摘要：Although the scalar components methodology used to build VARMA models is rather difficult, the VAR models application being easier in practice, the forecasts based on the first models have a higher degree of accuracy. This statement is demonstrated for variables like the 3-month Treasury bill rate and SHAPE * MERGEFORMAT the spread between the 10 year government bond yield, where the quarterly data are from the U.S. economy (horizon: first quarter of 2001 – second quarter of 2013). It was used a better measure of accuracy than those used in literature till now, the generalized forecast error of second moment, that was adapted to measure relative accuracy
关键词：macroeconomic forecasts; VARMA models; accuracy; scalar components methodology; full information maximum likelihood; canonical correlation.
