Forecasting the Total Energy Consumption in the United States: A Semiparametric Markov Switching Approach

Nademi, Younes; Nademi, Arash

doi:10.311581/JAMDA.2508.1002.1.1.4

Forecasting the Total Energy Consumption in the United States: A Semiparametric Markov Switching Approach

Document Type : Research Article

Authors

Younes Nademi ¹

Arash Nademi ²

¹ Department of Economics, Faculty of Humanities, Ayatollah Boroujerdi University, Boroujerd, Iran

² Department of statistics, Ilam branch, Islamic Azad University, Ilam, Iran

10.311581/JAMDA.2508.1002.1.1.4

Abstract

We applied a semiparametric Markov switching AR-ARCH (SMSARCH) model to forecast the total U.S. energy consumption in the residential, commercial, industrial, transportation and electric power sectors. For this purpose, we compared several SMSARCH models containing different core functions with the models such as ARIMA, GARCH, EGARCH, Markov switching in mean and GARCH based on their abilities to forecast the total energy consumption. The time period from January 2000 to December 2015 was used for the in-sample estimation, while the period for the out-of-sample forecasting was from January 2016 to December 2016. The root mean square error (RMSE) criterion for both in-sample and out-of-sample periods indicates that the forecasting abilities of the SMSARCH models in all the U.S. energy sectors are better than those of the other studied parametric models. Furthermore, the results of Diebold and Mariano test showed that there is a significant difference between the values of RMSE for all models.

Keywords

Forecasting

Semiparametric Markov Switching Models

Total Energy