing patterns through time and sharply reduces size of forecast error.