Giovanni Cerulli at the 2025 UK Stata Conference in London
On 11 September Giovanni Cerulli (CNR-IRCrES) presented the paper “Optimal Policy Learning for Multi-Action Treatment and Risk Preference” at the 2025 UK Stata Conference, held at Westminster University in London (UK).
The paper addresses the topic of Optimal Policy Learning (OPL) with observational data, that is, data-driven optimal decision-making in multi-action (or multi-arm) settings where a finite set of decision options is available.
The contribution is organized into three parts. The first part reviews the main approaches to estimating the reward (or value) function and the optimal policy within this analytical framework, highlighting the identification assumptions and the statistical properties of offline OPL estimators. The second part examines decision-making under risk, showing how the optimal choice may be influenced by the decision-maker’s attitude toward risk, specifically in relation to the trade-off between conditional mean and conditional variance of rewards. The third part discusses the potential limitations of optimal data-driven decision-making by considering the failure of two key assumptions required to identify the optimal choice: overlapping and unconfoundedness.
The study concludes by reflecting on both the opportunities and the challenges that characterize Optimal Policy Learning approaches.
The UK Stata Conference provides Stata users worldwide with the opportunity to exchange ideas, experiences, and information on new applications of the software.
