Modeling economic growth for Nigeria using robust statistical models.

Abstract

Economic growth is one of the most important goals of macroeconomic policy-making, but measuring it is not easy. This study aimed at measuring economic growth for Nigeria using robust statistical models. In this study, Gross Domestic Product (RGDP) is used as a proxy for economic growth and is modelled using selected predictors, namely internal debt (INDT), external debt (EXDT), interest rate (RINR), an exchange rate (REXR), and trade openness (OPEN). Quarterly RGDP index collected from the Central Bank of Nigeria for the period 1986 to 2022 was used in this study. Exploratory data analysis (EDA) revealed the linear relationship between the RGDP and the predictors. EDA also revealed the presence of multicollinearity and outliers in the predictors. In the presence of outlier and multicollinearity, this study utilizes the ridge regression, robust principal component regression, partial least square regression, average centered penalized regression, gaussian process regression and the coupler FMKL-GLD quantile regression. The performance and the efficiency of the adopted methods were evaluated using forecasting accuracy metrics, namely the root mean square error (RMSE), mean absolute error (MAE) and mean absolute percentage error (MAPE). In using robust PCR, it can be asserted that the robust principal component regression (M-estimator) technique was an efficient and optimal technique for predicting RGDP. Specifically, PC1 and PC2 account for 35.39% and 22.15% in RGDP. In PLS, non-cross-validated and cross-validated PLS selection methods were used. Thus, 91.5% of the variance in economic growth drivers were explained by the five components generated and selected from the non-cross-validated PLS method. While, 72.6% of the variance in economic growth drivers were explained by the two components generated and selected from the cross-validated PLS method. Hence, after the cross-validation and extraction, the first and second components were efficient and optimally predicted 63.1% and 18.4% economic growth. In the average centered penalized regression model, the performance of the LASSO, ridge and elastic net techniques were compared. Using the least value of the forecasting metric values, the LASSO-average centered penalized regression was robust. The result of the best-performing average-centered penalized regression model indicated that INDT, RINR, REXR and OPEN positively contributed to the RGDP by 4.27%, 0.40%, 0.49% and 0.52% respectively. while, EXDT decreases RGDP by 0.97%. In the fitted gaussian process regression, the main effect of INDT, EXDT, RINR, REXR and OPEN for predicting RGDP were 38.30%, 12.20%, 1.10%, 2.00%, and 1.20% respectively which were increased after the independent re-sampling to 56.30%, 6.90%, 3.10%, 2.80%, and 2.10% for predicting RGDP. The estimated performance of FMKL-GLD quantile model techniques revealed that FMKL-GLD 50Q model was efficient for examining and predicting economic growth in Nigeria. Thus, INDT, RINR, REXR and OPEN positive contribution to RGDP were 17.94%, 29.42%, 7.99% and 145.10% respectively. Meanwhile, RGDP, as a result of EXDT was reduced by 3.92%. Therefore, government and policymakers should properly harness the benefit of trade openness to engender international patronage for economic growth. Also, coupler FMKL-GLD 50Q quantile regression technique and gaussian process regression method are the most efficient predictive statistical methods to deal with multicollinearity and outliers. However, Nigeria’s economy had gone through various seasons, thus, a further study can be done to investigate structural breaks and propose appropriate model(s).