Enhancing the Iraqi Oil Revenue Forecasting: Comparative Evaluation of the ARIMA Model Extensions and Hybrid Models

Abstract

The current study provides the necessary response to the problem of oil revenue prediction errors in Iraq because it explicitly analyzes the extensions of the autoregressive integrated moving average model (ARIMA) and the hybrid methodology. Although the results of the existing studies have already proven the efficacy of simple ARIMA structures, there remains the gaps in the understanding on how the more sophisticated alterations and hybrid approaches could further enhance predictive accuracy. We engage in a full comparative study of extensions of ARIMA models such as seasonal adjustment, integration of exogenous variables, and hybrid models that integrate machine learning and the traditional time series models. The suggested approaches are operated on monthly data on Iraqi oil revenues between 2021 and 2023 and tested with harsh methods of validation in form of calculation including AIC, BIC, MAE and MAPE. The findings indicate that the hybrid models are more successful than the single ARIMA models by reducing the MAPE by 15%. Moreover, residual diagnostics verify the stability of these hybrid methods where there is no severe autocorrelation and the properties of error distribution are improved. The research adds to the literature in that it provides empirical data on effectiveness of hybrid forecasting method in volatile commodity markets giving decision makers more valid instruments to use in financial planning. In addition, the results also underscore the need to have a combination of statistical and machine learning paradigm to identify complex non-linear trends on oil revenue data. Besides the contribution to the methodological discussion of time series forecasting, the given research also offers some practical insights.