Base metal prices, especially steel, play a significant role in industrial economics, making them worth knowing about future values. In most cases, we expect superior performance from multivariate forecasting models comparing univariate methods due to the involvement of explanatory variables in the system. Standard vector auto regressive model can only capture short-run dynamics because of the differencing process for non-stationary series that eliminates the possible long-run relationship. Instead, performing non-stationary series on levels through the vector auto-regressive framework does not suffers such loss. Moreover, the vector error correction model can define both short-term and long-run dynamics explicitly. These models can yield more robust forecasts in the mid-term and long-term by investigating short-run and long-run relationships simultaneously. The current study aims to perform an out-of-sample forecast for the United States steel prices index 18 months ahead using cointegrated variables. The results suggest that the non-stationary vector auto-regressive model outperforms the vector error correction model regarding mean absolute percentage error and root mean square error as forecast accuracy measures.

No abstract