Essays on double/debiased machine learning and policy evaluation

Much of empirical economics research seeks to establish causal relationships among variables. A significant and rapidly evolving literature in econometrics is focusing on refining Machine Learning (ML) methodologies to handle complex or high-dimensional data and enhance the credibility of causal conclusions, especially in observational analysis. However, while ML methods excel at prediction, their improper use in estimating a structural relationship can produce statistically invalid and misleading inferences. One method that offers a robust solution across various empirical settings is the Double/Debiased Machine Learning (DML). DML is a semi-parametric framework that employs Neyman orthogonalization and cross-fitting procedures to debias estimates derived from flexible, high-dimensional ML models and deliver consistent, asymptotically normal estimators. Hitherto, DML has primarily found application in microeconomic analysis, while its potential for macroeconomic identification remains an open research avenue. The challenge this dissertation seeks to address is to provide empirical evidence for the benefits of integrating causal ML techniques into the widely used macroeconometric framework of Local Projections (LP). By explicitly accounting for high-dimensional confounders, this methodological integration yields robust and less model-dependent dynamic impulse response functions. The manuscript is structured into three chapters, all connected by the adoption of DML methodology to analyze macroeconomic phenomena. Chapter 1 evaluates the effectiveness of vaccination policy on COVID-19 mortality in the European Union. Chapter 2 tests the ability of the DML methodology to deliver robust monetary policy identification. This study estimates the dynamic effect of monetary policy on the labor share of income in a panel of Euro Area countries. Chapter 3 analyzes the heterogeneous effects of ECB monetary policy across member states, specifically studying how country-specific economic conditions govern the transmission to output and inflation. Methodologically, it advances the DML framework by adapting the strategy to panel-data settings via the novel time-series Reverse Cross-Fitting (RCF) technique. In chronological terms, each chapter of this thesis represents a modest advancement over the preceding one. While Chapter 1 applies the DML method to a critical public health question, Chapter 2 integrates DML within the macroeconomic framework for the first time, and Chapter 3 further innovates by adapting the cross-fitting procedure to the panel case.