Assessing the Robustness of Ordinary Least Squares and Double Weighted M-Estimation Methods for Predicting Crude Oil Prices in Nigeria: A Study of Predictive Accuracy and Generalization.

Abstract:

This study evaluates the robustness of Ordinary Least Squares (OLS) and Double Weighted M-Estimation (DWME) methods for predicting crude oil prices in Nigeria, focusing on predictive accuracy and generalization. Using 192 monthly data points (2006–2021) from the Central Bank of Nigeria (CBN) and Nigerian National Petroleum Corporation (NNPC), the dataset included crude oil prices, production, crude oil production, and exchange rates, with synthetic datasets simulated via multivariate normal distribution for varying dimensions (n = 10 to 1,000). The performance measures such as Mean Squared Error (MSE), Mean Absolute Error (MAE), and R-squared were assessed. Results showed comparable MSE values for training data, with OLS_TRAIN ranging from 172.85 to 694.56 and DWME_TRAIN from 173.03 to 699.27. Testing data revealed DWME's marginal superiority, with slightly lower MSE (e.g., DWME_TEST median 548.68 vs. OLS_TEST median 543.85). MAE trends indicated consistency for both methods, with DWME showing marginally better stability across dimensions. R-squared values highlighted improved generalization for smaller datasets, with DWME_TEST peaking at 0.7043 and OLS_TEST at 0.7544 for the 10x3 dimension. Both methods struggled with generalization as dimensions increased but exhibited stable training performance. In conclusion, DWME demonstrated slightly better robustness, especially in testing scenarios, affirming its suitability for predictive tasks involving economic and energy-related variables.