This article presents a comparison of forecasting performance for a va
riety of linear and nonlinear time series models using the U.S. unempl
oyment rate. Our main emphases are on measuring forecasting performanc
e during economic expansions and contractions by exploiting the asymme
tric cyclical behavior of unemployment numbers, on building vector mod
els that incorporate initial jobless claims as a leading indicator, an
d on utilizing additional information provided by the monthly rate for
forecasting the quarterly rare. Comparisons are also made with the co
nsensus forecasts from the Survey of Professional Forecasters. In addi
tion, the forecasts of nonlinear models are combined with the consensu
s forecasts. The results show that significant improvements in forecas
ting accuracy can be obtained over existing methods.