This paper introduces “Terraced” Vector Autoregressive (VAR) models, an innovative twist on traditional VAR modeling, which allows the econometrician to simultaneously forecast both exogenous and endogenous variables and the confidence intervals around those forecasts. In an application of our Terraced VAR framework, we have estimated coincident indices of economic activity for the United States, New York State and the six largest metropolitan areas of New York State and incorporated them into Terraced VARs, which forecast the unemployment rate, total non-farm employment, real wages and average hours worked in manufacturing in those regions.
Doviak, E., & MacDonald, S. (2010). Forecasting the New York State Economy with" Terraced" VARs and Coincident Indices. New York Economic Review, 41(1), 14-34. http://econpapers.repec.org/RePEc:nye:nyervw:v:41:y:2010:i:1:p:14-34