Comparing Multiple Linear and Robust Regression Models in Assessing Economic Indicators’ Impact on GDP

Abstract

Macroeconomic data for 142 countries in 2022 and 2023 were obtained from the World Bank, International Monetary Fund, and Trading Economics, covering interest rate, inflation rate, unemployment rate, public debt, trade balance, gross fixed capital formation, fiscal balance, and current account. Correlation analysis reveals that most indicators exhibit weak linear relationships with GDP, while trade balance shows the strongest and statistically significant negative correlation with GDP (2022: r = −0.314, p < 0.001; 2023: r = −0.301, p < 0.001). Stepwise MLR identifies unemployment rate, public debt, and current account as significant predictors of GDP in both years. In the MLR models, unemployment rate consistently shows a negative effect on GDP (2022: β = −0.0768; 2023: β = −0.0776), while public debt and current account exhibit positive effects. Model evaluation indicates that both MLR and RMLR provide comparable predictive accuracy. The MLR model achieves slightly lower prediction errors, with MAE values of 1.1458 (2022) and 1.3923 (2023), and RMSE values of 1.8092 (2022) and 1.7877 (2023). The corresponding R² values for the MLR model are 20.09% in 2022 and 22.03% in 2023, marginally higher than those of the RMLR model. These results suggest that, after logarithmic transformation and assumption checking, the MLR model is sufficient for GDP estimation, while RMLR serves as a valuable robustness check that confirms the stability of the findings. Future research may extend this study by incorporating additional years of data, exploring nonlinear or panel regression frameworks, and examining regional or income-group heterogeneity to further enhance the explanatory power of GDP models.