This paper investigates the existence of seasonal patterns in the quarterly merchandise export and import data of Pakistan from 1982:1 to 2002:1. Unit root tests are applied to determine whether the seasonal component in each variable exhibits stochastic non-stationarity. Deterministic and stochastic effects are isolated and quantified. Few alternate DGP specifications are identified, fitted and tested for their outof- sample forecasting performance. A tentative finding is that deterministic effects are relatively more important than stochastic ones. However, integrated models, i.e., ARIMA, mixed ARIMA, and ARIMA-GARCH, outperform deterministic models with respect to forecasting.