Raw materials prices have historically been highly volatile. The main reason behind the raw materials prices volatility is that they are influenced by other volatile factors, such as global economic conditions, electricity prices, growth of developing countries, the weather conditions, currency exchange rates, interest rates and other unpredictable phenomena, such as pandemics ans wars. A specific category of raw materials, which is characterized by highly volatile prices, are the non-ferrous metals.
The volatility of raw material prices constitutes a serious challenge for businesses, since the cost of raw materials is a major part of their total production costs. Unstable raw material costs translate into unstable profit margins for businesses, which in turn results in decreasing operating capital and put in danger the survival of the firm. In order to successfully choose and implement pricing strategies, manufacturers first need to forecast the non-ferrous metals prices.
In this context, the main research objective of this repo is to support manufacturing firms in their endeavors to manage non-ferrous metals prices volatility by comparing the contributions of different non-ferrous metals prices forecasting methods to some risk management strategies and deciding on which method yields the best results. The comparison of the contributions of the different prices forecasting methods on the implementation of the risk-mitigation strategies is done through forecasting aluminium, copper, and zinc prices for five different lead times using three machine learning methods and two deep learning methods.
Based on the existing literature, the machine learning methods that are chosen in this study are the Extreme Gradient Boosting, the Random Forest, and the K-Nearest Neighbors , while the two deep learning methods are the Long-Short Term Memory and the Multilayer Perceptron. The findings of this study show that the best performing model for aluminium, copper, or zinc price forecasting is, for all the different lead times, the multivariate Extreme Gradient Boosting.