Abstract: On the basis of the time series data, machine learning can also be used for predicting the future development of export in various states. It offers, of course, to measure trade between the world´s two largest economies – China and the USA, which has an impact on the global world economy. Therefore, the objective of this contribution is to predict the USA export to the People´s Republic of China in the context of mutual sanctions using machine learning. The data set contains monthly data on the development of the USA export to China between January 2000 and July 2019. Regression is carried out using neural networks. There are generated three sets of multilayer perceptron networks considering the time series lag of 1 month, 5 months, and 10 months. A total of 10,000 neural structures are generated, out of which 5 with the best characteristics are retained. Export values between August 2019 and December 2020 are predicted and subsequently, the results of all three experiments are compared. The result closes to the ideal one is with the time series lag of 10 months; the networks are also able to capture the trend and fluctuations of the time series. Yet, there is certain overfitting notable, mainly due to the gradation of mutual trade war between the USA and the PRC.
Authors: Tomáš Krulický, Eva Kalinová, Jiří Kučera
Keywords: machine learning, export, prediction, artificial neural networks, time series