Research

Working Papers

• “Traffic Congestion and Labor Supply: Evidence from Smartphone data in Mexico”
  (presented at 2022 CREEA Annual Conference, TWEEDS 2022, PacDev 2023, 2023 AERE Summer Conference, 2023 Emerging Mobility Scholars Conference, 20th International Economic Association World Congress, and the 2023 LACEA and LAMES Conference)
  Abstract

  Does traffic congestion affect time allocation? I use highly granular smartphone data from Mexico City to study empirically how traffic congestion affects work- time allocation. I find that traffic increases hours worked. The effect is driven by workers leaving work later, not by changes in arrival time. There is modest evidence that labor income does not increase although total hours do. These results highlight an avoidance mechanism consistent with bottleneck models that has been overlooked when estimating the costs of congestion.
• “When Simpler is Better: Task-Specific Machine Learning vs. Commercial LLMs for Labor Force Classification in Developing Countries”, with Jose Incio and Carlos Ospino
  
  Abstract

  Identifying which workers are exposed to outdoor environments is critical for climate adaptation, health, and labor policy, yet this information is rarely collected in household surveys. Using 234,104 observations from Peru’s national household survey (ENAHO, 2004–2018), we evaluate eight models–spanning traditional machine learning, a fine-tuned language model (Flan-T5), and commercial LLM APIs (GPT-4o, Claude 3.5 Sonnet)–for classifying workers as indoor or outdoor based on short Spanish task descriptions. Traditional ML models substantially outperform commercial LLMs on both accuracy and cost. Our findings demonstrate that AI can dramatically accelerate labor force characterization for research and policy, and that sophisticated commercial AI is not always superior to simpler task-specific alternatives.
• “How Do Workers Respond to Generative AI? Evidence from the Hiring Process of an Online Labor Market”, with Estrella Gomez-Herrera
  
  Abstract

  Most evidence on the labor-market effects of generative AI concerns demand: whether clients post fewer jobs or hire fewer workers in exposed occupations. We study the supply-side response of workers. On a large European online labor market for skilled freelancers, we observe every stage of the hiring funnel, from the client's contact through the worker's response and proposal to the contract and earnings, which separates the demand that reaches each worker from the worker's own decision to compete for it. Following a fixed cohort of 64,537 freelancers every month from June 2022 to January 2024, and comparing job categories with high and low exposure to AI before and after the release of ChatGPT, we find that the demand reaching exposed workers did not rise but their behavior changed. Exposed workers became 19% more likely to send a proposal in a given month and, as a result, 14% more likely to win a contract. The response occurs on the extensive margin, proposals are prepared faster, and the effect grows over time. A four-channel model shows that these patterns are expected when generative AI mainly lowers the cost of competing for work, rather than the return to labor.

Work in Progress

• “Migration and Environmental Quality: Evidence from Facebook data in Colombia”, with Pablo Ordonez and Julieth Santamaria.
  (presented at 2022 AAEA Annual Meeting)
  Abstract

  How would the projected massive migration to cities in the developing world impact the environment? We study the effect of urban growth on air pollution using the exogenous Venezuelan diaspora to Colombia. We track migrants using the Facebook logins of users who created an account in Venezuela and are currently in Colombia. Using a difference-in-difference approach and an instrumental variable approach, we find that migration increases PM 2.5. This result highlights the need for labor market policies that allow newcomers to relocate to less polluting industries.

• “The effect of crime concentration on deforestation: Evidence from Colombia”, with Rodrigo Londono.

• “Climate Shocks and Informal Transitions”, with Nicolas Guida-Johnson.

Journal Publications

3. "Is it Too Hot to Work? Evidence from Peru", Environment and Development Economics, 1-20, 2025 paper (open access)


2. "COVID-19 impacts on Amazon Deforestation", Trees, Forests and People, Volume 20, 2025 (with Jerico Fiestas and Javier Montoya) paper (open access)


1. "The Persistent Effects of COVID-19 on Labor Outcomes: Evidence from Peru", Applied Economics Letters, 30:8, 1065–1076, 2023 (with Fernando Aragon and Carlos Ospino) paper - working paper