Regional disparities in the European Union. A machine learning approach

We investigate the hypothesis of regional convergence in the per-capita GDP in 242 European regions (NUTS2) during the 2000–2021 period. The literature shows mixed results, from absolute convergence towards a joint long-run distribution to multiple regimes (convergence club). Our results show a broad convergence to an unimodal distribution. Although the GDP distribution was characterized by a twin-peak property in 2000, it tends to disappear over time, bringing, in 2021, to an unimodal distribution. Physical and human capital is the most responsible for the convergence process and the EU cohesion funds. To empirically investigate the question, we first apply alternative techniques of cluster identification. Later, we assess whether clusters and covariates affect the per-capita GDP. We use a novel machine learning algorithm (GPBoost) instead of the more traditional techniques used in the current literature. The results show that a convergence process is at work; physical and human capital are mainly responsible for the gdp explanation. but eu funds play a relevant role as well. moreover, complementarities do exist among these variables.