The accuracy of forecasting neural networks and the impact of using fuzzy sets for the currency market

Abstract

The aim of the article is to check the accuracy of forecasts of neural networks on the currency market and the impact of fuzzy sets on their accuracy. The study presented in this article uses an original approach that considers the use of neural networks and fuzzy sets in the mechanism of investment decision making. The empirical study is based on projections of the three currency pairs of the Swiss franc, British pound, and the dollar against the euro. These currencies are forecasted using three different neural networks - ELM, MLP and LSTM, for ten different forecast horizons (from 1 to 10 days). In forecasting, neural networks use historical data, both for price levels and rates of return. The research carried out confirmed that the presented method is in many cases more accurate than the methods compared to it in this study

Keywords

• currency exchange rate
• forecasting
• neural networks
• fuzzy sets