A random coefficient approach to seasonal adjustment of economic time series

Abstract A new seasonal adjustment procedure based on a random coefficient model that permits vector serial correlation in the coefficient errors is introduced. The principal advantages of the model are (i) it allows both deterministic and stochastic trend and seasonal components and their interactions to be simultaneously (rather than sequentially, by repeated filter application) identified and estimated (rather than assumed known), (ii) it allows continuous change in both first and second moments of all components ( a propo Anderson's (1971, pp. 46–49) argument in favor of moving average deterministic trend models for changing processes), and (iii) it makes explicit the identification problem associated with individual parameters (often concealed in sequential techniques) and establishes instead estimable functions appropriate for seasonal adjustment. In addition, the model is sufficiently general as to nest virtually every previous seasonal adjustment procedure as a subcase. The final section of the paper applies the procedure to a monthly demand deposit series for the U.S.