On the use of Singular Spectrum Analysis for Forecasting U.S. Trade before, during and after the 2008 Recession.

Abstract

This paper is aimed at introducing the powerful, nonparametric time series analysis and forecasting technique of Singular Spectrum Analysis (SSA) for trade forecasting via an application which evaluates the impact of the 2008 recession on U.S. trade forecasting models. This research is felicitous given the magnitude of the structural break visible in the U.S. trade series following the 2008 economic crisis. Structural breaks resulting from such recessions might affect conclusions from traditional unit root tests and forecasting models which makes use of these tests. As such, it is prudent to evaluate the sensitivity and reliability of parametric, historical trade forecasting models in comparison to the relatively modern, nonparametric models. In doing so, we introduce the SSA technique for trade forecasting and perform exhaustive statistical tests on the data for normality, stationarity and change points, and the forecasting results for statistical significance prior to reaching the well-founded conclusion that SSA is less sensitive to the impact of recessions on U.S. trade, in comparison to an optimized ARIMA model, Exponential Smoothing and Neural Network models. Ergo, we conclude that SSA is able to provide more accurate forecasts for U.S. trade in the face of recessions, and is therefore presented as an apt alternative for U.S. trade forecasting before, during and after a future recession.