Application of AI for Oil Price Forecasting

This paper examines methods for forecasting oil prices, comparing traditional autoregressive models (ARIMA, SARIMAX) with machine learning approaches (LSTM). The target variable is the price of WTI crude oil. The dataset covers 2015–2019 and includes both WTI price data and a set of exogenous variables: the Wilshire 5000, Dow Jones, and DXY indices; the Baltic Dirty Tanker Index (BDTI); the Brent–WTI price spread; volumes of crude oil produced and refined in the US; the number of active drilling rigs; US crude oil inventories; US petroleum product consumption; US net oil imports; and the number of days US crude oil inventories can cover demand, excluding the strategic petroleum reserve. Forecasts were generated over a five-day horizon. The results indicate that machine learning outperforms autoregressive models. The LSTM model incorporating the Dow Jones index as an exogenous variable showed the best performance (RMSE: 1.5). Autoregressive models performed less effectively, with RMSE values of 1.668 for ARIMA and 1.51 for SARIMAX. Even the LSTM model without exogenous variables (RMSE: 1.51) outperformed ARIMA, underscoring the relative advantage of machine learning in short-term oil price forecasting.