Document Type : Research Paper

Authors

• alireza Erfani ¹
• elmira asl roosta ²
• abdolmohamad kashian ²

¹ Economics department,, faculty of economics, management and administrative sciences

² semnan university

Abstract

The exchange rate has always been one of the most important economic indicators, influenced by various factors. Some of these factors are reflected in economic variables, while others are manifested in political-economic news. An important question that has not yet been precisely answered is whether it is possible to have a comprehensive model for modeling and predicting the exchange rate that encompasses all effective variables and factors. In this research, as an answer to this question, a comprehensive model based on deep learning is presented using machine learning and a data fusion approach. The model supports various types of data, and for its training, news affecting the exchange rate from 10 major domestic and international sources were collected over the period from 2014 to 2023. These were provided to the model along with exchange rate data and other economic indicators. To find the best model, eight machine learning models, two statistical models, and a large language model (LLM) were trained and tested in both regression and classification modes. To avoid bias and random results, cross-validation techniques and repeated training and testing of the models with different random initial values were employed. The results indicate that the proposed approach, by directly considering all influential factors, has significantly outperformed previous approaches.

Keywords

• Exchange Rate Forecasting
• Data Fusion
• Comprehensive Model
• Machine Learning
• Artificial Intelligence

Main Subjects

• Monetary economy