Vietnam-EU Monthly CN8 Trade Data Pipeline
Overview
This project builds a reproducible pipeline for EU member-state imports originating in Vietnam at monthly CN8 product detail. It retains member-state rows, produces an EU-total aggregation, and documents the reporting grain, variable definitions, raw-data retention, and quality controls.
The verified public artifact is a partial January 2024 release. The repository documentation describes a broader 2018-to-latest target, but that target is not presented as completed.
Research Question
How can detailed monthly EU imports from Vietnam be transformed into a versioned CN8 panel without confusing classification, unit, revision, or coverage changes with economic movements?
Economic or Technical Motivation
Detailed trade releases are revised over time, product codes change across nomenclatures, unit values can be noisy, and missing product-country-month records are not automatically zero trade. These issues can create false entry, exit, price, or diversification signals if they are not made explicit.
Data
The verified member-state Parquet file contains 15,096 rows and 17 columns. The verified EU-total file contains 3,061 rows and 18 columns. Both cover January 2024 only. Unit value is missing in 874 member-state rows and 66 EU-total rows where the necessary quantity information is unavailable; supplementary-unit codes themselves are present.
The main processed grain is reporter by partner by period by CN8 product by supplementary unit. The data dictionary includes dataset schema, pipeline outputs, quality checks, EU27 reporter codes, and sources.
Methodology
The crawler processes one month at a time, validates each output, and resumes from valid Parquet parts. Transformations preserve leading zeroes in CN8 codes, aggregate statistical procedures, retain raw filtered rows, and compute member-state and EU-total outputs separately. The project avoids hard-coding an unverified CBAM mapping.
Main Findings
The current files establish a usable one-month release and an explicit data contract. They do not establish trends, trade-policy effects, product upgrading, or complete 2018-to-latest coverage.
Robustness, Validation, or Model Assessment
Documented checks cover schema, unique grain, reporters, periods, numeric types, product-code preservation, quantities, and aggregation. The current transform test does not collect successfully because of a dynamic-import/dataclass initialization error, so test-suite completion is not claimed.
Tools and Technologies
Python, pandas, Parquet, Eurostat Comext bulk data, JSONL manifests, and spreadsheet data dictionaries are directly demonstrated.
Limitations
Only January 2024 is verified. CN revisions require year-specific concordances, chapter 99 and confidentiality can suppress detail, and EU-total imports are not final-consumption destinations. The failed test collection still requires repair.
Deliverables
- Monthly acquisition and transformation scripts
- January 2024 member-state and EU-total Parquet releases
- Data dictionary and quality rules
- Download manifest and raw/interim/processed directory contract