Type

Status

completed

Author

Motahar, Seyed Armin

Examiner

Prof. Dr. Erwin Amann

Download

https://10.17185/duepublico/84757

Abstract

This dissertation examines the impact of wealth inequality on economic growth and addresses the persistent lack of consensus in the empirical literature, which largely stems from an overreliance on income-based measures, the neglect of cross-country heterogeneity, and the widespread assumption of linear relationships. Using a comprehensive panel dataset covering 150 countries over the period 1980–2020, the thesis consists of three interconnected empirical studies that combine machine-learning techniques with econometric analysis. The first study identifies the most relevant inequality measures for forecasting economic growth under heterogeneity and nonlinearity by clustering countries using K-means and applying XGBoost models to determine inequality indicators with the highest predictive importance within each cluster, showing that wealth inequality is the dominant predictor of growth for most country groups, while income inequality is more relevant only in a limited subset of economies. The second study focuses explicitly on the relationship between wealth inequality and economic growth, employing interpretable machine-learning tools such as SHAP values and surrogate models and revealing a nonlinear, inverted U-shaped relationship, whereby moderate levels of wealth inequality may support growth, while excessive concentration becomes growth-dampening; these results are further corroborated using fixed-effects and dynamic panel GMM estimations. The third study investigates the transmission channels through which wealth inequality affects economic growth by combining mediation analysis with Generalized Additive Models, allowing for a flexible assessment of nonlinear mechanisms related to human capital accumulation, investment, and institutional quality, and demonstrating that different channels exert opposing effects whose relative importance varies across countries and inequality levels. Overall, the dissertation contributes to the inequality–growth literature by prioritizing wealth inequality over income proxies, explicitly accounting for heterogeneity and nonlinearity, and integrating machine-learning methods with causal econometric approaches, yielding nuanced policy implications that emphasize the context-dependent nature of inequality’s growth effects.