Gold price forecasting using multivariate stochastic model

Commodities prices are pivotal to the mineral investment decision and have a considerable impact on mining companies' financial performance and countries that depend on mineral resources. Therefore, understanding the future mineral price movement is critical for revenue-based planning both for the company and the country. In this article, the Autoregressive Distribution Lag (ARDL) model was used to forecast annual gold prices using gold demand, treasury bills rates, and lagged gold prices as covariates. Augmented Dickey Fuller and the Phillips Perron methods were used to test for unit roots and found that all the variables were integrated of order one. Subsequently, the cointegration test was undertaken, which indicated that there is no cointegration between the variables. This entailed application of the short-run version of the ARDL to forecasts and consequent analysis. A Granger causality analysis show that gold demand Granger causes gold price; and that treasury bill rates do not Granger cause gold price. Lastly, the ARDL (4, 4, 2) model, which provides best ARDL forecast results, was evaluated against two other forecasting methods namely stochastic mean reverting, and Autoregressive Integrate Moving Average (ARIMA). Results showed that the ARDL model emerged as a best of all the three forecasting methods to forecast annual gold prices.

Automated summary

This article uses the ARDL model to forecast annual gold prices, and through Granger causality analysis and comparison with two other forecasting methods, it concludes that the ARDL model is the best method for predicting annual gold prices.