This paper explores the hypothesis that the seasonal patterns of macro
economic variables vary with expansions and contractions. Graphical te
chniques and generalized predictive tests for structural stability are
used to identify and test patterns of changing seasonality. A Monte C
arlo exercise shows the power of the tests against interesting alterna
tives. The empirical results suggest that seasonal patterns are unstab
le and that in many cases changes are linked to the stages of the busi
ness cycle. The forecasting costs incurred by treating seasonality as
constant are discussed and evaluated.