Replication package for: Tax-Base Concentration and Granular Fiscal Risk: Evidence from Macau's Gaming Sector
Description
Replication package for "Tax-Base Concentration and Granular Fiscal Risk: Evidence from Macau's Gaming Sector" (Economic Modelling). The package contains the public source files as downloaded from the statistical agencies (data/raw/original/, with the cells used highlighted), the data management code that builds the estimation dataset directly from them (code/build_analysis_data.py), and one script per exhibit named after it: code/Table2.py produces Table 2, code/Figure5.py produces Figure 5. Running VERIFY.py checks the whole paper in one command. Scripts write only to output/; the deposited data are never modified. Firm-level revenue for the six Macau concessionaires is licensed Compustat Global data and is not redistributed. A synthetic panel preserving the original identifiers, variable names and data structure is supplied in its place, calibrated to aggregate moments of operator revenue growth, so that every script runs without a WRDS subscription. README section 2.3 gives the query needed to obtain the licensed panel. Extracted into an empty directory with no licensed data present, the package reproduces all 268 published values in Tables 1-9 and every figure except Figure 6(a), which plots one line per operator; the published panel is included in published/Figure6a.pdf.
Files
Steps to reproduce
Requires Python 3.11 (pandas, numpy, statsmodels, linearmodels, matplotlib, openpyxl, xlrd); Figures 9 and 11 additionally require R 4.4 with the packages BVAR, readxl, dplyr and janitor. Unzip the package and run every command from its root folder. To check the whole paper in one command: python VERIFY.py (about a minute; add --bvar to also estimate the two Bayesian VARs behind Figures 9 and 11). It rebuilds every table and figure, compares each published value with the reproduced one, and prints a verdict. Each exhibit is one script named after it: python code/Table2.py produces Table 2 and writes output/Table2.xlsx, whose vs_paper sheet compares every value with the published one; python code/Figure5.py produces output/Figure5.pdf. The same pattern covers Tables 1-9 and Figures 2-11; README sections 5 and 6 map each exhibit to its script and its page in the paper. To rebuild the dataset from the public sources: python code/build_analysis_data.py reads the agency files in data/raw/original/ and writes output/analysis_data.xlsx, then compares it with the deposited data/analysis_data.xlsx. No script writes to data/. Firm-level Compustat Global data is licensed and not included. A synthetic panel with the same identifiers, variable names and structure is supplied in its place (python code/make_synthetic_firm_panel.py regenerates it). Every reported value reproduces without the licensed data; only Figure 6(a), which plots one line per operator, needs it, and the published panel is included in published/Figure6a.pdf. README section 2.3 explains how to obtain the licensed panel.
Institutions
- Bank of KoreaSeoul, Seoul