Antonio Coppola, Harvard University ([email protected])
Matteo Maggiori, Stanford University Graduate School of Business ([email protected])
Brent Neiman, University of Chicago Booth School of Business ([email protected])
Jesse Schreger, Columbia Business School ([email protected])

This README describes the overall structure of the replication package for this paper. The uppermost directory of the replication folder contains the following objects:

1. README.md (the file you are reading right now)
2. README.pdf (the file you are reading right now)
3. CMNS_Data_Guide.pdf
4. aggregation_data_sources (a folder)
5. main_analysis (a folder)
6. cmns_aggregation (a folder)
7. data.zip (a compressed folder)
8. CMNS_Master.sh
9. CMNS_Master.do

CMNS_Master.sh, found in the uppermost directory of this repository, is the primary executable script, which runs the replication from start to finish. It does this by calling the CMNS_Master.do file, written in Stata, which in turn runs the various steps of the code and also defines project globals. Since a few of the steps are written in Python and R as well, CMNS_Master.sh also calls the Python and R executables in order to run these. The individual jobs are split into three folders, which are executed in the following sequence:

1. The folder aggregation_sources contains jobs that read in security- and issuer-level information from various commercial data sources that are used as inputs into our aggregation algorithm, which associates the universe of traded equity and debt securities with their issuers' ultimate parents, including those issued in tax havens.

2. The folder cmns_aggregation contains files that implement our aggregation procedure. They take as input the processed files produced by the jobs in aggregation_sources (whose structure is described in the script cmns_aggregation/UP_Aggregation.py). This code can also be used in a stand-alone manner by researchers interested purely in our aggregation procedure. To that end, the script cmns_aggregation/UP_Aggregation.py provides configurable parameters that allow to bypass certain source files if these are missing, giving a user the option of only using a subset of the commercial data sources imported in the previous steps in their own research. For the purpose of replicating our paper, however, all input sources should be used. We also recommend that stand-alone users of the aggregation code employ all the aggregation sources for their own projects if possible, since the quality of the final mapping will deteriorate as sources are removed.

3. The folder main_analysis contains files that implement the rest of the analysis in our paper, which includes building the rest of our data sources and producing the figures and tables in the paper. The following files in main_analysis reproduce the tables and figures in the paper:

   • Table 1: Reallocation_Matrices.do
   • Tables 2-5: Restatement_Tables.do
   • Tables 6-7: Firm_Level_Tables.do
   • Tables 8-9: Restatement_Tables.do
   • Figure 1: Cross_Country_Graphs.do
   • Figures 2-3: Network_Charts.R
   • Figure 4: BRICS_Times_Series.py
   • Figure 5: Cross_Country_Graphs.do
   • Figure 6: Separately-produced illustration, no .do file
   • Figures 7-8: China_VIE_NFA.do
   • Figure 9: Representativeness_Analysis.do
   • Figure 10: Home_Bias.do
   • Figure 11: Sales_Analysis.do

The bash script CMNS_Master.sh is the main executable file for the procedure. Launching CMNS_Master.sh will run the full replication. This file should be called as:

bash CMNS_Master.sh

• The code is built for Unix systems and assumes that the Stata, R, and Python interpreters are configured on your executable path. The required versions are version 15+ for Stata, version 3.6+ for Python, and version 4.0.2+ for R. Packages may need to be installed using a package manager (e.g. pip) as necessary.

• Prior to running the procedure, please be sure to perform a find-and-replace in the code folder for the following expressions. These are user- and system-specific procedure parameters that will need to be filled in accordingly. Individual files also point out these parameters whenever they are present:

  <CODE_PATH>: Path to the copy of this code repository on the host system
  <DATA_PATH>: Path to the folder containing the data, in which the procedure is executed
  <API_KEY>: If using the script in `FIGI_Download.py`, an OpenFIGI API key is required. This should be
             obtained directly from OpenFIGI and used in place of this placeholder
  

Most of the data used in this project are commercially available, so cannot be included in this replication packet. The file CMNS_Data_Guide.pdf gives an overview of all input files used. In the cases in which they are public, the actual files can be found in data.zip. For some private data files, in the cases in which the structure of the data might otherwise be ambiguous, the data.zip archive also contains sample version of the files that only include sample rows: these are provided in order to demonstrate the structure of the data. Given there are a large number of private data files that are required to run the code, we acknowledge that most readers of this packet will not use the code to literally replicate our results, but rather can look to the code to understand what calculations we made for each result.