Longevity of Trade Partnerships: A Duration Analysis of Ethiopia's International Trade Flows
Description
This dataset provides a comprehensive panel of Ethiopia’s bilateral trade flows spanning from 1997 to 2023, specifically curated to examine the longevity and stability of international trade partnerships. Moving beyond traditional trade volume analysis, this data is structured for survival analysis (duration analysis) to track the lifecycle of export relationships, identifying the specific timing of trade initiation, persistence, and dissolution. The dataset integrates annual bilateral export values with a robust set of gravity-based variables, including geographic distance, contiguity (shared borders), and regional classifications, with a specific focus on Sub-Saharan African partners.
Files
Steps to reproduce
Steps to Reproduce 1. Data Merging and Pre-processing The final analytical dataset was constructed by merging records from UN Comtrade (1997–2023), World Bank (WDI/WGI), and CEPII Gravity databases using Stata 18. All monetary values were converted to real terms, and institutional scores were aggregated using Principal Component Analysis (PCA) via the pca command, where the first principal component served as the summary index for Institutional Quality. 2. Survival Data Setup Using the stset command in Stata 18, the panel data was declared as survival-time data. Trade "spells" were identified as continuous sequences of non-zero export values. Multiple spells for the same country-pair were treated as independent to maintain robust survival estimates. Relationships still active in 2023 were identified as right-censored. 3. Model Estimation and Validation Non-parametric Analysis: Baseline survival and hazard rates were visualized using Kaplan-Meier (sts graph) and Nelson-Aalen (sts graph, cumhaz) estimates. Parametric Modeling: We employed the streg command to fit random-effects parametric survival models. A Weibull distribution was specified to capture the monotone hazard rate of trade dissolution. Unobserved Heterogeneity: The frailty(gamma) option was utilized to address unobserved heterogeneity (intra-pair correlation) across Ethiopia's trade partners. Goodness of Fit: Model selection was confirmed using the estat gofplot command, a new feature in Stata 18, which provided Cox-Snell residual plots to verify that the Weibull distribution appropriately fits the trade duration data. Quick Metadata Checklist for Mendeley Software used: Stata 18 Main Commands: stset, streg, sts graph, pca, estat gofplot Data Format: csv (Standard Data)