Finansal Zaman Serilerinin Derin Öğrenme Algoritmaları ile Tahminlenmesi

Abstract

Stock market index data, foreign currency, and gold have an important place in financial time series. Therefore, value or direction of movement estimation studies on this subject attracts the attention of both investors and researchers. This study aims to estimate the daily value of the US Dollar, Gold, and Borsa Istanbul (XU) 100 index using deep learning methods: Recurrent Neural Networks and Long-Short-Term Memory. A data set consisting of 2280 business days between 2013-2022, which includes the date, US Dollar, Gold, and XU 100 closing data, was used in the study. Mean absolute error, mean square error, root mean square error, and coefficient of determination were used to evaluate the performance of the developed prediction models. When the results were examined, it was seen that the Long-Short-Term Memory algorithm performs better than the Recurrent Neural Network algorithm and achieved a determination coefficient value of over 95% for the US Dollar, Gold, and XU 100 index. Moreover, the findings obtained in the study indicate that deep learning algorithms can show high prediction performance on financial time series without using extra independent variables.