Foreign Exchange Rates Forecasting Using Deep Learning

  • โสรยา แพสุวรรณ์
  • สมพร ปั่นโภชา
  • บำรุง พ่วงเกิด
Keywords: Deep Learning, Recurrent Neural Networks, Convolutional Neural Networks, Convolutional with Recurrent Neural Networks

Abstract

The objectives of this study are to study the system of Recurrent Neural Networks (RNN), Convolutional Neural Networks (CNN) and Convolutional with Recurrent Neural Networks (C-RNN), the methods in Deep Learning, and to investigate the most suitable forecasting model of four foreign exchange rates (USD, EUR, GBP and JPY) which are time series from 1 January 2010 to 31 December 2019. The data is divided into 2 groups, the first for the training set in order to find the appropriate model for forecasting of foreign exchange rates, which is selected from the top 3 models with the lowest mean absolute error (MAE). And the second for the test set in order to testing the 3 selected models to find the forecasting effectiveness with the root mean square error (RMSE).

The results show that the most suitable forecasting models of USDTHB are LSTM[1,5] , CNN[9,2] and C-RNN[7,6] , that the models gives forecasting effectiveness with the RMSE of 0.1877 0.2719 and 0.1620 respectively. The most suitable forecasting models of EURTHB are LSTM[6,12] , CNN[6,3] and C-RNN[6,4], that the models gives forecasting effectiveness with the RMSE of 0.5160 0.6948 and 0.4978 respectively. The most suitable forecasting models of GBPTHB are LSTM[2,1] , CNN[5,9] and C-RNN[3,11], that the models gives forecasting effectiveness with the RMSE of 0.6344 0.7991 and 0.5210  respectively. The most suitable forecasting models of JPYTHB are LSTM[4,1] , CNN[2,7] and C-RNN[11,1], that the models gives forecasting effectiveness with the RMSE of 0.0023 0.0027 and 0.0022 respectively. The comparison of the forecasting method shows that the C-RNN has higher effectiveness.

Published
2020-08-19

Most read articles by the same author(s)

1 2 3 4 > >>