project
How a partnership between UPE and the Fernando de Noronha Territorial Authority mobilized applied econometrics and computational modeling to decompose price differentials between the archipelago and mainland Brazil — and what it taught about applying data science where the problem is public policy, not product.
Case snapshot
Fernando de Noronha has a chronically high cost of living compared to the mainland. The Territorial Authority and UPE needed a rigorous diagnosis: how much of the differential is unavoidable logistical cost, and how much is avoidable inefficiency — and where to intervene first.
Structure the research as an econometric decomposition problem: build a representative price basket, model the determinants of the differential by component (transport, scale, taxation, market structure) and simulate logistical intervention scenarios with computational modeling.
Ongoing research funded by the Territorial Authority via PROPEGI/UPE, producing structured diagnostics that feed public policy decisions on island logistics and supply.
Fator Noronha uses applied econometrics and computational modeling to explain why prices on the archipelago are higher and which part of the gap can actually be addressed through policy.
Situation
Fernando de Noronha is one of Brazil's most prized tourist destinations — and one of the most expensive places to actually live. Permanent residents, service workers, and civil servants face prices for basic consumer goods, food, building materials, and services that are systematically higher than on the mainland.
The immediate cause is well known: the archipelago depends almost entirely on mainland supply, with access limited to air and sea, and without the market scale to sustain meaningful local production. But the immediate cause is not the complete answer. Part of the differential is genuine structural cost — freight, packaging, supply chain losses. Another part may be logistical inefficiency, market concentration, or poorly calibrated taxation.
Decomposition problem
The research question sounds straightforward: why are prices in Noronha higher? But answering it rigorously requires solving a decomposition problem with several layers.
The observed differential has no single cause and reflects transport, scale, taxes, and market structure all at once.
Primary data collection is unavoidable, so the project had to design a representative price basket from scratch.
Methodological decisions
Models isolate transport, scale, taxation, and market structure while controlling for the other components.
The goal is decomposition and interpretability for public policy, not just predictive accuracy.
Computational modeling tests how freight subsidies, route changes, or tax regimes would change prices.
The project is coordinated by Guilherme Nunes Martins and involves researchers across undergraduate, master's, and doctoral levels. My contribution focuses on computational modeling and data analysis.
What I learned
The biggest difference between this project and the ML or MLOps work I've done in healthcare and product environments isn't technical — it's about who uses the output and what they need to do with it.
The end user isn't a system; it's a decision-maker. In software products, model output feeds another automated process. Here, the output feeds a public policy discussion between researchers, managers, and authorities.
Real interdisciplinarity requires epistemic humility. Bringing data science tools into applied econometrics is useful — but only when there's genuine respect for the foundations of the destination field.
Ongoing research · Started June 2024 · Funded by the Territorial Authority of the State District of Fernando de Noronha via PROPEGI/UPE