Titolo della tesi: Modern Nonparametric Methods Applied to Economics
The field of economics has experienced a paradigm shift in recent years due to the availability of large-scale datasets and advancements in computational power. This has prompted researchers to explore alternative approaches beyond the traditional parametric models used in economic analysis. In this paper, we explore modern nonparametric methods and their applications in economic modelling. Specifically, we compare the performance of prominent nonparametric techniques, including XGBoost and Random Forest (tree-based ensemble methods), and hybrid methods (with the Super Learner R package) with more traditional models such as the linear Gaussian model and generalized linear regressions. By evaluating these methods within the context of economic analysis, we aim to shed light on their potential benefits and limitations, ultimately contributing to a more comprehensive understanding of their applicability in economic research. Notably, ensemble learning techniques which combine the best of both worlds achieve remarkably good performances, comparable with those achieved by domain experts that uses fine-tuned forecasting models. This is discussed in greater detail in the fourth section, where we present a case study on forecasting major tax revenues in Italy, but similar conclusions can be extended outside the scope of time series analysis, like in the study on cross-sectional tax justice data which is discussed in the final section.