2  Overview

2.1 Course philosophy

NREC6006 is not a traditional international trade theory course. It is an applied research course on gravity models in Python. The final objective is a reproducible empirical project that can develop into a publication-style gravity paper.

The course begins with replication. Students replicate a Post-Soviet gravity paper before designing their own regional extension. This sequence teaches how empirical trade research actually works: the paper, data, code, tables, and interpretation must be consistent.

2.2 Why replication comes first

Replication is the starting point because it forces students to answer practical research questions:

  • What is the unit of observation?
  • Which variables are required?
  • Which sample restrictions matter?
  • Which estimators are used?
  • Which coefficients are robust?
  • Which discrepancies require explanation?

The reference dataset contains 5,253 observations, 15 exporters, 15 importers, and years 1992-2020. The core institutional variables are \(wto\_joint\), \(EU\_joint\), and \(EAEU\_joint\).

2.3 Why Python is used

Python is used because it supports a complete applied workflow in one environment:

  • data inspection and cleaning with pandas;
  • numerical transformations with numpy;
  • OLS, fixed-effects-style formulas, PPML, and GPML with statsmodels;
  • descriptive plots with matplotlib;
  • network analysis with networkx;
  • reproducible notebooks that combine code, tables, and interpretation.

The course also helps students read gravity examples from other software traditions and translate them into Python-based workflows.

2.4 What students produce

By the end of the course, students produce:

  1. a documented gravity dataset;
  2. a reproducible Python notebook;
  3. replication tables for the Post-Soviet paper;
  4. robustness and sensitivity notes;
  5. a presentation;
  6. a publication-style gravity paper.

The paper is not written after the econometrics are finished. It is built throughout the course.

2.5 Replication workflow

The course follows a research workflow:

research question -> data -> descriptive analysis -> OLS -> FE -> PPML -> robustness -> paper

In practice, the workflow is iterative. Descriptive analysis may reveal a sample issue. Fixed effects may change the institutional interpretation. PPML may raise questions about zero trade flows. Robustness checks may require returning to the data.

2.6 Model sequence

Students learn the estimators in the order needed for the replication:

Stage Estimator Purpose
Baseline OLS Teach the log-linear gravity model
Heterogeneity FE OLS Control for exporter and importer differences
Transformation DDM Show the mechanics behind two-way fixed effects
Normalization BVU Study trade intensity relative to economic size
Multiplicative model PPML Estimate gravity in levels
Robustness GPML Compare alternative multiplicative assumptions
Structural gravity Structural PPML Add exporter-year and importer-year fixed effects

The goal is not to memorize estimators. The goal is to understand what each estimator controls for and how that changes interpretation.

2.7 Output standard

Every student output should be reproducible. A table in the paper must be traceable to a dataset, notebook cell, model formula, and interpretation note.

Note

The course standard is not “the coefficient looks right.” The standard is: the data, code, estimate, and interpretation can be checked by another researcher.