Forex Rate Forecasting Based on Deep Learning Ensembled Predictions

Medina Salas, Luis Daniel (2024) Forex Rate Forecasting Based on Deep Learning Ensembled Predictions. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (991kB) | Preview
Preview	PDF (Configuration Manual) Download (627kB) | Preview

Abstract

The aim of this paper is to improve foreign exchange rate forecasting by evaluating the capabilities of Recurrent Neural Networks (RNNs), which are known for their ability to analyze sequential data. Four LSTM models with different architectures were created and compared to discover the most suitable architecture based on their performances.

Accurate forecasting of Forex rates is critical across industries, influencing the valuation of foreign investments and driving the dynamics of international trade. Given the complexity and volatility inherent in Forex markets, which are defined by detailed time-series data, traditional forecasting approaches hardly capture the full range of market dynamics and other macroeconomic factors and behaviours.

By overcoming frequent RNN time series predictors that only consider historical data from the predicted variable as input, the models in this paper perform a multidimensional approach that aims to provide a more comprehensive understanding of the factors that influence the Mexican currency (MXN) price against the American dollar (USD). The most distinctive aspect of this approach is the use of the Keras Functional API, which allows more flexible model architectures, including the integration of different input streams to allow for a wider range of factors that influence the MXN. In this case, there are included external macroeconomic variables such as crude oil international prices and the performance of the Mexican stock market.

The provided findings are that by considering the previously mentioned macroeconomic factors correlated to the Mexican currency, into the LSTM models, it was possible to get more reliable predictions with a lower error rate, than the models that do not consider these macroeconomic factors and only consider historical data from the predicted currency.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Bradford, Michael UNSPECIFIED
Uncontrolled Keywords:	Forex rate forecasting; time series; Recurrent Neural Network; Keras functional API
Subjects:	Q Science > QA Mathematics > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning H Social Sciences > HG Finance > Investment > Stock Exchange
Divisions:	School of Computing > Master of Science in Data Analytics
Depositing User:	Ciara O'Brien
Date Deposited:	05 Jun 2025 14:52
Last Modified:	05 Jun 2025 14:52
URI:	https://norma.ncirl.ie/id/eprint/7767

Actions (login required)

View Item