Comparison of price-prediction models in forecasting commodity prices
Date
2020
Authors
Jiyana, Thelma Thobile
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Commodity price is one of the vital inputs in mining projects valuations. If high incorrect price is used, the project will be overvalued. Subsequently, an uneconomic project may be commissioned and fail to yield expected targets. If a low incorrect price is used, the project will be undervalued. Consequently, an economic project may be shelved or abandoned due to an incorrect price being used. Mining companies are price takers; therefore, it is important to be able to apply an appropriate commodity price during valuation. However, it is difficult to predict commodity prices due to numerous uncertain factors that influence the price movements such as technology, supply, demand and macro-economics. The impact of global events further complicates the price prediction process. There are various price prediction models that can be used to predict commodity prices with a certain degree of confidence such as mean reversion, autoregressive moving average, variants of conditional variance, dynamic model averaging and dynamic model selection. However, these models are based on different assumptions yielding different results. Copper and gold commodities were selected for this study in order to compare the forecast accuracy of the commonly used price-prediction models. Copper was selected because it is regarded as a reliable indicator for the strength of the market due to its widespread application in all sectors. A rising copper price suggests a strong economy and the converse is true. Gold was selected because it is affected by various factors such as business cycle, exchange rate, stock price and interest rate. Given that gold price movement is affected by numerous factors, it is important to investigate if there is a price prediction model that can be able to forecast the price of gold. The prices of these commodities were sourced from Market Index website. This research study selected the most commonly used models to predict copper and gold prices. Python and MATLAB programming languages were used to apply these models because of availability and simplicity. The price-prediciton models used were Autoregressive Integrated Moving Average (ARIMA) and Glosten, Jagannathan, and Runkle Model (GJR). The selected model parameters were ARIMA (9, 1, 9) and ARIMA (7, 1, 7) for copper and gold price prediction. GJR (1, 1) model was used for both copper and gold volatility forecasting. The measurement of forecast accuracy used was MAPE since it varies between 0 and 1, thus it is not influenced by the scale of the time series. Both ARIMA and GJR models considerably failed to forecast the commodity prices. For the purpose of comparison, the results showed that ARIMA (9, 1, 9) and ARIMA (7, 1, 7) models are only suitable to forecast copper and gold prices over a short-term, that is, periods less than three years. It was also found that GJR (1, 1) model yielded superior results when forecasting copper and gold prices conditional variances for periods over five years. Based on the findings of this study it is recommended that ARIMA (9, 1, 9) and ARIMA (7, 1, 7) models be used to forecast copper and gold prices over one-year and three-year periods. When forecasting price movements over three years, then GJR (1, 1) is recommended to forecast price volatility up to a seven-year period
Description
A research report submitted to the Faculty of Engineering and the Built Environment, University of the Witwatersrand, in partial fulfilment of the requirements for the degree of Master of Science in Engineering, 2020