Deep learning based hybrid gold index (XAU/USD) direction forecast model

Abstract

Forecasting the direction of the gold index, which determines the dollar value of 1 ounce of gold, is an attractive research topic for investors, researchers and analysts. In this study, it is aimed to construct a model that predicts the direction of the gold index based on deep learning. The proposed model is obtained as a result of blending the numerical dataset of the gold index with the textual data. This is a hybrid prediction model both in this aspect and in terms of hybrid deep learning methods used in direction prediction. As far as we know, this is the first attempt in the literature that uses the social media platform as a source for financial sentiment analysis and constructs a deep learning-based direction prediction model for the gold index by blending it with numerical data. The contribution of the study to the literature is summarized in four stages: In the first stage, in order to carry out the sentiment analysis, the dataset is cleaned and ready for modeling by methods such as parsing the dataset collected from the Twitter environment, finding the correct forms of the words in the dictionary, finding the roots of the words, normalizing the words, and cleaning the unused characters. Afterwards, the dataset is classified using 14 different deep learning-based methods. Secondly, the results of the sentiment analysis are blended with the numerical data of XAU/USD. Third, the XAU/USD direction prediction model is constructed with deep learning models. Fourth, the performance of the results from five different forecasting models in predicting the direction of XAU/USD is presented. As a result, the performance of the proposed model is significantly superior with high accuracy when compared to the state-of-the-art studies.