Pascuel Plotkin

We exploit a novel identifier of outsourcing events present in Italian administrative data. This information permits us to estimate the effects of outsourcing across a wide set of occupations without restricting the analysis to workers who remain employed after being outsourced. We find that outsourcing leads to substantial earnings losses, primarily driven by the extensive margin---a margin not fully analyzed by previous research---as several outsourced workers become non-employed shortly after joining the contracting firms. Outsourced workers in non-routine manual jobs have the largest earnings losses, while those in jobs involving abstract, cognitive tasks experience some earnings gains following the outsourcing event. Our evidence is consistent with some Italian firms using outsourcing to bypass the country's strict employment protection legislation.

We estimate the impact of a labour demand shock on Canadian recently graduated bachelor students. We leverage a unique Canadian administrative data that features links between individuals post-secondary education and their tax-files. We estimate that a standard deviation increase in school-major specific labour demand, increased earnings immediately in a magnitude between 2.3 and 2.6 log points, with persistent and increasing effects throughout the oil shock. Additionally, the labour demand shock had effects on other labour market outcomes such as unemployment or self-employment. In terms of schooling, school-major specific labour demand had a positive effect on dropouts, which remarks the importance of considering the outside option of students when studying school enrollment.

This project investigates in detail the labor supply of gig workers and the role of information frictions in shaping labor supply of these workers in Brazil. For that, we combine administrative data from the largest food delivery platform in Latin America, with primary survey data from a sample of riders in the platform. Previous studies on labor supply have assumed that workers have a full knowledge of the wage distribution and form expectations rationally. Yet in the context of gig employment where earnings present a high volatility, it is possible that workers form inaccurate expectations. In this project we overcome this constraint by surveying a sample of platform workers to elicit their reservation wages and expected wages. We first document the error in workers beliefs, contrasting survey data and administrative records. We then move on to an experimental research design where we expose workers to unbiased information on expected wages and study how they update their beliefs and behavior on the platform. By linking the experimental research design with platforms administrative records and employer-employee data, we study how unbiased information can potentially modify labor supply decisions and adjust labor allocation in a more efficient way - as measured by the surplus generated between the actual earnings in the platform and workers reservation wages.

We investigate the potential impact of the diffusion of large language models on hiring discrimination by comparing the biases in AI-driven and human-driven resume screening. To do so, we reanalyze data from previous field experiments by reconstructing resumes and comparing human call-back rates with LLM-generated results. We find that LLMs exhibit significantly less racial bias than human decision-makers. However, LLMs are not perfectly unbiased. In a more stringent test, we fail to reject the null hypothesis that LLMs do not use information contained in names.

We study the effects of deunionization–characterized by the weakening of labor unions–on firm compensation, labor inequality, and earning gaps. Specifically, we leverage a reform in Brazil that ended the automatic renewal of collective bargaining agreements (CBAs) upon their expiration, significantly curtailing the bargaining power of unions over wages and job amenities. Using a dataset on the universe of CBAs alongside an administrative linked employer-employee dataset, we estimate the causal effect of deunionization. Our approach compares firms with expired CBAs, where workers are no longer covered by a CBA, to those with active CBAs at the time of the reform’s implementation. We extend our analysis to explore the mechanisms through which deunionization affects wage and amenity distributions, applying textual analysis to the CBAs to explain changes in employee welfare. Additionally, the analysis examines the heterogeneity in the reform’s effects across different industries, occupations, and worker skill levels, aiming to uncover whether the reform widened inequality in earnings and employment amenities.