Performance evaluation of Bootstrap-Linear recurrent formula and Bootstrap-Vector singular spectrum analysis in the presence of structural break.

The Singular Spectrum Analysis (SSA) forecasting method has been used widely recently. SSA has characteristics and advantages in decomposing data into trends, oscillations, and noise, which are for forecasting. Forecasting withSSA can be done by several ways e.g. Linear Recurrent Formula (LRF), Vector, and Simultaneous. The Bootstrapping process is implemented in those forecasting methods to increase the accuracy and builds the interval forecast. This study focuses on testing the sensitivity and accuracy of the combination of the Bootstrap-LRF and Bootstrap-Vector methods. The tests are applied to data that contains structural breaks, namely Indonesia's trade monthly data (exports and imports) from 1993 to 2019. The test results show that Bootstrap-Vector has a smaller forecasting range and is more accurate than Bootstrap-LRF in long-horizon forecast. Moreover, the Bootstrap-Vector is more stable in the presence of structural break in the data, meaning that this method is less sensitive to structural changes. Meanwhile, Bootstrap-LRF is more accurate in short horizon forecast. [ABSTRACT FROM AUTHOR]