A Neural Network Approach to Pricing in the Precious Metals Market

The modern development of the science of artificial neural networks (ANN) has allowed to use their nature and properties in various applied fields of science. One of the most important applications of ANN is the modeling of prices in the precious metals market. Just like in any other market, based on the prediction of current prices, because the ability of ANN to learn like a true biological neural network, relying on the input with subsequent testing of the output, provides a significant advantage in the prediction tasks compared to the classical mathematical algorithms. Predicting the price of precious metals with relatively high precision and low error is in great demand among all individuals and legal entities that carry out transactions which are directly related to the purchase and sale of these precious metals, since accurate knowledge of the future price of a particular metal will bring maximum benefits of these operations. Numerous methods have been developed [2-4] for the use of neural networks in the modeling of price forecasts, which make the prediction of the rate of exchange for a particular currency (rather objective). The applied methods make the prediction using the classical perceptron along with astrological cyclic indices [2], recursive neural networks [3], and/or using elements of mathematical statistics (for example, use of U-statistic and the coefficient of determination ) [4]. The goal of this paper is the attempt to usethe ANN in the forecasting problem that allows predicting the price of precious metals in the near future, based on an algorithm that makes predictions by learning based on an array of input data and does not depend on the said elements of mathematical statistics. The paper presents a new method for using an artificial neural network in forecasting problems. Experimental studies of this method were carried out on the basis of the precious metals pricing rate on the Ukrainian Interbank Exchange. The corresponding conclusions are made regarding the effectiveness of the method and the possibilities for its further improvement based on the results of these studies. It is expected that such an algorithm will give a prediction as close as possible to the real value.