This repository contains scripts for working with the CEPII BACI trade dataset. It shows how the yearly csv files can be compiled into one table and exported as a space-efficient parquet file. It also demonstrates how this may be queried and manipulated with SQL using DuckDB's Python API.
The BACI (Base pour l’Analyse du Commerce International) dataset is an annual record of bilateral merchandise trade flows published by the French think tank CEPII (Centre d’Etudes Prospectives d’Informations Internationales). Flows are recorded at the product level according to different editions of the Harmonized System. The dataset builds from the raw entries found in UN COMTRADE, fixing as much as possible missing and inconsistent data (Gaulier and Signago 2010).
Each BACI table is a csv file corresponding to one year of trade. It has the following columns:
t: yeari: ISO numeric code of the exporting countryj: ISO numeric code of the importing countryk: HS product codev: value of trade flow in thousands of USDq: weight of trade flow in metric tons
Each row is therefore the dollar amount v of product k sent by country i to country j in the year t. Dictionaries for the country and product codes are included when downloading an edition of BACI. Note that q is missing in about 2% of entries.
See usage.ipynb. See also eda.ipynb for some exploratory data analysis.
G. Gaulier and S. Signago. (2010). BACI: international trade database at the product-level. The 1994-2007 version. CEPII Working Paper No. 2010-23.
@misc{ksreyes2023baciscripts,
title = {Working with the BACI dataset},
author = {Kenneth S. Reyes},
url = {https://github.com/ksreyes/baci},
year = 2023,
}