BUILDING EXPORT FORECAST MODEL USING A KERNEL-BASED DIMENSION REDUCTION METHOD

The purpose of this article is to build Vietnam's export turnover forecast model at the monthly frequency on a large data set of potentially time-series predictors. The model is built based on the Dynamic factor model, where factors are extracted from the data set of non-redundant and relevant variables in potential predictors by a variable dimension reduction method using kernel tricks and based on an RMSE-best model. The results show that the percentage of absolute error of out-of-sample forecasts by the built model is less than 2%, and the out-of-sample forecast accuracy of this model is much higher than that of the model built based on the export demand model.