Articolo su “Machine Learning with Applications”

Pubblicato in open-access sulla rivista Machine Learning with Applications (Elsevier) l’articolo “Risk-adjusted policy learning and the social cost of uncertainty: Theory and evidence from CAP evaluation” a cura di Giovanni Cerulli e Federico Caracciolo (Università degli Studi di Napoli Federico II)

The paper extends Optimal Policy Learning (OPL) to settings where policymakers care not only about maximizing expected welfare, but also about risk and uncertainty. Building on Roy’s safety-first principle, we develop a risk-adjusted policy learning framework that explicitly balances expected outcomes against the risk of undesirable realizations. We apply the approach to EU Common Agricultural Policy (CAP) subsidies using Italian FADN microdata. The broader message is simple: good policy targeting should consider not only who is expected to benefit most, but also how uncertain those benefits are. A step toward making policy learning more useful for real-world decisions under uncertainty.

#MachineLearning #PolicyEvaluation #Agriculture #CAP #DataScience #Research #OpenScience