Comparative Analysis of Economy-Based Multivariate Oil Price Prediction Using LSTM

Predicting unrefined petroleum prices is crucial for decision-makers, financiers, and academics in the energy industry. This paper presents an economic system-based multivariate oil price forecasting model that utilizes LSTM, a sophisticated deep learning algorithm that could capture tricky and nonlinear interactions in data with temporal order. Our model takes WTI crude price and numerous economic variables, containing US Dollar Index Futures, Gold Futures, Ten-year US Bond Yield, and S&P 500, as input characteristics. Before model training, we executed data pre-processing to eliminate outliers and enhance model accuracy. Via exploratory file study, we diagnosed and identified 482 outliers between 2007 and 2009, which were influenced by the financial crisis. We also expelled data from 2020 and after, as the hurricane of COVID-19 should distort our judgments. To choose our model’s efficiency, we used metrics such as square (R2), Mean Square Error (MSE), and Mean Absolute Error (MAE). Our findings display that the LSTM version has powerful predicting talents and can correctly forecast oil price bases on economic indicators. This review paper also stresses the significance of macroeconomic indicators in oil price forecasting.