Evaluating and forecasting conventional gasoline price fluctuations using Garch models and machine learning methods

Conventional gasoline price can affect the government and society as a strategic commodity in the community. Conventional gasoline price fluctuations have economic, political, social, cultural, and environmental effects. Thus, the prediction of its volatility is essential. This study aims to hybridize and propose different Garch models based on two distributions and various algorithms in machine learning, such as random forest, ridge regression, support vector regression (SVR), and elastic-net for predicting weekly gasoline price volatility. The results depict Garch and GJRgarch models based on t-student distribution can predict volatility. The combination of ridge regression and GJRgarch model can better predict volatility for the seven-step-ahead. The RMSE scale has been used to compare results that the scale value is ۰.۰۱۴۷۵ in the hybrid method. In fact, combining the ridge regression with t-student-GJR garch model has the slightest error prediction or the most accuracy among different Garch models and machine learning algorithms.

Keywords:

Garch models , machine learning , gasoline price , volatility , distribution