Predicting forex pair movements: Integrating sentiment analysis, technical, and fundamental indicators using machine learning and deep learning models

Loading...
Thumbnail Image
Supplementary material
Other Title
Authors
Dave, Yash
Author ORCID Profiles (clickable)
Degree
Master of Applied Technologies (Computing)
Grantor
Unitec, Te Pūkenga – New Zealand Institute of Skills and Technology
Date
2024
Supervisors
Varastehpour, Soheil
Shakiba, Masoud
Type
Masters Thesis
Ngā Upoko Tukutuku (Māori subject headings)
Keyword
foreign exchange market
forex
sentiment analysis (SA)
deep-learning algorithms
computer modelling
algorithms
Citation
Dave, Y. (2024). Predicting forex pair movements: Integrating sentiment analysis, technical, and fundamental indicators using machine learning and deep learning models (Unpublished document submitted in partial fulfilment of the requirements for the degree of Master of Applied Technologies (Computing)). Unitec, Te Pūkenga - New Zealand Institute of Skills and Technology https://hdl.handle.net/10652/6489
Abstract
RESEARCH QUESTIONS 1. How do technical indicators compare with models, including fundamental indicators and sentiment analysis, in predicting forex prices? 2. How does sentiment analysis affect the accuracy of forex price predictions? 3. Which machine learning or deep learning models most effectively integrate these different data types for forex prediction? ABSTRACT The foreign exchange (forex) market is one of the largest and most liquid financial markets globally, making accurate price prediction crucial for traders, policymakers, and financial institutions. This research focuses on improving forex price prediction by integrating sentiment analysis with technical and fundamental indicators. Historical forex data, economic indicators, and sentiment data extracted from financial news were employed to develop predictive models. The primary models used include Long Short-Term Memory (LSTM) networks, Extreme Gradient Boosting (XGBoost), ensemble models combining these approaches, and Transformer models. Large Language Models (LLMs), such as GPT-4 and GEMINI Advanced, were also utilised for sentiment analysis. The performance of these models was evaluated using comprehensive metrics, including Mean Absolute Error, Mean Absolute Percentage Error, Root Mean Square Error, and R-squared. The analysis was conducted on two different time frames: Fourhour and one-day data for forex pairs EUR/USD, NZD/USD, EUR/NZD, and AUD/NZD. Key findings indicate that integrating sentiment analysis improves the predictive accuracy of models, particularly in the Four-hour time frame. The XGBoost model, when combined with technical, fundamental, and sentiment data, achieved the lowest Root Mean Square Error across various currency pairs, outperforming other configurations. In contrast, technical indicators alone delivered the most accurate predictions for the one-day time frame, although combining technical and fundamental indicators also resulted in solid performance. These results highlight the importance of integrating diverse data types, particularly sentiment analysis, for accurate and timely forex price prediction. The findings have important implications for traders, analysts, and policymakers, suggesting that incorporating sentiment analysis into forecasting models can enhance accuracy. Future research should explore further refinements of these models and investigate the integration of additional data sources to continue improving predictive capabilities. ii
Publisher
Link to ePress publication
DOI
Copyright holder
Author
Copyright notice
All rights reserved
Copyright license
Available online at