Port digitalization, information quality and operational times: evidence from Brazil's Porto Sem Papel single window
Warning
Work in progress. This is a research compendium under active development. Results, figures and text are preliminary, not peer reviewed and may change without notice. Please do not cite the numbers as final findings.
© 2026 Valber Gregory. All rights reserved. The repository is public for transparency and reproducibility review only; see LICENSE and LICENSING.md. Citation is welcome (how to cite).
Trabalho em andamento — todos os direitos reservados. Resultados preliminares, sem revisão por pares; sujeitos a mudança. Veja LICENSE.
Between July 2011 and April 2013, Brazil made the Porto Sem Papel (PSP) maritime single window mandatory, port by port. This project combines
- 1.18 million commercial vessel calls with cargo (ANTAQ waterway statistics, 2010–2026),
- foreign-trade flows with observed freight costs (Comex Stat, 2010–2026), and
- a curated registry of the adoption dates, read from the full text of the ordinances in the Official Gazette (DOU): 23 ports dated, 21 with high confidence,
into a reproducible information system (DuckDB + R) and a staggered difference-in-differences evaluation of what the single window changed.
Preliminary reading (subject to change): the clearest change is in what the state records, not in how long ships wait. In Santos, the share of vessel calls with the post-operation waiting time recorded jumps from 0 % to 93 % with the single window; across all adopting ports, however, the average change in recording is not robust, so this is a case rather than a programme-wide effect. Aggregate waiting times do not move. The apparent documentary effect for cabotage vessels continues a pre-existing downward trend and disappears once port-specific trends are allowed (HonestDiD and trend-adjusted DiD).
| Staggered adoption, dated from official acts | Recording of |
|---|---|
 |
 |
| Event study: probability that |
Event study: recurrent cabotage vessels |
 |
 |
| Estimates by sample, with Lee bounds | Sensitivity to pre-adoption trends (HonestDiD) |
 |
 |
Event studies: Sun–Abraham estimator, port and month fixed effects, standard errors clustered by port; shaded bands are 95 % uniform (sup-t) bands, thin lines pointwise 95 % intervals.
Everything in the manuscript — every table, figure and every number quoted in
the text (article/latex/numbers.tex) — is produced by one targets pipeline.
No number is typed by hand.
1. Environment (R 4.4.3; packages pinned in renv.lock)
# from the project root
renv::restore() # recreates the exact package library
source("scripts/00_check_environment.R") # checks R, packages, DuckDB, LaTeX2. Raw data (public sources, not redistributed; each download is logged with
URL, date and SHA-256 in data/metadata/download_log.csv)
| Source | File | How |
|---|---|---|
| ANTAQ waterway statistics 2010–2026 | data/raw/antaq/estatistico.zip (909 MB) |
R/03_download_antaq.R |
| Comex Stat (NCM, imports and exports) | data/raw/comex/*.csv |
R/04_download_comex.R + python/validate_downloads.py |
| IBGE state boundaries 2024 | data/raw/ibge/BR_UF_2024.zip (14.7 MB) |
URL in download_log.csv |
| PSP ordinances (DOU) | versioned registry data/metadata/psp_portarias_dou.csv |
python/scan_dou_legacy.py |
3. Pipeline
targets::tar_make() # full rebuild (~50 min on a laptop once DuckDB is built)
targets::tar_outdated() # what is out of date
targets::tar_visnetwork() # dependency graphThe DAG (_targets.R) runs scripts/09 → 27 in dependency order: DuckDB
ingestion → descriptives → event studies and DiD at the vessel-call level →
few-cluster inference (wild cluster bootstrap, permutation) → coverage
(information-quality) results → heterogeneity, placebo and HonestDiD → Lee
bounds → pre-trend sensitivity → figures → tables and macros → outputs/overleaf.zip. Each script also
runs on its own (Rscript scripts/NN_*.R).
4. Tests
testthat::test_dir("tests/testthat") # registry, cohorts, sample rules, WCB/permutationpython -m pytest tests/python # DOU ordinance parserFull details, timings and known pitfalls: docs/reproducibility_guide.md.
Non-obvious choices are dated in docs/decisions_log.md.
| Path | Content |
|---|---|
R/ |
functions: data building, DiD models, inference, figure theme |
scripts/ |
numbered, runnable steps orchestrated by _targets.R |
sql/ |
DuckDB schema |
python/ |
DOU mining and download validation |
data/metadata/ |
versioned curated metadata (ordinances, crosswalk, reporting-gap registry, download log) |
article/latex/ |
manuscript sources (Elsevier elsarticle, anonymised for review) |
app/ |
Shiny observatory (Rscript scripts/07_launch_dashboard.R, port 4200) |
docs/ |
protocol, data inventory, identification strategy, writing guide, gallery |
The work has not been published yet. If you need to refer to it, cite the
repository with its version and access date (metadata in CITATION.cff;
GitHub's Cite this repository button produces APA and BibTeX):
Gregory, V. (2026). Brazilian Port Digital Transformation Observatory: Port Digitalization, Information Quality and Operational Times (Version 0.2.0, work in progress) [Research compendium]. GitHub. https://github.com/valbergregory/Port-Digitalization-Observatory
@misc{gregory2026observatory,
author = {Gregory, Valber},
title = {Brazilian Port Digital Transformation Observatory: Port Digitalization,
Information Quality and Operational Times},
year = {2026},
note = {Version 0.2.0, work in progress. All rights reserved},
howpublished = {\url{https://github.com/valbergregory/Port-Digitalization-Observatory}}
}Once a paper or an archived release with a DOI exists, this section will point to it.
Compêndio de pesquisa sobre o Porto Sem Papel: 1,18 milhão de atracações comerciais da
ANTAQ, comércio exterior do Comex Stat com frete observado e datas de adoção
lidas nas íntegras das portarias do DOU. Tudo é reproduzível em R com
renv::restore() e targets::tar_make(). Trabalho em andamento; resultados
preliminares; todos os direitos reservados. Contato: valber.gregory@gmail.com