Short-Term Forecasting and Seasonal Adjustment

This paper explores the stochastic properties and prediction performance of several economic time series both before and after adjustment by the U.S. Bureau of the Census XII seasonal adjustment program. The results suggest that within the class of auto- regressive-integrated-moving average models, seasonally adjusted data do not lead to consistently improved predictions and in many circumstances produce forecasts which are less accurate than those produced using the unadjusted data.