Seasonal Adjustment and Relations between Variables

Abstract

This article studies the effect of official seasonal adjustment procedures on the relations between variables. By considering time-invariant linear filters, and in particular a linear approximation to the Census Method II adjustment program, the effect of adjusting one or both of the variables in a distributed lag relation is examined, and the distortions which can arise are described. Applying the actual (nonlinear) adjustment procedure to artificial data indicates that at least for the particular x-series used, the results of the linear filter analysis provide a good guide to the behavior of estimates obtained from data adjusted by the official method.