This is the replication package for my paper "Self-employment and Labor Market Risks" published in the International Economic Review.

If you find this package helpful in your work, please consider citing the paper:

@article{audoly2025self,
  title={Self-Employment And Labor Market Risks},
  author={Audoly, Richard},
  journal={International Economic Review},
  volume={66},
  number={2},
  pages={661--686},
  year={2025},
  publisher={Wiley Online Library}
}

The code was last run on:

• Stata 18 MP
• Matlab 2020a

The corresponding scripts are contained in the data folder.

download_sipp.py downloads the SIPP data from the NBER website for survey years 1996, 2001, 2004, and 2008. The script only works with Python 2 for some obscure reason. To get them into Stata format, navigate to the 1996 folder and run 1_make1996.do Proceed similarly for survey years 2001, 2004, and 2008.

Set the data paths in globals.do to the SIPP data and run the do-files in the order suggested by the prefix:

1_extract_variables.do extracts the main variables from the raw data.

2_prep_panels.do prepares monthly panels and defines wealth variables.

3_prep_spells.do constructs employment spells and assign status as paid- or self-employed

4_prep_samples.do prepare the analysis samples. This file calls the matlab script clustering.m, which groups workers using k-means. You will need to adjust the macro matlab_command in globals.do to run the script in one go. This script produces Figure 2.

5_analysis.do produces the plots and tables reported in the data section: Figure 1 and Tables 1-6.

6_model_inputs.do computes and stores all inputs for the model.

7_bootstrap.do bootstraps the moment computations in 6_model_inputs.do to get the standard errors for the targeted moments.

A_macro_trends.do benchmarks the macro trends in unemployment and self-employment derived from the SIPP to the CPS.

B_additional_checks.do performs a series of additional checks on the data requested during the revision process. Not all of them made it into the final manuscript.

The other do-files in the folder are called by the scripts previously described.

The corresponding scripts are in the model folder.

main.m produces all model output: Figures 3-5 and Tables 7-9. Most of the heavy lifting is done by the matlab functions in the lib folder.