MULTIVARIATE FORECASTING OF STEEL PRICES

Abstract

Steel products are the most used raw material for many industries regarding their accessibility, strength, and relatively low costs comparing to the other base metals with similar characteristics. In the fast-paced economic environment of the post-world war II era with the growing expansion of the economy in the world, steel usage and consequently, the prices of steel become an essential concern for the countries and organizations. From the first decade of the 21st century, with the launch of the online trades for steel products in the commodity markets, the importance of steel prices has become even more critical. The practice of the price series forecast is conducted by various statistical and data-driven models in the literature. However, there is a lack of investigation to find practical and user-friendly statistical models in forecasting steel prices where, besides simplicity, can perform realistic and precise forecasts. The VAR and VEC models are newly introduced models comparing to the conventional models in econometrics. While the exogeneity in the conventional models can cause several difficulties in model specifications, the VAR systems, by treating all the variables as endogenous variables, can overcome this issue. Also, the VEC model that is a particular case of the VAR model, can assess the short-run and long-run dynamics by the cointegration relations in a single model. The data in this study are ranged from Jan. 2009 to Jun. 2020. The forecast evaluation is through the out-of-sample approach, which is more compatible with the real-world setting. The results of this study suggest the dominance of the VAR model over the VEC model in the forecast horizon of 18 months attributed to a mid-term forecasting horizon.