Considering seasonal fluctuations in equalizing time series by means of artificial neural networks for predicting development of USA and People´s Republic of China trade balance

Balance of payments is an accounting identity of each country. The ability to make a qualified and accurate prediction of trade balance of huge world economies such as the USA and the People´s Republic of China economies can have influence on the world´s economy. An enormous expansion and advancement of artificial intelligence offers a possibility to measure and predict also this indicator. The aim of the contribution is to propose a methodology for taking into account the seasonal fluctuations in equalizing time series by means of artificial neural networks on the example of the USA and People´s Republic of China trade balance. For the analysis, the trade balance data of the two countries from the period between 1985 and 2018 are used. Regression by means of neural structures is carried out in two alternatives, where the second calculation takes into account the monthly seasonality of the time series. The result indicates that the additional variable in the form of the month in which the value was measured enables more order and accuracy. The other experimental calculation indicates that especially the fourth and the fifth retained neural networks are able to capture the whole course of the trade balance. They are thus able to identify and maintain local fluctuations of the time series, that is, to maintain its seasonal course. An interesting fact is also that in the case of the first alternative, the most successful networks were only the radial basis function neural networks, while in the second alternative those were only the multilayer perceptron networks.