15  Paper Writing

15.1 Introduction

The final paper adapts the Post-Soviet replication workflow to a new region or policy question. Students should not start from a blank page. They should start from the replication architecture: data, descriptive evidence, baseline gravity, fixed effects, PPML, robustness, and cautious policy interpretation.

The goal is a publication-style empirical paper supported by a reproducible Python notebook.

15.2 Choosing a region

A good region has a clear economic motivation and enough variation for empirical analysis. The region should be large enough to form bilateral pairs and long enough in time to study policy variation.

Possible regions include Africa, GCC, MENA, COMESA, ECOWAS, EAC, ASEAN, and the Western Balkans.

The region must be justified. Do not choose a region only because data are available. Explain why the region matters for trade policy, integration, food security, logistics, or development.

15.3 Choosing an institutional or policy variable

The policy variable should be measurable at the country-pair-year level whenever possible. Examples include:

  • joint WTO membership;
  • joint membership in a regional economic community;
  • shared trade agreement;
  • customs union membership;
  • EU integration status;
  • trade facilitation or logistics indicators.

The variable should have variation. A dummy that is always 0 or always 1 cannot explain bilateral trade differences.

15.4 Turning a policy issue into a gravity question

Weak question:

Does regional integration matter?

Better question:

Is joint membership in a regional economic community associated with higher bilateral exports after controlling for GDP, distance, language, contiguity, and fixed effects?

The better question identifies the outcome, policy variable, controls, and empirical comparison.

15.5 Defining the dependent variable

The dependent variable should be a bilateral trade flow. Students must state:

  • whether the flow is exports or imports;
  • whether it is total trade or sector-specific trade;
  • whether it is reported by exporter, importer, or reconciled source;
  • whether zero flows are included;
  • the currency and scale.

For log-linear OLS, use positive trade flows. For PPML, use trade in levels and keep zeros when the dataset includes them.

15.6 Choosing the sample period

The sample period should match the policy question. If the policy variable changes in 2015, a one-year cross-section cannot identify before-and-after variation. If the region experienced major shocks, the paper should explain whether those years are included or excluded.

A good data section reports:

  • country list;
  • years;
  • number of observations;
  • missing values;
  • zero-flow treatment;
  • source of trade flows and covariates.

15.7 Selecting estimators

The final paper should use a minimum model sequence:

  1. baseline OLS;
  2. fixed-effects OLS;
  3. PPML;
  4. at least one robustness check.

Additional estimators may include DDM, BVU, GPML, pair fixed effects, or structural PPML. Use them because they answer a research question, not because more columns look stronger.

15.8 Designing robustness checks

Robustness checks should be theoretically motivated. Common choices include:

  • alternative trade-flow measure;
  • alternative distance measure;
  • positive-flow sample versus all valid flows;
  • different fixed-effect structures;
  • PPML versus log-linear OLS;
  • excluding extreme trade-flow observations;
  • region or subperiod checks.

Each robustness check should test whether the main finding depends on a fragile modeling choice.

15.9 Interpreting institutional coefficients cautiously

Institutional variables are not self-interpreting. A positive coefficient on a trade agreement dummy does not automatically prove that the agreement caused trade to increase.

Students should write:

Joint membership in [policy variable] is associated with higher bilateral trade conditional on the controls and fixed effects.

Avoid:

[Policy variable] caused trade to increase.

Causal language requires a research design that supports causal identification.

15.10 Regional paper examples

Region Possible question Key variable Main estimator
Africa Is joint WTO or REC membership associated with intra-African trade? WTO or REC joint membership PPML with fixed effects
GCC Is regional integration associated with oil-linked bilateral trade? GCC membership or sector indicator OLS and PPML
MENA Do WTO membership or trade agreements predict bilateral trade? WTO or agreement dummy FE OLS and PPML
COMESA Does shared bloc membership increase regional exports? COMESA joint membership PPML
ECOWAS Is regional bloc membership associated with higher intra-bloc trade? ECOWAS joint membership FE OLS and PPML
EAC Does deeper regional integration predict bilateral exports? EAC joint membership PPML with robustness checks
Western Balkans Does EU integration matter more than WTO membership? EU alignment and WTO joint membership FE OLS and PPML
ASEAN Are regional trade agreements associated with higher trade? RTA joint membership PPML

These are research-design examples, not pre-written findings.

15.11 From replication to original paper

The move from replication to original paper should be systematic:

  • keep the workflow;
  • change the region;
  • change the policy variable;
  • rebuild the dataset;
  • rerun descriptive statistics;
  • rerun OLS, fixed-effects, PPML, and robustness models;
  • rewrite the interpretation for the new setting.

The Post-Soviet paper teaches the architecture. The student’s paper applies that architecture to a new empirical question.

15.12 Manuscript structure

A region-specific gravity paper should include:

  1. introduction and motivation;
  2. literature and policy context;
  3. data and variables;
  4. gravity model and estimators;
  5. descriptive evidence;
  6. main results;
  7. robustness checks;
  8. policy interpretation;
  9. conclusion and limitations.

The paper should be concise. Tables and figures should support the argument, not replace it.