Applying Machine Learning Data Imputation Techniques for Missing Data in the Analysis of Financial Development and Trade Openness in Africa: Evidence From CFA Countries

This study addresses the critical challenge of missingness in African macroeconomic data, by employing advanced machine learning imputation to enhance the reliability of economic analysis. Utilizing panel data from 13 Communauté Financière Africaine (CFA) countries spanning 1980-2023, the research tackles a 2.5% missingness rate through a hybrid supervised-unsupervised learning framework. The methodology first applies K-means clustering to group data into optimal homogenous subsets, followed by the k-Nearest Neighbor (kNN) algorithm for precise imputation. The accuracy of this approach is validated by low Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) scores. Econometric analysis was then conducted using the Feasible Generalized Least Square (FGLS) Methodology, which is robust to the detected cross-sectional dependence, heteroskedasticity, and autocorrelation in the panel data. The main findings demonstrate a positive and statistically significant relationship between financial development, trade openness, and economic growth. Crucially, the analysis of the interaction term between financial development and trade openness revealed a negative coefficient. This suggests that insufficient financial development acts as a moderator, diminishing the positive impact of trade openness on economic development in the CFA region. In conclusion, the study successfully illustrates the value of machine learning in producing more reliable economic insights and highlights the necessity of strengthening financial systems to fully capitalize on the benefits of trade liberalization in Africa.