Research

This paper asks whether autonomous LLM research agents perform better when generating papers that follow research paradigms already well represented in the literature. I study this question using evidence from the Autonomous Policy Evaluation (APE) experiment, where LLM agents generate empirical economics papers and compete in a tournament-style evaluation against human-written benchmarks. I measure literature support by locating each paper abstract in the semantic space of economics abstracts from OpenAlex, then test whether papers in denser regions of the literature perform better. Literature support is positively associated with tournament performance for AI-generated papers, but not for human-written papers. The result suggests that autonomous research agents are most effective when operating close to existing research paradigms, and that their current limits are tied to the distribution of prior literature and model training data.

This paper evaluates the long-run impact of the Cassa per il Mezzogiorno (CasMez), a state development bank active from 1950 to 1984, which aimed to reduce the economic divide between Northern and Southern Italy. Leveraging newly digitized microdata from CasMez archives, municipal statistics, voting patterns, geographic data, and a custom-reconstructed historical road network, the study estimates the causal effects of CasMez's dual funding streams-industrial credit to private firms and public infrastructure investment. Using Marginal Structural Models, Propensity Score Matching, and Subclassification, the analysis finds that industrial lending generated sustained industrial growth, while infrastructure investments had limited impact unless implemented through coordinated, multi-stakeholder plans targeting clearly measurable outcomes. The results highlight the greater effectiveness of market-aligned interventions over standalone public investment in driving local development.

This paper examines the effectiveness of subsidized industrial credit in promoting firm growth in postwar Southern Italy, where long-term loans were provided exclusively by public institutions at capped interest rates. Using custom machine learning tools, I develop an automated pipeline to construct a unique 15-year panel of firm-level financial data from archival records. Leveraging event-study methods robust to staggered treatment timing, I find that subsidized credit led to lasting improvements in firms' ability to leverage fixed assets to generate profits. These effects are only detectable when analyzing cumulative outcomes, which motivates the development of a simple theoretical framework to formalize their identification. This framework enables clear estimation of causal effects, showing that subsidized credit led firms to expand their asset base and improve their ability to generate profits. However, complementary evidence indicates that the program operated under binding credit rationing constraints, limiting access for many potentially eligible firms and thereby dampening its broader impact on local economic growth.

With Sara Bagagli, LSE

This paper studies how large-scale transport infrastructure reshapes industrial specialization and spatial integration. Exploiting the staggered expansion of Italy's highway network between 1955 and 1975 and industrial census data from 1951 to 2001, we show that highway access causes a large and persistent decline in municipal specialization in both tradable and non-tradable sectors. The effect is driven primarily by a more even distribution of employment across existing activities rather than by the emergence of new sectors. To interpret these findings, we reconstruct Italy's historical road network using a custom algorithm, compute travel times between municipalities, and identify evolving travel-time communities based on mutual accessibility. Highway access anchors municipalities more firmly within these communities, reducing transitions into more specialized communities and increasing persistence in more diversified ones. Overall, transport infrastructure reshapes local economies through two linked channels: reallocating activity within places and reorganizing their connections within the spatial economy.