Comparison of Neural Networks and Regression Time Series When Estimating the Copper Price Development

Abstract

In recent years, the primary copper ore stock has been cut sharply and the price of crude copper has been rising. On the other hand, thanks to a huge industrial interest, the production of copper products has increased significantly over recent years. It is therefore clear that the prediction of the copper price is very important. A variety of techniques, such as statistical methods—regression time series or artificial neural networks—are used for prediction. The aim of this contribution is to perform a regression analysis of the copper price development on the New York Stock Exchange using the mentioned linear regression and neural networks, expertly compare both methods, and identify the more suitable one for a possible prediction of future copper price developments. Input data includes copper price data from January 2006 to April 2018. First, linear regression is performed, and then, neural networks are used for regression analysis. A total of 1000 neuron structures are generated, five of which with the best characteristics are kept, and these are then further worked with. From the linear regression, the curve obtained by the spline function appears to be best, and the neural networks have all been proven to be usable in practice.