Forecasting Foreign Currency Exchange Rate

B.Sc. FEVEN SHIMELIS, BSc

Master's thesis

Abstract:

The foreign currency exchange rate(forex) is the most traded market in the financial world. Forex forecasting can be performed by analysing past data and gather patterns. Different methods for forex forecasting have been proposed in the research world. Every forex pair has it is own characteristics, and the forecast depends on the algorithms utilized. Thus, in this study a deep learning model which predicts a forex price has been developed using collected historical dataset only. The proposed system has four components: preprocessing, feature analysis and selection, normalization, and finally the forecasting phase. In this study, 22 years of a daily real-world historical dataset of CZK/Euro and CZK/USD forex pairs are collected from investing.com financial platform. Simple RNN and LSTM based forecasting models are proposed, the models are implemented using Keras in Python and tested using 20 days dataset. Mean squared error and accuracy are used to evaluate the performance of the models, MSE was taken as a metrics and calculate the error based on actual and predicted prices. Simple RNN achieved an accuracy of 89.91% and 82.81% for CZK/Euro and CZK/USD respectively. LSTM achieved forecasting accuracy of 97.64% and 93.68% for CZK/Euro and CZK/USD respectively. As a result, LSTM outperforms the simple RNN with higher accuracy and minimum mean squared error.

Keywords

Exchange Rate, Deep Learning, Forecasting, Artificial Neural network (ANN), Recurrent Neural Network (RNN, Czech Koruna

Language used: English

Date on which the thesis was submitted / produced: 1. 4. 2021

Thesis defence

• Supervisor: Ing. Pavel Šimek, Ph.D.

Czech University of Life Sciences Prague

Faculty of Economics and Management

Master programme / field:
Systems Engineering and Informatics / Informatics