Abstract

Forecasting global foreign trade is essential for developing government trade policies and management strategies for multinational corporations. However, achieving an accurate trade forecast is challenging because of the complex structural relationships between exports, imports and other economic variables. Many traditional forecasting models, such as time series, econometric, and machine learning, provide less accurate forecasts for trade data. This paper proposes an ensemble learning approach to improve forecasting performance by hybridizing the structural relationships between trade and deep learning models to predict foreign trade for ten major countries. The proposed method first establishes a cointegration relationship between exports and imports and their structural variables. The cointegrated models are then used to predict the future of trade, which is used as a benchmark model for comparison. A hybrid deep learning algorithm uses the cointegrated variables as input variables to predict trade data, and then are compared with time-series forecasts and economic structural models. The experimental results reveal that the ensemble learning method can achieve excellent forecasting performance for the tested periods of trade data. In most cases, the root means square error and mean absolute percentage error values are smaller than the time series and economic structural models.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call