A note on normalization schemes: The case of generalized forecast error variance decompositions