Forex Forecasting

This repository contains models designed to forecast foreign exchange currencies using Neural Networks. Two hybrid methods are presented (SCNN_model and es-rnn) while a purely statistical one (V-AR-model) was used as a benchmark. A detailed analysis of implementations, methodology, and results can be found in the report.

The project was carried out as a key part of the curriculum for the 'Μ401 - Deep Neural Networks' course, as taught at the National and Kapodistrian University of Athens (NKUA), during the Fall of 2022.

Project Structure

$PROJECT_ROOT
¦
+-- SCNN_model 
¦   # Forecasting using a simple Exponential Smoothing
¦   # and Convolutional Neural Networks
¦
+-- V-AR-model 
¦   # Forecasting using Vector Autoregression
¦   # (used as a benchmark)
¦
+-- es-rnn 
¦   # Forecasting using Holts-Winters and Recurrent Neural Networks
¦
+-- presentation 
¦   # Presentation summarising implementations, results, and conclusions 
¦
+-- report 
    # Comprehensive report detailing implementations, results, and conclusions

Algorithms

Before diving into the individual methodologies, ensure you have the necessary dependencies installed. Each implementation directory (SCNN_model, V-AR-model, es-rnn) contains its own README.md file, detailing specific requirements and dependencies; so have a look at it before running the respective model.

Data Source

The data is sourced from the European Central Bank and can be downloaded using a script of the repo. The dataset contains historical exchange rates of various currencies against the Euro, and is published around 16:00 CET.

These rates represent the official currencies of non-euro area Member States of the European Union and world currencies with the most liquid active spot FX markets. The dataset can be downloaded from here, and an overview can be seen bellow.

Date	USD	JPY	BGN	...	THB	ZAR
2023-09-15	1.0658	157.50	1.9558	...	38.145	20.2968
2023-09-14	1.0730	158.13	1.9558	...	38.387	20.3109
2023-09-13	1.0733	158.28	1.9558	...	38.397	20.3300
...	...	...	...	...	...	...
1999-01-04	1.1789	133.73	NaN	...	NaN	6.9358

Authors

• Darmanis Michael - mdarm
• Efstathios Kotsis - staks1
• Vasilis Venieris - vasilisvenieris

License

This project is licensed under the MIT License - see the LICENSE file for details.