Predicting Precious Metal Markets using Custom Deep-Learning NARX Network: A Portfolio Application

Abstract

The thesis aims to predict precious metal market prices using a deep learning model known as Nonlinear AutoRegressive with eXogenous input. Market forecasts of the 58 assets selected are evaluated through portfolio techniques such as Mean-Variance and Conditional Value-at-Risk to demonstrate the real-world application. This investigation provides a framework for future value projection, including the preprocessing stage, feature selection, and dataset construction. Additionally, a novel error measure is proposed to comprehensively assess the estimation accuracy.