Forecasting foreign exchange rate with machine learning techniques

Date

2023-9-07

Author

Küçük, Rohat

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

444
views

684
downloads

Cite This

Since the collapse of the Bretton Woods system, international macroeconomics has grappled with the challenges of explaining exchange rate behavior under flexible regimes. The exchange rate disconnect puzzle underscores the elusive relationship between exchange rates and their underlying macroeconomic determinants. Drawing on this, groundbraking work by Meese and Rogoff (1983), have posited the potential superiority of the simple random walk model over complex economic theories in forecasting exchange rates. To further probe into this, our study employs a diverse set of models—ARIMA, Extreme Gradient Boosting, Support Vector Regression, Long Short Term Memory, Convolutional Neural Network, and notably, an ARIMA-LSTM Hybrid—with the objective of capturing both the linear and non-linear dynamics inherent in the data. We examine the forecasting landscape for the Russian Ruble, Euro, and Turkish Lira over time horizons of 3, 7, 14, and 30 days. Utilizing models grounded in classical economic theories such as the Uncovered Interest Rate Parity, Purchasing Power Parity, and the Monetary Model and contrasted their performance with models that did not incorporate these theoretical constructs. The effectiveness of these models was assessed based on Root Mean Square Error, Mean Absolute Percentage Error, and Theil's U metrics. Our findings underscore the nuanced role of economic variables in forecasting, the absence of a universally optimal model, and the consistent phenomenon of benchmark Random Walk models holding their own, particularly with the Euro. Conclusively, the realm of forex forecasting is underscored as multifaceted, emphasizing the need of a diverse, evolving toolkit in navigating the ever-shifting global financial landscape.

Subject Keywords

Exchange rates, Forecasting, Machine learning, Random walk, Time series

URI

https://hdl.handle.net/11511/105480

Collections

Graduate School of Natural and Applied Sciences, Thesis