Traiberman (2019) Replication Project

Project Description

This repository contains a partial computational replication of the structural labor model presented in Traiberman (2019). Because the original Danish administrative data are strictly confidential, this project translates the core Expectation-Maximization (EM) first-stage algorithm into Julia and validates its numerical stability and convergence using a synthetic Data Generating Process (DGP) build exactly to match all the First Stage code requirements

• Paper: "Occupations and Import Competition: Evidence from Denmark" by Sharon Traiberman (AER 2019). DOI: 10.1257/aer.20161925
• Original Replication Package: DOI: 10.3886/E231387V1

Course and Author

• Course: Computational Economics (PhD program at Collegio Carlo Alberto) taught by Prof. Florian Oswald
• Author: Debora Candela - Senior Allieva at CCA

Computational Requirements & Data

• Software: Tested on Julia 1.12.4 (and Quarto 1.9.37 for the report)
• Runtime & Memory: The full pipeline (DGP + EM algorithm) takes only a few seconds to run on a standard laptop, i.e. it requires minimal RAM.
• Data Access: No external data downloads are required. Because the original Danish administrative data is strictly confidential, this package dynamically generates a synthetic 12,000 worker-year panel upon execution.

Installation and Environment Setup

This project ensures full reproducibility by utilizing a localized Julia environment (Project.toml and Manifest.toml).

To download the repository and start Julia, run the following commands in your standard terminal:

git clone https://github.com/Debora-Candela/TraibermanReplication.git
cd TraibermanReplication
julia 

Then, to replicate easily the outputs or test the environment, run the following commands in the Julia REPL in the package environment (pkg>) using ]:

activate .
instantiate

Press backspace to return to the standard Julia prompt (julia>), then load the package:

using TraibermanReplication

TraibermanReplication.run()
TraibermanReplication.run_tests()

Compile the Technical Report

The project includes a detailed Quarto report (report.qmd) containing the theoretical background, implementation details, and replication exhibits. To render the report to HTML, that you can anyhow already see here exit Julia and run the following command in your standard terminal:

quarto render report.qmd

Repository Structure

The replication package is a modular Julia project with the following structure:

├── src/
│   ├── DGP.jl                     # Generates the synthetic 12,000 worker-year panel and covariates
│   ├── FirstStage.jl              # Implements the core Expectation-Maximization (EM) algorithm
│   ├── Results.jl                 # Processes EM outputs to build the synthetic Table 4 analogue
│   └── TraibermanReplication.jl   # Main module orchestrating the end-to-end pipeline
├── test/
│   └── runtests.jl                # Unit testing suite asserting DGP integrity and First Stage mathematical bounds
├── report.qmd                     # Quarto source file containing the technical replication report
├── Project.toml                   # Julia environment dependencies
├── Manifest.toml                  # Exact package versions for strict reproducibility
└── README.md                      # This file

File Descriptions:

• TraibermanReplication.jl: The entry point of the package. It exports the run() function which sequentially triggers the data generation and the EM estimator;

• DGP.jl: Contains generate_full_dgp(), which simulates worker demographics, calculates structural human capital profiles, and pre-allocates the transition matrices (XPI, YPI, dex) required by the EM algorithm;

• FirstStage.jl: Contains estimate_first_stage(), the computational engine of the project. It alternates between Weighted Least Squares (WLS) estimation of wage/transition parameters and Bayesian updating of latent type probabilities until convergence;

• Results.jl: Contains table4_like(), a post-processing function that aggregates the raw EM outputs into structural income and comparative advantage exhibits;

• runtests.jl: Executes the full pipeline on a reduced parameter set to automatically verify data dimensions, probability distribution constraints and estimator stability.