Deseasonalized State-Space Time Series Forecasting with Application to the US Salmon Market

Abstract

An approach that combines seasonality removal with a multivariate, state-space, time series forecasting model is developed to provide shortrun forecasts for the US salmon market. Time series included in the model are: US fresh Atlantic salmon wholesale price index; fresh salmon (Atlantic, coho and chinook) monthly US import quantities and prices; and US chum and sockeye salmon monthly export prices. Four versions of the state-space forecasting model are compared in terms of their statistical performance during out-of-s ample forecasts. Out-of-sample 3-, 6- and 12-month ahead directional predictions are generated to test the models' performance in terms of direction. Under identical modeling conditions, out-of-sample statistical and directional tests indicate that deseasonalization improves the overall performance of the state-space model. As a result, a linear, deseasonalized, state-space forecasting model is selected to provide twelve monthly out-of-sample forecasts for all series.