Replication exercise for Computational Economics

Authors: Guillaume POUSSE, Vivan SHARMA

Overview and Data Availability

This is the ReadMe for a replication of Clemens, Lews & Postel, "Immigration Restrictions as Active Labor Market Policy: Evidence from the Mexican Bracero Exclusion." (2018). The citation can be found below. The original code was in STATA, and can be found here alongside the data used.

Clemens, Michael A., Ethan G. Lewis, and Hannah M. Postel. 2018. "Immigration Restrictions as Active Labor Market Policy: Evidence from the Mexican Bracero Exclusion." American Economic Review, 108 (6): 1468–87.

Computational Requirements

• Julia (produced using version 1.11.2)
• The following packages:
  • Test
  • HypothesisTests
  • Statistics
  • Plots
  • StatFiles
  • CSV
  • DataFrames
  • Dates
  • ReadStatTables
  • ReadStat
  • PanelDataTools
  • FixedEffectModels
  • GLM
  • RegressionTables
  • Distributions
  • PrettyTables
  • CategoricalArrays

Scripts

The files are named after their respective purposes. Data_cleaning.jl cleans the raw data and exports the cleaned csv.'s to the output folder. Table1.jl, Table2.jl, Table3.jl all produce regression tables by the same number as in the paper. Similarly, Figure2.jl, Figure3.jl do the same for the graphs. Functions.jl is a script wherein intermediate functions are defined. These functions are used in several of the other scripts. Lastly, bracero_pkg.jl is the main file of interest - it defines the package.

Instructions for replication

1. Go to your computer's terminal.
2. Enter: git clone https://github.com/glpousse/bracero_pkg.jl
3. Enter: cd bracero_pkg.jl
4. Open the Julia REPL.
5. Enter:

   using Pkg
   Pkg.activate(".")
   Pkg.instantiate()

6. Enter: using bracero_pkg
7. Enter: bracero_pkg.run()
8. This should run all of the code and produce all of the output.

Notes on replication

For the most part, we were able to fully replicate the main findings of the paper. There are discrepancies in Table 2, most likely due to how missing values were dealth with. In their STATA code, the authors generate a variable domestic_seasonal as the rowtotal of three other variables. They defined it such that it took the value 0 if all three inputs were missing and added whatever values were available, even if one or two of the others were missing. They converted all observations outside of a certain time frame to missings. This may be a problematic approach, and can bias the estimates from the regression. While there will be a loss in explanatory power, the regression results could change significantly if we turned the zeros summed from missing inputs into missing values. This perhaps warrants further research.

Also, Table 1 is split into two halves; the first half is exported to a .txt file, and the second half is printed on the screen when the code is run. The second half contains the information on semi-elasticities seen in the paper.