(A comparative study of a sample of currency prices for the period 2015-2024)

A Dissertation Submitted To

The Council of College of Administration and Economics at Karbala University, As Partial Fulfillment of the Requirements for

PH.D. Degree Financial and Banking Sciences.

Submitted by:

Zahraa Kadhim Majeed Al-Shammous

Supervised by

        Assistant Professor                                                      Professor   

Dr. Haider Abbas Abdullah                                        Dr. Amir Ali Khalil

This study aims to evaluate the effectiveness of deep learning models, specifically the Long and Short Term Memory (LSTM) model and the Gate Recurrent Unit (GRU) model, in predicting the prices of two distinct categories of financial currencies: the cryptocurrency Bitcoin, a highly volatile digital asset, and the Euro/Dollar currency pair, one of the world’s major exchange rates. This contributes to providing more accurate predictive estimates that can support decision-makers in the currency markets, based on a ten-year weekly time series, from January 4, 2015, to December 29, 2024, with 522 observations. In order to analyze and test the study’s hypotheses, the Long-Term Memory Network (LSTM) model structure and the Gate Recurring Unit (GRU) model were built using the R programming language and a set of R libraries, most notably the well-known Keras library in the field of machine learning, The study adopted an analytical approach to compare two models and differentiate between them to select the optimal model according to performance metrics represented by Mean Absolute Error (MAE), Mean Square Error (MSE), and Root Mean Square Error (RMSE). The study reached several results, the most important of which is the superiority of the LSTM model over the GRU model in predicting Bitcoin and Euro/Dollar prices in this study. In light of this, the study presented several recommendations, the most important of which is relying on the Long Memory Network Model (LSTM) in forecasting in future studies, especially cryptocurrencies. It also recommended the necessity of relying on modern technical methods, especially artificial intelligence and machine learning applications, in forecasting currency prices, especially cryptocurrencies, due to their high capabilities in dealing with various types of data without preconditions, as in other methods. This study contributes to enriching the financial literature by providing a comparative (applied) analysis between two radically different markets in terms of structure, organization and risk level, as well as highlighting the growing role of deep learning techniques as a supportive tool for decision-making and risk management.

Keywords: (Deep learning, cryptocurrencies, Bitcoin, fiat currencies, long-term memory model (LSTM), Gate Recurring Unit (GRU)

(A comparative study of a sample of currency prices for the period 2015-2024)

A Dissertation Submitted To

The Council of College of Administration and Economics at Karbala University, As Partial Fulfillment of the Requirements for

PH.D. Degree Financial and Banking Sciences.

Submitted by:

Zahraa Kadhim Majeed Al-Shammous

Supervised by

        Assistant Professor                                                      Professor   

Dr. Haider Abbas Abdullah                                        Dr. Amir Ali Khalil

This study aims to evaluate the effectiveness of deep learning models, specifically the Long and Short Term Memory (LSTM) model and the Gate Recurrent Unit (GRU) model, in predicting the prices of two distinct categories of financial currencies: the cryptocurrency Bitcoin, a highly volatile digital asset, and the Euro/Dollar currency pair, one of the world’s major exchange rates. This contributes to providing more accurate predictive estimates that can support decision-makers in the currency markets, based on a ten-year weekly time series, from January 4, 2015, to December 29, 2024, with 522 observations. In order to analyze and test the study’s hypotheses, the Long-Term Memory Network (LSTM) model structure and the Gate Recurring Unit (GRU) model were built using the R programming language and a set of R libraries, most notably the well-known Keras library in the field of machine learning, The study adopted an analytical approach to compare two models and differentiate between them to select the optimal model according to performance metrics represented by Mean Absolute Error (MAE), Mean Square Error (MSE), and Root Mean Square Error (RMSE). The study reached several results, the most important of which is the superiority of the LSTM model over the GRU model in predicting Bitcoin and Euro/Dollar prices in this study. In light of this, the study presented several recommendations, the most important of which is relying on the Long Memory Network Model (LSTM) in forecasting in future studies, especially cryptocurrencies. It also recommended the necessity of relying on modern technical methods, especially artificial intelligence and machine learning applications, in forecasting currency prices, especially cryptocurrencies, due to their high capabilities in dealing with various types of data without preconditions, as in other methods. This study contributes to enriching the financial literature by providing a comparative (applied) analysis between two radically different markets in terms of structure, organization and risk level, as well as highlighting the growing role of deep learning techniques as a supportive tool for decision-making and risk management.

Keywords: (Deep learning, cryptocurrencies, Bitcoin, fiat currencies, long-term memory model (LSTM), Gate Recurring Unit (GRU)