A machine learning-based framework for exchange rate analysis and prediction

Abstract

Interests towards exchange rate (ER) analysis and prediction have increased in past decades. Theoretical and empirical literature support the relationship between ER and trading goods activity (imports and exports). However, there has not been any approach identifying the most important trading goods in order to forecast the ER based on these goods. Through a case study using the Guatemalan quetzal against the US dollar, this paper aims to predict the ER based on relevant trading goods and provide an organised database of the desired results. Using principal component analysis (PCA), we predict the ER with a low approximation error and determine which production sector should be the focal point for the country. We also provide theoretical foundations for a proper ER forecasting implementation using genetic algorithm (GA), lower–upper decomposition, and Lagrange and spline interpolation. The empirical results show manufacturing industries and exports to El Salvador and Honduras as the most relevant trading goods. This work could be beneficial and useful to national institutions, policy makers, and other academics or analysts who may utilise the data for further analysis.