Module tax_deficit_simulator.utils
This module provides several functions useful to run the simulations defined in "calculator.py" or "firm_level.py".
It also provides various utils for the "app.py" file, especially to add file download buttons.
Expand source code
"""
This module provides several functions useful to run the simulations defined in "calculator.py" or "firm_level.py".
It also provides various utils for the "app.py" file, especially to add file download buttons.
"""
# ----------------------------------------------------------------------------------------------------------------------
# --- Imports
import base64
import os
import numpy as np
# ----------------------------------------------------------------------------------------------------------------------
# --- Paths to files that can be downloaded from the simulator
path_to_files = os.path.dirname(os.path.abspath(__file__))
path_to_files = os.path.join(path_to_files, 'files')
# ----------------------------------------------------------------------------------------------------------------------
# --- Utils for the calculator.py file
COUNTRIES_WITH_MINIMUM_REPORTING = ['KOR', 'NLD', 'IRL', 'FIN']
COUNTRIES_WITH_CONTINENTAL_REPORTING = ['AUT', 'NOR', 'SVN', 'SWE']
def rename_partner_jurisdictions(row):
"""
In the OECD data, each reporting country displays a line "Foreign Jurisdictions Total", which displays the sum of
revenues, profits, corporate income taxes paid, etc. for the parent country across all foreign partner jurisdic-
tions. In most cases, we want to eliminate this total to avoid any double-counting. But some countries (see list of
alpha-3 codes above) only display a domestic vs. foreign breakdown and in these cases, it is important that we do
not erase the "Foreign Jurisdictions Total" line. Therefore, we slightly rename it for these countries.
"""
if row['Parent jurisdiction (alpha-3 code)'] in COUNTRIES_WITH_MINIMUM_REPORTING:
if row['Partner jurisdiction (whitespaces cleaned)'] == 'Foreign Jurisdictions Total':
return 'Foreign Total'
else:
return row['Partner jurisdiction (whitespaces cleaned)']
else:
return row['Partner jurisdiction (whitespaces cleaned)']
def manage_overlap_with_domestic(row, kind):
"""
When cleaning and preprocessing the OECD data, we introduce several indicator variables:
- one that takes value 1 if and only if the partner jurisdiction is a tax haven;
- another that takes value 1 if and only if the partner jurisdiction is a non-haven country;
- and a last one that indicates whether the parent and partner jurisdictions coincide.
In the breakdown of the total tax deficit into domestic, tax-haven-based and non-haven tax deficits, we need to
avoid any double-counting and two of these indicator variables cannot take the value 1 simultaneously.
We therefore give the priority to the "Is domestic?" indicator variable: no matter whether the jurisdiction is a tax
haven or not, this indicator variable takes the value 1 if the parent and partner are the same, while the other in-
dicator variables are set to 0.
"""
if row['Is domestic?']:
return 0
else:
if kind == 'haven':
return row['Is partner jurisdiction a tax haven?']
elif kind == 'non-haven':
return row['Is partner jurisdiction a non-haven?']
def impute_missing_carve_out_values(
row,
avg_carve_out_impact_non_haven, avg_carve_out_impact_tax_haven,
avg_carve_out_impact_domestic, avg_carve_out_impact_aggregate
):
"""
This function allows to impute missing carve-out values based on two inputs: pre-tax profits and the average redu-
ction in pre-tax profits due to carve-outs observed in OECD data. The factor used is determined by the partner ju-
risdiction group in which profits are booked: tax havens, non-havens, domestic, aggregate partner jurisdictions.
"""
if not np.isnan(row['CARVE_OUT']):
return row['CARVE_OUT']
elif row['Is domestic?'] == 1:
return row['Profit (Loss) before Income Tax'] * avg_carve_out_impact_domestic
elif row['Is partner jurisdiction a non-haven?'] == 1:
return row['Profit (Loss) before Income Tax'] * avg_carve_out_impact_non_haven
elif row['Is partner jurisdiction a tax haven?'] == 1:
return row['Profit (Loss) before Income Tax'] * avg_carve_out_impact_tax_haven
else:
return row['Profit (Loss) before Income Tax'] * avg_carve_out_impact_aggregate
# ----------------------------------------------------------------------------------------------------------------------
# --- Utils for the firm_level.py file
ADDITIONAL_ETRs = {
'GEO': 0.15,
'BLR': 0.18,
'SOM': 0.05,
'PNG': 0.3,
'TLS': 0.1,
'ABW': 0.25,
'MOZ': 0.32,
'GAB': 0.3,
'CRI': 0.3
}
def compute_ETRs(row, kind):
"""
This function is used in the "firm_level.py" file to determine, for each partner jurisdiction where the multinatio-
nal is active, what effective tax rate should be used. When we can, we compute the ETR on a cash basis as the ratio
of the amount of corporate income taxes paid to profit before tax.
If this computation yields an error (0 profit before tax or NaN value at the numerator or denominator):
- for non-bank multinationals, we take the statutory corporate income tax rate of the partner jurisdiction;
- for banks, we take the average ETR faced by the multinational in this jurisdiction over the last 6 years.
Besides, for banks, if the computed ETR is negative, we replace it by the pre-computed average ETR.
"""
if kind == 'mne':
# In the case of a non-bank multinational
if row['Partner jurisdiction code'] in ADDITIONAL_ETRs.keys():
return ADDITIONAL_ETRs[row['Partner jurisdiction code']]
else:
try:
effective_tax_rate = row['CIT paid'] / row['Profit before tax']
except:
# If the usual computation yields an error, we use the statutory corporate tax rate
effective_tax_rate = row['Statutory CIT rate']
return effective_tax_rate
elif kind == 'bank':
# In the case of a bank
try:
effective_tax_rate = row['CIT paid'] / row['Profit before tax']
except:
# If the usual computation yields an error, we use the average ETR computed over the last 6 years
effective_tax_rate = row['Average ETR over 6 years']
if effective_tax_rate < 0:
# If the ETR computed on a cash basis is negative, we replace it with the average ETR
effective_tax_rate = row['Average ETR over 6 years']
return effective_tax_rate
# ----------------------------------------------------------------------------------------------------------------------
# --- Utils for the app.py file
def get_table_download_button(df, scenario, effective_tax_rate, company=None, taxing_country=None, carve_out_rate=None):
"""
This function is used in the "app.py" file to generate the HTML code that instantiates the download button on the
following pages: "Case study with one multinational" (scenario=0), "Multilateral implementation scenario" (scenario
=1), "Partial cooperation scenario" (scenario=2) and "Unilateral implementation scenario" (scenario=3).
The HTML code is then injected in a st.markdown() component, the "allow_unsafe_html" argument being set to True.
This sort of hack was found on Streamlit user forums.
"""
# We output the DataFrame to a csv format
csv = df.to_csv(index=False)
# We encode the csv file in the right format
b64 = base64.b64encode(csv.encode()).decode()
href = f'<a href="data:file/csv;base64,{b64}"'
if scenario == 0:
# "Case study with one multinational" page
company_name = company.lower().replace(' ', '_')
href += f' download="{company_name}_{effective_tax_rate}_perc.csv">'
elif scenario == 1:
# "Multilateral implementation scenario" page
href += f' download="multilateral_scenario_{effective_tax_rate}_perc.csv">'
elif scenario == 2:
# "Partial cooperation scenario" page
href += f' download="partial_cooperation_scenario_{effective_tax_rate}_perc.csv">'
elif scenario == 3:
# "Unilateral implementation scenario" page
taxing_country = taxing_country.lower().replace(' ', '_')
if 'china' in taxing_country:
taxing_country = 'china'
href += f' download="unilateral_scenario_{taxing_country}_{effective_tax_rate}_perc.csv">'
elif scenario == 4:
# Substance-based carve-outs page
href += f' download="min_ETR_{effective_tax_rate}_perc_CO_rate_{carve_out_rate}_perc.csv">'
else:
raise Exception('Value not accepted for the scenario argument.')
href += '<input type="button" value="Click here to download the table" class="download-button table"></a>'
return href
def get_report_download_button():
"""
Following the same principle, this function builds the HTML code that instantiates the download button allowing the
user to obtain the full-text version of the study in PDF format.
"""
# We fetch and read the .pdf file from the files folder
path = os.path.join(path_to_files, 'EUTO2021.pdf')
with open(path, 'rb') as file:
report_content = file.read()
# We encode it to the right format
b64 = base64.b64encode(report_content).decode()
# And we build the HTML code
href = f'<a href="data:application/octet-stream;base64,{b64}" download="{os.path.basename(path)}">'
href += '<input type="button" value="Click here to download the full-text report (PDF)" '
href += 'class="download-button pdf"></a>'
return href
def get_carve_outs_note_download_button():
"""
Following the same principle, this function builds the HTML code that instantiates the download button allowing the
user to obtain the note on substance-based carve-outs in PDF format.
"""
# We fetch and read the .pdf file from the files folder
path = os.path.join(path_to_files, 'carve_outs_note.pdf')
with open(path, 'rb') as file:
note_content = file.read()
# We encode it to the right format
b64 = base64.b64encode(note_content).decode()
# And we build the HTML code
href = f'<a href="data:application/octet-stream;base64,{b64}" download="{os.path.basename(path)}">'
href += '<input type="button" value="Click here to download the note on substance-based carve-outs (PDF)" '
href += 'class="download-button pdf"></a>'
return hrefFunctions
def compute_ETRs(row, kind)-
This function is used in the "firm_level.py" file to determine, for each partner jurisdiction where the multinatio- nal is active, what effective tax rate should be used. When we can, we compute the ETR on a cash basis as the ratio of the amount of corporate income taxes paid to profit before tax.
If this computation yields an error (0 profit before tax or NaN value at the numerator or denominator):
-
for non-bank multinationals, we take the statutory corporate income tax rate of the partner jurisdiction;
-
for banks, we take the average ETR faced by the multinational in this jurisdiction over the last 6 years.
Besides, for banks, if the computed ETR is negative, we replace it by the pre-computed average ETR.
Expand source code
def compute_ETRs(row, kind): """ This function is used in the "firm_level.py" file to determine, for each partner jurisdiction where the multinatio- nal is active, what effective tax rate should be used. When we can, we compute the ETR on a cash basis as the ratio of the amount of corporate income taxes paid to profit before tax. If this computation yields an error (0 profit before tax or NaN value at the numerator or denominator): - for non-bank multinationals, we take the statutory corporate income tax rate of the partner jurisdiction; - for banks, we take the average ETR faced by the multinational in this jurisdiction over the last 6 years. Besides, for banks, if the computed ETR is negative, we replace it by the pre-computed average ETR. """ if kind == 'mne': # In the case of a non-bank multinational if row['Partner jurisdiction code'] in ADDITIONAL_ETRs.keys(): return ADDITIONAL_ETRs[row['Partner jurisdiction code']] else: try: effective_tax_rate = row['CIT paid'] / row['Profit before tax'] except: # If the usual computation yields an error, we use the statutory corporate tax rate effective_tax_rate = row['Statutory CIT rate'] return effective_tax_rate elif kind == 'bank': # In the case of a bank try: effective_tax_rate = row['CIT paid'] / row['Profit before tax'] except: # If the usual computation yields an error, we use the average ETR computed over the last 6 years effective_tax_rate = row['Average ETR over 6 years'] if effective_tax_rate < 0: # If the ETR computed on a cash basis is negative, we replace it with the average ETR effective_tax_rate = row['Average ETR over 6 years'] return effective_tax_rate -
-
Following the same principle, this function builds the HTML code that instantiates the download button allowing the user to obtain the note on substance-based carve-outs in PDF format.
Expand source code
def get_carve_outs_note_download_button(): """ Following the same principle, this function builds the HTML code that instantiates the download button allowing the user to obtain the note on substance-based carve-outs in PDF format. """ # We fetch and read the .pdf file from the files folder path = os.path.join(path_to_files, 'carve_outs_note.pdf') with open(path, 'rb') as file: note_content = file.read() # We encode it to the right format b64 = base64.b64encode(note_content).decode() # And we build the HTML code href = f'<a href="data:application/octet-stream;base64,{b64}" download="{os.path.basename(path)}">' href += '<input type="button" value="Click here to download the note on substance-based carve-outs (PDF)" ' href += 'class="download-button pdf"></a>' return href -
Following the same principle, this function builds the HTML code that instantiates the download button allowing the user to obtain the full-text version of the study in PDF format.
Expand source code
def get_report_download_button(): """ Following the same principle, this function builds the HTML code that instantiates the download button allowing the user to obtain the full-text version of the study in PDF format. """ # We fetch and read the .pdf file from the files folder path = os.path.join(path_to_files, 'EUTO2021.pdf') with open(path, 'rb') as file: report_content = file.read() # We encode it to the right format b64 = base64.b64encode(report_content).decode() # And we build the HTML code href = f'<a href="data:application/octet-stream;base64,{b64}" download="{os.path.basename(path)}">' href += '<input type="button" value="Click here to download the full-text report (PDF)" ' href += 'class="download-button pdf"></a>' return href -
This function is used in the "app.py" file to generate the HTML code that instantiates the download button on the following pages: "Case study with one multinational" (scenario=0), "Multilateral implementation scenario" (scenario =1), "Partial cooperation scenario" (scenario=2) and "Unilateral implementation scenario" (scenario=3).
The HTML code is then injected in a st.markdown() component, the "allow_unsafe_html" argument being set to True.
This sort of hack was found on Streamlit user forums.
Expand source code
def get_table_download_button(df, scenario, effective_tax_rate, company=None, taxing_country=None, carve_out_rate=None): """ This function is used in the "app.py" file to generate the HTML code that instantiates the download button on the following pages: "Case study with one multinational" (scenario=0), "Multilateral implementation scenario" (scenario =1), "Partial cooperation scenario" (scenario=2) and "Unilateral implementation scenario" (scenario=3). The HTML code is then injected in a st.markdown() component, the "allow_unsafe_html" argument being set to True. This sort of hack was found on Streamlit user forums. """ # We output the DataFrame to a csv format csv = df.to_csv(index=False) # We encode the csv file in the right format b64 = base64.b64encode(csv.encode()).decode() href = f'<a href="data:file/csv;base64,{b64}"' if scenario == 0: # "Case study with one multinational" page company_name = company.lower().replace(' ', '_') href += f' download="{company_name}_{effective_tax_rate}_perc.csv">' elif scenario == 1: # "Multilateral implementation scenario" page href += f' download="multilateral_scenario_{effective_tax_rate}_perc.csv">' elif scenario == 2: # "Partial cooperation scenario" page href += f' download="partial_cooperation_scenario_{effective_tax_rate}_perc.csv">' elif scenario == 3: # "Unilateral implementation scenario" page taxing_country = taxing_country.lower().replace(' ', '_') if 'china' in taxing_country: taxing_country = 'china' href += f' download="unilateral_scenario_{taxing_country}_{effective_tax_rate}_perc.csv">' elif scenario == 4: # Substance-based carve-outs page href += f' download="min_ETR_{effective_tax_rate}_perc_CO_rate_{carve_out_rate}_perc.csv">' else: raise Exception('Value not accepted for the scenario argument.') href += '<input type="button" value="Click here to download the table" class="download-button table"></a>' return href def impute_missing_carve_out_values(row, avg_carve_out_impact_non_haven, avg_carve_out_impact_tax_haven, avg_carve_out_impact_domestic, avg_carve_out_impact_aggregate)-
This function allows to impute missing carve-out values based on two inputs: pre-tax profits and the average redu- ction in pre-tax profits due to carve-outs observed in OECD data. The factor used is determined by the partner ju- risdiction group in which profits are booked: tax havens, non-havens, domestic, aggregate partner jurisdictions.
Expand source code
def impute_missing_carve_out_values( row, avg_carve_out_impact_non_haven, avg_carve_out_impact_tax_haven, avg_carve_out_impact_domestic, avg_carve_out_impact_aggregate ): """ This function allows to impute missing carve-out values based on two inputs: pre-tax profits and the average redu- ction in pre-tax profits due to carve-outs observed in OECD data. The factor used is determined by the partner ju- risdiction group in which profits are booked: tax havens, non-havens, domestic, aggregate partner jurisdictions. """ if not np.isnan(row['CARVE_OUT']): return row['CARVE_OUT'] elif row['Is domestic?'] == 1: return row['Profit (Loss) before Income Tax'] * avg_carve_out_impact_domestic elif row['Is partner jurisdiction a non-haven?'] == 1: return row['Profit (Loss) before Income Tax'] * avg_carve_out_impact_non_haven elif row['Is partner jurisdiction a tax haven?'] == 1: return row['Profit (Loss) before Income Tax'] * avg_carve_out_impact_tax_haven else: return row['Profit (Loss) before Income Tax'] * avg_carve_out_impact_aggregate def manage_overlap_with_domestic(row, kind)-
When cleaning and preprocessing the OECD data, we introduce several indicator variables:
- one that takes value 1 if and only if the partner jurisdiction is a tax haven;
- another that takes value 1 if and only if the partner jurisdiction is a non-haven country;
- and a last one that indicates whether the parent and partner jurisdictions coincide.
In the breakdown of the total tax deficit into domestic, tax-haven-based and non-haven tax deficits, we need to avoid any double-counting and two of these indicator variables cannot take the value 1 simultaneously.
We therefore give the priority to the "Is domestic?" indicator variable: no matter whether the jurisdiction is a tax haven or not, this indicator variable takes the value 1 if the parent and partner are the same, while the other in- dicator variables are set to 0.
Expand source code
def manage_overlap_with_domestic(row, kind): """ When cleaning and preprocessing the OECD data, we introduce several indicator variables: - one that takes value 1 if and only if the partner jurisdiction is a tax haven; - another that takes value 1 if and only if the partner jurisdiction is a non-haven country; - and a last one that indicates whether the parent and partner jurisdictions coincide. In the breakdown of the total tax deficit into domestic, tax-haven-based and non-haven tax deficits, we need to avoid any double-counting and two of these indicator variables cannot take the value 1 simultaneously. We therefore give the priority to the "Is domestic?" indicator variable: no matter whether the jurisdiction is a tax haven or not, this indicator variable takes the value 1 if the parent and partner are the same, while the other in- dicator variables are set to 0. """ if row['Is domestic?']: return 0 else: if kind == 'haven': return row['Is partner jurisdiction a tax haven?'] elif kind == 'non-haven': return row['Is partner jurisdiction a non-haven?'] def rename_partner_jurisdictions(row)-
In the OECD data, each reporting country displays a line "Foreign Jurisdictions Total", which displays the sum of revenues, profits, corporate income taxes paid, etc. for the parent country across all foreign partner jurisdic- tions. In most cases, we want to eliminate this total to avoid any double-counting. But some countries (see list of alpha-3 codes above) only display a domestic vs. foreign breakdown and in these cases, it is important that we do not erase the "Foreign Jurisdictions Total" line. Therefore, we slightly rename it for these countries.
Expand source code
def rename_partner_jurisdictions(row): """ In the OECD data, each reporting country displays a line "Foreign Jurisdictions Total", which displays the sum of revenues, profits, corporate income taxes paid, etc. for the parent country across all foreign partner jurisdic- tions. In most cases, we want to eliminate this total to avoid any double-counting. But some countries (see list of alpha-3 codes above) only display a domestic vs. foreign breakdown and in these cases, it is important that we do not erase the "Foreign Jurisdictions Total" line. Therefore, we slightly rename it for these countries. """ if row['Parent jurisdiction (alpha-3 code)'] in COUNTRIES_WITH_MINIMUM_REPORTING: if row['Partner jurisdiction (whitespaces cleaned)'] == 'Foreign Jurisdictions Total': return 'Foreign Total' else: return row['Partner jurisdiction (whitespaces cleaned)'] else: return row['Partner jurisdiction (whitespaces cleaned)']