Seasonality in Dynamic Regression Models

We examine the implications of treating seasonality as an unobserved component which changes slowly over time. This approach simplifies the specification of dynamic relationships by separating non-seasonal from seasonal factors. We illustrate this approach using the consumption model of Davidson et al. (1978), estimating a stable error correction model between consumption, income and prices over the period 1958-92. More generally, we argue that autoregressive models are unlikely to model slowly changing seasonality successfully, and may confound seasonal effects with the dynamic responses of prime interest. Our approach can be used in a wide range of cases and involves little loss in efficiency even if seasonality is deterministic.