Policy Research Working Paper                      9973




   Globalization and Factor Income Taxation
                               Pierre Bachas
                             Matthew Fisher-Post
                               Anders Jensen
                              Gabriel Zucman




Development Economics
Development Research Group
March 2022
Policy Research Working Paper 9973


  Abstract
  How has globalization affected the relative taxation of                            countries since the 1990s, albeit from a low base. (3) Con-
  labor and capital, and why? To address this question, this                         sistently across a variety of research designs, the findings
  paper builds and analyzes a new database of effective mac-                         show that the rise in capital taxation in developing countries
  roeconomic tax rates covering 150 countries since 1965,                            can be explained by a tax capacity effect of international
  constructed by combining national accounts data with                               trade: trade openness leads to a concentration of economic
  government revenue statistics. Four main findings are                              activity in formal corporate structures, where capital taxes
  obtained. (1) The effective tax rates on labor and capi-                           are easier to impose. (4) At the same time, international
  tal have converged globally since the 1960s, due to a 10                           economic integration reduces statutory tax rates, due to
  percentage-point increase in labor taxation and a 5 per-                           increased tax competition. In high-income countries, this
  centage-point decline in capital taxation. (2) The decline in                      negative tax competition effect of trade has dominated,
  capital taxation is concentrated in high-income countries.                         while in developing countries, the positive tax-capacity
  By contrast, capital taxation has increased in developing                          effect of international trade appears to have prevailed.




 This paper is a product of the Development Research Group, Development Economics. It is part of a larger effort by the
 World Bank to provide open access to its research and make a contribution to development policy discussions around the
 world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may
 be contacted at pbachas@worldbank.org.




          The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development
          issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the
          names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those
          of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and
          its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent.


                                                        Produced by the Research Support Team
          Globalization and Factor Income Taxation∗
                    †                    ‡              §
       Pierre Bachas, Matthew Fisher-Post, Anders Jensen, Gabriel
               ¶
       Zucman




   ∗
     The database built in this paper is available online at http://globaltaxation.world. We acknowledge
financial support from the Weatherhead Center for International Affairs at Harvard University and the World
Bank Development Economics’ Research Support Budget. Zucman acknowledges support from the Stone
Foundation. We are indebted to Elie Gerschel, Rafael Proença, Roxanne Rahnama and Anton Reinicke for
excellent research assistance. We thank Damien Capelle, Denis Cogneau, Gordon Hanson, Sergey Nigai,
Steven Pennings, Thomas Piketty, Tristan Reed, Ariell Reshef, Bob Rijkers, Dani Rodrik, Emmanuel Saez and
Roman Zarate for insightful comments. The findings, interpretations, and conclusions expressed in this
paper are entirely those of the authors. They do not represent the views of the World Bank and its affiliated
organizations, or those of the Executive Directors of the World Bank or the governments they represent.
   †
     World Bank Research, pbachas@worldbank.org
   ‡
     Paris School of Economics, mfp@psemail.eu
   §
     Harvard Kennedy School and NBER, anders_jensen@hks.harvard.edu
   ¶
     UC Berkeley and NBER, zucman@berkeley.edu
1    Introduction
How has globalization affected the relative taxation of labor and capital, and why? Has
international economic integration eroded the amount of taxes effectively paid by capital
owners, shifting tax burdens to workers? If so, which countries have been most affected by
this process and through which mechanisms? Answering these questions is critical to better
understand the macroeconomic effects and long-run social sustainability of globalization.
    To address these questions, this paper builds and analyzes a database of effective
tax rates on labor and capital covering more than 150 countries since 1965. Constructed
following a common methodology that combines government revenue statistics with
national accounts data, these series allow us to study trends in labor and capital taxation
comprehensively, globally, and over a long period of time. Our database captures all taxes
paid at all levels of government: corporate income taxes, individual income taxes, payroll
taxes, property taxes, estate and inheritance taxes, consumption taxes, and other indirect
taxes. This makes it possible to estimate total tax wedges, for instance the gap between
what it costs to employ a worker and what the worker receives. Because our series are
based on national accounts data that are harmonized across countries, they can be used to
meaningfully compare effective tax rates internationally and over time. Last, since capital
income is always more concentrated than labor income, the relative taxation of the two
factors of production is closely linked to the progressivity of the overall tax system. Our
database thus provides insights into changes in tax redistribution over the last half-century.
    To maximize the time and geographical scope of this database, we conducted a large-
scale digitization and harmonization of historical data published by national statistical
offices, which we combine with existing (but limited in coverage) series published by the
United Nations, the OECD, and the IMF. The construction of our effective tax rates proceeds
in three steps. Using national accounts data we first compute total labor and capital income
in each country. Using government revenue statistics we then classify all government
revenue sources into either labor taxes, capital taxes, or indirect taxes. Combining these
two inputs, we compute effective macroeconomic tax rates on labor and capital by dividing




                                              1
labor or capital taxes paid by the corresponding income flow. The database—including
detailed decompositions by type of tax—is available online at https://globaltaxation.world.
   From this database, we are able to make two main contributions. The first is to establish
a set of facts on the evolution of factor income taxation. Taking a global perspective, we find
that average effective labor and capital tax rates have converged globally since the 1960s,
due to a 10 percentage-point increase in labor taxation and 5 percentage-point decrease in
capital taxation. This decline in capital taxation is driven by a collapse in the taxation of
corporate profits, from close to 30% in the 1960s to less than 20% in the late 2010s. The rise
in labor taxation owes primarily to the expansion of payroll taxes.
   Our most striking findings involve the evolution of capital taxation. We uncover an
asymmetric evolution of capital taxation across countries of different development levels.
In high-income countries, effective capital tax rates collapsed, from close to 40% in the
post-World War II decades to about 30% in 2018. For instance, in the United States, the
average effective capital income tax rate fell from more than 40% in the 1960s to 25% in 2018.
By contrast, in developing countries effective capital tax rates have been on a rising trend
since the 1990s, albeit starting from a low level. Effective capital tax rates rose from about
10% in the 1990s to 20% in 2018, with the increase happening primarily in large economies.
Between 1995 and 2018, for example, the effective capital tax rate rose from 10% to 30% in
China, 18% to 28% in Brazil, 7% to 11% in India, and 5% to 10% in Mexico. This increase is
one factor explaining the rise in the overall tax-to-GDP ratio of developing countries, along
with the increase of indirect taxes and a slow but steady rise in labor taxation.
   This rise of capital taxation in low- and middle-income countries had not been noted in
the literature before, due to the lack of data on the evolution of tax structures in developing
countries. The finding appears to be robust. It holds when we exclude China and oil-rich
countries; when we restrict the analysis to a balanced sample of countries; and under
different weighting schemes. It holds with alternative approaches to computing capital and
labor income in non-corporate businesses, where factor shares are not directly observable.
It is also robust to alternative ways of assigning certain taxes to capital versus labor.
   Why did effective tax rates on capital rise in developing countries while they fell in
high-income countries in the era of hyper-globalization? Our second main contribution


                                              2
is to formulate and test a hypothesis that sheds light on this puzzle. Our hypothesis is
motivated by the observation that the increase in capital taxation in developing countries
coincides with their trade liberalization. Between the late 1980s and the early 2000s, many
countries opened their markets and reduced tariffs. This policy revolution, combined
with technological improvements (e.g., the rise of container shipping), led to a boom in
international trade and reshaped the economy of countries such as Mexico, India, and
China. We hypothesize that trade liberalization exerts a positive effect on developing
countries’ ability to raise tax revenue: by increasing the concentration of economic activity
in formal corporate structures at the expense of smaller informal businesses, it facilitates the
imposition of taxes, particularly of corporate taxes—a pro-tax-capacity effect. Meanwhile
globalization exacerbates tax competition and create new opportunities for tax avoidance,
putting downward pressure on capital tax rates—a race-to-the-bottom effect. Our evidence
suggests that in high-income countries the race-to-the-bottom effect has dominated, while in
developing countries the tax-capacity effect appears to have prevailed since the mid-1990s.
    To establish these results, we implement three research designs. First, we run non-
parametric estimations of the five-year relation between changes in effective tax rates and
changes in trade openness. Second, we analyze major trade liberalization events which
occurred in seven large developing countries. These events are those that caused the largest
and most sudden reduction in trade barriers, including for instance the often-discussed
WTO accession of China in 2001 (Goldberg and Pavcnik, 2007; Goldberg and Pavcnik, 2016;
Brandt et al., 2017). We use synthetic control methods to create counterfactuals for each
country’s event, and present event-study graphs. Last, we extend the two instruments for
trade openness presented in Egger, Nigai, and Strecker (2019), to estimate the effect of
trade on factor taxation.
    In each case we find that trade openness leads to a large rise in effective capital taxation
in developing countries, and a smaller increase in effective labor taxation. On the contrary,
trade integration has a null or negative effect on capital taxation in high-income countries,


     Trade also encourages the adoption of modern accounting practices and leads to growth in firm size and
the expansion of value chains. As large corporations are more visible and generate information trails (Kleven,
Kreiner, and Saez, 2016; Basri et al., 2019), the literature is consistent with the hypothesis that trade-induced
economic change could make the tax base more enforceable in low-tax-capacity states.


                                                       3
but a positive effect on labor taxation. Although the sources of variation and identification
strategies involved are different in our three empirical specifications, our results are
consistent across them and robust to a range of sensitivity checks.
   To better understand these results, we study potential mechanisms using event studies
and the instrumental variable research designs. Consistent with the tax-capacity hypothesis,
we find that trade liberalization leads to a rise in the fraction of domestic product that
originates from the corporate sector (at the expense of the non-corporate business sector)
and to an increase in salaried employment (at the expense of self-employment). These
changes lead to a growing fraction of output being produced and income being earned in
sectors that are more visible and easier to tax. We also find that the positive impact of trade
on capital taxation, in addition to being concentrated in developing countries, is stronger
in populous countries and in countries with restrictions on capital flows. This finding
is consistent with the notion that large countries and countries managing their capital
accounts are less exposed to the race-to-the-bottom effect that has pushed capital taxation
down in high-income countries. Last, trade liberalization is actually associated with a
decline in statutory corporate tax rates across all countries, but more so in high-income
countries. On net, the trade-induced increase in tax capacity dominates the statutory tax
rate reduction in developing countries, and vice-versa in rich countries.
   The rest of the paper proceeds as follows. In Section 2, we relate our work to the
existing literature. Section 3 describes the methodology and data collection. Section 4
presents our findings on the evolution of effective tax rates over the long-run. In Section 5,
we present graphical evidence on the association between trade openness and effective
tax rates. Section 6 studies the impacts of major trade liberalization events. Section 7
presents instrumental variable estimates of the effect of trade liberalization and investigates
heterogeneity and mechanisms. In Section 8 we analyze episodes of capital liberalization.
Section 9 concludes. The paper is supplemented by an Online Appendix that provides
step-by-step details of data construction and additional results.




                                              4
2     Related literature

2.1    Globalization and taxation

Our paper first relates to the literature on globalization and taxation. Since Adam Smith
(1776), economists have conjectured that increased openness pushes governments to reduce
taxes on the most mobile factors of production (e.g., high-wage workers, capital) and
recover the revenue shortfalls by increasing the taxation of less mobile factors (Bates,
Da-Hsiang, and Lien, 1985; Rodrik, 1997). This effect is thought to be particularly strong
for countries competing for capital (Wilson, 1999; Kanbur and Keen, 1993).
    In an important contribution, Egger, Nigai, and Strecker (2019) show that globalization
led to a decline in the progressivity of labor taxation in OECD countries since 1994 (namely,
an increase in labor taxation for the middle class and a decline for the top 1 percent). Our
approach, which focuses on the changing balance of labor vs. capital taxation in both
OECD and non-OECD countries, is complementary. For OECD countries, our findings
reinforce Egger, Nigai, and Strecker (2019): we show that in addition to reducing labor tax
progressivity, globalization has been associated with a sharp decline in capital taxation.
This decline reduces overall tax progressivity (above and beyond the decline due to falling
labor tax progressivity) given the concentration of capital income at the top.
    Our paper also adds to the macroeconomic literature on the link between trade and
taxes. Due to the lack of systematic statistics on the evolution of tax structures in developing
countries, this literature has focused on high-income countries or a single tax (e.g., the
corporate income tax). Our contribution to this literature is to build and analyze a new
global dataset of effective tax rate, extending prior work which focused on developed
countries (Mendoza, Razin, and Tesar, 1994; Carey and Rabesona, 2004; McDaniel, 2007).
This allows us to uncover new trends (most importantly, the rise in capital taxation in




     Rodrik (1997) finds that trade openness is associated with a decline in effective capital tax rates and an
increase in effective labor tax for 14 OECD countries with high levels of capital mobility, between 1965 and
1991. In a sample of 14 OECD countries between 1981 and 1995, Swank and Steinmo (2002) find that trade
is not associated with changes to effective tax rates, neither on capital nor on labor. Over the same period
Slemrod (2004) finds that trade is not associated with changes to the statutory corporate income tax rate.


                                                     5
developing countries in the era of hyper-globalization) and to formulate an hypothesis that
can explain this dynamic.


2.2    Tax capacity in developing countries

Our paper also relates to the growing literature on tax capacity in developing countries.
This literature highlights a number of factors driving the rise of taxation over the path of
development, including change in employment structure (Jensen, 2022), growing capacity
to observe income (Pomeranz, 2015), the threat of whistle-blowing in large firms (Kleven,
Kreiner, and Saez, 2016), and administrative investments in tax capacity (Besley and
Persson, 2014). We complement these studies by investigating the role of a new channel,
international trade. The increase in effective tax rates we document in response to trade
liberalization is consistent with previous studies showing that trade has a positive effect
on growth (e.g., Goldberg and Pavcnik, 2016; Wacziarg and Welch, 2008) and growth
is associated with higher tax rates (e.g., Besley and Persson, 2014). Our approach goes
further by showing the direct role of trade openness. Moreover, we provide evidence
on mechanisms through which trade liberalization affects effective tax rates – namely an
increase in the corporate share of domestic product and a transition from self-employment
to salaried employment. While the literature on tax capacity has emphasized theoretically
and descriptively the importance of these mechanisms along the development path, we
provide some of the first well-identified evidence.
    Last, we add to a body of work in economic history documenting the long-run evolution
of tax revenue and tax capacity (Cogneau et al., 2021; Albers, Jerven, and Suesse, 2020). A
strand of that literature studies the tax revenue effects of trade liberalization and the extent
to which lost tariff revenues were compensated by other taxes, in particular the value-added
tax (Baunsgaard and Keen, 2009; Cage and Gadenne, 2018; Buettner and Madzharova, 2018);
Benzarti and Tazhitdinova (2021) also study indirect taxes and international trade. Our

      We also find that trade liberalization increases the effective tax rate on labor, suggesting that it raises
labor formality. Recent work shows this is the case for the tradable sector (e.g., Dix-Carneiro, Goldberg,
et al., 2021), but the evidence of the impact of trade on overall labor formality in developing countries is mixed
(e.g., McCaig and Pavcnik, 2018; Dix-Carneiro and Kovak, 2017; Attanasio, Goldberg, and Pavcnik, 2004).
      These studies cover a century of tax revenue, dating from pre-independence, respectively for French
colonial Africa, and all of Africa.


                                                        6
paper complements this work by focusing on effective labor vs. capital tax rates in addition
to total tax revenues, and by implementing several identification strategies to capture the
causal effect of trade.



3       Construction of factor shares and effective tax rates
This section describes the construction of our database of effective tax rates on labor
and capital. The data covers the 150 most populous countries from 1965 to 2018, with
exceptions only for pre-independence, civil war, and command economy eras. The database
is available online at http://globaltaxation.world along with country-specific visualizations
and notes. Here we focus on the general methodological principles.


3.1     Conceptual framework and methodology

3.1.1    Factor shares

We begin by decomposing each country’s output into a labor and capital component.
Following standard national accounts definitions, net domestic output Y at factor prices
(i.e., before indirect taxes) can be expressed as:


                               Y = CE + OSCORP + OSHH + OSP U E                                          (1)


where CE is compensation of employees (wages, salaries, plus supplements to wages and
salaries such as contributions to pensions); OSCORP is the operating surplus of corporations
(profits, net of depreciation); OSHH is the operating surplus of households (actual and
imputed rental income); and OSP U E is the operating surplus of private unincorporated
enterprises, or mixed income.



     Cage and Gadenne (2018) find that trade liberalization led to a decrease in overall tax revenue pre-1995.
Our paper highlights positive revenue effects of trade liberalization but looking at different outcomes
(effective capital and labor tax rates, as opposed to indirect tax revenues) and sample periods (the positive
effects we obtain are concentrated after 1995). We also complement Buettner and Madzharova (2018) who
show that lost tariff revenue from WTO accession events in the 1990s were fully compensated.


                                                     7
   The capital share of net domestic output, denoted α is computed as:

                           YK   OSCORP + OSHH + (1 − φ) · OSP U E
                      α=      =                                                            (2)
                           Y     CE + OSCORP + OSHH + OSP U E

where φ is the labor share of mixed income. The labor share of net domestic output, 1 − α
equals compensation of employees plus a share φ of mixed income:

                              YL         CE + φ · OSP U E
                     1−α=        =                                                         (3)
                              Y    CE + OSCORP + OSHH + OSP U E

   Four points are worth noting. First, our output measure is net domestic product, that is,
operating surplus is measured net of capital depreciation. Throughout this paper we focus
on net-of-depreciation output concepts, as in, e.g., Karabarbounis and Neiman (2014) and
Guerriero (2019). Second, as is standard in the literature (see Browning, 1978; Saez and
Zucman, 2019b), we do not allocate indirect taxes to labor or capital; we instead compute
factor shares of domestic product net of indirect taxes. Third, public-sector enterprises are
usually included in the corporate sector (see Lequiller and Blades, 2014). Last, we compute
factor shares of domestic output (as opposed to national income). For example, residents
in Lesotho may earn labor income in South Africa, and corporations resident in France may
book profits in Luxembourg. Wages earned in South Africa are included in YL for South
Africa (not Lesotho); profits booked in Luxembourg are included in YK for Luxembourg
(not France). This is the most logical thing to do for our purposes, since countries typically
try to tax domestic output.


The labor share of mixed income.      In the data we collected (discussed in Section 3.2), we
observe all components of equations (3) and (2), except for the labor share of mixed income,
φ. Measuring the labor component of self-employment and unincorporated enterprises’
income is challenging, as discussed in Gollin (2002) and Karabarbounis and Neiman (2014).
For our benchmark series we follow the literature in assuming that φ = 70% (see Blanchet,
Chancel, et al., 2021). To test for robustness, we also implement the method discussed in
Young (1995), Gollin (2002), and Guerriero (2019) and developed further in Cette, Koehl,
and Philippon (2020) and ILO (2019). This method imputes to the self-employed a labor

                                              8
income similar to the wage they would have earned in an employer-employee relation,
based on observable characteristics. We extend the estimates in ILO (2019), using ILOSTAT
(2021) data on self-employment shares of the workforce, to all countries since 1991, and
impute the series backwards to complete it. Details are in the data Appendix B.


3.1.2   Effective tax rates on capital and labor

We allocate each tax revenue source to labor, capital, or a mix of the two. Specifically,
(1) Corporate income taxes, wealth taxes, and property taxes are allocated to capital. (2)
Payroll taxes and social security payments are allocated to labor. (3) Personal income
taxes are allocated partly to labor and partly to capital (see below), reflecting the fact that
personal income is composed of salaries, capital income, and mixed income. (4) Indirect
taxes are treated as a separate category (i.e., are assigned neither to labor nor to capital).
Table A1 summarizes our allocation.


Allocation of the personal income tax to factors of production. The main empirical
difficulty in assigning taxes to labor or capital concerns the allocation of the personal
income tax (PIT). In most countries, both labor income and capital income are subject to
the PIT. As labor income accounts for about 70% of national income and capital income for
about 30%, a naive procedure would allocate 70% of the PIT to labor and 30% to capital. In
practice, however, not all labor and capital income is subject to personal income taxation,
due both to the fact that some individuals are not required to file an income tax return
and to legal exemptions for some forms of income. Exemptions for capital (e.g., imputed
housing rents, undistributed corporate profits, investment income earned on retirement
accounts) are typically larger than for labor (e.g., pension contributions, health insurance
contributions); see, e.g., Saez and Zucman (2019b). In the United States, Piketty, Saez, and
Zucman (2018) find that 75% of labor income is subject to the individual income tax in
2015, as opposed to only a third of capital income. This would call for allocating about
15% of the personal income tax to capital and about 85% to labor. A last difficulty is that

     If 75% of labor income is taxable and labor income is 70% of national income (resp. 33% and 30% for
capital income), then 75% × 70%/(75% × 70% + 33% × 30%) = 84% of the PIT base is labor income and 16%
is capital income.


                                                   9
labor and capital income are not necessarily subject to the same tax rate. Starting in the
1990s, a number of economies have adopted dual-income tax systems, whereby labor and
capital incomes are subject to different schedules, with labor income typically subject to a
progressive tax schedules and (some) capital income subject to flat tax rates.
   To account for these facts, we proceed as follows. We start with the baseline assumption,
consistent with US data, that 15% of PIT revenues derive from capital and 85% from labor.
We then use data from Jensen (2022), which documents the location of the PIT threshold
across countries, to adjust for the share of capital income in the PIT at the country-year level.
Since richer taxpayers derive a larger share of their income from capital, countries with a
high PIT exemption threshold—and thus more exempted taxpayers—have a larger capital
share of PIT revenue. Finally, we use data from the OECD to account for dual income
tax systems. Specifically, when dividends face a lower tax rate than ordinary income, we
compute the ratio of the statutory rate on dividends to that of the top rate on labor income,
and adjust the capital share of PIT revenue down when this ratio is below one. Appendix
B.2 discusses these adjustments in further detail.
   The resulting share of PIT revenue allocated to capital varies between 7% and 35%,
depending on countries and years. Over time, this share falls from a global average of 19%
in 1965 to 14% in 2018, due to both a reduction in PIT exemption thresholds and to the
adoption of lower tax rates on dividends in some countries.


Effective tax rates. The total tax revenue assigned to labor and capital is:


                      TL =      [λic · τi ] and        TK =   [(1 − λic ) · τi ]              (4)

   where λic is the allocation to labor of each type of tax τi in country c (see Table A1).
   The effective tax rates (ETR) on labor and on capital, ET RL and ET RK , are computed
by dividing tax revenue collected by the size of the respective labor and capital income
flows:

                                       TL                           TK
                             ET RL =          and        ET RK =                              (5)
                                       YL                           YK


                                                  10
    These measures of macroeconomic effective taxation capture the overall tax burden on
labor and capital, building on the work of Mendoza, Razin, and Tesar (1994) and Carey
and Rabesona (2004). These ETRs have a number of advantages. First, they capture the
economically relevant tax wedges on each factor of production (i.e., the wedges that matter
for production decisions), such as the difference between the costs to employ a worker
and what the worker receives. Second, because national account statistics are compiled
following harmonized concepts and methods, they are conceptually comparable over time
and across countries. Third, since they rely on the amount of tax effectively collected by
governments, they incorporate the net effects of all tax rules— such as base reductions,
exemptions, and tax credits—and of tax avoidance and evasion.


3.2     Data sources

3.2.1   National income components

To estimate factor shares for 156 countries since the 1960s, we create a harmonized panel of
national accounts which combines data from the UN System of National Accounts, the
World Inequality Database, and other sources.
    From the World Inequality Database (WID, 2020), we retrieve United Nations (UN)
System of National Accounts (SNA) data that covers 4,000 country-years. These data come
from the production and income accounts of the online UN SNA (2008), “Main Aggregates
and Detailed Tables.” In addition, the UN Statistics Division provided us access to their
archival data on the components of GDP, with over 2,000 country-year observations from
the 1960s and 1970s, presented following the 1968 System of National Accounts (UN




      Compared to pre-existing work, our ETRs are global, cover over a half-century, systematically integrate
mixed income, and account for heterogeneity in the capital component of PIT revenue using new data on the
location of the PIT exemption threshold and on the relative taxation of dividends vs. labor income.
      The computation of these economically relevant wedges does not require one to make assumptions
about behavioral responses (and hence about the incidence of taxes). See Mendoza, Razin, and Tesar (1994)
and Saez and Zucman (2019a) for a conceptual discussion.
      The ETRs computed in (5) are called backward-looking ETRs in the literature. A separate literature tries
to model all statutory features of the tax system at a point in time in order to measure forward-looking ETRs
(see, e.g., King and Fullerton, 1984; Devereux, 2004).


                                                     11
SNA, 1968).      When these accounts are incomplete (e.g., a component of GDP is missing, or
there is no data on depreciation), we recover missing values using accounting identities or
by following the imputation procedures used in the World Inequality Database (WID, 2020).
To ensure comparability with the more recent data, we recast the historical series into the
2008 System of National Accounts framework.                To our knowledge, this is the first factor
income shares dataset that harmonizes data from the 2008 and 1968 System of National
Accounts. In countries and years when the two systems overlap (typically in the 1970s,
when countries transitioned from the old to the new framework), the series match well.
    Our work expands the dataset in Karabarbounis and Neiman (2014) along two dimen-
sions. First, the integration of the 1968 System of National Accounts data extends coverage
in time and space.       Second, while Karabarbounis and Neiman (2014) focuses on factor
shares in the corporate sector, we compute factor shares of total domestic output.


3.2.2   Tax revenue data

We construct a new tax revenue dataset that includes disaggregated tax revenue data by
type of tax. Our database includes all taxes—personal income taxes, corporate income
taxes, Social Security payroll taxes, property taxes, wealth taxes, estate and inheritance
taxes, consumption and other indirect taxes —at all levels of government. We integrate
previously unused historical data from developing countries to obtain a global coverage.
    We first gathered existing high-quality data from OECD (2020) and ICTD/UNU-WIDER
(2020) for recent years, and from the IMF GFS (2005) for older years. Second, we retrieved
thousands of country-year observations of historical revenue data from the Harvard
University Library archives, as well as online data from national statistical offices and


     The variables include value added; compensation of employees; operating surplus of corporations;
operating surplus of unincorporated enterprises; consumption of fixed capital; and indirect taxes net of
subsidies.
     Specifically, 1968 SNA data always include compensation of employees and operating surplus of
corporations, but do not disaggregate mixed income and operating surplus of the household sector (which
are lumped into a single aggregate). We impute the split of mixed income vs. household operating surplus
according to the split seen in the 2008 SNA at the time of switching (typically in the 1970s). We also follow
the United Nations guidelines to stitch these series together (UN, 2018).
     Karabarbounis and Neiman (2014) restrict their sample to online UN SNA 2008 data, and to countries
with at least 15 consecutive years of complete-case data.
     Lamont Library, Government Documents section.


                                                     12
finance ministries. Third, we classify each revenue source following the OECD’s tax
classification (see OECD, 2020). Table B1 details the data sources used.
    When available, OECD tax revenue data is our preferred source, because it covers and
classifies all types of tax revenues, usually back to 1965 for OECD countries. OECD data
accounts for 41% of the country-year observations in our dataset. Its drawback is its limited
coverage of non-OECD countries: in total it covers 93 countries, and only over the past two
decades.
    To increase coverage, we augment the OECD data with the tax revenue data from the
ICTD/UNU-WIDER (2020) (17% of observations). This dataset achieves near worldwide
coverage but, for our purposes, faces limitations: it only starts in the 1980s; it does not
follow the tax classification of the OECD; it sometimes mixes personal and corporate
income taxes; and it often lacks payroll taxes and decentralized taxes. To address these
shortcomings, we use historical public finance data from government reports, primarily
from the Harvard Library archives (30% of country-year observations) and from the IMF
GFS (2005) offline historical database (10% of observations).
    To stitch together country-by-country time series of tax revenues, we follow three
principles. First, we aim to only rely on a maximum of two data sources by country: the
OECD when it exists, and the alternative source with the best coverage over time and by
tax type. Archival data is our second in priority since it often dis-aggregates revenue by
source, and goes back to the 1960s. Our data hierarchy choice also depends on which
source best matches the OECD data over their shared time frame. Second, we interpolate
series with gaps, but only up to four years between two data points. Finally, we check
country-specific policy reports and scholarly studies to triangulate across data sources and
to identify events which may explain discordance across sources.
    Tax revenues are disaggregated as finely as possible by source, according to the OECD
tax classification (OECD, 2020). To allocate taxes to capital and labor, we pay attention
to three dimensions. First, we systematically separate income taxes into personal and


     The ICTD/UNU-WIDER data draws principally from the IMF Government Finance Statistics online
data, which covers the past few decades well. Our use of the IMF data is restricted to the offline historical
dataset, which covers 1972-89 and fills gaps from the OECD and historical archives data. The ICTD does not
report pre-1980 data.


                                                    13
corporate income.         Second, we always include payroll taxes, which requires at times
to add new data sources (e.g., we digitized payroll tax revenues from the UN System of
National Accounts and from Fisunoglu et al. (2011)). Third, we always include taxes on
property, often the main source of local government tax revenues.


3.2.3    Data coverage

Figure A1 shows that our dataset covers 86% of world GDP in 1965 and 98% in 2015, as the
number of countries grows from 78 to 156.             The main change in the sample of countries
covered corresponds to the entry of ex-communist countries in the early 1990s. We also
include China from 1994 on, the date of the creation of the modern Chinese tax system
(World Bank, 2008).       Late decolonization and end of civil wars are other reasons to enter
the panel later than 1965.
    The dataset is thus composed of two (quasi) balanced panels: the first covers the
years 1965-1993 and excludes communist regimes. It accounts for 85-90% of world GDP
during those years. The second covers 1994-2018 and includes former communist countries
and China; it accounts for 98% of world GDP. At their time of entry into the dataset,
ex-communist countries account for 8% of world GDP (4.5% for China and 3.3% for Russia).



4       Global trends in tax revenues, factor shares and
        effective tax rates
With this new dataset, we document the global evolution of tax revenues, factor shares,
and effective taxation on capital and labor from 1965 to 2018. Our objective is to show
time series for each outcome which can be interpreted as the global value worldwide, in
each year. For example, the global effective tax rate on capital equals worldwide capital tax



     In some cases, this split is not available in the headline, official income tax revenue aggregates, so we
look to expert studies elsewhere.
     In the most recent years we do not cover 100% of world GDP, as we did not try to collect data from
countries with fewer than 1 million inhabitants when these were not available through online sources.
     See Appendix B.3 for a case study of China.


                                                    14
revenue divided by worldwide capital income in the same year.                    For each outcome we
first show the global trends, and then show separately high vs. low and middle-income
countries.


Tax revenues.      Figure 1 shows the time series of tax revenue as a share of net domestic
product (NDP), separated into its main components: corporate income taxes, property
and asset taxes, personal income taxes, payroll taxes, and indirect taxes (VAT and tariffs).
Globally, tax revenue as a share of NDP increased from 26% to 32% between 1965 and 2015.
This is driven by an increase in payroll and personal income taxes, which went from 11%
to 16%. Indirect taxes slightly rose over the past 50 years from 8% to 9%, while revenues
from taxes on capital (corporate and property) stagnated at around 6%.
    We observe two differences in tax revenue patterns between high-income versus low-
and middle-income countries. First, tax revenue as a share of NDP is much higher in
rich countries than in developing countries (37% vs. 23% in 2018). Second, in developing
countries, all types of taxes increase their revenue collection over time (particularly from
1990 onward), including those on capital. By contrast, corporate income tax revenue
decreased over time in high-income countries, and revenues from property taxes stagnated.
Rising tax revenue in rich countries came primarily from the expansion of payroll taxes
between 1965 and 1985.


Factor shares.     Figure 2 shows the capital share of net domestic product over time (solid
line) and the capital share within the corporate sector (dotted line). The capital share of
world income increased from 20% to 26%. This global trend is due to rises in the capital
share within both rich countries (from 25% to 28%) and developing countries (from 36% to
38%) and to the increasingly large weight of developing countries in world income. The
capital share within the corporate sector followed the same evolution as the aggregate
capital share: it increased from around 19% in 1965 to 28% in 2015.


     Global figures depend on countries’ changing shares of world GDP: Figure A2 shows how the weight of
different countries evolved over time, highlighting the growth of China’s weight in the past 20 years, and to
a lesser extent that of other developing countries. The weight of China in world GDP was far less in the
pre-1994 era during which it is excluded. Appendix B provides further discussion.


                                                    15
Effective tax rates.       Figure 3 shows our key time series: the evolution of the effective
tax rates on labor (red) and capital (blue); and, within capital income, the evolution of
the effective tax rate on corporate profits (dashed blue).                   Globally, the ETRs on labor
and capital converged between 1965 and 2018. This is due to a large increase in labor
taxation and a mild decrease in capital taxation. The global ET RL increased from 16% to
approximately 25%, while over that same period, the ET RK decreased from an average of
32% in the mid-1960s to 27% in the late 2010s. Within the corporate sector, the global ETR
on corporate profits saw a more pronounced decline, from 27% in 1965 to 18% in 2018.
The decline in global effective capital taxation captures both changes within countries
(most importantly the reduction in corporate tax rates in most high-income countries) and
changes in the allocation of profits across countries, such as the rise of profit shifting to tax
havens (Tørsløv, Wier, and Zucman, 2020).
    The patterns are robust to the two main sources of uncertainty in the computation of
effective tax rates. Figure A5 shows the evolution of ETRs over time when we allocate
mixed income to labor and capital following the ILO country-specific method, instead of
the benchmark 70/30 split. Figure A6 varies the allocation of the personal income tax (PIT)
revenue to labor versus capital taxes, comparing the benchmark to two extreme scenarios
which allocate 100% to labor (0% to capital), or 70% to labor (30% to capital).
    These global trends mask heterogeneity by development level. First, the ET RL increased
by more in developed than in developing countries, even though the starting point was
already higher in rich countries (18% in 1965) than in poorer countries (6%). Second,
and most importantly, the decline in the effective tax rate on capital is concentrated in
high-income countries, where it went from close to 40% in 1965 to about 30% by 2018.


     The ETR on corporate profits is computed as the ratio of the revenue from the corporate income tax over
the operating surplus of the corporate sector.
     Figure A4 shows the ETR series in a fully balanced panel of all countries since 1965, where missing
values are imputed to control for the changing sample composition over time (most importantly China and
Russia missing pre-1994). Imputing missing values has limited impact on the global ETRs series, since the
countries entering in 1994 only represent 8% of global net product at that time, and Russia’s ETRs are close to
the global average upon entry. China’s ETR are lower, which explains the slight drop in adjusted ET RK and
ET RL series pre-1994. Focusing on developing countries only (1994 entrants now represent a third of total
net product), the imputation of missing country-years raises the pre-1994 ETR on labor and on capital by 2
percentage points, due to Russia’s higher levels, while China’s ETRs match developing countries’ average in
1994.


                                                      16
In contrast, the ET RK increased in developing countries, albeit from a low base: it rose
from 10% to 20%, with most of the increase happening after 1995. Despite this rise,
effective capital tax rates in emerging countries remain significantly lower than in rich
countries in 2018. The increase in effective capital taxation within low- and middle-income
countries starting around 1990 appears to be a broad-based phenomenon. Figure 4 shows
the evolution of ETRs in several sub-samples of developing countries. First, the rise
remains—but is more muted—when excluding China: in that case ET RK rises from 10%
in 1995 to 13% in 2018. The lower increase highlights the importance of China’s rising
capital taxation and its extraordinary growth (and thus rising global weight) over the past
25 years. Second the rise in ET RK remains when excluding oil-rich countries (defined as
deriving 7% or more of their GDP from oil); in that case the rise in ET RK (from 10% in 1990
to 24% in 2018), is in fact larger than in our benchmark series. Removing oil-rich countries
also leads to more stable series and a flat trend in effective capital taxation pre-1990. If we
exclude both ex-communist countries and oil-rich countries (mid-right panel), we again
observe a large rise in ET RK .
   The rise in capital taxation in developing countries thus reflects more than China’s
rise or changes in tax collection in oil-rich countries. Figure A7 shows the evolution of
effective tax rates in the most populated developing countries. In a majority of large
countries, ET RK increased between 1990 and 2018: for example, India’s rose from 6 to
11%, Indonesia’s from 10 to 16%, and Brazil’s from 10 to 28%. The bottom panels of Figure
4 divide the sample between the 18 largest (non-oil rich) developing countries whose
population exceeds 40 million, and the 55 countries with population under 40 million.
The rise in ET RK is much more pronounced in large countries, where it increases from
approximately 10% to 25%, while the rise is modest in smaller countries (from 8% to 12%).




                                             17
5      Correlation in trade and factor income taxation

5.1     Motivation

How can we explain the evolution of effective tax rates on labor and capital since the 1960s?
A natural starting point is the large literature which focuses on the role of globalization.
Cross-border trade in goods and services has grown (relative to GDP) in both developed
and developing countries since the 1960s; this increase was driven in part by the rise
of global value chains (Feenstra and Hanson, 2001), where the production process is
fragmented across borders and firms rely on foreign subsidiaries and contractors. The
literature argues that firms’ ability to shift production processes across borders limits
governments’ capacity to tax mobile production factors. The long-run decline in ET RK
in rich countries is consistent with this hypothesis.                 However, we saw that since the
1990s—the onset of the hyper-globalization period—developing countries experienced a
rise in ET RK .
    Focusing on developing countries, we observe that the positive association between
trade and capital taxation runs deeper: when we separate countries based on their initial
level of trade in the pre-1995 period, early globalized countries saw trade and the ETR on
capital rise in tandem prior to the 1990s, and stagnate thereafter (Figure A9). Developing
countries which participated in the second wave of globalization (after the early-1990s
proliferation of trade agreements) saw an increase in their trade and capital taxation in
1995-2018 period. These heterogeneous trends motivate our systematic analysis of the
impact of globalization in the remainder of the paper.
    Trade and capital flows are both important dimensions of globalization and correlate
with each other. Because internationally comparable data are more widely available for
trade than for capital flows, and the literature focuses primarily on causal determinants
of trade openness, we focus on trade as the main measure of globalization. We return to
capital openness in Section 8.




      The long-run decline in statutory corporate tax rates is also consistent with this hypothesis (Figure A8).


                                                       18
5.2    Correlation over time

Our first empirical strategy exploits the within-country association between trade and our
outcomes of interest: factor shares and effective tax rates on labor and capital. We measure
trade as the share of imports and exports relative to GDP. We create 5-year growth rates
within countries in both the trade measure and our outcomes of interest. To visualize these
associations, we plot binned scatters of each outcome against trade, after residualizing
all variables against year fixed effects. Each dot in the figure corresponds to a ventile (20
equal-sized bins) of the residualized trade openness distribution; we add back the mean of
each variable to ease interpretation.
    Figure 5 non-parametrically shows the medium-run within-country association, con-
ditional on global time trends, but without any other controls or weights. We observe
a positive association between the within-country growth in trade openness and ET RK .
Trade openness is also positively correlated with ET RL , although the slope is smaller than
for capital. We also observe a positive association between trade openness and the capital
share of income; this association is almost twice as large for the corporate capital share.
Trade may thus positively affect capital taxation, both through increasing capital’s share of
aggregate income and by raising ET RK .
    Previous studies on ETRs and globalization mainly focus on rich countries. In Figure 6,
we find that the association between trade and ET RK differs between high- and low- and
middle-income countries. The relationship between trade openness and ET RK has a mild


       An interesting extension would be to study imports and exports separately, noting that import competition
may impact factor taxation differently than export opportunities (see Goldberg and Pavcnik, 2016; McCaig
and Pavcnik, 2018). In the event-study and instrumental variables designs, our estimating variation affects
both imports and exports. More generally, imports and exports are strongly correlated in the data, making it
hard to estimate separate effects with precision. This extension may be better suited to using micro-data
which could leverage the fact that different firms within the same economy are exposed either to import or
export shocks.
       The positive association between trade and the capital share is not consistent with classical trade models
such as Heckscher-Ohlin (Ohlin, 1933), which predict that trade favors a country’s abundant factor (labor in
poor countries). Rather, it is consistent with bargaining models, in which opportunities to produce abroad
improve capital owners’ bargaining position (Rodrik, 1998a; Harrison, 20050; Rodriguez and Ortega, 20060).
It is also consistent with the global value chains theory (Feenstra and Hanson, 2001): high-income countries
focus on capital-intensive portions of the global value chain and outsource labor-intensive processes to
developing countries. Outsourced processes are still relatively capital-intensive for developing countries.
Thus, trade integration benefits capital in both groups of countries, despite capital being the scarce resource
in poorer ones.


                                                      19
negative slope in high-income countries, but a steeply positive slope in developing countries.
While the negative slope in high-income countries is consistent with the cross-border
mobility hypothesis (see Section 2), the positive slope in developing countries suggests
that other channels—such as a pro-tax capacity effect of globalization—could be at play.



6     Event-studies around large trade liberalization events

6.1    Empirical design

In this section, we analyze trade liberalization events in key developing countries. To
discern sharp breaks from trends in our outcomes, we search for events which caused
large trade barrier reductions. We focus on the six events studied in the review papers by
Goldberg and Pavcnik (2007) and Goldberg and Pavcnik (2016) (Colombia in 1985, Mexico
in 1985, Brazil in 1988, Argentina in 1989, India in 1991, Vietnam in 2001), and add the often
discussed World Trade Organization accession of China in 2001 (Brandt et al., 2017). These
events share two key features. First, they are characterized by large reductions in tariffs,
the easiest trade barrier to measure. For instance, Brazil reduced average tariffs from 59%
to 15% percent, India from 80% to 39%, and China from 48% to 20%. Second, these events
have been studied exhaustively before. Since trade liberalization events do not occur in a
vacuum and are often accompanied by other reforms, we can rely on the existing in-depth
narrative of the conditions surrounding trade reforms to gauge threats to identification and
to our results’ interpretation.      Appendix C.1 details all seven trade liberalization events.
    For each event and outcome, we construct a synthetic control country following the
methodology in Abadie, Diamond, and Hainmueller (2010). The synthetic control is
created as a weighted average over the donor pool of countries. To construct the weights,
we match on the level of each outcome in the 10 years prior to the event, to minimize
the mean squared prediction error between the event-country and the synthetic control
countries in pre-event years. We then create event-study graphs showing the average of the
outcome variable for treated countries vs. synthetic controls by relative time to the event.

    The reductions in trade barriers are sometimes implemented over several years. To be conservative, we
focus on the earliest start year for each event as defined in published studies.


                                                   20
   We also implement the event-study in a regression setting, where we include country
and calendar year fixed effects, using the seven treated countries and their synthetic
controls in the 10 years before and after the events:

                            10
                Yit =                  βj ∗ 1(j = t)t ∗ Di + θt + κi + πY ear(it) +   it   (6)
                        j =−10,j =−1

   where θt are event-time fixed effects, κi are country fixed effects, and πY ear(it) are year
fixed effects. The year fixed effects control for common shocks to factor shares and taxation
which may be correlated with clusters of reforms. Di is a dummy equal to one if country
i is treated. βj captures the difference between treated and synthetic control countries
in event time j , relative to the pre-reform year j = −1 (omitted period). Since statistical
inference based on small samples should be approached with caution (Abadie, Diamond,
and Hainmueller, 2010), we plot 95% confidence bounds based on the wild bootstrap
method (Cameron, Gelbach, and Miller, 2008), clustered at the country-event level.
   We run two more specifications to attenuate potential issues with synthetic control event
studies. First, in addition to the dynamic effects model, we estimate the simpler difference-
in-differences model, where the coefficient measures the average treatment effect over the
first 10 years post-liberalization. We compute coefficients based on the imputation method
of Borusyak, Jaravel, and Spiess (2021), which addresses estimation issues from two-way
fixed effects and heterogeneous event-times (Chaisemartin and D’Haultfœuille, 2020).
Second, we are interested in the impact of trade liberalization on several outcomes (trade,
factor share, factor taxation). Our baseline approach creates a separate synthetic control
for each event and each outcome, which increases the likelihood of obtaining similar
pre-trends (Akcigit et al., 2021), but implies that for a given country-event, the synthetic
control countries might differ across outcomes. Therefore we also implement a design
where we simultaneously match on all outcomes of interest for each country-event (similar
to Jaeger, Noy, and Schoefer, 2021). All methodological details are in Appendix C.2.




                                                    21
6.2    Results

Figure 7 shows the results for the main outcomes. The left-hand panels display the event-
studies in levels, while the right-hand panels display the regression-based event-studies.
The top panels show that for trade openness the synthetic control matches the average
treated country closely during the 10 years prior to the event.             Trade openness increases
in the year of the event and its trend changes in post-reform years, compared to the stable
pre-trends. The absence of a dip in the immediate pre-reform years limits concerns about
intertemporal substitution, although some liberalization events were predictable (especially
in China and Mexico where the event is World Trade Organization accession). Overall, as
expected, trade increases substantially when countries slash their import tariffs.
    Turning to our outcomes, we see that trade liberalization events coincide with a positive
break from trend in the capital share of domestic product. The synthetic control continues
on its slight upward trend.       The impacts on factor taxation are displayed in the bottom set
of panels of Figure 7. We observe that ET RK sharply increases following the liberalization
event. Both ET RK and ET RL break from the stable pre-trend at the time of liberalization,
but the effect on capital taxation is about double that on labor. Despite the small sample
size, the dynamic post-treatment effect coefficients are typically significant at the 5% level.
The p-value for the joint significance of all post-reform dummies are well below 0.05. Based
on the difference-in-differences model, the liberalization events led to a 10 percentage point
rise in trade openness over 10 years and a 4.8 (2.0) percentage point increase in the effective
tax rate on capital (labor) (Table A2).


6.3    Robustness, interpretation and limitations

To test for robustness of the event-study results, we conduct several checks. First, we
jointly match on all four outcomes for each event to create synthetic controls, instead of
creating outcome specific synthetic controls. Figure A10 shows that this leads to a small
deterioration of the pre-trends, but to very similar point-estimates. Second, to check that

    Table C1 details the synthetic control matches for each event and outcome.
    The stability of these patterns helps alleviate concerns that the true counterfactual level would be
overstated if trade flows and returns to capital were diverted away from countries in the synthetic control.


                                                    22
one specific event does not drive the results, we remove one treated country at a time;
Figure A11 shows robust dynamic treatment effects for all subsets of treated countries.
Finally, the last row of Table A2 shows that the results are similar when we re-estimate the
difference-in-differences coefficient following the imputation method of Borusyak, Jaravel,
and Spiess (2021) to attenuate issues with the two-way fixed effects estimation.
    We recognize that our set of treated countries is small and that liberalization events do
not occur in a vacuum. The timing of the events could coincide with unobserved changes
in determinants of factor shares and factor taxation. Yet the relatively stable trends in
treated countries pre-liberalization imply that these confounding changes would have to
sharply coincide with the events. The narrative analyzes of the reforms, reproduced from
past studies in Appendix C, do not reveal obvious confounding shocks.
    Even if the events are primarily trade related, our interpretation of the dynamic
coefficients depends on whether other reforms or confounding economic shocks occurred
in post-reform years (Rodriguez and Rodrik, 2001). As discussed in Appendix C, some
countries implemented further trade reforms following the initial liberalization event:
Mexico joined NAFTA in 1994; Argentina and Brazil joined MERCOSUR in 1991. Some
countries also liberalized cross-border capital flows (Mexico removed capital inflow
restrictions in 1989; India liberalized foreign direct investment rules in 1993). These reforms
often occurred several years after trade liberalization, but we observe sharp effects in the
first few years. The short-run results showing a swift break from stable pre-trends are
thus more likely to be directly attributable to trade liberalization. We caution, however,
against attaching too much importance to the specific medium-run coefficients as those
incorporate further cross-border liberalization reforms, general equilibrium impacts, and
other systemic reforms (Goldberg and Pavcnik, 2007).
    Finally, we note that our results are based on a selected set of trade liberalization events
characterized by sharp tariff cuts, in large developing countries with constraints on capital


     Wacziarg and Wallack (2004) study if trade liberalization events in developing countries coincide with
domestic reforms. Among our seven events, only Mexico has a confounding domestic privatization reform
within the first five years of our event-year; Brazil (privatization) and Colombia (broad market-oriented
reforms) had confounding reforms between 5 and 10 years after liberalization; and, the remaining four
countries had no confounding reforms. The results are robust to excluding Mexico (Figure A11).


                                                    23
mobility (Chinn and Ito, 2006). The impacts of trade liberalization are more likely to carry
over to countries with similar characteristics (see also Section 7.3).



7     Regressions with instrumental variables for trade

7.1    Empirical design

In this section, we study the impact of trade in a regression setting, which permits the study
of mechanisms and of heterogeneity by income levels. We use instruments to alleviate
endogeneity concerns. We estimate how trade impacts factor shares and factor taxation:


                             yct = µ ∗ tradect + Θ ∗ Xct + βc + πt +       ct                          (7)

    where yct is the outcome of interest in country c in year t, tradect is the share of import
and exports in net domestic product and µc and πt are country and year fixed effects. We
cluster the error term,     ct ,   at the country level. We also estimate models which include,
in Xct , proxies for confounding determinants of factor shares and factor taxation: the
exchange rate, gross capital formation, log of population, log of GDP per capita, and capital
openness (Rodrik, 1997; Harrison, 2005).
    OLS estimation may be biased due to reverse causality and unobservable confounding
factors which correlate with changes in trade. Since we are interested in uncovering causal
effects, the challenge is to find exogenous trade variation. This leads us to focus on the two
instruments in Egger, Nigai, and Strecker (2019). The first instrument relies on the general
structure of quantitative general equilibrium models of trade (Eaton and Kortum, 2002;
Arkolakis, Costinot, and Rodriguez-Clare, 2012). Under the standard gravity model
assumptions, this instrument uses the average bilateral trade frictions between exporting
and importing countries as the source of variation (aggregated to the country-year level).
In our context, this instrument is valid if the distribution (not the level) of trade costs




    Other studies which leverage the structure of the gravity model to create instruments for trade include
Frankel and D. Romer (1999), Wacziarg (2001), and Anderson and Wincoop (2003).


                                                    24
among individual country-trading pairs is not influenced by the level of factor shares or
factor taxation in the import or export country.
    The second instrument exploits the time-series variation in global oil prices interacted
with a country-specific measure of access to international markets. Specifically, access at
the country-level is captured by the variance of distance from the three most populated
cities to the closest maritime port. Intuitively, this time-invariant measure captures the
internal geography of a country which is an important component of transportation costs.
Following a global shock to oil prices, the transportation costs will be larger in countries
with less concentrated access to maritime ports, leading to a larger drop in imports and
exports.    Conceptually, both instruments aim to capture variation in trade costs driven by
exogenous economic forces. They are detailed in Appendix D.
    We extend the data coverage of these instruments to our full set of countries and
time periods. Since the IV estimate of equation (7) recovers a local average treatment
effect (LATE), it is important to understand the relevance of each instrument across our
full sample. Figure A12 shows that each instrument is relevant in different subsamples:
the oil-distance instrument has a strong first stage in recent decades and in high-income
countries, while the gravity instrument has a stronger first-stage in earlier time periods
and in lower-income countries.            This reveals that an IV estimate based on either of the
individual instruments will be driven by first-stage compliers with characteristics that differ
from the average country in the full sample. But, restricting the analysis to subsamples
where an individual instrument has a strong first stage biases the IV estimates upwards
(Abadie, Gu, and Shen, 2019). To guard against this, we combine the two instruments,
which also raises statistical power (Mogstad, Torgovitsky, and Walter, 2020), and estimate
a LATE that is representative across income levels and time periods. The LATE identified




     In the transport logistics literature, oil prices are a key determinant of transportation costs (Gross,
Hayden, and Butz, 2012; Storeygard, 2016).
     Conceptually, the oil-distance instrument may be stronger in high-income countries if economic
development is associated with improvements in domestic road networks (holding the physical distances
from cities to maritime ports constant). We measure transportation networks in the latest year available; this
introduces possible measurement error which weakens the instrument’s relevance in earlier periods.


                                                     25
with multiple instruments retains an intuitive interpretation: it is a weighted combination
of the instrument-specific LATEs using the instruments one at a time.
    Finally, an attractive feature of these instruments is that they impact cross-border trade
in different ways: Table A3 shows that the gravity instrument causes on average an increase
in trade, while the oil-distance instrument reduces trade. Moreover, both instruments
have significant impacts on imports as well as exports. As such, our IV-estimate reflects
the broad impacts of cross-border trade through its increases and decreases of goods and
services in and out of the country.


7.2    Main results

Table 1 presents the OLS and IV estimation of equation (7) for our core outcomes. Panel A
shows the OLS results, while Panels B through D show different IV specifications. The OLS
and IV coefficients display the same sign, but the IV coefficients are always larger. In Panels
B-D we estimate the IV model. Panel B shows the IV weighted by countries’ yearly national
domestic product (NDP), our benchmark to mirror the global trends shown in section 4.
The 1st stage shows a strong F-statistic of 26.07. The IV estimation yields a positive impact
of trade on the capital share, both in national income and in the corporate sector.
    Turning to the effective tax on factor shares, the IV-results indicate that trade leads to
statistically significant increases in the effective tax rate of both capital and labor, but the
effect on capital (0.375) is twice as large as the effect on labor (0.163). The IV-coefficient on
ET RL is more precise (p-value = 0.003) than that on ET RK (p-value=0.081), which, as we
will see later, masks large heterogeneity across income levels.
    The IV estimates in Panel B are globally representative, since they include country
weights, but Panel C shows that the results are robust to removing these weights: the
1st stage strength is reduced (F-statistic=8.415), but the results are broadly similar. In


      The weights are a function of the strength of each instrument in the first-stage regression in the full
sample (Angrist and Imbens, 1995). In our setting, a stronger weight is placed on the oil-distance instrument
(see first-stage regressions in Table A7).
      The re-allocation towards capital inside the corporate sector implies that our results are not confounded
by a positive impact of trade on the corporate share of national income. Moreover, we relate to the previous
literature on global trends in factor incomes which focuses on shares within the corporate sector.


                                                      26
particular, trade’s impact on the effective tax on capital (0.250, p-value=0.018) remains
positive, and larger than that on labor taxation (0.133, p-value=0.013).
   In Panel D, we include the set of country-year varying controls contained in Xct in
addition to the NDP weights. The inclusion of controls can help improve the precision of
the estimates and could increase the likelihood that the exclusion restriction holds. Indeed,
the controls lower the p-values and trade continues to have a positive impact on the capital
factor share, and a larger positive impact on capital factor taxation than on labor taxation.
Throughout the panels, we see in column 5 that the coefficients on the ETR on corporate
profits mirror that of ET RK , and are more precise.
   The IV-results are robust to a battery of checks. First, we show that they hold with
different measures of trade intensity (Table A4). Second, since one of the instruments relies
on oil price variation, we allow oil-rich countries to be on a separate non-parametric time
path; this addresses the concern that our estimating variation is correlated with trends in
factor shares and effective taxation specific to oil-producing countries, and re-enforces the
results (Table A5). Third, the results are broadly similar when we change our measurement
assumptions to construct factor shares and ETRs (Table A6). The results are robust to
the alternative assignment of taxes to capital versus labor proposed by Mendoza, Razin,
and Tesar (1994); to changing the share of the PIT assigned to capital vs labor; and to
using the ILO (2019) method to attribute mixed-income to labor vs capital, although
the coefficients on ET RK and ET RL are now closer. Fourth, the results based on each
individual instrument are comparable to the joint IV results (Table A7).
   Finally, Table A3 directly reports the reduced-form impact of trade on our outcomes,
leveraging the fact that the two instruments have opposite effects on trade. We find that
the effects of globalization are symmetric: expanded openness increases both ET RL and
ET RK , while reduced cross-border trade decreases the effective taxation of both factors.


7.3   Mechanisms and heterogeneity by income levels

These results re-enforce the findings of the previous sections, that trade raises effective
tax rates, especially on capital. One conjecture to explain these results is that trade exerts



                                             27
a pro-tax capacity effect: trade openness changes the structure of labor markets and
corporations by concentrating economic activity in large capital intensive firms; in turn
this relaxes tax enforceability constraints.         Although the tax capacity channel has not been
studied in-depth before, a wide literature argues that trade exacerbates tax competition
and increases tax avoidance opportunities, thus exerting a downward pressure on capital
tax rates in rich countries (a race-to-the-bottom effect).
    To shed light on each of these mechanisms we look at how trade impacts outcomes
which more directly relate to each hypothesis (self-employment shares for tax-capacity,
statutory corporate tax rates for the race-to-the-bottom). We then revisit previous results
to check for heterogeneous impacts across income levels, since we expect the tax capacity
effect to mainly operate in low and middle-income countries.


Outcomes linked to mechanisms                Table 2 repeats the benchmark IV specification but
looks at additional outcomes. Panel A shows that trade leads to a reduction in the
statutory corporate income tax rate, thus supporting a race-to-the-bottom effect.                       At the
same time, Panels B and C show that trade causes a reduction in the share of workers in
self-employment, and an increase in the corporate share of GDP. Thus, two countervailing
forces appear to be at play: the growth in employee-employment and of the corporate
sector raises tax enforceability, while active government policies in the form of reductions
of the statutory corporate tax rate lower the tax burden on capital.


Heterogeneity by development level Motivated by the contrasting long-run trends of
ET RK in high versus low and middle-income countries, we investigate if trade impacts

     The literature convincingly shows that third-party information trails are key for tax enforcement
(Pomeranz, 2015; Naritomi, 2019). Activities with limited third-party data, such as self-employment, lead to
high tax evasion rates (Kleven, Knudsen, et al., 2011), and the movement from self-employment to formal
wage employment is associated with growth in tax enforcement capacity (Jensen, 2022).
     In Appendix Table A8, we verify that corporate income tax rate changes are significantly associated
with changes in corporate income tax revenue (% of GDP) and with ET RK . Consistent with past studies on
the determinants of tax policies (C. Romer and D. Romer, 2010), the outcome variable is the first-difference of
the CIT rate: ∆CITt,t−1 . Results are robust to alternative outcomes, including: the level of the CIT rate while
controlling for the lagged CIT rate; and a reform-tracker which changes value when the CIT rate changes.
Results available upon request.
     Table A9 also shows that trade primarily causes a transition from agriculture to industry, with a small
positive impact on services. Thus, in this empirical setting, trade induces a transition from a commonly
identified ’hard-to-tax’ sector (agriculture) to an ’easy-to-tax’ sector (industry).


                                                      28
taxation differentially across development levels. We estimate heterogeneous IV effects by
interacting the trade variable with a high-income country dummy:


           yct = µ ∗ tradect + κ ∗ tradect ∗ 1(HighIncome) + Θ ∗ Xct + βc + πt +              ct       (8)

    The results are presented in Table 3. In Column (1), we find that trade has a strong
positive effect on ET RK in developing countries, but a null effect in rich countries. Column
(2) shows that trade increases ET RL in both samples, but that this effect is stronger in
high-income countries. Column (3) finds that trade decreases the statutory corporate tax
rate in both samples, although by more so in rich countries. In contrast, Columns (4)-(6)
show that the positive effect of trade on the employee-share and the corporate share is
entirely concentrated in developing countries, with null effects in high income countries.
These results point to heterogeneous mechanisms depending on countries’ income levels:
globalization might have limited capital taxation in rich countries by putting downward
pressure on statutory corporate tax rates; while in developing countries although statutory
rates also fell, they were more than counteracted by the expansion of the capital tax base
which became more enforceable. On net this led to a rise in effective capital taxation in
developing countries.
    The trade liberalization events (Section 6) took place exclusively in developing countries.
In Appendix Figure A13, we find that the trade-events led to growth in the corporate
share but had no important impacts on corporate tax policy. These mechanism results are
strongly consistent with the IV analysis, and reinforce the plausible role of enforceability
in mediating trade’s impact on capital taxation in developing countries.


Heterogeneity by country size and mobility of capital Beyond the split by development
level, we estimate additional sources of heterogeneity which mediate the impact of trade on
taxation. Concerns related to capital flight are more pronounced in small-market economies
(Wilson, 1999) and in countries with few restrictions on capital mobility (Rodrik, 1997;

     The two instruments leverage distinct variation; as such, interpreting heterogeneous IV-coefficients is
challenged by the possibility that each instrument captures different LATEs (Mogstad, Torgovitsky, and
Walter, 2020). However, as previously discussed, the IV-estimates based on using each instrument separately
are in the range of the estimates based on using both instruments simultaneously (Table A7).


                                                    29
Chinn and Ito, 2006). We test for these mechanisms, by looking at heterogeneous treatment
effects for statutory tax rates and effective tax rates using equation (8).
    Table 4 shows the results. In Panel A, we find that increased trade openness leads to a
reduction in the statutory CIT rate which is stronger in smaller countries and in countries
with limited restrictions on capital mobility. Mirroring this result, panel B, shows that the
positive effect of trade on ET RK only occurs in large countries (population over 40 million)
and in countries with capital restrictions. These results support the conjecture that the
pro-tax capacity effects of trade happen simultaneously with the race-to-the-bottom effects:
only countries that can limit capital mobility and tax avoidance are able to increase ET RK
when they open to trade. A further hypothesis is that countries which collect less revenue
from capital due to trade liberalization, compensate by taxing more the immobile factor,
labor, to balance their budget (Rodrik, 1998b). In Panel C, we indeed find that the rise in
ET RL is qualitatively larger in small countries and in countries without capital restrictions.


7.4    Impacts on tax revenue

To further substantiate the tax capacity hypothesis, we look at the impact of trade on overall
tax collection as a share of GDP, including capital, labor and indirect taxes.                 Table 5 shows
the impacts of trade on different taxes by development level. Column (1) shows that in
developing countries, trade openness leads to a significant increase in overall tax to GDP
as opposed to a null effect in rich countries. The positive result for developing countries
re-enforces the hypothesis that trade produces an increase in overall tax capacity, while the
null result for rich countries is expected given their already high tax capacity.
    Further, Table 5 breaks down the impact of trade on different tax revenue sources. In
developing countries, the increase in total tax revenue with trade is primarily due to the
significant rise in corporate income tax. All other tax sources slightly rise with trade, but no
coefficient is significantly different from zero. In high income countries the effect of trade
on CIT collection is slightly negative (but insignificant). Among taxes mainly assigned to
labor, the PIT collection does not change with trade, as opposed to payroll taxes which

    Looking directly at tax revenue also alleviates potential concerns of weak statistical capacity in developing
countries, which could bias our measures of national income components and thus ET RL and ET RK .


                                                       30
increases significantly in high-income countries. These results on labor taxes echo the
literature: Egger, Nigai, and Strecker (2019) shows that trade shifts the tax burden of the
personal income tax away from the top earners and towards the median worker without
changing overall collection. The rise in payroll tax revenues re-enforces Rodrik (1997)’s
insurance argument: to protect themselves against the economic fluctuations brought by
trade openness, workers demand further social protection, financed by payroll taxes.


7.5    Quantitative importance of trade openness

How should we think of the quantitative importance of trade in accounting for the rise
in capital taxation in developing countries? First, we note that although the IV and
event-study estimations (Section 6) rely on entirely different identifying assumptions and
methodologies, they yield comparable results in magnitude. Under the strong assumption
that the trade liberalization events only impact factor taxation via trade, the event-study
results imply an impact on ET RK of 0.489, compared to the IV-coefficient of trade on ET RK
in developing countries of 0.44 (Table 3). Taken at face value, this means that increasing
trade by 10 percentage points raises ET RK by 4 to 5 percentage points. Second, we can
combine our ET RK coefficient with the change in trade openness in developing countries
between 1965 and 2018: this back of the envelope calculation implies that trade accounts
for a rise in ET RK of 3 percentage points. This number should be taken with caution, but
suggests that a third of the long-run rise in ET RK is explained by trade globalization.



8     Capital liberalization events
Until now, we have studied one key dimension of globalization, in the form of trade
openness, and its impact on factor income taxation. Given our interest in the taxation of
capital, another relevant dimension of globalization is capital openness. However, due to
differences in countries’ reporting requirements for capital flows, data on capital openness


     Concretely, the long-run increase in trade openness is 7.01 percentage points (Figure A9) and the
trade-coefficient for ET RK is 0.44 (Table 3), hence 7.01 ∗ 0.44 = 3.08ppt. The long-run increase in ET RK is
10.1ppt (Figure 3), thus yielding 3.08/10.1 = 0.305


                                                    31
is not as internationally comparable and available than data on trade (Egger, Nigai, and
Strecker, 2019). Further, to our knowledge, the literature has not identified a credible
instrumental variable for capital openness (Magud, Reinhart, and Rogoff, 2011; Alfaro,
Kalemli-Ozcan, and Volosovych, 2005).            Notwithstanding, we provide here some evidence
on the impact of capital liberalization based on an event-study design.
    We rely on Chari, Henry, and Sasson (2012) who identified capital liberalization events
in 25 developing countries corresponding to the date when foreign investment in the
domestic stock market was first allowed.             The paper shows that these seemingly narrow
events actually greatly expand foreign capital flows into the country, including foreign
direct investment (FDI), and raise the import of capital goods.             Compared to other reforms
aimed at lifting restrictions on FDI, opening the domestic stock market internationally
occurs at a precise point in time (other policies are often less precise and staggered); is
not marked by policy-reversal or by net capital outflow; and is unambiguously related to
capital liberalization (Henry, 2007; Eichengreen, 2001).
    We employ the same empirical design as in in Section 6, and create a synthetic control for
each treated country and outcome of interest (see Appendix C.3 for details). We measure
capital openness as the sum of foreign assets and liabilities, as a share of GDP (Lane
and Milesi-Ferreti, 2017). Figure 8 reports the results.             Starting from a stable pre-trend,
we observe a rise in capital openness, precisely at the time of the event, which keeps on
building over the post-event decade. The ET RK also increases, but with a few years lag
relative to the event: in the medium-run, the effect is precisely estimated and significant



     Trade and capital flows may also exhibit strong co-movement. Per example, as the cost of importing
intermediate goods decreases, a firm can decide to outsource the production of intermediate goods abroad –
which may result in an outflow of capital if the firm decides to purchase the production process abroad. This
co-movement makes it challenging to estimate separate effects of capital and trade flows with precision.
     Removing restrictions on the stock market constitute a liberalization of the capital account in relation
to domestic financial markets. Lane and Milesi-Ferretti (2009) find that capital account liberalization and
domestic financial development are strongly correlated with financial integration across countries
     FDI includes both green field investments (building of plants from scratch) and cross-border mergers
and acquisitions; the latter is directly impacted by stock market liberalization and makes up 40-60% of FDI in
recent times in developing countries. It is likely that the increased foreign ownership on the stock market
subsequently triggers an increase in green field investments.
     In Appendix Figure A14, we show that the results are robust to creating synthetic controls that are
based on simultaneously matching all outcomes for each treated country.


                                                     32
at the 5% level.     There is no discernible effect on ET RL . The absence of an increase in
the capital share is intriguing; we note that Chari, Henry, and Sasson (2012) find large
wage effects, suggesting a proportionate (and high) growth of both factor incomes. Foreign
inflow of capital, as well as any subsequent increase in capital goods import and aggregate
investment, may positively impact ET RK by contributing to the growth of large, complex
firms with employees and thereby raising the tax-enforceable share of capital income
(Section 7.3). Consistent with a role for tax capacity, we find that the capital liberalization
events led to a decrease in the non-corporate sectors (Appendix Figure A15).
    Qualitatively, these results are consistent with those from the previous sections, sug-
gesting that the positive impact of globalization on effective capital taxation in low and
middle income countries is robust to using capital instead of trade openness. However,
given the inherent limitations with the measurement of capital flows, we consider that
our results based on trade openness provide more meaningful and robust insights into
globalization’s impacts on factor taxation.



9    Concluding remarks and perspectives
In this paper, we combine a new global database with several empirical strategies to provide
novel evidence on trends and causal effects of globalization on tax structures worldwide.
Our starting point is the systematic harmonization of novel historical national accounts
data and dis-aggregated government revenue statistics. This data collection permitted the
construction of a new measure of effective tax rates on capital and labor in 156 countries
between 1965 and 2018.
    Using this database, we make two contributions. First, we establish new facts. Taking a
global perspective, the average effective tax rates on labor and capital have converged, due
to a increase in labor taxation and a fall in capital taxation. We find differences between
developed and developing countries: while the effective tax rate on capital fell in OECD



     Consistent with our result, Quinn (1997) finds a positive correlation between de-jure capital account
openness and corporate taxation as a share of GDP. Note that the events considered here remove restrictions
on capital inflows; it is possible that increased capital outflows may, conversely, reduce ET RK .


                                                    33
countries, it increased in the rest of the world (albeit starting from a very low level) in the
post-1995 period of hyper-globalization.
   Our second contribution is to formulate and test a new hypothesis that sheds light on
these diverging global trends: that trade liberalization exerts a pro-tax-capacity effect, by
increasing the concentration of economic activity in large, formal corporate structures.
Using a verity of research designs, we show that trade leads to a higher effective taxation
of capital, but only in developing countries. For these countries, the base expansion
channel has been quantitatively large enough to offset the negative tax-competition effect
of globalization. In high income countries, by contrast, the tax-competition effects has
dominated, leading to a decline in capital taxation.
   In this paper we have taken a global and macroeconomic perspective on tax systems
and inequality, focusing on factor income shares and effective tax rates on labor and
capital. In future research, our database (available online) could be used to study the
effects of globalization on tax progressivity and inequality between groups of individuals.
By combining our macroeconomic tax rates on labor and capital with estimates of the
progressivity of labor and capital taxes (for instance using tax simulators, as in Egger,
Nigai, and Strecker (2019), one could estimate changes in the progressivity of the entire
tax system. Moreover, these changes in tax progressivity could be compared to the effects
of globalization on the distribution of pre-tax income. This would make it possible to
quantify the extent to which changes in taxation caused by globalization have curbed or
exacerbated the unequalizing effects of international economic integration.




                                              34
References
Abadie, Alberto, Alexis Diamond, and Jens Hainmueller (2010). “Synthetic control methods
   for comparative case studies: Estimating the effect of California’s tobacco control
   program”. In: Journal of the American Statistical Association 105(490), pp. 493–505.
Abadie, Alberto, Jiaying Gu, and Shu Shen (2019). Instrumental Variable Estimation with
   First-Stage Heterogeneity. Tech. rep.
Akcigit, Ufuk et al. (2021). “Taxation and Innovation in the 20th Century.” In: NBER Working
   Paper Series( 24982).
Albers, Thilo, Morten Jerven, and Marvin Suesse (2020). “The Fiscal State in Africa: Evidence
   from a century of growth”. In: African Economic History Working Paper Series.
Alfaro, Laura, Sebnem Kalemli-Ozcan, and Vadym Volosovych (2005). “Capital Flows in a
   Globalized World: The Role of Policies and Institutions”. In: NBER Working Paper Series(
   11696).
Anderson, James and Eric van Wincoop (2003). “Gravity with Gravitas: A Solution to the
   Border Puzzle”. In: American Economic Review 93(1), pp. 170–92.
Angrist, Joshua and Guido Imbens (1995). “Two-Stage Least Squares Estimation of Average
   Causal Effects in Models With Variable Treatment Intensity”. In: Journal of the American
   Statistical Association 90(430), pp. 431–45.
Arkolakis, Costas, Arnaud Costinot, and Andres Rodriguez-Clare (2012). “New Trade
   Models, Same Old Gains?” In: American Economic Review 102(1), pp. 94–130.
Attanasio, Orazio, Pinelopi Goldberg, and Nina Pavcnik (2004). “Trade reforms and
   wage inequality in Colombia”. In: Journal of Development Economics 74(2), pp. 331–
   366.       : 0304-3878.      : https : / / doi . org / 10 . 1016 / j . jdeveco . 2003 . 07 . 001. :
   https://www.sciencedirect.com/science/article/pii/S0304387804000185.
Basri, Chatib et al. (2019). “Tax Administration vs. Tax Rates: Evidence from Corporate
   Taxation in Indonesia”. In: NBER Working Paper( 26150).
Bates, Robert, Da-Hsiang, and Donald Lien (1985). “A Note on Taxation, Development and
   Representative Government”. In: Politics and Society 14(1), pp. 53–70.
Baunsgaard, Thomas and Mick Keen (2009). “Tax revenue and (or?) trade liberalization”.
   In: Journal of Public Economics 94(9), pp. 563–577.
Benzarti, Youssef and Aliza Tazhitdinova (2021). “Do Value-Added Taxes Affect Interna-
   tional Trade Flows? Evidence from 30 Years of Tax Reforms”. In: American Economic
   Journal: Economic Policy 13(4), pp. 469–89.
Besley, Timothy and Torsten Persson (2014). “Why Do Developing Countries Tax So Little?”
   In: Journal of Economic Perspectives 28(4), pp. 99–120.
Blanchet, Thomas and Lucas Chancel (2016). “National Accounts Series Methodology.”
       : https://wid.world/document/1676/.
Blanchet, Thomas, Lucas Chancel, et al. (June 2021). Distributional National Accounts
   Guideliens: Methods and Concepts Used in the World Inequality Database. Tech. rep. World
   Inequality Lab.       : https://wid.world/document/distributional-national-accounts-
   guidelines-2020-concepts-and-methods-used-in-the-world-inequality-database/.
Borusyak, K, X Jaravel, and J Spiess (2021). Revisiting Event Study Designs: Robust and Efficient
   Estimation. Tech. rep. Working Paper.


                                                 35
Brandt, Loren et al. (2017). “WTO Accession and Performance of Chinese Manufacturing
   Firms”. In: American Economic Review 107(9), pp. 2784–2820.
Browning, Edgar K. (1978). “The Burden of Taxation.” In: Journal of Political Economy 86(4),
   pp. 649–671.       : 10.1086/260703.
Buettner, Thiess and Boryana Madzharova (2018). “WTO membership and the shift to
   consumption taxes”. In: World Development 108(1), pp. 197–218.
Cage, Julia and Lucie Gadenne (2018). “Tax revenues and the fiscal cost of trade liberalization,
   1792-2006”. In: Explorations in Economic History 70, pp. 1–24.
Cameron, Colin, Jonah Gelbach, and Douglas Miller (2008). “Bootstrap-Based Improvements
   for Inference with Clustered Standard Errors”. In: Review of Economics and Statistics
   90(3), pp. 414–27.
Carey, David and Josette Rabesona (2004). “Tax Ratios on Labor and Capital Income and on
   Consumption.” In: Measuring the Tax Burden on Capital and Labor. Ed. by Peter B. Sørensen.
   MIT Press.       : 10.7551/mitpress/4522.001.0001.
Cette, Gilbert, Lorraine Koehl, and Thomas Philippon (2020). “Labor share”. In: Economics
   Letters 188, p. 108979.     : 0165-1765.   : https://doi.org/10.1016/j.econlet.2020.108979.
       : https://www.sciencedirect.com/science/article/pii/S0165176520300239.
Chaisemartin, Clément de and Xavier D’Haultfœuille (Sept. 2020). “Two-Way Fixed Effects
   Estimators with Heterogeneous Treatment Effects”. In: American Economic Review 110(9),
   pp. 2964–96.       : 10.1257/aer.20181169.       : https://www.aeaweb.org/articles?id=10.
   1257/aer.20181169.
Chari, Anusha, Peter Henry, and Diego Sasson (2012). “Capital Market Integration and
   Wages”. In: American Economic Journal: Macroeconomics 4(2), pp. 102–132.
China Daily (2018). Taxation: Historical development.        : https://govt.chinadaily.com.cn/s/
   201809/20/WS5c08d6fe498eefb3fe46e52c/historical-development.html.
Chinn, Menzie and Hiro Ito (2006). “What Matters for Financial Development? Capital
   Controls, Institutions, and Interactions”. In: Journal of Development Economics 81(1),
   pp. 163–192.
CIAT and IDB (2019). IDB-CIAT Revenue Collection Database.            : https://www.ciat.org/idb-
   ciat-revenue-collection-database/?lang=en.
Cogneau, Denis et al. (2021). “Taxation in Africa from Colonial Times to Present Evidence
   from former French colonies 1900-2018”. In: Working Paper.
Devereux, Michael P. (2004). “Measuring Taxes on Income from Capital.” In: Measuring
   the Tax Burden on Capital and Labor. Ed. by Peter B. Sørensen. MIT Press.            : 10.7551/
   mitpress/4522.001.0001.
Dix-Carneiro, Rafael, Pinelopi Goldberg, et al. (Jan. 2021). Trade and Informality in the Presence
   of Labor Market Frictions and Regulations. Working Paper 28391. National Bureau of
   Economic Research.         : 10.3386/w28391.        : http://www.nber.org/papers/w28391.
Dix-Carneiro, Rafael and Brian Kovak (2017). “Trade Liberalization and Regional Dynamics”.
   In: American Economic Review 107(10), pp. 2908–46.
Eaton, Jonathan and Samuel Kortum (2002). “Technology, Geography, and Trade.” In:
   Econometrica 70(5), pp. 1741–79.
Egger, Peter H., Sergey Nigai, and Nora M. Strecker (2019). “The Taxing Deed of Globaliza-
   tion.” In: American Economic Review 109(2), pp. 353–90.            : 10.1257/aer.20160600.


                                               36
Eichengreen, Barry (2001). “Capital Account Liberalization: What do Cross-Country Studies
   Tell Us?” In: World Bank Economic Review 15(3), pp. 341–365.
Feenstra, Robert and Gordon Hanson (2001). “Global Production Sharing and Rising
   Inequality: A Survey of Trade and Wages”. In: NBER Working Paper Series( 8372).
Fisunoglu, Ali et al. (2011). Relative Political Capacity Dataset. Version 2.4 (August 2020).     :
   10.7910/DVN/NRR7MB.
Frankel, Jeffrey and David Romer (1999). “Does Trade Cause Growth?” In: American
   Economic Review 89(13), pp. 379–399.
Goldberg, Pinelopi and Nina Pavcnik (2007). “Distributional Effects of Globalization in
   Developing Countries.” In: Journal of Economic Literature 45(1), pp. 39–82.
Goldberg, Pinelopi and Nina Pavcnik (2016). “The Effects of Trade Policy”. In: NBER
   Working Paper Series( 21957).
Gollin, Douglas (2002). “Getting Income Shares Right”. In: Journal of Political Economy
   110(2).     : 10.1086/338747.
Gross, Wendelin, Cristina Hayden, and Christian Butz (2012). “About the impact of rising
   oil price on logistics networks and transportation greenhouse gas emission”. In: Logistics
   Research 4, pp. 147–56.
Guerriero, Marta (2019). “The Labor Share of Income around the World: Evidence from a
   Panel Dataset.”        : https://www.adb.org/publications/labor-share-income-around-
   world-evidence-panel-dataset.
Gygli, Savina et al. (2019). “The KOF GLobalization Index – revisited”. In: The Review of
   International Organizations 14, pp. 543–574.
Harrison, Ann (2005). Has Globalization Eroded Labor’s Share? Some Cross-Country Evidence.
   Working Paper.
Henry, Peter (2007). “Capital Account Liberalization: Theory, Evidene and Speculation”.
   In: Journal of Economic Literature 45(4), pp. 887–935.
ICTD/UNU-WIDER (2020). Government Revenue Dataset.                    : https://www.wider.unu.
   edu/project/government-revenue-dataset.
ILO (2019). “The Global Labour Income Share and Distribution.”                 : https://www.ilo.
   org/ilostat-files/Documents/Labour%5C%20income%5C%20share%5C%20and%
   5C%20distribution.pdf.
ILOSTAT (2021). Statistics on labour income and inequality.       : https://ilostat.ilo.org/topics/
   labour-income/.
IMF GFS (2005). Historical Government Finance Statistics (1972-1989 in GFSM 1986 format).
   CD-ROM.          : https://www.imf.org/external/pubs/cat/longres.aspx?sk=18674.0.
Jaeger, Simon, Shakked Noy, and Benjamin Schoefer (2021). “What Does Co-Determination
   Do?” In: NBER Working Paper Series( 28291).
Jefferson, Gary (2016). “State-Owned Enterprise in China: Reform, Performance, and
   Prospects”. In: Working Paper.
Jensen, Anders (2022). “Employment Structure and the Rise of the Modern Tax System”.
   In: American Economic Review 112(1), pp. 213–34.
Kanbur, Ravi and Michael Keen (1993). “Jeux Sans Frontier: Tax Competition and Tax
   Coordination when Countries Differ in Size”. In: American Economic Review 83(1),
   pp. 877–92.


                                                37
Karabarbounis, Loukas and Brent Neiman (2014). “The Global Decline of the Labor Share.”
   In: Quarterly Journal of Economics 129(1), pp. 61–103.      : 10.1093/qje/qjt032.
King, Mervyn A. and Don Fullerton (1984). “The Theoretical Framework.” In: The Taxation
   of Income from Capital: A Comparative Study of the United States, the United Kingdom, Sweden,
   and Germany. National Bureau of Economic Research, pp. 7–30.                 : 9780226436319.
       : 10.7208/chicago/9780226436319.001.0001.
Kleven, Henrik, Martin Knudsen, et al. (2011). “Unable or Unwilling to Cheat? Evidence
   from a Tax Audit Experiment in Denmark”. In: Econometrica 79(3), pp. 651–692.
Kleven, Henrik, Claus Kreiner, and Emmanuel Saez (2016). “Why Can Modern Governments
   Tax so Much? An Agency Model of Firms as Fiscal Intermediaries”. In: Economica 83(330),
   pp. 219–246.
Kornai, János, Eric Maskin, and Géard Roland (Dec. 2003). “Understanding the Soft
   Budget Constraint”. In: Journal of Economic Literature 41(4), pp. 1095–1136.             : 10 .
   1257/002205103771799999.           : https://www.aeaweb.org/articles?id=10.1257/
   002205103771799999.
Lane, Philip and Gian Maria Milesi-Ferreti (2017). “International Financial Integration in
   the Aftermath of the Global Financial Crisis”. In: IMF Working Paper 17(115).
Lane, Philip and Gian Maria Milesi-Ferretti (2009). “The Drivers of Financial Globalization”.
   In: American Economic Review P P 98(2), pp. 327–32.
Lequiller, François and Derek Blades (2014). Understanding National Accounts. 2nd ed. OECD.
       : 10.1787/9789264214637-en.
Lotz, Joergen R and Elliott R Morss (1970). “A theory of tax level determinants for developing
   countries”. In: Economic Development and cultural change 18(3), pp. 328–341.
Magud, Nicolas, Carmen Reinhart, and Kenneth Rogoff (2011). “Capital controls: Myth
   and reality”. In: NBER Working Paper Series( 16805).
Mahdavi, Paasha (2020). Power Grab: Political Survival through Extractive Resource Nationaliza-
   tion. Cambridge University Press.        : 10.1017/9781108781350.
McCaig, Brian and Nina Pavcnik (2018). “Export Markets and Labor Allocation in a
   Low-Income Country”. In: American Economic Review 108(7), pp. 1899–1941.
McDaniel, Cara (2007). “Average tax rates on consumption, investment, labor and capital
   in the OECD 1950-2003”. In:
Mendoza, Enrique G., Assaf Razin, and Linda L. Tesar (1994). “Effective tax rates in macroe-
   conomics: Cross-country estimates of tax rates on factor incomes and consumption.” In:
   Journal of Monetary Economics 34(3), pp. 297–323.       : 10.1016/0304-3932(94)90021-3.
Mogstad, Magne, Alexander Torgovitsky, and Christopher Walter (2020). “The Causal
   Interpretation of Two-Stage Least Squares with Multiple Instrumental Variables”. In:
   NBER Working Papers( 25691).
Naritomi, Joana (2019). “Consumers as Tax Auditors”. In: American Economic Review 109(9),
   pp. 3031–72.
OECD (2020). Revenue Statistics.        : https://www.oecd.org/tax/revenue- statistics-
   2522770x.htm.
OECD-UCLG (2019). World Observatory on Subnational Government Finance and Investment.
        : http://www.sng-wofi.org/data/.
Ohlin, Bertil (1933). “Interregional and International Trade”. In:


                                               38
OPEC (2021). Average annual OPEC crude oil price from 1960 to 2021.           : https://www.
   statista.com/topics/1830/opec/.
Piketty, Thomas, Emmanuel Saez, and Gabriel Zucman (2018). “Distributional National
   Accounts: Methods and Estimates for the United States”. In: The Quarterly Journal of
   Economics 133(2), pp. 553–609.
Pomeranz, Dina (2015). “No Taxation without Information: Deterrence and Self-Enforcement
   in the Value-Added Tax”. In: American Economic Review 105(8), pp. 2539–69.
Quinn, Dennis (1997). “The Correlates of Change in International Financial Regulation”.
   In: American Political Science Review 91(3), pp. 531–551.
Rodriguez, Francisco and Daniel Ortega (2006). “Are Capital Shares Higher in Poor
   Countries? Evidence from Industrial Surveys”. In: Working Paper.
Rodriguez, Francisco and Dani Rodrik (2001). “Trade Policy and Economic Growth: A
   Skeptic’s Guide to the Cross-National Evidence”. In: NBER Macroeconomic Annual 15,
   pp. 261–338.
Rodrik, Dani (1997). “Trade, Social Insurance, and the Limits to Globalization”. In: NBER
   Working Paper Series( 5905).
Rodrik, Dani (1998a). “Why do More Open Economies Have Bigger Governments?” In:
   Journal of Political Economy 106(5), pp. 997–1032.    : 10.1086/250038.
Rodrik, Dani (1998b). “Why do More Open Governments Have Bigger Governments?” In:
   Journal of Political Economy 106(15), pp. 997–1032.
Romer, Christina and David Romer (2010). “The Macroeconomic Effects of Tax Changes:
   Estimates Based on a New Measure of Fiscal Shocks”. In: American Economic Review
   100(1), pp. 763–801.
Ross, Michael and Paasha Mahdavi (2015). “Oil and gas data, 1932–2014”. In: Harvard
   Dataverse 2.
Saez, Emmanuel and Gabriel Zucman (2019a). Clarifying Distributional Tax Incidence: Who
   Pays Current Taxes vs. Tax Reform Analysis. Tech. rep. UC Berkeley.
Saez, Emmanuel and Gabriel Zucman (2019b). The Triumph of Injustice: How the Rich
   Dodge Taxes and How to Make Them Pay. W.W.Norton.            : 9780393531732.        : https:
   //taxjusticenow.org/#/.
SeaRates (2021). Online Freight Shipping and Transit Time Calculator.         : https://www.
   searates.com/services/distances-time/.
Slemrod, Joel (2004). “Are Corporate Tax Rates, Or Countries, Converging?” In: Journal of
   Public Economics 88(6), pp. 1169–1186.
Storeygard, Adam (2016). “Farther on down the Road: Transport Costs, Trade and Urban
   Growth in Sub-Saharan Africa”. In: Review of Economic Studies 83(1), pp. 1263–95.
Swank, Duane and Sven Steinmo (2002). “The New Political Economy of Taxation in
   Advanced Capitalist Democracries”. In: American Journal of Political Science 46(3), pp. 642–
   655.
Topalova, Petia and Amit Khandelwal (2011). “Trade Liberalization and Firm Productivity:
   The Case of India”. In: Review of Economics and Statistics 93(3), pp. 995–1009.
Tørsløv, Thomas, Ludvig Wier, and Gabriel Zucman (Apr. 2020). The Missing Profits of Nations.
   Working Paper 24701. National Bureau of Economic Research.              : 10.3386/w24701.
       : http://www.nber.org/papers/w24701.


                                              39
UN (2018). Handbook on Backcasting.      : https://unstats.un.org/unsd/nationalaccount/
   aeg/2018/M12_8iiib_Backcasting.pdf.
UN DESA (2019). World Urbanization Prospects: The 2018 Revision. United Nations, Depart-
   ment of Economic and Social Affairs, Population Division.
UN SNA (1968). A System of National Accounts.           : https : / / unstats . un . org / unsd /
   nationalaccount/docs/1968SNA.pdf.
UN SNA (2008). A System of National Accounts.           : https : / / unstats . un . org / unsd /
   nationalaccount/docs/sna2008.pdf.
UNSD (2018). Handbook on Backcasting.      : https://unstats.un.org/unsd/nationalaccount/
   aeg/2018/M12_8iiib_Backcasting.pdf.
Wacziarg, Romain (2001). “Measuring the Dynamic Gains from Trade”. In: World Bank
   Economic Review 15(3), pp. 393–429.
Wacziarg, Romain and Jessica Wallack (2004). “Trade Liberalization and Intersectoral Labor
   Movements”. In: Journal of International Economics 64, pp. 411–439.
Wacziarg, Romain and Karen Horn Welch (2008). “Trade Liberalization and Growth: New
   Evidence”. In: World Bank Economic Review 22(2), pp. 187–231.
WID (2020). World Inequality Database.      : http://wid.world.
Wilson, John (1999). “Theories of Tax Competition”. In: National Tax Journal 52(2), pp. 269–
   304.
Wong, Christine P. W. and Richard M. Bird (2008). “China’s Fiscal System: A Work in
   Progress”. In: China’s Great Economic Transformation. Ed. by Loren Brandt and Thomas G.
   Rawski. Cambridge University Press, pp. 429–466.        : 10.1017/CBO9780511754234.013.
World Bank (2008). Public Finance in China: Reform and Growth for a Harmonious Society.
   Ed. by Jiwei Lou and Shuilin Wang. The World Bank.
Young, Alwyn (1995). “The Tyranny of Numbers: Confronting the Statistical Realities
   of the East Asian Growth Experience”. In: The Quarterly Journal of Economics 110(3),
   pp. 641–680.




                                               40
                 Figure 1: Tax Revenue as a Share of Domestic Product




Notes: This figure plots the time series of tax revenue as a share of net domestic product (NDP), separated
into five revenue sources. The top left panel corresponds to the global average, weighting country-year
observations by their share in that year’s total NDP, in constant 2019 USD (N=156). The bottom-left
panel shows the results for high-income OECD countries (N=37), and the bottom right for low- and
middle-income countries (N=119). We consider as high-income, all OECD countries that meet the World
Bank’s classification of high-income. Tax revenues are separated into five main categories: indirect taxes
(including domestic consumption taxes, excises, and tariffs), payroll taxes, taxes on personal income,
taxes on property and wealth, and taxes on corporate income. The dataset is composed of two (quasi)
balanced panels: the first covers the years 1965-1993 and excludes communist regimes. It accounts
for 85-90% of World GDP during those years. The second, covers 1994-2018 and integrates former
communist countries, and in particular China and Russia, and accounts for 98% of World GDP.




                                                 41
                       Figure 2: Capital Share of Domestic Product




Notes: This figure plots the time series of the capital share as a percentage of net domestic product
(NDP). The solid line corresponds to the overall capital share, and the dotted line to the capital share
within the corporate sector. The top left panel corresponds to the global average, weighting country-year
observations by their share in that year’s total NDP, in constant 2019 USD (N=156). The bottom-left
panel shows the results for high-income OECD countries (N=37), and the bottom right for low- and
middle-income countries (N=119). We consider as high-income, all OECD countries that meet the World
Bank’s classification of high-income. The dataset is composed of two (quasi) balanced panels: the first
covers the years 1965-1993 and excludes communist regimes. It accounts for 85-90% of World GDP
during those years. The second, covers 1994-2018 and integrates former communist countries, and in
particular China and Russia, and accounts for 98% of World GDP.




                                                 42
                    Figure 3: Effective Taxation of Capital and Labor




Notes: This figure plots the time series of average effective tax rates on labor (blue) and capital (red), as
well as the effective tax rate on corporate profits (red dashed line). The top-left panel corresponds to the
global average, weighting country-year observations by their share in that year’s total NDP, in constant
2019 USD (N=156). The bottom-left panel shows the results for high-income OECD countries (N=37),
and the bottom-right panel for low- and middle-income countries (N=119). High-income countries are
OECD countries that meet the World Bank’s income threshold of high-income. The dataset is composed
of two (quasi) balanced panels: the first covers the years 1965-1993 and excludes communist regimes. It
accounts for 85-90% of World GDP during those years. The second, covers 1994-2018 and integrates
former communist countries, and in particular China and Russia, and accounts for 98% of World GDP.
Figure shows how the entry into our panel in 1994 of these countries impact the results, by imputing
their pre-1994 data with a regression procedure.




                                                 43
Figure 4: Effective Taxation of Capital and Labor in Low- and Middle-Income Countries




  Notes: This figure plots the time series of average effective tax rates on labor, capital, and corporate profits,
  in the 118 low- and middle-income countries. Compared to the full sample of 156 countries, it excludes
  OECD countries classified as high-income by the World Bank. The top-left panel is our benchmark
  result, taken from 3. The top-right panel excludes former communist countries, most notably China and
  Russia. The mid-left panel excludes oil-rich countries (the 33 countries where average oil production
  since 1990 has exceeded 6.5% of GDP, per Ross and Mahdavi (2015)). The mid-right panel excludes both
  ex-communist and oil-rich nations. Finally the bottom panels show the results separately for the 18 large
  (non-oil rich) countries to the left, and the 68 small (non-oil rich) countries to the right. Large (small)
  countries are defined as having a population above (below) 40 Million in 2018.




                                                     44
   Figure 5: Change in Capital Shares and Factor Taxation vs. Change in Trade

              (a) Capital Share                           (b) Capital Share of Corporate Sector




             (c) ETR on Capital                                      (d) ETR on Labor




Notes: These figures show the relationship between trade and the capital share of domestic product (a);
the capital share of the corporate sector (b); the effective tax rate on capital income (c); and the effective
tax rate on labor income (d). Trade is measured as the sum of import and exports as a share of NDP.
Both the x-axis and y-axis are measured as within-country percent changes over 5 years. Each graph
shows binned scatter plots of each outcome against trade, after residualizing all variables against year
fixed effects. Each dot corresponds to a ventile (20 equal-sized bins) of the residualized trade variable.
For ease of interpretation, we add back the (non-residualized) mean of the given variable. Linear trend
lines are unweighted, with year fixed effects, and are estimated based on the underlying country-year
panel data. The corresponding slope and standard error are shown top-left in each panel.




                                                  45
     Figure 6: Change in Factor Taxation vs. Change in Trade, by Income Level

          (a) ETR K: High-Income                           (b) ETR K: Low & Middle-Income




          (c) ETR L: High-Income                            (d) ETR L: Low & Middle-Income




Notes: These figures show the association between changes in trade and changes in effective tax rates of
capital (panels a and b) and labor (panels c and d), respectively for high income OECD countries and for
low and middle income countries. Trade is measured as the sum of import and exports as a share of NDP.
Both the x-axis and y-axis are measured as within-country percent changes over 5 years. Each graph
plots binned scatter plots of the outcome against trade, after residualizing all variables against year fixed
effects. Each dot corresponds to a ventile (20 equal-sized bins) of the residualized trade variable. For
ease of interpretation, we add back the (non-residualized) mean of the given variable. Linear trend lines
are unweighted, with year fixed effects, and are estimated based on the underlying country-year panel
data. The corresponding slope and standard error are shown top-left in each panel.




                                                  46
                                                                 Figure 7: Event Study of Trade Liberalization Reforms




                                                                                                                                           .4
                                                       Treated Countries                                                                                Joint significance of event dummies:
                                                       Synthetic Control                                                                                 4.52
                                      .6
                                                                                                                                                        (.006)




                                                                                                                                           .3
                                                                                                                  Coefficient on Trade
      Trade (% of NDP)
                                      .5




                                                                                                                                           .2
                                                                                                                                           .1
                                      .4




                                                                                                                                           0
                                      .3




                                                                                                                                           −.1
                                      .2




                                                                                                                                           −.2
                                            −10   −8      −6     −4        −2   0   2       4   6   8   10                                        −10    −8        −6       −4       −2        0   2   4   6   8   10
                                                                      Years to/from Event                                                                                        Years to/from Event




                                                                                                                                           .12
                                                       Treated Countries
                                                                                                                                                        Joint significance of event dummies:
                                                       Synthetic Control                                                                                 6.6
                                                                                                                                                        (.002)
                                      .38




                                                                                                                                           .08
                                                                                                                  Coefficient on K−share
      Capital Share




                                                                                                                                           .04
                                      .34




                                                                                                                                           0
                                      .3




                                                                                                                                           −.04
                                      .26




                                            −10   −8      −6     −4        −2   0   2       4   6   8   10                                 −.08   −10    −8        −6       −4       −2        0   2   4   6   8   10
                                                                      Years to/from Event                                                                                        Years to/from Event
                                                                                                                                           .12




                                                       Treated Countries
                                                       Synthetic Control                                                                                Joint significance of event dummies:
                                                                                                                                                         2.5
                                      .16




                                                                                                                                                        (.059)
      Effective Tax Rate on Capital




                                                                                                                                           .08
                                                                                                                  Coefficient on ETR_K
                                                                                                                                           .04
                                      .12




                                                                                                                                           0
                                      .08




                                                                                                                                           −.04
                                                                                                                                           −.08
                                      .04




                                            −10   −8      −6     −4        −2   0   2       4   6   8   10                                        −10    −8        −6       −4       −2        0   2   4   6   8   10
                                                                      Years to/from Event                                                                                        Years to/from Event
                                                                                                                                           .12




                                                       Treated Countries
                                                                                                                                                        Joint significance of event dummies:
                                                       Synthetic Control                                                                                 4.66
                                      .16




                                                                                                                                                        (.005)
      Effective Tax Rate on Labor




                                                                                                                                           .08
                                                                                                                  Coefficient on ETR_L
                                                                                                                                           .04
                                      .12




                                                                                                                                           0
                                      .08




                                                                                                                                           −.04
                                                                                                                                           −.08
                                      .04




                                            −10   −8      −6     −4        −2   0   2       4   6   8   10                                        −10    −8        −6       −4       −2        0   2   4   6   8   10
                                                                      Years to/from Event                                                                                        Years to/from Event


Notes: These figures show event-studies for trade liberalization in seven large developing countries: Argentina,
Brazil, China, Colombia, India, Mexico and Vietnam. The panels correspond to different outcomes: trade;
capital share; effective tax rate on capital, and on labor. The left-hand graphs show the average level of
the outcome in every year to (since) the event for the treated group and for the group of synthetic control
countries. The right-hand graphs show the coefficients on the ‘to’ (‘since’) dummies, in a regression with
country fixed effects, year ‘to’ (‘since’) fixed effects, and calendar year fixed effects. The bars represent the
95% confidence intervals. Standard errors are clustered at the country-reform level and estimated with the
wild bootstrap method. The top-left corners report the F-statistic on joint significance of the post-reform
dummies, with the p-value in parentheses below. Details on methodology in Section 6.1 and Appendix C.2.


                                                                                                             47
                                                                                    Figure 8: Event Study of Capital liberalization Reforms




                                                                                                                                      Coefficient on Log capital openness (% of NDP)
                                                          13




                                                                                                                                                                                       .8
                                                                           Treated Countries
                                                                                                                                                                                              Joint significance of year−since event dummies:
                                                                           Synthetic Control
      Log of capital openness (% of NDP)
                                                                                                                                                                                              2.24
                                                                                                                                                                                              (.029)




                                                                                                                                                                                       .6
                                                          12




                                                                                                                                                                                       .4
                                                                                                                                                                                       .2
                                                          11




                                                                                                                                                                                       0
                                                                                                                                                                                       −.2
                                                          10




                                                                −10   −8      −6     −4        −2   0   2       4   6   8   10                                                                −10     −8       −6       −4        −2       0    2   4   6   8   10
                                                                                          Years to/from Event                                                                                                                 Years to/from Event
      Capital share of national domestic product (in %)
                                                          .42




                                                                                                                                                                                       .12
                                                                           Treated Countries
                                                                                                                                                                                              Joint significance of year−since event dummies:
                                                                           Synthetic Control                                                                                                  2.12
                                                                                                                                                                                              (.038)




                                                                                                                                                                                       .08
                                                                                                                                      Coefficient on Ksh_ndp
                                                          .4




                                                                                                                                                                                       .04
                                                          .38




                                                                                                                                                                                       0
                                                          .36




                                                                                                                                                                                       −.04
                                                                                                                                                                                       −.08
                                                          .34




                                                                −10   −8      −6     −4        −2   0   2       4   6   8   10                                                                −10     −8       −6       −4        −2       0    2   4   6   8   10
                                                                                          Years to/from Event                                                                                                                 Years to/from Event
                                                                                                                                                                                       .12
                                                          .2




                                                                           Treated Countries                                                                                                  Joint significance of year−since event dummies:
                                                                           Synthetic Control                                                                                                  1.37
      Effective tax rate on capital (in %)




                                                                                                                                                                                              (.222)
                                                                                                                                                                                       .08
                                                                                                                                      Coefficient on ETR_K
                                                          .16




                                                                                                                                                                                       .04
                                                                                                                                                                                       0
                                                                                                                                                                                       −.04
                                                          .12




                                                                                                                                                                                       −.08
                                                                                                                                                                                       −.12
                                                          .08




                                                                −10   −8      −6     −4        −2   0   2       4   6   8   10                                                                −10     −8       −6       −4        −2       0    2   4   6   8   10
                                                                                          Years to/from Event                                                                                                                 Years to/from Event
                                                                                                                                                                                       .12
                                                          .2




                                                                           Treated Countries                                                                                                  Joint significance of year−since event dummies:
                                                                           Synthetic Control                                                                                                  1.63
                                                                                                                                                                                              (.122)
      Effective tax rate on labor (in %)




                                                                                                                                                                                       .08
                                                                                                                                      Coefficient on ETR_L
                                                          .16




                                                                                                                                                                                       .04
                                                                                                                                                                                       0
                                                                                                                                                                                       −.04
                                                          .12




                                                                                                                                                                                       −.08
                                                                                                                                                                                       −.12
                                                          .08




                                                                −10   −8      −6     −4        −2   0   2       4   6   8   10                                                                −10     −8       −6       −4        −2       0    2   4   6   8   10
                                                                                          Years to/from Event                                                                                                                 Years to/from Event


Notes: These figures show event-studies for trade capital reforms in the 25 developing countries of Chari,
Henry, and Sasson (2012). The panels correspond to different outcomes: capital openness; capital share;
effective tax rate on capital, and on labor. Capital openness is the log of total foreign assets and liabilities as a
percentage of GDP. The left-hand graphs show the average level of the outcome in every year to (since) the
event for the treated group and for the group of synthetic control countries. The right-hand graphs show the
coefficients on the ‘to’ (‘since’) dummies, in a regression with country fixed effects, year ‘to’ (‘since’) fixed
effects, and calendar year fixed effects. The bars represent the 95% confidence intervals. Standard errors are
clustered at the country-reform level and estimated with the wild bootstrap method. The top-left corners
report the F-statistic on joint significance of the post-reform dummies, with the p-value in parentheses below.
More details are in Section 8 and Appendix C.3.

                                                                                                                                 48
                 Table 1: Trade Impacts on Factor Shares and Factor Taxation

                                     Capital Share                         Effective Tax Rate
                                 overall     corp. sector          labor      capital     corp. profits
          Panel A: OLS

    Trade                       0.0195*         0.0217           0.0246**      0.0168         0.0120
                                (0.0109)       (0.0148)          (0.0101)     (0.0302)       (0.0220)

          Panel B: IV estimate (NDP-weighted)

    Trade                        0.151**        0.184**          0.163***      0.375*        0.342***
                                (0.0698)       (0.0800)          (0.0538)      (0.213)        (0.121)
    First-stage F-statistic       26.07          26.07            26.07         26.07          26.07

          Panel C: IV estimate (unweighted)

    Trade                        0.118*         0.122             0.133**     0.250**        0.359***
                                (0.0681)       (0.0826)          (0.0526)     (0.105)        (0.0870)
    First-stage F-statistic       8.42           8.42               8.42        8.42           8.42

          Panel D: IV estimate (NDP-weighted, with controls)

    Trade                        0.115**        0.142**          0.226***     0.400***        0.205*
                                (0.0475)       (0.0546)          (0.0551)      (0.112)        (0.129)
    First-stage F-statistic       19.02          19.02            19.02         19.02          19.02

               N                  4518           4518              4518         4518           4518
Notes: This table presents results from estimating the effect of trade on factor shares and factor taxation. In
Panel A, we present results from estimating equation (7) using OLS, while Panels B and C and D present
IV estimates—weighted by National Domestic Product (NDP); unweighted; and weighted with controls,
respectively. Across columns, the outcome is the capital share of national domestic product and within the
corporate sector, and the effective tax rate on labor, capital and corporate profits. Trade is measured as the
sum of export and imports divided by NDP. IV estimates in panels B,C,D instrument for trade using the
oil-price and the gravity-instruments from Egger, Nigai, and Strecker (2019). All estimates include country
and year fixed effects and observations are weighted by net domestic product in constant 2019 USD at PPP
(except in Panel C). The controls included in Panel C are: USD exchange rate; gross fixed capital formation
(as a percentage of NDP); (log) population; (log) GDP per capita; and de jure capital accounts mobility. For
more details, see Section 7. * p<0.10 ** p<0.05 *** p<0.01. Standard errors in parentheses are clustered at the
country level.




                                                     49
                         Table 2: Trade Impacts on Additional Outcomes

                                               OLS                               IV
                                                             Weighted       Unweighted        Controls
                                                (1)            (2)             (3)              (4)
          Panel A: Statutory Corporate Income Tax Rate

    CIT rate                                   -0.002         -0.064***        -0.051*        -0.061***
                                              (0.003)          (0.017)         (0.028)         (0.017)

          Panel B: Self-Employment as a Share of the Workforce

    Self-employment                           -0.0117          -0.220*        -0.185***       -0.174***
                                             (0.0145)          (0.126)         (0.0460)        (0.0560)

          Panel C: National Income Components

    Corporate profits                        0.0339***          0.175**        0.124***        0.206***
                                             (0.0128)         (0.0767)        (0.0321)        (0.0726)

    Employee compensation                    0.00848           -0.0749        -0.0964          0.0485
                                             (0.0175)         (0.0904)        (0.0669)        (0.0785)

    Mixed income                              -0.0231          -0.0685        -0.0391         -0.202**
                                             (0.0182)          (0.105)        (0.0301)        (0.0816)

    Household operating surplus               0.0002           0.0145          0.0072          0.0171
                                             (0.0039)         (0.0146)        (0.0159)        (0.0137)

    Corporate-sector value-added             0.0396*            0.164         0.0917***       0.274***
                                             (0.0210)          (0.109)         (0.0342)       (0.0943)

    First-stage F-statistic                                     26.07           19.02           8.415
                   N                           4518             4518            4518            4518
Notes: This table reports estimates of the impact of trade on additional outcomes. Each cell corresponds to a
coefficient on trade from a regression model which varies in the outcome (across rows) and estimation model
(across columns). Across columns, the coefficients are based on estimating equation 7 using, respectively:
OLS; IV; IV without weights, IV with weights and controls. The controls included in column (4) are: USD
exchange rate; gross fixed capital formation (as a percentage of NDP); (log) population; (log) gross domestic
product per capita; and de jure capital accounts mobility. Weighted regressions are weighted by annual net
domestic product in constant 2019 USD at PPP. Across panels, the outcome is: the statutory corporate income
tax rate (Panel A); the self-employed share of the active workforce (Panel B); the share in national income of
corporate profits, employee compensation, mixed income, household operating surplus, and the share of the
corporate sector in the economy (Panel C). Trade is measured as the sum of export and imports divided by
net domestic product. We instrument for trade using the oil-price and the gravity-instruments from Egger,
Nigai, and Strecker (2019). All estimates include country and year fixed effects. For more details, see Section
                                                       50
7. * p<0.10 ** p<0.05 *** p<0.01. Standard errors in parentheses are clustered at the country level.
               Table 3: Heterogeneous Impacts of Trade by Development Level

                       ET RK      ET RL       CIT           Capital Self         Mixed      CorporateCorporate
                                              rate          share    em-          in-       profits   value-
                                                            of in-  ploy-        come                added
                                                            come    ment
                          (1)        (2)         (3)          (4)     (5)           (6)        (7)         (8)

 Trade                 0.444**     0.145      -0.043*       0.182**   -0.252** -0.124       0.219***    0.220***
                       (0.181)    (0.093)     (0.024)       (0.077)    (0.107) (0.097)       (0.063)     (0.083)
 Trade∗1(High-inc.)     -0.441     0.120       -0.032        -0.219     0.232  0.374*       -0.299**    -0.381**
                       (0.347)    (0.194)     (0.047)       (0.137)    (0.209) (0.205)       (0.146)     (0.176)
 Implied coef. for      0.003     0.265**    -0.075***       -0.036    -0.021    0.250*       -0.080      -0.160
 Trade∗1(High-inc.)    (0.231)    (0.122)     (0.457)       (0.083)   (0.151)    (0.144)     (0.102)     (0.129)

          N             4518        4518        3810         4518      4518        4518       4518        4518
Notes: This table presents results from the heterogeneous IV analysis based on estimating equation (8). The
top row denotes the outcome variable: effective tax rate on capital; effective tax rate on labor; statutory
corporate income tax rate; capital share of domestic product; self-employed share of workforce; mixed
income as a share of domestic product; corporate profits; and share of the corporate sector in the economy.
The regression coefficients for T rade as well as the interaction with a dummy for high-income countries,
T rade ∗ 1(High − income) are presented. The bottom row reports the coefficient for the linear combination
of T rade and the interaction term. Trade is measured as the sum of export and imports divided by net
domestic product. We instrument for trade using the oil-price and the gravity-instruments from Egger, Nigai,
and Strecker (2019). All estimates include country and year fixed effects and observations are weighted by
net domestic product in constant 2019 USD at PPP. For more details, see Section 7. * p<0.10 ** p<0.05 ***
p<0.01. Standard errors in parentheses are clustered at the country level.




                                                       51
                      Table 4: Additional Heterogeneity Impacts of Trade

                        Heterogeneity Hc :               Small             Capital
                                                       population         openness
                        Panel A: CIT rate
                    Trade                                -0.053***        -0.063***
                                                          (0.014)          (0.018)
                    Trade∗Hc                               -0.034           -0.034
                                                          (0.054)          (0.079)

                    Coefficient on Trade in Hc             -0.088*           -0.094
                                                         (0.049)           (0.072)
                         Panel B: ET RK
                    Trade                                0.357**           0.617**
                                                         (0.177)           (0.274)
                    Trade∗Hc                             -0.491            -0.483
                                                         (0.544)           (0.456)

                    Coefficient on Trade in Hc              -0.134            0.133
                                                          (0.456)          (0.224)
                         Panel C: ET RL
                    Trade                                0.169***           0.144
                                                          (0.061)          (0.158)
                    Trade∗Hc                               0.145            0.159
                                                          (0.282)          (0.275)

                    Coefficient on Trade in Hc               0.314           0.304**
                                                          (0.242)          (0.139)
Notes: This table presents results from the heterogeneous IV analysis based on estimating equation (8). The
top row denotes the source of heterogeneity Hc , respectively across columns: a dummy for small population
size (below 40 million); the Chinn-Ito index of capital account openness (Chinn and Ito, 2006), which is a
continuous variable between 0 and 1; and, a dummy indicator for the post-1995 period. Across Panels, we
estimate the effects of trade on the statutory corporate income tax rate (Panel A), the effective tax rate on
capital (Panel B), and the effective tax rate of labor (Panel C). At the bottom of each panel, we report the
coefficient on trade (and standard error) in the heterogeneity sub-sample as the linear combination of the
coefficients on T rade and T rade ∗ Hc . Trade is measured as the sum of export and imports divided by net
domestic product. We instrument for trade using the oil-price and the gravity-instruments from Egger, Nigai,
and Strecker (2019). All estimates include country and year fixed effects and observations are weighted by
net domestic product in constant 2019 USD at PPP. For more details, see Section 7. * p<0.10 ** p<0.05 ***
p<0.01. Standard errors in parentheses are clustered at the country level.




                                                    52
         Table 5: Trade Impacts by Tax Source (% of GDP) and Development Levels

                               total taxes    indirect        CIT       property        PIT      payroll
                                   (1)          (2)           (3)          (4)          (5)        (6)

     Trade                       0.218*         0.002      0.102***       0.025        0.010       0.055
                                 (0.112)       (0.047)      (0.028)      (0.025)      (0.025)     (0.045)

     Trade∗1(High-Inc)           -0.270         -0.146     -0.128**       -0.012       -0.061      0.090
                                 (0.251)       (0.132)      (0.061)      (0.042)      (0.062)     (0.093)



     Implied coef. for            -0.052        -0.144       -0.026       0.013       -0.0506    0.145**
     Trade in High-Inc           (0.188)       (0.105)      (0.036)      (0.026)      (0.042)    (0.056)

               N                  4518          4518         4518         4518         4518        4518
Notes: This table presents results for the impact of trade on different sources of taxation, estimated using the
IV. Across columns, the outcome is the revenue collected by each tax as a percentage of NDP: total taxes, then
indirect taxes, corporate income tax, property and wealth taxes, personal income tax, and social security taxes.
The regression coefficients for T rade as well as the interaction with a dummy for high-income countries,
T rade ∗ 1(High-income) are presented. The bottom row reports the coefficient for the linear combination
of T rade and the interaction term. Trade is measured as the sum of export and imports divided by GDP.
We instrument for trade using the oil-price and the gravity-equation instruments from Egger, Nigai, and
Strecker (2019). All estimates include country and year fixed effects and observations are weighted by net
domestic product in constant 2019 USD at PPP. We use dummy variable controls for significantly interpolated
revenue data (rare) or imputed factor share data (frequent) in all columns. * p<0.10 ** p<0.05 *** p<0.01.
Standard errors in parentheses are clustered at the country level.




                                                      53
Appendices
Appendix A             Additional Figures and Tables

                     Figure A1: Data Coverage for Effective Tax Rates




 Notes: This figure shows the coverage of our effective tax rate data between 1965 and 2018, globally and
 for high vs. low- and middle-income countries. The solid lines plot the percentage of total population
 and GDP that is covered in our data (left axis). The dashed lines show the number of countries in the data
 (right axis). The missing ‘missing’ income (and population) prior to the 1990s corresponds to communist
 countries, particularly China, Russia and the ex-Soviet republics, and Vietnam. In addition to limited
 data on public revenue, communist country present a conceptual mismatch with our framework for factor
 income taxation (see Appendix B). Other missing country-years correspond to conflicts, independence
 post 1965, and in a few cases to missing data.




                                                  54
                   Figure A2: Share of Worldwide GDP by Country




Notes: This figure plots the relative weight in world GDP of important countries. The countries are the
USA, Eurozone 19, China, Russia, the remaining high-income countries and the remaining low and
middle-income countries (less China and Russia). By construction these lines add to 100% of world GDP
in each year. The dotted lines for China and Russia highlight that we do not have data on effective tax
rates for these countries pre-1994, as discussed in Appendix B and Figure A1, but we know for every
year their contribution to world GDP.




                                               55
            Figure A3: Tax Revenue in Low- & Middle-Income Countries




Notes: This figure plots the time series of tax revenue as a share of net domestic product (NDP),
separated by revenue source, for the 119 low- and middle-income countries (sample countries that are
not high-income OECD countries), weighting country-year observations by their share in that year’s
total NDP, in constant 2019 USD. The top left-panel repeats the benchmark figure for low and middle
income countries. The top-right panel excludes former communist countries, most notably China and
Russia. The mid-left panel excludes oil-rich countries (the 33 countries where average oil production
since 1990 has exceeded 6.5% of GDP, per Ross and Mahdavi (2015)). The mid-right panel excludes both
ex-communist and oil-rich nations.




                                               56
                Figure A4: Impact of Imputations on Effective Tax Rates




Notes: This figure plots the time series of effective tax rates for labor and capital series with and without
imputations for missing country-years. This allows us to control for the sample composition changing
over time and shows the plausible impact that having a balanced panel might have had on the time-series
of ETRs. The solid lines correspond to the ETR series as shown in the paper and are derived from
the unbalanced panel of countries, where ex-communist countries enter in 1994, including China and
Russia. The dotted lines show the results of imputing values for the missing country-years in the
pre-1994 data, as to have a fully balanced panel. We impute ETR values of the missing years based on
a simple prediction model, which extends the trends observed in the data to the missing observation.
Concretely, we decompose the ETR on capital and on labor into year and country fixed effects, separately
for high versus low and middle-income countries: ET Ri,t = βt + βi + i,t , where βt and βi are year
and country fixed effects. Each country-year is weighted by its share of worldwide income (in constant
2019 USD). We impute missing data by adding, to the first year of available data for a given country,
the difference between the year fixed-effect of a missing year and that of the first available year in the
country’s time series. For example, the imputed 1993 value of ET RK for China (whose series begins in
                              K                K            ˆ        ˆ
1994) is constructed as ET RCHN,   1993 = ET RCHN,1994 + β1993 − β1994 .




                                                 57
        Figure A5: Impact of Mixed Income Allocation on Effective Tax Rates




Notes: This figure plots the time series of effective tax rates for labor and capital varying the method
which assigns mixed income in national accounts (self-employment income) to labor vs capital. The
benchmark method assigns 70% of mixed-income to labor and 30% to capital. The alternative method
(ILO mixed income) imputes to the self-employed a labor income similar to the wage they would have
earned in an employer-employee relation, based on observable characteristics. In practice, we extend the
estimates in ILO (2019), using ILOSTAT (2021) data on self-employment shares of the workforce, to all
countries since 1991, and impute the series backwards to complete it. Further details are in section 3.1.1
and the data Appendix B.




                                                 58
  Figure A6: Impact of Personal Income Taxes Allocation on Effective Tax Rates
                                       (a) ETR on Capital




                                        (b) ETR on Labor




Notes: These figures plot the time series of effective tax rates for capital (panel a) and labor (panel b)
as a function of how we allocate personal income tax (PIT) revenue to each factor. Tax revenue from
PIT represents a mix of taxes on labor and capital, since it includes salaries, as well as capital income
and capital gains. Our baseline, consistent with US data, starts by assuming that 15% of PIT revenue
derives from capital and 85% from labor. It is then adjusted to take into account that some forms of
capital income face reduced rates compared to labor income, and that in some countries the PIT only
covers the very top of the income distribution (see Section 3.1.2). We then show two extreme bound
scenarios: as an upper bound for the ETR on capital, we assign 30% of PIT tax revenue to capital and
70% to labor. As a lower bound we assign 0% to capital and 100% to labor.

                                                59
            Figure A7: Effective Tax Rates in Large Developing Countries




Notes: This figure shows the evolution of effective tax rates on labor and capital for the 17 largest low
and middle income countries. Countries are displayed when they rank in the top 20 both in terms of
population and GDP, in 2018.




                                                60
 Figure A8: Trends in statutory corporate income tax rates by development level




Notes: This figure plot the time series of statutory corporate income tax rates by level of development:
the light blue X-line is for middle and low income countries, while the dark blue circle-line is for high
income countries. Each line plots the year fixed effects from an unweighted OLS regression, in the
relevant sub-sample of countries, on country and year fixed effects. The inclusion of country fixed effects
eliminates the influence of countries entering and leaving the sample. The fixed effects are normalized to
equal the level of the outcome variable in the relevant sub-sample in 1965.




                                                61
Figure A9: Trends in Factor Taxation by Initial Trade Openness, in Developing Countries




   Notes: This figure plot the time series of trade openness (top-left panel), average effective tax rates on
   capital (top-right panel) and labor (bottom-left panel). The sample is limited to low- and middle-income
   countries. Within each panel, the green line (orange line) traces the evolution of the group of developing
   countries which had relatively high (low) trade openness prior to 1995. Specifically, high (low) trade
   openness is defined as having average yearly trade openness which lies above (below) the global yearly
   average between 1965 and 1995. Trade openness is measured as the share of imports and exports in
   national domestic product; note that this share can exceed a value of 1.Each line plots the year fixed
   effects from an unweighted OLS regression, in the relevant sub-sample of the outcome, on country and
   year fixed effects. The inclusion of country fixed effects eliminates the influence of countries entering
   and leaving the sample. The fixed effects are normalized to equal the level of the outcome variable in the
   relevant sub-sample in 1965. The shaded area highlights the notable 1990-1995 period, which marks the
   beginning of the ‘second wave’ of globalization, featuring a proliferation of bilateral and multilateral
   trade agreements (Egger, Nigai, and Strecker, 2019).




                                                    62
Figure A10: Event Studies of Trade Liberalization, Simultaneous Matching on Outcomes




                                                                                                                                             .4
                                                         Treated Countries                                                                                Joint significance of event dummies:
                                                         Synthetic Control                                                                                 6.38
                                        .6
                                                                                                                                                          (.001)




                                                                                                                                             .3
                                                                                                                    Coefficient on Trade
        Trade (% of NDP)
                                        .5




                                                                                                                                             .2
                                                                                                                                             .1
                                        .4




                                                                                                                                             0
                                        .3




                                                                                                                                             −.1
                                        .2




                                                                                                                                             −.2
                                              −10   −8      −6     −4        −2   0   2       4   6   8   10                                        −10      −8       −6       −4        −2      0   2     4   6   8   10
                                                                        Years to/from Event                                                                                          Years to/from Event




                                                                                                                                             .12
                                                         Treated Countries
                                                                                                                                                          Joint significance of event dummies:
                                                         Synthetic Control                                                                                 29.18
                                                                                                                                                          (0.001)
                                        .38




                                                                                                                                             .08
                                                                                                                    Coefficient on K−share
        Capital Share




                                                                                                                                             .04
                                        .34




                                                                                                                                             0
                                        .3




                                                                                                                                             −.04
                                        .26




                                                                                                                                             −.08
                                              −10   −8      −6     −4        −2   0   2       4   6   8   10                                        −10      −8       −6       −4        −2      0   2     4   6   8   10
                                                                        Years to/from Event                                                                                          Years to/from Event




                                                                                                                                             .12
                                                         Treated Countries
                                                                                                                                                          Joint significance of event dummies:
                                                         Synthetic Control                                                                                 1.40
                                        .16




                                                                                                                                                          (.468)
        Effective Tax Rate on Capital




                                                                                                                                             .08
                                                                                                                    Coefficient on ETR_K
                                                                                                                                             .04
                                        .12




                                                                                                                                             0
                                        .08




                                                                                                                                             −.04
                                                                                                                                             −.08
                                        .04




                                              −10   −8      −6     −4        −2   0   2       4   6   8   10                                        −10      −8       −6       −4        −2      0   2     4   6   8   10
                                                                        Years to/from Event                                                                                          Years to/from Event
                                                                                                                                             .12




                                                         Treated Countries
                                                                                                                                                          Joint significance of event dummies:
                                                         Synthetic Control                                                                                 26.58
                                        .16




                                                                                                                                                          (0.001)
        Effective Tax Rate on Labor




                                                                                                                                             .08
                                                                                                                    Coefficient on ETR_L
                                                                                                                                             .04
                                        .12




                                                                                                                                             0
                                        .08




                                                                                                                                             −.04
                                                                                                                                             −.08
                                        .04




                                              −10   −8      −6     −4        −2   0   2       4   6   8   10                                        −10      −8       −6       −4        −2      0   2     4   6   8   10
                                                                        Years to/from Event                                                                                          Years to/from Event


  Notes: These figures show event-studies for trade liberalization reforms in seven countries, over four
  outcomes: trade (as a percentage of domestic product); capital shares; and effective tax rates (on capital
  and labor). The left-hand graphs show the average level of the outcome in every year relative to the event,
  for the treated group and for the group of synthetic controls. The right-hand graphs show the coefficients
  on the ‘to’ (‘since’) dummies, in a regression model with country fixed effects; year ‘to’ (‘since’) fixed
  effects; and calendar-year fixed effects. The bars represent the 95% confidence intervals for ‘to’ (‘since’)
  reform coefficients, while standard errors are clustered at the country-reform level and estimated using
  the wild bootstrap method. In the top-left corner, we report the F-statistic on joint significance of the
  post-reform dummies, with the p-value in parentheses below. These graphs are constructed similar
  to Figure 7, with the exception that the synthetic control for each event-country is based on matching
  simultaneously on all outcomes.




                                                                                                               63
Figure A11: Robustness of Trade Liberalization to Changing Sample of Event Countries
                                               (a) Trade as % of NDP                                                                                             (b) Capital Share of NDP




                                                                                                                              .06
                 .15




                                                                                                         Capital share of natonal domestic product
                                                                                                                                        .04
                  .1
 Trade (% of NDP)




                                                                                                                            .02
       .05




                                                                                                                 0
                 0
                 −.05




                                                                                                                              −.02
                                   −10    −8   −6   −4    −2      0      2     4   6   8   10                                                        −10    −8    −6   −4    −2      0      2     4   6   8   10
                                                         Years to/from Event                                                                                                Years to/from Event


                                         (c) Effective Tax Rate on Capital                                                                                  (d) Effective Tax Rate on Labor
                                                                                                                        .07
                 .07
   Effective Tax Rate on Capital




                                                                                                       Effective Tax Rate on Labor
                                                                                                                           .05
                        .05




                                                                                                                 .03
             .03




                                                                                                     .01
 .01             −.01




                                                                                                                        −.01




                                   −10    −8   −6   −4    −2      0      2     4   6   8   10                                                        −10    −8    −6   −4    −2      0      2     4   6   8   10
                                                         Years to/from Event                                                                                                Years to/from Event


         Notes: These figures show event studies for trade liberalization reforms in seven countries, over four
         outcomes: trade (as a percentage of domestic product); capital share (as a percentage of domestic
         product); and effective tax rates on capital and labor. In each figure, the solid green line displays the
         estimated coefficients for the interaction between a treatment dummy and a year ‘to’ (‘since’) dummy
         [note the omitted period is t − 1], corresponding to the graphs displayed in the right column of Figure 7.
         Each lightly-shaded gray line repeats the estimation procedure based on a sample that removes one of
         the seven treated countries, one at a time. All the gray lines thus represent the dynamic treatment effects
         but for different subsets of the treated countries. More details can be found in Appendix C.2.




                                                                                                64
         Figure A12: Strength of Individual Instruments Across Subsamples
                               (a) Sub-samples of NDP per capita




                                 (b) Sub-samples of time-periods




Notes: These figures show the individual statistical strength of the two instruments, denoted ’oil-distance’
and ’gravity’. The outcome is the first-stage F-statistic from a regression of trade openness on the
individual instruments (see Section 7). The outcome is shown across subsamples of log GDP per capita
(Panel A) and years (Panel B). To construct each figure, the x-axis is first partitioned into ten deciles (ten
bins of equal size). The first-stage F-statistic is then separately estimated in samples centered on each
decile. The estimation is done in increments of one decile, and the bandwidth uses one decile of data on
either side of the decile-center. To maintain an equal size in all estimation samples, estimation centered
on the first and the tenth decile are therefore dropped. The value on the x-axis is the average value of the
partitioning variable in each estimation sample.

                                                  65
                            Figure A13: Event Study of Trade Liberalization – Mechanism Outcomes

                     .1




                                                                                                                                .1
                                       Treated Countries
                                       Synthetic Control                                                                                     Joint significance of event dummies:
                                                                                                                                              5.55
                                                                                                                                             (.002)




                                                                                                  Coefficient on CIT change
                     .05




                                                                                                                                .05
      CIT change
                     0




                                                                                                                                0
                     −.05




                                                                                                                                −.05
                     −.1




                                                                                                                                −.1
                            −10   −8      −6     −4        −2   0   2       4   6   8   10                                             −10      −8       −6       −4        −2      0   2     4   6   8   10
                                                      Years to/from Event                                                                                               Years to/from Event
                     .65




                                                                                                                                .1
                                       Treated Countries
                                       Synthetic Control                                                                                     Joint significance of event dummies:
                                                                                                                                              14.11
                                                                                                                                             (0)




                                                                                                  Coefficient on selfemployed
                                                                                                                                .05
                     .6
      Selfemployed
                     .55




                                                                                                                                0
                                                                                                                                −.05
                     .5




                                                                                                                                −.1
                     .45




                            −10   −8      −6     −4        −2   0   2       4   6   8   10                                             −10      −8       −6       −4        −2      0   2     4   6   8   10
                                                      Years to/from Event                                                                                               Years to/from Event
                     .3




                                                                                                                                .1




                                       Treated Countries
                                       Synthetic Control                                                                                     Joint significance of event dummies:
                                                                                                                                              8.33
                                                                                                                                             (0)
                     .25




                                                                                                                                .05
                                                                                                  Coefficient on os_corp
      os_corp
                     .2




                                                                                                                                0
                                                                                                                                −.05
                     .15




                                                                                                                                −.1
                     .1




                            −10   −8      −6     −4        −2   0   2       4   6   8   10                                             −10      −8       −6       −4        −2      0   2     4   6   8   10
                                                      Years to/from Event                                                                                               Years to/from Event
                     .25




                                                                                                                                .1




                                       Treated Countries
                                       Synthetic Control                                                                                     Joint significance of event dummies:
                                                                                                                                              4.23
                                                                                                                                             (.008)
                                                                                                                                .05
                                                                                                  Coefficient on mi_hh
      mi_hh
                     .2




                                                                                                                                0
                                                                                                                                −.05
                                                                                                                                −.1
                     .15




                            −10   −8      −6     −4        −2   0   2       4   6   8   10                                             −10      −8       −6       −4        −2      0   2     4   6   8   10
                                                      Years to/from Event                                                                                               Years to/from Event


Notes: These figures show event-studies for trade liberalization in seven large developing countries: Argentina,
Brazil, China, Colombia, India, Mexico and Vietnam. The panels correspond to different outcomes: corporate
income tax rate; self-employed share of workforce; corporate profits share of national income; household
mixed income share of national income. The left-hand graphs show the average level of the outcome in
every year to (since) the event for the treated group and for the group of synthetic control countries. The
right-hand graphs show the coefficients on the ‘to’ (‘since’) dummies, in a regression with country fixed
effects, year ‘to’ (‘since’) fixed effects, and calendar year fixed effects. The bars represent the 95% confidence
intervals. Standard errors are clustered at the country-reform level and estimated with the wild bootstrap
method. The top-left corners report the F-statistic on joint significance of the post-reform dummies, with the
p-value in parentheses below. Details on methodology in Section 6.1 and Appendix C.2.

                                                                                             66
Figure A14: Event Study of Capital Liberalization – Simultaneous Matching on Outcomes




                                                                                                                                         Coefficient on Log capital openness (% of NDP)
                                                             13




                                                                                                                                                                                          .8
                                                                              Treated Countries
                                                                                                                                                                                                 Joint significance of year−since event dummies:
                                                                              Synthetic Control
         Log of capital openness (% of NDP)


                                                                                                                                                                                                 1.31
                                                                                                                                                                                                 (.248)




                                                                                                                                                                                          .6
                                                             12




                                                                                                                                                                                          .4
                                                                                                                                                                                          .2
                                                             11




                                                                                                                                                                                          0
                                                             10




                                                                                                                                                                                          −.2
                                                                   −10   −8      −6     −4        −2   0   2       4   6   8   10                                                                −10     −8       −6       −4        −2       0    2   4   6   8   10
                                                                                             Years to/from Event                                                                                                                 Years to/from Event
         Capital share of national domestic product (in %)
                                                             .42




                                                                                                                                                                                          .12
                                                                              Treated Countries
                                                                                                                                                                                                 Joint significance of year−since event dummies:
                                                                              Synthetic Control                                                                                                  .99
                                                                                                                                                                                                 (.469)




                                                                                                                                                                                          .08
                                                             .4




                                                                                                                                         Coefficient on Ksh_ndp
                                                                                                                                                                                          .04
                                                             .38




                                                                                                                                                                                          0
                                                             .36




                                                                                                                                                                                          −.04
                                                             .34




                                                                                                                                                                                          −.08
                                                                   −10   −8      −6     −4        −2   0   2       4   6   8   10                                                                −10     −8       −6       −4        −2       0    2   4   6   8   10
                                                                                             Years to/from Event                                                                                                                 Years to/from Event




                                                                                                                                                                                          .12
                                                             .2




                                                                              Treated Countries                                                                                                  Joint significance of year−since event dummies:
                                                                              Synthetic Control                                                                                                  1.69
         Effective tax rate on capital (in %)




                                                                                                                                                                                          .08    (.106)
                                                                                                                                         Coefficient on ETR_K
                                                             .16




                                                                                                                                                                                          .04
                                                                                                                                                                                          0
                                                                                                                                                                                          −.04
                                                             .12




                                                                                                                                                                                          −.08
                                                                                                                                                                                          −.12
                                                             .08




                                                                   −10   −8      −6     −4        −2   0   2       4   6   8   10                                                                −10     −8       −6       −4        −2       0    2   4   6   8   10
                                                                                             Years to/from Event                                                                                                                 Years to/from Event
                                                                                                                                                                                          .12
                                                             .2




                                                                              Treated Countries                                                                                                  Joint significance of year−since event dummies:
                                                                              Synthetic Control                                                                                                  2.73
                                                                                                                                                                                                 (.009)
         Effective tax rate on labor (in %)




                                                                                                                                                                                          .08
                                                                                                                                         Coefficient on ETR_L
                                                             .16




                                                                                                                                                                                          .04
                                                                                                                                                                                          0
                                                                                                                                                                                          −.04
                                                             .12




                                                                                                                                                                                          −.08
                                                                                                                                                                                          −.12
                                                             .08




                                                                   −10   −8      −6     −4        −2   0   2       4   6   8   10                                                                −10     −8       −6       −4        −2       0    2   4   6   8   10
                                                                                             Years to/from Event                                                                                                                 Years to/from Event


   Notes: These figures show event-studies for capital liberalization reforms over four outcomes: capital
   openness; capital shares; and effective tax rates (on capital and labor). The left-hand graphs show the
   average level of the outcome in every year relative to the event, for the treated group and for the group
   of synthetic controls. These graphs are constructed similar to Figure 8, with the exception that the
   synthetic control for each event-country is based on matching simultaneously on the four outcomes.
   The right-hand graphs show the coefficients on the ‘to’ (‘since’) dummies, in a regression model with
   country fixed effects; year ‘to’ (‘since’) fixed effects; and calendar-year fixed effects. The bars represent the
   95% confidence intervals for ‘to’ (‘since’) reform coefficients, while standard errors are clustered at the
   country-reform level and estimated using the wild bootstrap method. In the top-left corner, we report
   the F-statistic on joint significance of the post-reform dummies, with the p-value in parentheses below.




                                                                                                                                    67
                             Figure A15: Event Study of Capital Liberalization – Mechanism Outcomes

                          .1




                                                                                                                                          .1
                                            Treated Countries
                                            Synthetic Control                                                                                          Joint significance of event dummies:
                                                                                                                                                        2.2
                                                                                                                                                       (.051)




                                                                                                       Coefficient on CIT change
                          .05




                                                                                                                                          .05
      CIT change
                          0




                                                                                                                                          0
                          −.05




                                                                                                                                          −.05
                          −.1




                                                                                                                                          −.1
                                 −10   −8      −6     −4        −2   0   2       4   6   8   10                                                  −10      −8       −6       −4        −2      0   2     4   6   8   10
                                                           Years to/from Event                                                                                                    Years to/from Event
                          .65




                                                                                                                                          .1
                                            Treated Countries
                                            Synthetic Control                                                                                          Joint significance of event dummies:
                                                                                                                                                        2.51
                                                                                                                                                       (.015)




                                                                                                       Coefficient on selfemployed
                          .6




                                                                                                                                          .05
      Selfemployed
                          .55




                                                                                                                                          0
                          .5




                                                                                                                                          −.05
                          .45




                                 −10   −8      −6     −4        −2   0   2       4   6   8   10                                           −.1    −10      −8       −6       −4        −2      0   2     4   6   8   10
                                                           Years to/from Event                                                                                                    Years to/from Event
                          .3




                                                                                                                                          .1




                                            Treated Countries
                                            Synthetic Control                                                                                          Joint significance of event dummies:
                                                                                                                                                        1.17
                                                                                                       Coefficient on Corporate profits




                                                                                                                                                       (.331)
                          .25




                                                                                                                                          .05
      Corporate profits
                          .2




                                                                                                                                          0
                                                                                                                                          −.05
                          .15




                                                                                                                                          −.1
                          .1




                                 −10   −8      −6     −4        −2   0   2       4   6   8   10                                                  −10      −8       −6       −4        −2      0   2     4   6   8   10
                                                           Years to/from Event                                                                                                    Years to/from Event
                          .25




                                                                                                                                          .1




                                            Treated Countries
                                            Synthetic Control                                                                                          Joint significance of event dummies:
                                                                                                                                                        2.59
                                                                                                                                                       (.012)
                                                                                                       Coefficient on Mixed income
                                                                                                                                          .05
      Mixed income
                          .2




                                                                                                                                          0
                                                                                                                                          −.05
                                                                                                                                          −.1
                          .15




                                 −10   −8      −6     −4        −2   0   2       4   6   8   10                                                  −10      −8       −6       −4        −2      0   2     4   6   8   10
                                                           Years to/from Event                                                                                                    Years to/from Event


Notes: These figures show event-studies for trade capital reforms in the 25 developing countries of Chari,
Henry, and Sasson (2012). The panels correspond to different outcomes: corporate income tax rate; self-
employed share of workforce; corporate profits share of national income; household mixed income share of
national income. The left-hand graphs show the average level of the outcome in every year to (since) the
event for the treated group and for the group of synthetic control countries. The right-hand graphs show the
coefficients on the ‘to’ (‘since’) dummies, in a regression with country fixed effects, year ‘to’ (‘since’) fixed
effects, and calendar year fixed effects. The bars represent the 95% confidence intervals. Standard errors are
clustered at the country-reform level and estimated with the wild bootstrap method. The top-left corners
report the F-statistic on joint significance of the post-reform dummies, with the p-value in parentheses below.
More details are in Section 8 and Appendix C.3.

                                                                                                  68
 Table A1: Allocation of Taxes to Factor Incomes, by Type of Tax (Benchmark Estimates)

                    Type of tax                           Series        Allocation to labor (λτ )

     Panel A: Direct Taxes
 Personal income tax (PIT)                                 1100 Share λ [70,100%] to L (See 3.1.2)
 Corporate income tax (CIT)                                1200     0% to L ⇒ all CIT to K
   Other (unallocable) income tax                          1300     50% to L, 50% to K; rare
 Social security & payroll taxes                           2000             100% to L
 Property & wealth taxes                                   4000  0% to L ⇒ all asset taxes to K

     Panel B: Indirect Taxes & Other Revenue
 Indirect taxes                                            5000    excluded, assumed proportional
 Other taxes                                               6000           excluded ; minor
 Non-tax revenue                                           7000           excluded; non-tax

Notes: This table shows our benchmark assignment of statutory tax incidence λτ on labor (where the
assignment to capital is 1 − λτ ), for each of the types of taxes in our modified OECD (2020) classification. For
the purposes of assigning tax incidence onto factor incomes (see 3.1), we consider here only direct taxes,
and implicitly assume that indirect taxation falls proportionally onto labor and capital factor incomes (cf.
Browning, 1978; Saez and Zucman, 2019b). We treat ‘other taxes’ similarly (these are rare and insignificant),
and ignore non-tax revenue. For income tax revenues whose provenance cannot be understood as either
personal income tax (PIT) or corporate income tax (CIT), we assign them as a 50-50% split between the two;
these ‘unallocable’ income tax revenues are rare in occurrence and small in magnitude. Taxes in the 4000
series (largely property taxes) also include wealth and financial transaction taxes.




                                                     69
              Table A2: Synthetic Difference-in-Difference of Trade Liberalization

                                                          Capital Share Capital Share
                                        Trade                                                    ETR Capital            ETR Labor
                                                           of Income    Corp. Income
Panel A
Synthetic control for each outcome separately
Post*Treat                             0.0992∗∗               0.0271∗             0.0283            0.0485∗∗              0.0199∗
                                       (0.0530)               (0.0165)           (0.0202)           (0.0230)              (0.0111)
Imputed treatment effect                0.107∗∗∗              0.0269∗∗            0.0284∗∗          0.0490∗∗∗             0.0198∗∗∗
                                       (0.0291)              (0.0115)            (0.0142)          (0.0146)              (0.00512)
Panel B
Synthetic control for all outcomes jointly
Post*Treat                             0.0994∗               0.0292∗∗∗           0.0335∗∗∗          0.0339∗                0.0127
                                       (0.0561)              (0.00697)           (0.0106)           (0.0201)             (0.00974)
Imputed treatment effect                0.109∗∗∗              0.0272∗∗∗          0.0311∗∗∗          0.0341∗∗∗             0.0121∗∗∗
                                       (0.0292)              (0.00455)          (0.00661)          (0.00647)             (0.00430)
N                                        284                    290                292                282                   290
Notes: This table shows the result from the difference in difference regression of our outcomes on interest in event countries (treated),
compared to synthetic control countries. Panel A shows the results when the synthetic control matching is done for each event-country
and outcome separately. Panel B shows the results when the synthetic control matching is done jointly on all outcomes (but still separately
for each event-country). In practice we run the following regression: Yit = β DiD ∗ 1(j ≥ 0)t ∗ Di + θt + κi + πY ear(it) + it Where, the
β DiD coefficient is the difference-in-difference estimate, representing the average treatment effect from period 0 through 10 post the trade
liberalization event. We also present an additional difference-on-difference estimate proposed by Borusyak, Jaravel, and Spiess (2021). This
estimate is imputed by first estimating country and time fixed effects, using non-treated countries as well as treated countries before their
respective event. Those unit and year specific estimates are then used to impute the treatment effect for every treated country, and the
imputed coefficient is then the average of the individual treatment effects. Due to the small sample size, we present wild bootstrap standard
errors in parentheses (Cameron, Gelbach, and Miller, 2008), except for the imputed treatment effect according to Borusyak, Jaravel, and
Spiess (2021), where we report the default standard errors produced by the Stata command did_imputation. * p<0.10 ** p<0.05 *** p<0.01.




                                                                   70
                 Table A3: Reduced Form Regressions

                  Trade        Imports       Exports        ET RK        ET RL
                   (1)           (2)           (3)           (4)          (5)

Z Gravity         0.040**      0.034***      0.023***       0.012**      0.005**
                   (.017)       (.009)       (0.007)        (0.004)      (0.002)

Z Oil−Distance   -0.012***    -0.005***      -.004***      -0.002***     -.001***
                  (0.003)      (0.001)        (0.001)       (0.000)       (0.000)

      N            4518          4518         4518            4518        4518
  Notes: This table shows additional impacts of the two instruments Z Gravity
  and Z Oil−Distance . Both the gravity and the oil-distance instrument is from
  Egger, Nigai, and Strecker (2019). For ease of comparison, in this table we use
  standardized transformations of each instrument. Across columns, we regress
  both instruments simultaneously on: trade, measured as the sum of imports
  and exports (% of GDP); imports (% of GDP); exports (% of GDP); the effective
  tax rate on capital, ET RK ; and, the effective tax rate on labor, ET RL . * p<0.10
  ** p<0.05 *** p<0.01. Standard errors in parentheses are clustered at the country
  level.




                                        71
        Table A4: Robustness of IV Results to Changing the Measurement of Trade

                                                       Capital Share                 Effective Tax Rate
                                                  overall      corp. sector        on capital     on labor

  Trade in G&S (%NDP)                             0.151**         0.184**            0.375*        0.163***
                                                 (0.0698)        (0.0800)            (0.213)       (0.0538)
     First-stage F-statistic                       26.07           26.07              26.07         26.07

  Trade in G&S (%NDP), winsorized                 0.145**         0.178**            0.366*        0.168***
                                                 (0.0666)        (0.0764)            (0.207)       (0.0515)
     First-stage F-statistic                       29.05           29.05              29.05         29.05

  Trade in G&S (log levels)                      0.0328***       0.0380**            0.0699*       0.00607
                                                  (0.0124)       (0.0156)            (0.0393)      (0.0166)
     First-stage F-statistic                       9.025           9.025              9.025         9.025

  Trade in Goods Only (%NDP)                      0.196**         0.236**            0.469*         0.165**
                                                 (0.0975)         (0.112)            (0.276)       (0.0814)
     First-stage F-statistic                       21.14           21.14              21.14          21.14

                       N                           4518            4518                4518          4518
Notes: This table shows the robustness of the main IV results to changing the measure of trade. Across
columns, the outcomes of interest are: capital share of net domestic product (NDP, column 1); capital share
within the corporate sector (2); the effective tax rate on capital income (3); and the effective tax rate on labor
income (4). Across panels, the measurement of trade varies as follows: trade in goods and services (G&S),
expressed as a percentage of NDP; then the same, but winsorized at the 95th percentile; then log of the value
of trade in G&S, expressed in constant 2019 USD; and finally trade in goods only, expressed as a percentage
of NDP. At the bottom of each panel, we report the first-stage F-statistic. All estimates include country and
year fixed effects, with errors clustered at country level. We use dummy variable controls for significantly
interpolated revenue data or imputed factor share data. * p<0.10 ** p<0.05 *** p<0.01. Standard errors in
parentheses.




                                                      72
               Table A5: Robustness to Controls & Oil-Rich*Time Fixed Effects

                                                      Capital Share                Effective Tax Rate
                                                 overall     corp. sector        on capital     on labor

   IV: without controls                           0.151**       0.184**            0.375*       0.163***
                                                 (0.0698)      (0.0800)            (0.213)      (0.0538)
     First-stage F-statistic                       26.07         26.07              26.07         26.07

   IV: with controls                              0.115**       0.142**           0.400***      0.226***
                                                 (0.0475)      (0.0546)           (0.112)       (0.0551)
     First-stage F-statistic                       19.02         19.02             19.02          19.02

   IV: with controls & oil-rich time FE          0.261**        0.275**            0.573**       0.386**
                                                 (0.121)        (0.131)            (0.232)       (0.151)
     First-stage F-statistic                      3.896          3.896              3.896         3.896

                       N                           4518          4518               4518          4518
Notes: This table shows the robustness of the main IV results to including controls and then additional time
fixed effects for countries in which oil is an important part of the domestic product, exceeding 6.5% of GDP
(Ross and Mahdavi, 2015; Mahdavi, 2020). Across columns, the outcomes of interest are: capital share of net
domestic product (NDP, column 1); capital share within the corporate sector (2); the effective tax rate on
capital income (3); and the effective tax rate on labor income (4). Across panels, the first set of results show
our benchmark specification from Table 1, then to a vector of control variables, then adding additional time
fixed effects for the oil-rich countries. The control variables are as follows: USD exchange rate; gross fixed
capital formation (as a percentage of NDP); (log) population; (log) gross domestic product per capita; and de
jure capital accounts mobility (Chinn and Ito, 2006). In these IV specifications, we instrument for trade (in
goods and services, expresses as a percentage of net domestic product) with both price-distance and gravity
instruments (as discussed in Section 7). Estimates are weighted by net domestic product in constant 2019
USD at PPP. All estimates include country and year fixed effects, with errors clustered at country level. We
use dummy variable controls for significantly interpolated revenue data or imputed factor share data. *
p<0.10 ** p<0.05 *** p<0.01. Standard errors in parentheses.




                                                     73
        Table A6: Robustness to Alternative Measurements of Effective Tax Rates

                                       OLS                           IV               IV: with controls           IV: unweighted
                                       (1)                           (2)                     (3)                        (4)
                               ET RK        ET RL          ET RK       ET RL          ET RK        ET RL          ET RK       ET RL

Mendoza et al. (1994)         -0.00287     0.0125           0.261       0.138**      0.288***     0.183***        0.155**     0.0741
                              (0.0185)    (0.00899)        (0.158)     (0.0527)      (0.0889)     (0.0457)       (0.0676)    (0.0451)



PIT = 0% on capital           0.00830     0.0285***        0.387*      0.158***      0.408***     0.222***       0.272**      0.123**
                              (0.0249)     (0.0108)        (0.198)     (0.0559)       (0.107)     (0.0565)       (0.107)     (0.0532)

PIT = 15% on capital          0.00817      0.0265**        0.390*      0.158***      0.419***     0.218***       0.267**      0.125**
                              (0.0282)     (0.0103)        (0.214)     (0.0540)       (0.116)     (0.0538)       (0.108)     (0.0515)

PIT = 30% on capital          0.00685      0.0245**        0.394*      0.157***      0.430***     0.214***       0.260**      0.126**
                              (0.0317)    (0.00979)        (0.231)     (0.0524)       (0.127)     (0.0513)       (0.111)     (0.0501)



ILO factor shares             0.00126      0.0214**        0.231*      0.194***       0.237**     0.264***        0.234*      0.0865
                              (0.0332)     (0.0103)        (0.137)     (0.0652)      (0.0990)     (0.0622)        (0.126)    (0.0591)

  First-stage F-statistic                                   26.07          26.07       19.02       19.02          8.415        8.415
           N                    4518         4518           4518           4518        4518        4518           4518         4518
  Notes: This table shows the robustness of the main results to three alternative measures of effective tax rates: (1) with the Mendoza,
  Razin, and Tesar (1994) ETR definition. (2) with the capital share of PIT revenues bounded at levels that are either unrealistically low
  (0%), at an intermediate fixed parameter of (15%), or unrealistically high (30%). Finally, with the ILO (2019) factor share method,
  which adjusts the labor share of mixed-income a based on observable characteristics of the self-employed. Specification (1) presents
  the OLS estimate. We present the benchmark IV specification in column (2), and then either add controls in (3) or remove NDP
  weights in (4). Specifications (1)-(3) are weighted by NDP in constant 2019 USD at PPP. The additional control variables in (3) include:
  USD exchange rate; gross fixed capital formation (as a % of NDP); (log) population; (log) GDP per capita; and de jure capital accounts
  mobility (Chinn and Ito, 2006). All estimates include country and year fixed effects, with errors clustered at country level. * p<0.10 **
  p<0.05 *** p<0.01. Standard errors in parentheses.
                  Table A7: Individual Instruments vs. Both Instruments

                                              First Stage                                 IV
                                     (1)           (2)           (3)          ET RK        ET RL

  Panel A: Both Instruments Together

Z Gravity and Z Oil−Distance                                                  0.250**     0.133**
                                                                              (0.105)     (0.0526)
     First-stage F-statistic                                                  8.415       8.415



  Panel B: Each Instrument Separately

Z Gravity                        0.00769** 0.00749**                          0.294*      0.134*
                                 (0.00332) (0.00332)                          (0.163)     (0.0767)
     First-stage F-statistic                                                  5.101       5.101

Z Oil−Distance                   -0.118***                   -0.111***        0.151**     0.131***
                                 (0.0333)                    (0.0319)         (0.0645)    (0.0283)
     First-stage F-statistic                                                  12.14       12.14

             N                      4518          4518          4518            4518        4518
  Notes: This table shows the main results for both instruments together (Panel A), and then for each
  instrument separately (Panel B). In column (1), however, we show OLS estimates of trade (in goods and
  services, expressed as a percentage of NDP) regressed on both instruments together. Then columns (2)
  and (3) show OLS estimates of trade on each instrument separately. In the columns to the right, we
  present IV estimates for the effective tax rate on capital income ET RK and on labor income ET RL .
  See Section 7 and Appendix D for further discussion of these instruments. For ease of interpretation,
  these estimates are unweighted (therefore, these estimates match Panel B of Table 1 as well as Table ??).
  The price-distance instrument Z Oil−Distance has been multiplied by 1010 , for ease of interpretation. All
  estimates include country and year fixed effects, with errors clustered at country level. We use dummy
  variable controls for significantly interpolated revenue data or imputed factor share data. * p<0.10 **
  p<0.05 *** p<0.01. Standard errors in parentheses.




                                                   75
        Table A8: Association between statutory CIT rate and capital tax collection

                                  ET RK                                 CIT revenue (% GDP)
                    All       High inc     Mid-low inc            All      High-inc      Mid-low inc

     CIT rate    0.201***      0.216**        0.188***          0.037**       .036*          .038*
                  (0.055)      (0.097)         (0.067)          (0.016)      (0.020)         (.022)

        N          3653         1225            2428              3653        1225            2428
Notes: This table presents results from estimating the association between the corporate statutory tax rate
and the effective tax rate on capital, ET RK , as well as the share of corporate income tax revenue in GDP. For
both outcomes, the association is estimated for all countries, then separately for high income countries, and
for middle and low-income countries. All regressions include country and year fixed effects. * p<0.10 **
p<0.05 *** p<0.01. Standard errors in parentheses are clustered at the country level.




                                                     76
                       Table A9: Impacts of trade on sectoral composition

                                                  Industry     Services     Agriculture
                                                     (1)          (2)           (3)
                 Panel A: OLS                      0.041**      -0.018          -0.007
                                                   (0.016)      (0.014)        (0.012)

                 Panel B: IV weighted              0.152**       0.089          -0.189
                                                   (0.064)      (0.102)        (0.181)
                    First-stage F-statistic         33.40        33.40          33.40

                 Panel C: IV with controls        0.094***       0.069          -0.104
                                                   (0.032)      (0.063)        (0.093)
                    First-stage F-statistic         22.86        22.86          22.86

                 Panel D: IV unweighted             0.012        0.088          -0.074
                                                   (0.095)      (0.094)        (0.075)
                    First-stage F-statistic          7.75         7.75           7.75

                               N                    3849         3849           3849
Notes: This table shows the impacts of trade on sectoral composition. Across columns, the outcome variable
is (expressed as a % of national value-added): industry (manufacturing and non-manufacturing); services;
agriculture. Across panels, we present: OLS estimates (Panel A); benchmark IV (Panel B); IV with controls
(Panel C); IV without weights (Panel D). The weights used in all panels except D are net domestic product in
constant 2019 USD at PPP. In Panel C, the additional control variables include: USD exchange rate; gross
fixed capital formation (as a percentage of NDP); net foreign direct investment; (log) population; (log)
gross domestic product per capita; and de jure capital accounts mobility (Chinn and Ito, 2006). In all IV
specifications, we instrument for trade (in goods and services, expressed as a percentage of NDP) with both
price-distance and gravity instruments (as discussed in Section 7). At the bottom of each panel, we report
the 1st stage F-statistic on the joint significance of the two instruments. All estimates include country and
year fixed effects, with errors clustered at country level. * p<0.10 ** p<0.05 *** p<0.01. Standard errors in
parentheses.




                                                    77
Appendix B                Data construction

B.1      National income data

Data sources We used two datasets to compute factor shares of net domestic product.
Both come from the United Nations Statistics Division. The first is the SNA2008 (UN
SNA, 2008) online data repository.              The second is SNA1968 archival material (UN
SNA, 1968). While the SNA2008 data has extensive coverage, we were able to extend
further its coverage using SNA1968 observations. As shown in Table B1, we nearly double
our ‘complete’ coverage observations by adding the historical data. What we describe
as ‘complete’ observations are those that include a split of basic factor income concepts,
including the compensation of employees in the household sector and operating surplus in
the corporate sector. We briefly describe these national income accounting concepts here,
and how we attribute them to labor or capital factors of production, before returning to
discuss missing data and coverage.


Income Concepts Net domestic product at factor prices is (equation (1) in text):


                               Y = CE + OSCORP + OSHH + OSP U E


We detail below the national accounting concepts

   – ‘Compensation of employees’ (CE ) refers to formal labor income. It includes wages
       and salaries, employer and employee social contributions, and all payments from
       employers in a current or deferred income stream to their employees.

   – The ‘operating surplus of corporations’ (OSCORP ) includes all corporate income after
       paying employees and expenses, and after depreciation, and can also be thought of
       as formal-sector capital income.

   – The ‘operating surplus of private unincorporated enterprises’ (OSP U E ), or mixed
       income, includes income from self-employment, unincorporated professionals, house-

      We accessed this data via the World Inequality Database (WID, 2020).


                                                    78
      hold business owners, and the owners of private unincorporated enterprises (often
      informal enterprise, in developing countries). This income usually represents a
      return to both labor and capital.

   – ‘Household operating surplus’ (OSHH ) is essentially the imputed rental income
      accruing to homeowners who live in their own home (so that a country entirely of
      homeowners would not appear to have a lower national income than a country where
      landlords earn rental income on all homes).

   All operating surpluses—of corporations, of unincorporated enterprises, and of the
household sector—are estimated net of depreciation (consumption of fixed capital, CF C ).
This explains why we obtain factor shares of Net Domestic Product instead of Gross
Domestic Product (below we provide further detail on the data on depreciation).
   These concepts are sufficient for the decomposition of factor income into labor and
capital income flows, and our definition of factor shares follows that in most of the
literature. For completeness, we explain some of the other concepts that do not figure in
our calculations. We explain here why we decided not to include them and the role they
would play if included.
   (1) net indirect taxes (indirect taxes net of subsidies, N IT ) are excluded from consideration.
In our data these comprise 8-15% of national income. These indirect taxes are assumed to
be factor-neutral, i.e., levied on the returns to capital and labor proportional to each factor’s
share in the affected production process. (2) net foreign income—the net flow of labor and
capital incomes (usually compensation of employees or corporate profits) from abroad— is
ignored. One could move from ‘net domestic product’ to ‘net national income’ as our total
income concept. In practice, this seems to be a less relevant concept for our purposes (most
countries tax income earned domestically regardless of citizenship, whereas net foreign
income is taxed only with difficulty), but, this adjustment is minor for most countries.
In national accounting concepts following the ‘income approach’ to domestic product,




                                               79
government surplus is zero by construction and thus does need to be distributed to labor or
capital.   Nonprofit institutions are included in the household sector.               (3


Missing data and imputations           When values are still missing after we have appended
the two UN data sources, we follow a simple a transparent imputation procedure. In
the description here, we focus on mixed income, as this is the concept whose values are
most frequently missing in UN SNA data—in both UN SNA (2008) and especially UN
SNA (1968)—but the same procedure applies to other income concepts whose values are
occasionally missing in UN SNA data, as well.
   We follow the imputation procedure from Blanchet and Chancel (2016), which presents
a simple and transparent method to impute consumption of fixed capital (depreciation),
used in the World Inequality Database (WID) series on national income (contrarily to us
the WID does not systematically decompose income into factor shares). To explain our
imputation of mixed income OSP U E , we present the following stylized facts:

   1. Mixed income (OSP U E ) represents a lower share of national income in rich countries.
      This is due to the outsized role of the informal economy and household enterprises
      in developing countries.

   2. Some countries have structurally high (or low) levels of mixed income as a share of
      national income. These specific long-term country effects represent either (i) a ‘real’
      economic trend (e.g., the informal economy, as above, or the role of certain industries
      that that are characterized by self-employment); or (ii) a specific tendency or practice
      in the national statistical office (e.g., in some countries many household enterprises
      are closely followed by tax authorities and therefore mapped by statisticians into
      the corporate sector; or vice-versa for countries where some quasi-corporations are
      poorly accounted for).



     Public sector enterprises that operate on the market are included in the corporate sector.
     Their employees earn labor income compensation, while the institutions themselves (especially trusts
and institutions with significant endowments) do earn imputed rent from land ownership. In national
accounts parlance, these non-governmental organizations are referred to as ‘non-profit institutions serving
households’ (emphasis added).


                                                   80
  3. Mixed income as a share of national income is persistent and path-dependent, such
     that the value in year t will be closely correlated to the value in year t + 1. Exogenous
     shocks that affect OSP U E without similarly affecting total income per capita are rare,
     so its decrease is a slow process as income per capita increases.

Therefore, we can model OSP U E as a function of log national income per capita (at PPP),
with a random effect to capture constant country characteristics:


                            OSP U Eit = β0 + β1 N N Ipcit + ui + εit


where there is a random effect term u for each country i, and ε is the error term for each
country-year it. To account for persistence in OSP U E we model the error term εit as an
AR(1) process:
                                       εit = ρεi,t−1 + ηit

where ηit is i.i.d. white noise. As in Blanchet and Chancel (2016), when we know part
of the OSP U E series for a given country (observing it in later years), we can estimate the
country’s random effect ui and use that in the imputation. When no later value of OSP U E
is observed, we assume ui = 0. OSP U E returns to its expected long-run value at a rate of ρt .
   While mixed income OSP U E is the most frequently missing national income component,
others are occasionally missing, as well. When additional national income components
are missing, we run the same imputation procedure for those. Therefore, we are able to
generate an estimate of factor shares for every country-year in which we have data on
total-economy net national income per capita. The global coverage of factor share data, by
source and including imputed values, is shown in Table B1.


Mixed Income Adjustment As a robustness check (discussed above in Section 3.1, with
results in Appendix Table A6), we implement the method developed in Guerriero (2019)
and ILO (2019), to adjust for the share of labor income in mixed income. For each of
the three classes of self-employed workers for which data is available in ILO labor force
survey data (viz. self-employed employers, ‘own-account’ workers, and ‘contributing



                                               81
family members’), we estimate a country-year coefficient, computed as the ratio of the
average employee’s wage to that of the self-employed worker.
    Total labor income for a given country-year is retrieved as follows:

                                                         CE
                                      YL = CE +               · γi si
                                                         semp

where CE is the total compensation of employees in national accounts; semp is the share
                                                                                                    CE
of employees in the workforce (whose collective earnings equal CE ), such that                     semp
                                                                                                          is
the average employee wage; and si denotes the share in the workforce of each type of
self-employed worker. Each self-employment category i corresponds to a specific earnings
coefficient γi relative to the average employee compensation.              Capital income YK is Y − YL .
    Note, in Appendix Table A6, that the results from this robustness check do not affect
our estimates, as the factor shares estimated from this method do not systematically differ
from the factor shares calculated with a benchmark φ = 70% labor share of mixed income.
    For the United States, we use estimates of factor shares in net domestic product from
Piketty, Saez, and Zucman (2018) that incorporate a granular treatment of mixed income
reflecting specificities of the US non-corporate business sector. Specifically in the United
States, a number of large businesses (including listed firms) are organized as partnerships
(as opposed to corporations) and as such are classified as non-corporate businesses, while
they would be treated as corporations in other countries. Their income is counted as
mixed income in the US National Income and Product Accounts, while it would typically
be recorded as corporate profits in other countries. Since 1986, a growing number of
businesses have been organized as partnerships for tax reasons. The series we use take
these facts into account by: (1) assuming a higher capital share of income for partnerships
vs. other non-corporate businesses (sole proprietorships); and (2) factoring in the rising
capital intensity of partnerships since the 1980s.




     This is according to the ILO’s International Classification of Status in Employment, ICSE-93, with data
from 1991 to present.
     For country-years missing these concepts si and γi in raw ILOSTAT (2021) data—particularly for the era
prior to 1991—we impute observations using the same procedure as above.


                                                    82
B.2     Tax revenue data

Data     As described in Table B1, our tax revenue data draws from four types of sources:

    • OECD online Government Revenue Statistics (OECD, 2020)

    • ICTD/UNU-WIDER Government Revenue Dataset (ICTD/UNU-WIDER, 2020)

    • Historical archive work, principally from the Harvard Lamont Library.

    • IMF Government Finance Statistics (IMF GFS, 2005)

    While our use of the first three is more or less self-explanatory, and we retrieve off-the-
shelf data from these international organizations, the fourth fountainhead is perhaps the
most significant because it fills data gaps from these other three with extensive archival
work to stitch together harmonized long-run series for the more than 150 countries.
    Our archival work started in the Government Documents section of the Lamont
Library archive at Harvard University. For each country in our data-set, we scanned,
tabulated, processed, and unified official data from the public budget and national statistical
yearbooks, to retrieve the official statistics on revenue collection even in remote historical
eras. To complement hard-copy archival data, we also retrieved individual countries’
online datasets, usually published by a national statistical office or ministry of finance, or
sometimes cross-country datasets from international organizations and scholarly efforts
(e.g., CIAT and IDB, 2019; Fisunoglu et al., 2011; Lotz and Morss, 1970). Lastly, we would fill
remaining gaps by scouring scholarly sources for individual studies detailing the aggregate
and disaggregated levels of (central and decentralized levels of government) tax revenue
collection. To corroborate levels and trends of tax revenue across sources, in this way, also
helped in our triangulation and harmonization effort discussed below. These sources—and
their harmonization (see also ‘stitching’ section below)—are detailed country-by-country
in a forthcoming compendium of country case studies.



     Note, however, that we are using the IMF Historical Government Finance Statistics for the period 1972-89,
not the version that is readily available online for the period since then. The historical dataset is available as
a CD-ROM with dedicated software.


                                                       83
Revenue Concepts We classify each type of tax revenue following the OECD Government
Revenue Statistics (OECD, 2020). We consider the following simplified classification (see
also Table A1).

   • The ‘1000’ series of taxes, covers taxes on income and profits:

            This heading covers taxes levied on the net income or profits (i.e. gross
            income minus allowable tax reliefs) of individuals and enterprises. Also
            covered are taxes levied on the capital gains of individuals and enterprises.
            (OECD, 2020).

      We always separate the 100 series between personal income taxes (PIT, 1100) and
      corporate income taxes (CIT, 1200).

      While all corporate income taxes is allocated to capital taxation, PIT revenue is a mix
      of labor and capital income. Thus, within PIT, an important parameter is the share of
      revenue assigned to capital, denoted λP IT (as in Table A1). As discussed in Section
      3.1.2, we rarely observe directly the level of PIT revenue that derives from capital
      income and gains versus labor income.                In the United States, Piketty, Saez, and
      Zucman (2018) find that about 15% of the PIT revenue is from capital and about 85%
      from labor. To construct country-year specific λP IT , we start from the US benchmark
      (λP IT = 15%) to which we make two adjustments:

         1. First, how high the PIT exemption threshold is located in the income distribution,
            will impact λP IT since capital income is concentrated at the top of the income
            distribution. we retrieve PIT exemption thresholds from Jensen (2022) and
            extend these estimates to all country-years in our data. Countries with a higher
            PIT exemption threshold are assumed to have a higher λP IT than countries with
            a lower PIT exemption threshold. Since the US has a low exemption threshold
     The subtotal of PIT revenue from capital income includes: taxes on dividends; on the capital share
of self-employment income; and on capital gains. OECD data on statutory income tax rates tell us about
dual-income tax systems where the rate on dividends differs from the rate on salaries (used in our adjustments
below). OECD (2020) revenue data occasionally reports on PIT revenue from capital gains. From 2010-18,
among 27 countries that report some capital gains tax revenue within PIT revenue, this revenue averages just
under 4% of PIT revenue. In the US, that number is 7.5%. Of course, many countries do not tax capital gains
as such, and most countries tax other forms of capital income within the PIT system.


                                                    84
       and has λP IT = 15%, we similarly assign a 15% value to countries for which
       the PIT covers half or more of the population (usually high-income countries).
       For countries where the PIT covers 1% or less of the population, we assign a
       maximum capital share of 30%. For PIT thresholds with a coverage between 1
       to 50% of the population we do a linear extrapolation assigning a λ parameter
       between 15 and 30%.

    2. Second, we reduce λP IT in countries where a dual-PIT system sets capital income
       taxation to a lower— often flat— rate, while labor income continues to be taxed
       with (higher) progressive marginal tax rates. In these dual systems, we measure
       the percent difference between the capital tax rate (dividends) and the labor tax
       rate (this percent difference will be negative). Data on capital and labor rates are
       taken from OECD Revenue Statistics and country-specific tax code documents.
       We assume that 50% of capital in PIT comes from these sources that are taxed
       preferentially, reflecting the fact that self-employment income, and occasionally
       capital gains, are likely not to be taxed differently from labor. For this 50%, we
       multiply λ by the percent difference in capital versus labor tax rates.

• The ‘2000’ and ‘3000’ series of tax revenues, include respectively social security
  contributions and taxes on payroll and workforce:

       Classified here are all compulsory payments to general government that
       confer an entitlement to receive a (contingent) future social benefit. ...
       Contributions for the following types of social security benefits would,
       inter alia, be included: unemployment insurance benefits and supplements,
       accident, injury and sickness benefits, old-age, disability and survivors’
       pensions, family allowances, reimbursements for medical and hospital
       expenses or provision of hospital or medical services. Contributions may
       be levied on both employees and employers. ... Contributions to social
       insurance schemes which are not institutions of general government... are
       not considered as social security contributions. (OECD, 2020)



                                         85
     We also include here all payroll taxes, i.e., ‘taxes payable by enterprises assessed
     either as a proportion of the wages or salaries paid or as a fixed amount per person
     employed.’ We consider all such social contributions and payroll taxes attributable as
     taxes on labor income, as shown in Table A1.

   • The ‘4000’ series comprises property and wealth taxes:

            This heading covers recurrent and non-recurrent taxes on the use, ownership
            or transfer of property. These include taxes on immovable property or net
            wealth, taxes on the change of ownership of property through inheritance
            or gift and taxes on financial and capital transactions. (OECD, 2020)

     We attribute these revenues as taxes on capital income.

   • The ‘5000’ series, are indirect taxes on goods and services, including tariffs and
     trade taxes, value-added taxes, sales taxes, excises, and other forms of consumption
     taxes and taxes on production. Importantly, we consider these taxes both prior and
     proportional to factor income returns, such that the attribution of indirect taxes to
     factor incomes can be considered precisely proportional to factor incomes in the total
     economy.       If we were to include these taxes in our estimates of effective tax rates (see
     Section 3.1.2), we would only increase the levels of effective tax rates, but we would
     not change their trends relative to one another.

   Finally, note that we exclude other non-tax revenues and tax expenditures from
consideration. Once we classified each type of tax according to the simplified OECD (2020)
framework, we proceeded to stitch together long-run tax revenue series for all countries.


Appending and Stitching Data Sources In brief, our comprehensive tax revenue data
collection and harmonization process observed a simple set of rules. When more than one
source existed to document levels of tax revenue in a given country-year (with tax revenues
disaggregated as in the classification above), we compare statistics in the years of overlap
and prioritize sources as follows:

    In this sense, we also measure factor-price domestic product, i.e., net of indirect taxes.


                                                      86
   1. OECD revenue statistics take precedence.

   2. If OECD data is unavailable, then we use HA (historical archive, government
     documents from Harvard University Library archives and elsewhere [see above]).

   3. If remaining country-years are missing, we use complementary sources, including:
     ICTD/UNU-WIDER (2020) data that begins in 1980; IMF GFS (2005) that covers
     the period from 1972-1989; or, additional sources as described above. For social
     contributions in particular, two complementary sources provded helpful: the ‘D61’
     statistic on social contributions in the household sector in UN SNA (1968) and UN
     SNA (2008) are accurate (to the best of our knowledge) and extensive in its coverage;
     and, the data from Fisunoglu et al. (2011) provide additional coverage based on
     manually digitized versions of offline IMF archival documents.

We always referred to a variety of scholarly sources for every country, to corroborate against
each other and against these sources for the ‘stitch’ years across overlapping eras with data
from multiple sources.
   To check the robustness of long-run tax revenue series, country-by-country, and to
minimally interpolate where necessary (rarely), we applied the following rules:

   1. We do not include a country in our sample if there are fewer than 10 years of observed
     tax revenue data.

   2. We never interpolate more than four years without data during a time series.

   3. Only in rare cases do we use more than two data sources per country.

   4. If any discrepancies existed during ‘stitch’ years across data sources, we study the
     nature of the discrepancy in a review of scholarly sources—to then retrieve statistics
     on the revenues that are missing from the less-accurate source; and to ‘backcast’ using
     year-on-year growth rates of the less-accurate source, if necessary (see UNSD, 2018).

   5. To check that we were not missing significant sources of decentralized tax revenues,
     we use the recent OECD-UCLG (2019) study to find the countries with significant
    We interpolated some tax revenue data in only 5% of nearly 7000 observations.


                                                  87
       state and local public finance revenue collection, and collected more data for any
       countries where necessary.

  6. We make explicit any assumptions about PIT vs. CIT split during historical eras
       where the government documents do not clarify the nature and relative weights of
       individual vs. corporate collection of income tax revenue. In these cases, we rely on
       local scholarly sources that discuss the legislation.

  7. Similarly, in any (rare) cases where the OECD data did not measure the PIT vs. CIT
       split (and only ascribed certain revenues to a generic total income tax), we often
       find that actually the split can be made after studying the details of countries’ tax
       legislation. This only occurs rarely in practice.

  8. We check for significant regime change, political conflict, inflation episodes or other
       macro-economic crises that cast significant doubt on the credibility of estimates
       and the continuity of tax revenue time series. We do not interpolate between years
       characterized by such events.

  9. We exclude all tax revenue observations for territories prior to independence as well
       as all countries during the Communist era.

   Starting from this set of rules, we implement several checks on the quality of our factor
income and tax revenue datasets. They are outlined on a country-by-country basis in the
online appendix. As an illustration, below we reproduce the checks that were implemented
for the tax revenue series in China.


B.3      Case study: China’s establishment of a modern tax system in 1994

Based on the rules we established, we only include ex-command economies into our data
starting in 1994. Given China’s importance for the global economy, we discuss here the
reason for this choice.
   The tax revenue data for China covers most of our sample period, although it improves
a in quality in the 1980s.        Prior to the 1980s, China had a command economy’s model
      Official public finance statistics are available online at https://data.stats.gov.cn/english/index.htm.


                                                     88
of ‘profit delivery,’ in which the state spent directly the revenues of profitable SOEs and
subsidized unprofitable ones, an example of soft budget constraints (Kornai, Maskin, and
Roland, 2003). As part of the decade of economic reforms beginning in 1978, a corporate
income tax appears in China around 1983-84 (Wong and Bird, 2008). A decree from the
State Council in 1983 put a new 55% tax on the profits of enterprises, which were still
almost entirely state-owned. Starting in 1985, we observe CIT revenue in the data, although
it appears implausibly high (as a percentage of NDP, or of capital income). In addition,
not long after, the tax system was further reformed into a ‘fiscal contracting’ system
whereby firms negotiated a fixed tax revenue level (regardless of economic outcomes) to
local governments, who in turn delivered a share to the central government. This system
resembles a firm "poll" tax, and subsumed all types of tax revenues, including CIT, VAT
and payroll, making it impossible to assign taxes to capital or labor. For these reasons, we
exclude this sort of ‘pseudo’-CIT revenues dating from 1985 through 1993.
   China’s modern tax system arguably begins in 1994. A valuable resource to study the
introduction of a modern tax system is the book Public Finance in China (World Bank, 2008).
The book argues that prompted by low central government revenues, China established in
1994 a central tax administration; reformed the broken ‘fiscal contracting’ system; unified
the PIT; created a VAT; and reduced ‘extra budgetary’ (non-tax) revenues.                 Thus from
1994 onward we can categorize tax revenue by sources, assign them to capital or labor,
and thus construct ETRs. It should be noted, however, that to the extent that they were
measurable, tax revenue appeared higher in the 1980s than in the 1990s in China.




    The state-run media summarizes these policy reforms succinctly (China Daily, 2018), and agrees with
scholarly work (e.g., World Bank (2008) see esp. pp.13-22; or Wong and Bird (2008) pp.430-440)


                                                  89
                                    Table B1: Main Data Sources

                                            country-year obs.         %

                                      Panel A: Factor Share Data

                                SNA2008           2403 34.8%
                                SNA1968           1484 21.5%
                                composite/imputed 3016 43.7%

                                           N               6903     100%

                                      Panel B: Tax Revenue Data

                                OECD                       2881 41.7%
                                Harvard/archives           2092 40.2%
                                ICTD                       1276 18.5%

                                           N               6903     100%
Notes: For the N=6903 country-year observations in which we estimate effective tax rates on capital and labor
income (over 156 countries since 1965), Panel A presents the sources of our factor share data (on national
income components), while Panel B presents the sources our tax revenue data (on total revenues disaggregated
by type of tax). In the former, we use online data from UN SNA (2008) and archival data from UN SNA
(1968). In the latter, we draw tax revenue data from sources including OECD (2020), ICTD/UNU-WIDER
(2020), and IMF GFS (2005), as well as extensive archival research in the Harvard University Library, online
sources, and IMF historical data. (See Appendix B.)




                                                    90
Appendix C             Event studies

C.1    Description of trade events

Our selection of trade events is determined by three criteria: (i) the event can unam-
biguously be related to measurable policy reforms; (ii) the policy reforms induced large
changes in trade barriers; and (iii) the event has been studied in peer-reviewed academic
publications. The first criterion improves the transparency of the event-study design,
which rely on changes in outcomes around an explicitly defined policy event. The second
criterion increases the likelihood that we are able to observe sharp breaks in trend in our
macroeconomic outcomes which coincide with the timing of the policy event. The third
criterion allows for our results to be compared to previous work, and for anyone to study
the narrative surrounding these reforms. These prior papers have already established
positive effects of the reforms on cross-border trade (and other economic outcomes).


Selection of events These criteria lead us to focus on the six trade liberalization events
referenced in review articles by Goldberg and Pavcnik (2007) and Goldberg and Pavcnik
(2016), GP henceforth, to which we add China’s WTO accession event (studied in Brandt
et al., 2017). These liberalization events all feature reductions in tariff barriers, the most
commonly studied component of globalization, in part because it is easier to measure on a
consistent basis across space and time, as compared to non-tariff trade barriers and other
forms of globalization. These events also features reductions in non-tariff barriers which
are harder to measure; for example the number of products subject to import licences and
quotas fell. Fortunately, tariff and non-tariff barrier reductions seem highly correlated
(Goldberg and Pavcnik, 2007).
   All selected events feature large reductions in tariff rates. Most of the selected countries
did not participate in the early GATT/WTO negotiation rounds; consequently, tariffs
remained high in these countries prior to the events, such that reductions in tariff rates
remained an available policy lever.     These trade liberalization events were drastic: Brazil

    Some countries were not GATT members by the time of the event, such as Mexico; others (Brazil,
Colombia, India) were nominal GATT members, but were not forced to reciprocate tariff concessions
negotiated with GATT until the Uruguay Round (Goldberg and Pavcnik, 2016).


                                               91
cut tariff rates from 58.8 percent to 15.4 percent; India reduced rates from 80 percent to 39
percent; China reduced tariffs from 48% on average to 20%; Mexico reduced tariff rates
from 23.5% to 11.8%, while import licence requirements went from covering 92.5% of
national production to 25.4%; Colombia’s tariffs were reduced from 27% to 10% and import
requirements dropped from 72% of national production coverage to 1.1%. In the selected
countries, “tariff reductions constitute a ‘big part’ of the globalization process” (Goldberg
and Pavcnik, 2016). The timing of these events and references to additional papers which
study the events in detail are provided in Table C1.


Timing of events    Most studies provide detailed discussions of the context surrounding
the events. We repeat the rationale cited for the liberalization events and discuss why these
events are plausibly exogenous to the country’s economic circumstances at the time.

• Brazil The liberalization event of 1988 is detailed in Dix-Carneiro and Kovak (2017).
  The authors note that the high pre-reform average level of tariffs was driven by large
  cross-industry variation in protectionism and that the reform was unexpected:

       “In an effort to increase transparency in trade policy, the government reduced
       tariff redundancy by cutting nominal tariffs... Liberalization effectively began
       when the newly elected administration suddenly and unexpectedly abolished
       the list of suspended import licences and removed nearly all of the remaining
       special customs regimes.” (Dix-Carneiro and Kovak, 2017)

• Columbia Similarly to Brazil, tariff reductions in Colombia in 1985 were driven by the
  country’s commitment to impose uniform rates across products and industries under
  the negotiation commitments to the WTO. In Colombia’s case, Goldberg and Pavcnik
  (2007) note that the reform objective was to reduce cross-industry dispersion under WTO
  negotiations, thereby making “the endogeneity of trade policy changes less pronounced
  here [in Colombia] than in other studies.”

• China Brandt et al. (2017) note that trade openness reforms had gradually been im-
  plemented in China prior to the country’s WTO accession event in 2001, but that the


                                             92
  tariff reductions implemented in the immediate post-accession period were large, “less
  voluntary” and largely complied with the fixed WTO accession agreements. Importantly,
  the potential accession to WTO contributed to timing of the privatization initiatives
  in the pre-WTO years, in which the Chinese government restructured and reduced its
  ownership in state-owned enterprises. While the privatization efforts began in 1995 and
  were also gradual (Jefferson, 2016), given their importance in the national economy, it
  is possible that additional sell-offs in the immediate post-WTO years contribute to the
  observed break in trends in our outcomes.

• India The 1991 event in India occurred as a result of an IMF intervention that dictated
  the pace and scope of the liberalization reforms. Under the IMF program, the tariff
  rates had to be harmonized across industries, which, like in Brazil and Colombia, led
  to a large average reduction in tariffs. Topalova and Khandelwal (2011) provide an
  extensive discussion of the Indian reform, arguing that it “came as a surprise” and “was
  unanticipated by firms in India.” The reforms were implemented quickly “as a sort
  of shock therapy with little debate or analysis.” The IMF program was in response
  to India’s balance of payment crisis, which was triggered by“the drop in remittances
  from Indian workers in the Middle East, the increase in oil prices due to the Gulf War,
  and political uncertainty following the assassination of Rajiv Gandhi” (Topalova and
  Khandelwal, 2011).

• Vietnam The 2001 reform in Vietnam was implemented as a broad trade agreement
  that did not involve negotiations over specific tariffs (McCaig and Pavcnik, 2018). The
  reform was driven by the American government’s decision to reclassify Vietnam from
  ’Column 2’ of the US tariff schedule to the ’Normal Trade Relations’ schedule. Column
  2 was designed in the early 1950s for the 21 communist countries, including Vietnam,
  with whom the US did not have normal trading relations. McCaig and Pavcnik (2018)
  show that there are no differential trends between Vietnamese exports to the US relative
  to exports to other high-income countries. Vietnam’s case is compelling since the
  liberalization even was triggered by a foreign party, rather than by its own government.




                                           93
   These descriptions of reform context do not argue that liberalization events were
triggered by trends in taxation needs. Thus, this narrative analysis complements the
absence of a pre-trends result (Figure 7), to help alleviate endogeity concerns in the timing
of the events.


Post-event reforms       Yet, even if the reform timing is uncorrelated with confounding
trends, the interpretation of the event studies depends on whether other reforms and
macroeconomic shocks occurred in the immediate post-reform years. The detailed review
provided in (Goldberg and Pavcnik, 2007) is very helpful, as it notes all major further events
which followed the initial liberalization events. Argentina’s event in 1989 is followed by
accession to Mercosur in 1991; Brazil’s event in 1988 is followed by accession to Mercosur
in 1991 and the currency crisis in 1998; India’s event in 1991 is followed by foreign direct
investment liberalization in 1993; and Mexico’s 1985 WTO accession was followed by a
removal of capital inflow restrictions in 1989 and accession to NAFTA in 1994.
   Their discussion suggests that other reforms did occur, often a few years after the
trade liberalization event; and, these reforms served to reduce other non-tariff barriers to
cross-border flows. As such, while the immediate post-event impacts may more likely be
attributed to trade liberalization, the medium-run impacts might be better interpreted as
the reduced-form effects of globalization more generally, which includes an increase in the
flow of goods, services, and capital, as well as further policy responses.


C.2    Event study methodology

Sample Construction Our sample is constructed by applying a synthetic matching
procedure to every treated country, for each outcome of interest. The donor pool (the set
of all control countries from which to chose the synthetic control group) has to be fully
balanced in all pre-event periods. Thus, we discard all countries with data gaps before
1976. This gives us a sample of 103 countries for each outcome.             We have ten ‘pre’ event



    The exception to this rule is the trade variable, where we have more data gaps and consequently must
drop more countries from the donor pool. Here, the donor pool consists of 90 countries.


                                                  94
years for every country, except for Mexico and Colombia, where we have nine pre-period
years.   We then pool together all seven treated countries and their synthetic control units.


Empirical Strategy Using this panel, we estimate the following event study regression:

                              10
                  Yit =                  βj ∗ 1(j = t)t ∗ Di + θt + κi + πY ear(it) +   it
                          j =−10,j =−1

   where θt and κi represent, respectively, time relative to/from the event and country
fixed effects, and where πY ear(it) is a set of fixed effects for calendar years. Di is a dummy
equal to one if observation i is a treated country. Hence, βj capture the difference between
the treated countries and the group of synthetic controls across event time, with year t-1 as
the reference period.
   In addition to the event study regressions, we also use this setting to estimate a simple
difference-in-difference coefficient:


                      Yit = β DiD ∗ 1(j ≥ 0)t ∗ Di + θt + κi + πY ear(it) +       it


   Here, µDiD can be interpreted as an average treatment effect over the first 10 years post
treatment. We run both regressions—the event study and the difference-in-difference
regression—on a set of nine outcomes, and cluster standard errors at the country level. We
note that statistical inference based a small sample size should be approached with caution
(Abadie, Diamond, and Hainmueller, 2010). We therefore also report standard errors based
on the wild bootstrap method (Cameron, Gelbach, and Miller, 2008) for β DiD , in Table A2.
   Moreover, we use the imputation method developed by Borusyak, Jaravel, and Spiess
(2021) to report average treatment effects comparable to β DiD with a technique that deals
with issues with two-way fixed effects and heterogeneous timing of events (such as our
setting). The approach provides a transparent alternative method to the difference-in-
difference equation specified above. The average treatment effect τ is calculated in three
steps, detailed below.


    Moreover, China and Vietnam only have tax revenue data from 1994 onward, but its event happens in
2001. We therefore only match on seven pre-period years.


                                                      95
  1. We use untreated countries as well as treated countries in the years before treatment,
     to estimate unit and (relative) year fixed effects:



                                   Yit = θt + κi + πY ear(it) +   it



     if t < 0 or Di = 0. We note that to bring the approach developed by Borusyak, Jaravel,
     and Spiess (2021) closer to our estimation strategy, we include fixed effects for the
     years in relative time.

                                    ˆt and κ
  2. Equipped with the fitted values θ      ˆ i , we can now impute the unit specific
     treatment effect as:


                                               ˆt − κ
                                   ˆit = Yit − θ
                                   τ                ˆi − π
                                                         ˆY ear(it)


  3. The final step is to average over those coefficients to produce a treatment effect.
     We report unweighted averages, but heterogeneity in treatment effects could be
     accounted for by specifying weights.



Simultaneously Matching on Main Outcomes As we perform the synthetic matching
procedure for each event and outcome based on the outcome, we have a different ‘synthetic’
control for each country in every outcome. This means that while we use the same group
of treated countries in every regression, the set of control countries that feeds into the
synthetic control group varies across outcomes. We want to test that our results hold up in
a more rigid setting.
   To ensure robustness to a more restrictive synthetic control, we repeat all of the above
analysis, using a slightly different approach of constructing the synthetic control countries.
Specifically, we use our four main outcomes—trade (as a percentage of domestic product);
capital share of domestic product; and effective tax rate on labor and capital—to predict
one synthetic control group per treated country. The resulting weights for the three most
prominent countries in each control group are reported in Table ??. This exercise allows us


                                              96
to still run separate regressions for each outcome, but with the exact same composition of
the control group in each regression.
   Level and event study graphs for each outcome are reported in Figure A10. While it is
obvious that for some outcomes, the levels do not line up exactly in the ‘pre’ periods, the
graphs confirm the conclusions from Figure 7. In Table ??, we also report all estimated
coefficients (including the simple difference-in-difference estimate) of this alternative
approach. While we prefer the approach presented in the main body of this paper in
Section 6.1, this robustness exercise does demonstrate that the results there do not hinge
on the flexible nature of our synthetic control design.


C.3    Capital liberalization: Events and results

In this subsection, we provide details on the event-study analysis of capital liberalization in
Section 8. We use the 25 liberalization events in developing countries measured by Chari,
Henry, and Sasson (2012). The authors identify liberalization dates using the point in time
when countries first permitted foreigners to purchase shares of companies listed on the
domestic stock market. At first glance, this event may appear to be a narrow way to define
capital liberalization, but the authors provide several arguments that the dates in fact serve
as indicators for larger move towards open capital markets.
   First, net equity inflows soared in the periods where developing countries liberalized
their stock markets. While such flows were practically non-existent in the 1970s and
early 1980s, they accelerated sharply around the end of the 1980s when most countries
implemented their liberalization reforms. It is important to note that stock market
liberalizations account for a significant fraction of foreign direct investment (FDI). Indeed,
for purposes of official statistics, FDI includes any stock transaction (that is, a cross-border
merger or acquisition) that results in the purchaser owning 10 percent or more of the
voting shares. Chari, Henry, and Sasson (2012) quote data from Latin American and
East Asia which show that cross-border mergers and acquisitions account for 40% to 60%
of FDI. Second, by facilitating cross-border financial flows and ownership, stock market
liberalization induces large inflows of capital goods. The authors quote other studies



                                              97
which show that, following capital liberalization, the share of capital goods in total imports
rises by 9%. Through this channel, the liberalization events contribute to raising aggregate
investment. Third, policymakers often worry that increased net equity inflow can easily
convert into net outflow at the early signs of macro-economic instability. In turn, the
increased outflow can compound the instability and introduce volatility. But the authors
note that the liberalization events considered here were not marked by any reversal of
the liberalization policies or the equity inflows in the post-reform years. This allows us
to study the effects of liberalization reforms more cleanly and over longer time-horizons.
Fourth, there are difficulties in determining precise liberalization dates and the literature
has mainly chosen to ignore this measurement problem (Eichengreen, 2001). By focusing
on the initial date where stock markets were first opened to foreign capital, the events are
more cleanly measured than other candidate de-jure policy changes which may be harder
to pinpoint. For further details about the complexities of determining liberalization dates,
see Henry (2007).
   For these reasons, the authors consider the stock market liberalization events to be
meaningful indicators of countries’ progression towards freer capital inflows. The authors
select 25 liberalization events which: are recorded in official sources (Standard and Poor’s
Emerging Markets Database); are verified from primary sources; and, have been used
elsewhere in the literature. Table C2 lists the 25 countries with their liberalization dates.
   Our sample is constructed by applying a synthetic matching procedure to every treated
country for each outcome of interest. We follow the methodology and estimation techniques
described in detail in Appendix Section C2. Table C2 provides details on the main countries
that form the synthetic control group for each treated country and each outcome. Our
main measure of capital openness is the sum of foreign assets and liabilities. This stock
is expressed as a percent of GDP; given the strong right-skew, we use the log of this
ratio in the analysis. We find similar results when using alternative measures of capital
openness, including portfolio equity assets and liabilities (% of GDP) and the KOF financial
globalization index (Gygli et al., 2019). Moreover, we also find that the liberalization events
led to positive changes in the de jure index of capital openness produced by Chinn and



                                             98
Ito (2006). These results are available upon request. Finally, we note that due to limited
data availability, we were not able to study the impact on foreign direct investment.
   The level and event-study graphs for each outcome (capital openness; capital share of
national income; capital share of corporate income; ET RK ; ET RL ) are reported in Figure 8.
The statistic in the top-left corner reports the F-statistic and p-value for the joint significance
of all post-liberalization dummies. It is important to note that, when jointly considered, the
post-liberalization coefficients for ET RK are not statistically significant. The panels make
clear this is because the statistically significant effect on capital taxation materializes with a
few years’ lag to the liberalization event. Appendix Figure A14 shows that the results are
robust to creating synthetic controls for each event-country based on matching the four
outcomes simultaneously (capital openness; capital share; ET RK ; ET RL ). In particular,
the events continue to be associated with a large increase in capital openness, a positive
impact on capital taxation and a muted (null) impact on labor taxation.




                                               99
        Table C1: Weights in Synthetic Control for Trade Liberalization Events

Country       Event Year      Capital share (NDP) Weight              ETR on capital        Weight      ETR on labor       Weight      Trade Openness        Weight                                 Reference

                              Uruguay                     28.7 %      Bangladesh            41.6 %      Chile              35.9 %      Bangladesh            97.3 %
Argentina         1989        Oman                        18.6 %      Haiti                 14.1 %      Togo               31.6 %      United States         2.7 %               Goldberg and Pavcnik (2006)
                              Bolivia                     16.2%       Bolivia               13.4 %      Jordan             16.8 %      .                     .
                              ...                         ...         ...                   ...         ...                ...         .                     .
                              Bolivia                     9.1 %       Jordan                35.7 %      Panama             25.7 %      Bangladesh            59.8 %
                                                                                                                                                                               Goldberg and Pavcnik (2006),
Brazil            1988        Eswatini                    7.6 %       Sudan                 21.2 %      Guyana             21.7 %      United States         32.2 %
                                                                                                                                                                              Dix-Carneiro and Kovak (2017)
                              Mauritius                   7.0 %       Zimbabwe              12.7 %      Chile              14.5 %      Japan                 6.1 %
                              ...                         ...         ...                   ...         ...                ...         ...                   ...
                              Sweden                      37.0 %      Congo                 41.8 %      Kuwait             31.1 %      United States         36.2 %
China             2001        Switzerland                 24.3 %      Nicaragua             26.3 %      Pakistan           22.9 %      Bangladesh            36.0 %                        Brandt et al. (2006)
                              Sierra Leone                21.1 %      Gabon                 14.2 %      Uganda             20.2 %      Dominican Rep.        12.2 %
                              ...                         ...         ...                   ...         ...                ...         ...                   ...
                              Lesotho                     17.0 %      Kuwait                67.9 %      Paraguay           45.5 %      Bangladesh            50.7 %
Colombia          1985        Nicaragua                   9.5 %       Gabon                 14.6 %      Sudan              15.0 %      Iran                  22.6 %       Goldberg and Pavcnik (2006; 2016)
                              Bahrain                     8.1 %       Sierra Leone          12.6 %      Cameroon           11.5 %      Guatemala             12.5 %
                              ...                         ...         ...                   ...         ...                ...         ...                   ...
                              Nepal                       8.0 %       Uganda                41.4 %      Lebanon            37.9 %      United States         76.4 %
                                                                                                                                                                         Goldberg and Pavcnik (2006, 2016);
India             1991        Nicaragua                   6.9 %       Bolivia               14.0 %      Oman               17.6 %      Bangladesh            23.6 %
                                                                                                                                                                                     Topalova et al. (2009)
                              Iceland                     5.8 %       Haiti                 4.6 %       Jordan             16.2 %      .                     .
                              ...                         ...         ...                   ...         ...                ...         .                     .
                              Paraguay                    36.3 %      Sierra Leone          33.2 %      Tunisia            31.1        Bangladesh            72.0 %
                                                                                                                                                                               Feenstra and Hanson (1997);
Mexico            1985        Botswana                    18.7 %      Bahrain               23.6 %      Zimbabwe           25.8 %      Uruguay               9.6 %
                                                                                                                                                                          Goldberg and Pavcnik (2006, 2016)
                              Philippines                 14.3 %      Bolivia               14.7 %      Uruguay            15.9 %      Spain                 8.0 %
                              ...                         ...         ...                   ...         ...                ...         ...                   ...
                              Nepal                       27.5 %      Korea                 45.8 %      Bangladesh         72.8 %      Thailand              42.4 %
                                                                                                                                                                                Goldberg and Pavcnik (2016),
Vietnam           2001        United States               10.1 %      Luxembourg            19.2 %      Myanmar            22.6 %      Ghana                 22.6 %
                                                                                                                                                                                 McCaig and Pavcnik (2018)
                              Switzerland                 4.7 %       Trinidad & Tob.       17.3 %      Haiti              4.6 %       Venezuela             21.7 %
                              ...                         ...         ...                   ...         .                  .           ...                   ...
  Notes: This table shows the seven treated countries and the three countries with the largest weight in the synthetic control group for each outcome and each treated country. For
  each outcome, the pool of possible donor countries consists of 103 countries, with the exception of the trade variable, where we have only 90 countries with a balanced panel over the
  period considered. Note that the synthetic control method requires the panel of possible donor countries to be strictly balanced in all ‘pre’ periods that are used in the matching
  procedure. One additional restriction applies with respect to this sample. For the outcomes on trade and ET RK , the extrapolation of Vietnam to the years 1991-93 lead to outlier
  values in those years, so we do not use these imputed values. This results in the panel for these outcomes to be slightly unbalanced in the years t − 10 to t − 8.
Table C2: Weights in Synthetic Control for Capital Liberalization Events

                                  Event
                      Country             Capital share (NDP)   Weight   ETR on capital     Weight   ETR on labor      Weight   Capital Openness   Weight
                                  Year
                                          Uruguay               19.7 %   Bangladesh         39.7 %   Chad              35.8 %   United States      42.3 %
                      Argentina   1989    Qatar                 16.0 %   Paraguay           29.8 %   Togo              35.8 %   Papua New Guinea   33.3 %
                                          Bolivia               11.8 %   Nepal              7.5 %    Uruguay           28.2 %   Tanzania           11.8 %
                                          ...                   ...      ...                ...      ...               ...      ...                ...
                                          Malawi                9.6 %    Sudan              34.0 %   Guyana            30.3 %   United States      54.1 %
                      Brazil      1988    Swaziland             9.6 %    Mali               10.9 %   Congo             18.8 %   El Salvador        15.5 %
                                          Mauritius             7.8 %    Zambia             10.8 %   Togo              17.2 %   Iran               13.1 %
                                          ...                   ...      ...                ...      ...               ...      ...                ...
                                          Zambia                46.0 %   Bolivia            45.8 %   Madagascar        49.9 %   Kuwait             35.8 %
                      Chile       1987    Botswana              15.8 %   Uganda             25.2 %   Uruguay           42.3 %   Lebanon            16.7 %
               101




                                          Kuwait                13.0 %   Canada             19.0 %   Norway            7.6 %    Ireland            12.3 %
                                          ...                   ...      ...                ...      ...               ...      ...                ...
                                          Fiji                   29.6 %   Bangladesh         59.2 %   Sudan             60.1 %   Kuwait             31.7 %
                      Colombia    1991    Bolivia               17.1 %   Iceland            20.9 %   Trinidad and T.   24.5 %   Gabon              19.7 %
                                          Ireland               8.6 %    Papua New Guinea   11.2 %   Paraguay          14.9 %   Austria            13.0 %
                                          ...                   ...      ...                ...      ...               ...      ...                ...
                                          Qatar                 33.1 %   Sierra Leone       43.3 %   Nicaragua         50.7 %   Bolivia            45.6 %
                      Egypt       1991    Paraguay              18.2 %   Uruguay            24.3 %   Bahrain           23.1 %   United States      31.9 %
                                          Mauritius             10.9 %   Zambia             18.5 %   Luxembourg        13.0 %   Syria              18.0 %
                                          ...                   ...      ...                ...      ...               ...      ...                ...
                                          Canada                22.6 %   Swaziland          22.1 %   Lesotho           30.3 %   Italy              44.6 %
                      Spain       1993    Swaziland             18.0 %   Jamaica            16.5 %   Norway            29.8 %   Singapore          30.5 %
                                          Albania               8.3 %    New Zealand        8.9 %    Italy             18.7 %   Sweden             12.9 %
                                          ...                   ...      ...                ...      ...               ...      ...                ...
                                          Costa Rica            50.5 %   Myanmar            41.8 %   Panama            30.3 %   Botswana           13.5 %
                      Greece      1994    Mozambique            11.4 %   Bahamas            21.4 %   Austria           25.9 %   United States      9.4 %
                                          Albania               10.4 %   Malawi             19.2 %   Nicaragua         20.6 %   Saudi Arabia       8.6 %
                         ...             ...      ...                   ...      ...          ...      ...             ...
                         Mozambique      20.4 %   Benin                 17.1 %   Lebanon      36.0 %   Ireland         30.3 %
      Indonesia   1989   Sweden          5.7 %    Uganda                16.3 %   Bahamas      19.5 %   Oman            21.1 %
                         Gabon           4.8 %    Bangladesh            14.3 %   Myanmar      11.5 %   United States   19.3 %
                         ...             ...      ...                   ...      ...          ...      ...             ...
                         Mauritius       7.4 %    Uganda                29.3 %   Lebanon      64.2 %   Peru            49.7 %
      India       1986   Costa Rica      7.0 %    Bahrain               14.2 %   Chad         15.7 %   Denmark         19.9 %
                         Finland         6.1 %    Peru                  10.8 %   Peru         7.1 %    United States   14.6 %
                         ...             ...      ...                   ...      ...          ...      ...             ...
                         Denmark         34.9 %   Sweden                39.1 %   Sweden       78.8 %   United States   20.9 %
      Israel      1989   Algeria         27.5 %   Ecuador               17.8 %   Guyana       21.0 %   Mauritania      20.2 %
                         Bahrain         17.5 %   Barbados              16.3 %   ...          ...      Saudi Arabia    19.4 %
                         ...             ...      ...                   ...      ...          ...      ...             ...
                         Haiti           23.8 %   Lebanon               29.4 %   Nepal        70.2 %   Gabon           21.9 %
102




      Jordan      1995   Afghanistan     19.7 %   Ghana                 25.4 %   Ghana        16.4 %   Cameroon        19.7 %
                         Saudi Arabia    19.2 %   Honduras              22.3 %   Honduras     13.4 %   Botswana        19.4 %
                         ...             ...      ...                   ...      ...          ...      ...             ...
                         Lebanon         31.6 %   Uganda                38.2 %   Kuwait       14.3 %   United States   21.0 %
      Korea       1987   Cote d’Ivoire   28.4 %   Iceland               17.1 %   Madagascar   11.6 %   Qatar           19.5 %
                         Panama          11.9 %   Ethiopia              10.3 %   Bangladesh   10.0 %   Tanzania        15.2 %
                         ...             ...      ...                   ...      ...          ...      ...             ...
                         New Zealand     35.1 %   Niger                 41.8 %   Chad         68.0 %   Saudi Arabia    43.5 %
      Mexico      1989   Oman            22.3 %   Sierra Leone          31.2 %   Uruguay      15.3 %   United States   20.7 %
                         Panama          19.4 %   Central African Re-   10.4 %   Norway       8.9 %    Bolivia         20.2 %
                                                  public
                         ...             ...      ...                   ...      ...          ...      ...             ...
                         Paraguay        14.8 %   Denmark               19.3 %   Sudan        25. %    Syria           31.4 %
      Malaysia    1987   Kuwait          10.3 %   Ecuador               18.4 %   Mauritius    15.8 %   Guatemala       19.4 %
                         Luxembourg      5.9 %    Switzerland           15.8 %   Togo         11.2 %   United States   18.0 %
                         ...             ...      ...                   ...      ...          ...      ...             ...
                         Mozambique      40.0 %   Rwanda                60.3 %   Chad         26.4 %   Paraguay        36.1 %
      Nigeria     1995   Zambia          20.1 %   Sweden                17.2 %   Myanmar      22.3 %   Denmark         27.0 %
                           Saudi Arabia     17.9 %   Papua New Guinea      13.8 %   Nepal                 21.2 %   Oman            13.4 %
                           ...              ...      ...                   ...      ...                   ...      ...             ...
                           Nepal            7.5 %    Lebanon               39.3 %   Sierra Leone          40.9 %   Australia       18.2 %
      Pakistan      1991   Mozambique       4.8 %    Bahamas               9.7 %    Bangladesh            14.8 %   Iran            14.2 %
                           Tunisia          4.6 %    Malawi                8.3 %    Sri Lanka             14.0 %   Bolivia         8.7 %
                           ...              ...      ...                   ...      ...                   ...      ...             ...
                           Congo            67.1 %   Mauritius             52.4 %   Peru                  29.1 %   Cote d’Ivoire   21.7 %
      Philippines   1986   Iceland          15.5 %   Central African Re-   47.4 %   Chad                  22.8 %   Kuwait          16.6 %
                                                     public
                           Mozambique       11.1 %   ...                   ...      Bahrain               21.6 %   Saudi Arabia    16.2 %
                           ...              ...      ...                   ...      ...                   ...      ...             ...
                           France           94.3 %   Mozambique            40.0 %   Ecuador               32.6 %   Singapore       27.7 %
      Portugal      1993   Zambia           4.8 %    Guyana                26.8 %   Iceland               23.1 %   Iran            22.2 %
                           Mozambique       0.8 %    Singapore             14.3 %   Finland               22.6 %   Canada          21.6 %
103




                           ...              ...      ...                   ...      ...                   ...      ...             ...
                           Cote d’Ivoire    38.4 %   Bahamas               34.0 %   Lebanon               34.7 %   El Salvador     17.3 %
      Thailand      1987   Qatar            22.4 %   Gambia                20.2 %   Oman                  24.3 %   Qatar           16.8 %
                           Ireland          8.5 %    Haiti                 11.8 %   Jamaica               9.7 %    Japan           12.7 %
                           ...              ...      ...                   ...      ...                   ...      ...             ...
                           Oman             54.1 %   Honduras              65.4 %   Peru                  65.6 %   Niger           23.9 %
      Turkey        1989   Sierra Leone     12.3 %   Benin                 19.4 %   Norway                33.3 %   Germany         22.3 %
                           United Kingdom   9.8 %    Canada                9.8 %    Iceland               0.9 %    France          15.8 %
                           ...              ...      ...                   ...      ...                   ...      ...             ...
                           Paraguay         24.6 %   New Zealand           26.8 %   Bangladesh            35.1 %   Saudi Arabia    52.4 %
      Taiwan        1986   Congo            9.0 %    Bahrain               22.6 %   Haiti                 23.6 %   Qatar           16.6 %
                           Niger            7.1 %    Ethiopia              18.8 %   Central African Re-   11.2 %   Nicaragua       13.6 %
                                                                                    public
                           ...              ...      ...                   ...      ...                   ...      ...             ...
                           Sierra Leone     56.1 %   Gabon                 48.6 %   Jamaica               51.2 %   Tanzania        31.4 %
      Venezuela     1990   Kuwait           16.4 %   Oman                  39.1 %   Lesotho               34.2 %   Panama          19.9 %
                           Botswana         14.5 %   Mozambique            12.0 %   Barbados              14.5 %   United States   17.3 %
                            ...          ...      ...               ...      ...                  ...      ...             ...
                            Nepal        16.0 %   Sweden            57.3 %   Honduras             52.2 %   United States   24.0 %
      South Africa   1995   Honduras     15.1 %   Congo             16.5 %   Iceland              39.8 %   Panama          21.9 %
                            Jamaica      14.5 %   Zambia            9.3 %    Barbados             5.4 %    Senegal         16.4 %
                            ...          ...      ...               ...      ...                  ...      ...             ...
                            Belgium      35.6 %   Togo              52.9 %   Dominican Republic   42.3 %   Honduras        31.0 %
      Zimbabwe       1993   Austria      31.9 %   Barbados          20.2 %   Iceland              35.6 %   Benin           30.5 %
                            Mozambique   17.2 %   Trinidad and T.   12.3 %   Lesotho              13.4 %   Iran            10.5 %
                            ...          ...      ...               ...      ...                  ...      ...             ...
104
Appendix D              Instrumental variables for trade
In this section, we outline the construction of the two instrumental variables. Both
instruments are from Egger, Nigai, and Strecker (2019), which provide further details.


D.1    Instrument based on quantitative trade models

The first instrument leverages the general model structure of gravity models in general
equilibrium. These models permit the calibration of country pair-year-specific trade costs
from trade data. Models that use such models include Eaton and Kortum (2002); Anderson
and Wincoop (2003); Arkolakis, Costinot, and Rodriguez-Clare (2012). These models
rely on three specific assumptions. First, producers are either perfectly competitive and
make zero profits or chrge a constant mark-up; prices are independent of the location of
customers. Second, trade costs take the iceberg-form. Third, aggregate expenditure and
its allocation across products can be separated through a two-stage budgeting decision.
These three assumptions imply that bilateral consumption shares towards country j by
consumers in country i in year t, denoted πijt , are multiplicative components that are
exporter-year-specific (ejt ), importer-year-specific (ιit ) and pair-year specific (βijt ) as follows


                                 πijt =   ejt   × ιit    × βijt

   The component ejt is proportional to country j ’s suply potential and broadly captures
production costs and gross-of-tax factor income and is influenced possibly by both capital
and labor taxation. The component ιit is a function of the consumer price index, which
varies across years and countries. The key intuition is that both ejt and ιit may capture
country-year-specific factors, but the country pair-specific component βijt is free of any
country-year specific factor. Instead, βijt captures trade frictions across country-pairs and
time. Egger, Nigai, and Strecker (2019) exploit the multiplicative model structure about πijt
to recover measures of βijt . The authors assume that transaction costs between domestic
sellers and customers is constant, such that βiit = 1. Both the importer-year component
and exporter-year components can then be eliminated by normalizing import and export



                                                105
trade shares by the importer and exporters’ consumption from domestic sellers. In turn,
the product of the normalized shares gives the bilateral fractions of importing-exporting
country-pairs at a point in time:

                                           πijt πjit
                                               ·     = βijt · βjit
                                           πiit πjjt
                                     gravity
    Finally, the sum of these costs Zit      =              [βijt · βjit ] measures total trade frictions
for country i in year t and constitutes the instrument. Note that all exporter-year and
importer-year factors are removed from the instrument. This instrument is valid if the
distribution of trade costs among country-pairs (not its level) is not influenced by e.g.
changes in factor income shares or domestic labor and capital tax revenues. To construct
this instrument requires data on country-pair trade flows: we use the UN COMTRADE
database to construct as large a sample as possible with non-missing values for bilateral
consumption shares.
                                     gravity
    The first-stage regressions with Zit      are reported in Appendix Table A7. The
instrument is slightly stronger when using the log-level of trade or the share of goods-trade
as the endogenous measure of openness.


D.2     Instrument based on global oil prices and transportation distances

Our second instrument exploits spatial heterogeneity across countries in a way that
responds to oil price shocks. To build this instrument we require two parameters: global
oil prices over time, and within-country transportation distances. For the former, we collect
the long-run world price of oil from the well known ‘OPEC Reference Basket’ tracking
crude oil prices (OPEC, 2021). For the latter, we measure transportation distances from the
three largest three cities (from UN DESA, 2019) to their nearest ports, using international
shipping logistics calculators at SeaRates (2021). To calculate the distance of each city to
each port, we look at the map of the city and its distance to port, through the lens of these


      One could measure the distance from city to the nearest sea port, or to the nearest container terminal of
any kind. We made both measurements and ultimately make use of the latter. However, there is usually
little difference when calculating the within-country variance of these measurements across cities, and this
difference does not affect our results.


                                                     106
shipping logistics calculators. We manually enter and record the distances by road for each
city to its nearest port. These distances vary within a country to the extent that a country is
far from a port, and to the extent that cities (far from the port) are also far from one another.
    We then take the variance of the oil price poil           k
                                                t × distance di for each city k in country i

and year t:
                                                        3
                                     pricedist     1
                                    Zit          =           [(pt dk          2
                                                                   i − pt di ) ]
                                                   2   k=1

    This variance is increasing in countries whose principal population centers are far from
the nearest port and, more importantly, from each other, which implies a larger shock to
transportation costs in spread-out (and far from the port) countries than in countries with
concentrated populations (near to port).                It is this shock to trade that our instrument
captures. Alternatively, we can measure the variance in distance and then multiply it by
                                                               pricedist
the global price. The distribution of the variance instrument Zit        across country-years
would not change; the only impact would be a level-shift in factor p. While we consider the
main approach to more closely capture the sensitivity of spatial concentration to shocks in
transportation costs, results remain similar using this alternative measurement approach
(results available upon request).
    This second instrument is very different from the first instrument since it does not hinge
on any theoretical assumptions and is valid under very different assumptions. Specifically,
it relies on the assumption that the distribution of trade-costs induced by global oil price
shocks across countries with different domestic trade networks is not correlated with
contemporaneous changes in factor shares and tax revenues. Importantly, we verify that
the main results are robust to allowing major oil producing countries to be on a separate,
non-parametric time-path throughout the sample period.



     In this sense, our measure of the variance here does not here penalize a country for having its nearest
                                                                                  pricedist
port outside the national borders, or even thousands of kilometers away. The Zit            variance increases
when distance-to-port varies within a country, not when the average distance-to-port increases. One could
perhaps imagine an even stronger trade instrument that accounts for cross-country variance in the average
distance-to-port with an additional parameter. However, in the first-stage results discussed here, we find
sufficient evidence of a trade effect merely on the within-country variance of distance. An additional trade
outcome not discussed here, of course, is on domestic trade between cities. We only measure an international
trade effect, but one would similarly expect domestic trade to decrease as transportation costs increase.


                                                       107
                                pricedist
   Constructing the instrument Zit        as described above, the first stage results are
presented in Appendix Table A7. The instrument is strongly associated with changes in
trade within-country over time and is robust to using various measures of trade openness.




                                          108