Abstract
South Africa's persistent 'jobless growth' crisis, where economic expansion fails to reduce the country's critically high unemployment rate, highlights the limitations of traditional aggregate-level macroeconomic models like Okun's Law. This study addresses the need for a more granular diagnosis by conducting a definitive industry-level analysis of the growth-unemployment nexus. Using a hybrid econometric and machine learning framework, the primary aim is to disaggregate the South African economy to identify the key industry influences on its unemployment dynamics. The study employs a multi-stage methodology, utilizing Year-on-Year (YoY) quarterly data from Q1:2008 to Q4:2024. The analysis first establishes a foundational econometric baseline by implementing the ‘difference model’ of Okun’s Law, which relates the change in unemployment to the change in economic output. This baseline was estimated using dynamic Ordinary Least Squares (OLS) models to diagnose both the industry impacts on national unemployment and the direct employment elasticities. Subsequently, a comparative ‘horse race’ of machine learning models, including Ridge Regression, Random Forest, XGBoost and Support Vector Regression (SVR), is employed to robustly test the predictability of these relationships. Finally, SHapley Additive exPlanations (SHAP) is leveraged to provide a deep, interpretable diagnosis of the key industry influences. The findings are significant. The machine learning analysis reveals that the growth-unemployment relationship in South Africa is fundamentally unstable and only weakly predictable. This was validated by the aggregate-level ‘horse race’ where the linear Ridge Regression model emerged as the best predictor with a Test Set R-squared of 0.178, while the best-performing non-linear model, XGBoost, achieved a significantly lower R-squared of just 0.042. Despite this difficulty in forecasting, the explanatory SHAP analysis successfully identifies a clear hierarchy of influence among South Africa's industries. The Finance and Transport industries emerge as the most important industries on national unemployment. In stark contrast, traditional industrial engines like Manufacturing and Mining were found to have significantly diminished influence, a finding that provides powerful, data-driven proof of the economy's structural transformation. Critically, the SHAP analysis reveals an anomalous dynamic for the Finance industry, where high growth is a strong signal for a future increase in unemployment. In stark contrast, the Transport industry exhibits a conventional, unemployment-reducing relationship.
This study's primary contribution is the novel application of an interpretable machine learning framework to provide a deep, multivariate diagnosis of the South African economy. The...