A Machine Learning Framework for Cross-Market Dependency Analysis and Forecasting of Iranian Refinery Stocks

Abstract

The Iranian stock market is strongly influenced by developments in global energy markets due to the structural dependence of the national economy on oil revenues. This study proposes a machine learning framework for cross-market dependency analysis between global oil market indicators and the stock prices of major Iranian refinery companies listed on the Tehran Stock Exchange. Historical data, including Brent crude oil price, oil trading volume, exchange rate, and temporal variables, were collected and integrated into a unified dataset covering approximately eight years of trading activity. Three machine learning algorithms, including Random Forest Regression, XGBoost Regression, and Polynomial Regression, were evaluated and compared using R² score and RMSE metrics. Experimental results demonstrated that the Random Forest model achieved the best predictive performance among the evaluated methods, with R² values close to 0.99 for several target stocks. The proposed framework was further validated using out-of-sample market observations collected after the training period. Findings indicate that global oil market variables contain substantial predictive information regarding the behavior of Iranian refinery stocks and can support data-driven investment analysis in energy-dependent financial markets.