Policy Research Working Paper 10722 Two Decades of Top Income Shares in Honduras Giselle Del Carmen Santiago Garriga Wilman Nuñez Thiago Scot Development Economics A verified reproducibility package for this paper is Development Impact Group available at http://reproducibility.worldbank.org, March 2024 click here for direct access. Policy Research Working Paper 10722 Abstract This paper presents distributional national accounts for total income over the period, placing Honduras among Honduras over 2003–2019, using survey microdata, the most unequal countries in the world. Undistributed administrative tax records, and national account aggre- corporate profits are the overwhelming income source at gates. It assembles comprehensive data on formal income the very top of the distribution, highlighting its importance for high-income individuals, including information on in the measurement of income inequality. Finally, using a corporate shareholders, which allows corporate profits to panel of tax records, the paper also documents that not only be assigned to their owners. The estimates suggest a high is inequality persistent, but the same individuals are often and persistent inequality in the country: the top 1 percent observed at the top, suggesting that the observed inequality highest earners received approximately 30 percent of the has deep roots. This paper is a product of the Development Impact 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 tscot@worldbank.org. A verified reproducibility package for this paper is available at http://reproducibility. worldbank.org, click here for direct access. RESEA CY LI R CH PO TRANSPARENT ANALYSIS S W R R E O KI P NG PA 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 Two Decades of Top Income Shares in Honduras* Giselle Del Carmen Santiago Garriga Enodo S.A. CEFIP, IIE-FCE, Universidad Nacional de La Plata ˜ Wilman Nunez Thiago Scot ´ de Rentas (SAR) Servicio de Administracion DIME, World Bank JEL C ODES: H23, H31, H71, I38, J31, J32, J33 K EYWORDS: inequality, top income, administrative registries, Honduras * Del Carmen: gdelcarmen@enodohn.com; Garriga: santiago.garriga@econo.unlp.edu.ar; Nunez: ˜ wponce@sar.gob.hn; Scot: tscot@worldbank.org. This project was made possible by extensive collabo- ration with SAR, the tax authority in Honduras. We extend special thanks to Christian Duarte, Oziel Fern´ andez, Edgardo Enrique Hernandez, Milton Maldonado, Marlon Ochoa, Mario Enrique Pineda, ´ David Pinto, Jose Carlo Bermudez Sanchez and Alesandra D´ ıaz Tovar from SAR for their collaboration. Gabriel Oqueli, Malena Dolcet and Santiago Rodrigo Cesteros provided invaluable research assistance. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily 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. 1 Introduction Research on the measurement, drivers, and consequences of economic inequalities has blossomed since the early 2000s. Despite an active ongoing debate on precise levels and trends, there is an increasing consensus that income inequality has substantially increased in most high-income countries since the 1980s, often accompanied by a de- cline of labor share in total income (Auten & Splinter, 2023, Piketty, 2001, 2003, Piketty & Saez, 2003, Piketty & Zucman, 2014, Smith et al., 2022). For a long time, Latin America was seen as an exception to that global trend of ris- ing income inequality. After the ”lost decade” of the 1980s and the structural adjust- ment in the 1990s (Birdsall et al., 2011), economies in the region were buoyed by the commodity boom of the 2000s and, starting from very high levels, registered large de- creases in measured income inequality in the following decade (Alvaredo et al., 2023, Azevedo et al., 2013, Gasparini & Lustig, 2011). The fundamental reasons behind this trend are still up for debate, but most countries in the region faced strong wage in- creases for workers at the bottom of the distribution coupled with the introduction of robust cash transfer systems (Lustig et al., 2013). Nonetheless, the consensus on falling Latin American income inequality in the last two decades has been challenged recently. Most of the measurement of inequality in the period relied on household surveys, which are known to underestimate top incomes and fail to capture sources of income that are relevant for high-income earners such as dividends. A new set of studies has complemented household survey data with tax registries to provide a more detailed view of top earners in the region. In doing so, they have documented that the share of income flowing to top earners has often been stable – if not increasing – in many countries in the region since the 2000s (Flores et al., 2020, Medeiros et al., 2015). These results are often not contradictory with the observed compression in wage distribution documented previously – instead, they highlight the importance of correctly assessing other sources of income that are often not captured in surveys, particularly income from capital. Accounting for the full national income when measuring inequality often leads to much higher top income shares, leading De Rosa et al. (2022) to suggest that ”either the region is more unequal than previously thought or it is not as rich”. In this paper, we contribute to this debate by creating Distributional National Ac- counts (DINA) for Honduras, a lower-middle income country in Central America. We complement household survey data with administrative tax records and macroeco- nomic aggregates to provide measures of top income shares in the country for the 2003-2019 period. Our key findings are that the shares of pre-tax income flowing to top earners in Honduras are among the highest in the world, and they have been stable over the last two decades. The highest 1% of earners in 2019 received approximately 1 30% of total net national income, the same share received in 2003. The top 0.1% re- ceived approximately 20% of income throughout the period. Broader measures of top income shares show decreases in the period – the top 10% share decreased from 60% in 2003 to 55% in 2019. These levels of income concentration place Honduras among the most unequal countries in the world. According to ?, the richest 10% of the global population currently claims 52% of the world’s income. The Middle East and North Africa (MENA) region stands out as the most unequal region in the world (58%), while Latin America follows closely behind (55%). Moreover, Bergolo et al. (2023) show that in Latin America, the top 1% of earners receive approximately 20% of total income - this represents the highest concentration of income among regions globally. Within the region, only Mexico and Peru, where the top 1% receive 27% and 28%, respectively, register inequality levels similar to the ones we document in Honduras. As we discuss in more detail below, this might be partially driven by the lack of granular income data at the top of the distribution for other countries in the region. The detailed tax records we use allow us to provide a granular description of levels and composition of income at the very top of the distribution. We show that, while labor income represents close to 90% of total income for individuals all the way to the top 1%, within the top 1% highest earners, capital income becomes very important: it represents half of income for those in the top 0.1% and more than 80% for those within the top 0.01% highest earners. We also document that the trends in capital vs. labor share in the top 1% fully drive total income shares - the share of labor in total income decreased from 70% in 2011-2014 to close to 62% by 2019, with barely any change in the shares of the bottom 99% of earners but a large shift toward capital income in the top 1%. This paper contributes to the research agenda on measuring inequality at the top of the distribution in low- and middle-income countries. Many of the studies using tax registries, both in middle- and high-income countries, start from tabulated data of high-income earners provided by governments and use interpolation techniques to estimate income levels and composition at the top of the distribution. Working with SAR, the tax authority in Honduras, we use anonymized panel datasets at the taxpayer level, which lets us depict a very granular description of top earners. Furthermore, the panel data we use allows us to show that not only the level of income concentration remained stable over two decades in Honduras, but also the identity of individuals at the top is highly persistent: we show that more than half of the individuals belonging to the top 0.1% of the distribution in any given year will stay above that level in a three year horizon. Comparisons with the level of individual persistence at the top in other low- or middle-income countries is hard due to lack of data availability, but these rates are closer to 30-40% in high-income countries such as Norway and Australia (Al- stadsæter et al., 2016, H´erault et al., 2022). Additionally, we document the prevalence 2 of women at the top of the income distribution and discuss why female presence in the top 1% highest earners in Honduras seems to be larger than in many high-income countries (Neef & Robilliard, 2021). A second key contribution is the construction and use of a shareholder dataset to assign undistributed corporate profits to firm owners in proportion to their own- ership shares. Distributing corporate income to individuals is a challenging task – Alvaredo et al. (2020) note how being able to ”match the individual profit of firms to the tax record of the owners (...) is very rare” and Piketty et al. (2022), in reference to Bruil et al. (2022), mention their study is ”the most sophisticated, as it links individual business owners to their businesses and hence can assign business-level taxes to indi- vidual owners with great precision”. Other than a handful of exercises in high-income countries such as Norway (Alstadsæter et al., 2016), Chile (Fairfield & Jorratt De Luis, 2016) and Uruguay (De Rosa & Vil´ a, 2023), to our knowledge we provide the most detailed account of the importance of business income at the top of the distribution for a middle-income country. Finally, this paper builds on the broader DINA initiative, providing a new data point on top income shares using high-quality tax records combined with survey and national accounts data.1 Recent studies have provided valuable insights on income inequalities across the Middle-East (Alvaredo et al., 2019, Assouad et al., 2018), Asia, Africa (Czajka, 2020) and Europe (Bruil et al., 2022). In Latin America, only 11 countries are considered to have ”high-quality” measures of income inequality by the World Inequality Database project, due to the lack of available tax records.2 Recent research on top income inequality in the region includes ? on Colombia, ? on Costa Rica, ? on Mexico, ? on Uruguay, ? on Brazil and Flores et al. (2020) on Chile. As an illustration, the current level of top 1% income share in Honduras in the WID is 18.3%, obtained by ”regional imputation”.3 As with other recent studies in the region, we show that new estimates using detailed tax records might provide a very different perspective on the levels and trends of top income shares in Latin America. 2 Context Honduras is a lower-middle-income country in Central America with approximately 9.8 million inhabitants. Economic activity is linked to low value-added agricultural and manufacturing production, which is closely tied to the performance of its key 1 The existing literature in Honduras exclusively exploits survey data (see for instance Ham (2011), World Bank (2016) and S´ anchez & Urrea (2019)) 2 Still, this figure includes countries where there is no micro-data but tax-tabulations e.g., Brazil, and interpolation techniques are used to estimate incomes at the top. 3 Both values, the number of countries identified as with ’high-quality’ data, and the Honduran top income share, are obtained from the WID website (October 2023). 3 trading partners and the inflow of remittances. The country is particularly suscepti- ble to external shocks due to its heavy reliance on imports and exposure to extreme weather events like hurricanes, droughts and floods. In the past two decades, the Honduran economy has shifted from subsistence agri- culture to wage work in the service sector. While the share of jobs in agriculture dropped 10 percentage points (36% to 26%), the share of jobs in services increased from 41% to 52% (Michel & Walker, 2020). Nonetheless there is persistently high informal- ity across all sectors. Informal employment is among the highest in Latin America and the Caribbean (LAC), with approximately 4 out of 5 workers in the informal sector. The prevailing narrative of economic performance in the country over the last two decades is that the above-average economic growth at 4-5% per year registered, when compared to other countries in LAC, was not enough to raise welfare levels for a pop- ulation that grew at 2-3% yearly: poverty rates remain very high for regional levels, with almost half of Hondurans living on less than $6.85 a day (2017 PPP) (Figure A.1). Moreover, since the 1960s, Honduras has been diverging from wealthier countries, with per capita income (in constant US$) equal to only 4% of U.S. levels, below the shares of Guatemala (7.2%), El Salvador (6.8%) and Costa Rica (21.3%) (Hernandez Ore et al., 2015). At the same time, survey measures show a decrease in income inequality measured by the Gini coefficient of household income from .6 in 2005 to .5 by 2019 – which is similar to that observed by Brazil, a country that also reduced poverty rates by half in the same period. 3 Data sources The inequality measures discussed in this paper rely on comprehensive income dis- tributions for Honduras over the 2003-2019 period. To obtain these distributions, we combine several sources of information described in detail below. 3.1 EPHPM: Household survey The EPHPM (Encuesta Permanente de Hogares de Prop´ ´ ositos Multiples) is Honduras’ an- nual household survey since 1990. It surveys approximately 6,000 households, is na- tionally representative and allows for disaggregations for the two main cities (Distrito Central and San Pedro Sula), other urban and rural areas. This survey is used to track and monitor labor market outcomes as well as socioeconomic indicators of the Hon- duran population, including official poverty and inequality estimates. Similar to most household surveys, it includes detailed information on households’ dwelling characteristics, access to basic services, asset ownership, household compo- sition and individual characteristics. More importantly, it collects various sources of 4 income-related information including income from wages, self-employment, trans- fers, pensions and imputed rent.4 In addition, it allows identifying informal labor income, i.e., coming from an unregistered labor relation that does not pay taxes, and is not subject to any labor regulation. High levels of informality in the Honduran labor market limit the ability of ad- ministrative systems to document individual incomes. As of 2019, approximately 80 percent of the Honduran labor force was informal relative to an average of 54 percent in Latin America (Gasparini & Tornarolli, 2009, World Bank, 2020). This highlights the importance of relying on household survey information when measuring income outside the top of the distribution. In January 2020, the Government of Honduras updated its official income measure- ment methodology for the period 2014-2019, with the aim of creating a more accurate poverty profile for the country. The new official income aggregate includes improve- ments in data cleaning, annualized labor income for salaried workers receiving 13th and 14th salaries (commonly known as aguinaldo and catorceavo in Honduras) and im- puted rent for home owners. On average, these changes increased official household income by 16 percent, driven primarily by imputed rent (INE 2021). These changes have inevitably resulted in a break in comparability with previous years. In order to extend comparability for our analysis, we apply the new income measurement methodology for 2003-2013 (see Table A.1). 3.2 Administrative tax microdata We use a range of datasets that record income sources of individuals and corporations. These are administrative records from SAR (from the Spanish acronym for Servicio on de Rentas or Revenue Administration Service), the tax authority in de Administraci´ Honduras. 3.2.1 Personal Income Tax (PIT) data Honduras has a dual personal income tax system, with income from labor and mixed sources (operation of non-incorporated firms and service provision) being taxed at a progressive schedule and capital income (bank interests, dividends, capital gains) taxed at a flat rate. For the period we study, income from capital was taxed at a flat 10% rate. Income from labor and mixed sources face a progressive schedule with marginal rates between 15 - 25%, including an exemption threshold: any income below that 4 It is also relevant to document which incomes EPHPM does not cover: in the period we study, information on capital income was very limited, with specific questions on interests present in some but not all years, and no other detailed questions on other sources of capital income such as dividends. As we show below using tax data and, as has been widely documented (Alvaredo & Gasparini, 2015), capital income is particularly important at the very top of the income distribution. 5 level is not subject to PIT and also does not trigger withholding requirements. The exemption threshold throughout the period we study was typically set at around one and a half times the corresponding annual minimum wage, and this mostly defines the population for whom we observe labor and mixed income in this data-source.5 Unlike in other countries, individuals whose income is entirely withheld at source, such as capital income and wages, are not required to file the yearly PIT declaration. Income tax declarations are only required when individuals earn income that is not withheld at source, such as some forms of service provision and income from non- incorporated commercial enterprises. In order to assign income to individuals, we use both self-declared information on PIT declarations as well as third-party information through withholding mechanisms. We use datasets at the taxpayer level for each year in the period 2003-2019, includ- ing all possible income sources observed by the tax authority. Recovering information from several different data sources within the tax administration is possible since tax- payers are uniquely identified in all datasets using a personal identification number (RTN, for Registro Tributario Nacional in Spanish). We present a summary of taxpayer-level data availability in Table 1 where we high- light the following facts. First, the maximum number of individuals observed in the tax data is approximately 650,000 in 2019, representing less than 15 percent of the estimated adult population in that year. That is a direct result of the high levels of informality and of the high exemption rate for PIT, among other things.6 As discussed below, even in those years we only use a fraction of the administrative data to com- plement survey observations, since many of these observations in the tax data have very low incomes (e.g., only small declared amounts from interest in bank accounts). Second, information from withholding sources is very important: in 2019, for exam- ple, only 116,000 taxpayers filed the yearly income tax declaration. For all the other taxpayers observed in our data, some type of withholding allowed us to observe their income. Third, the total number of taxpayers observed each year steadily rose in the period, in part due to the increase in the yearly tax filing: it rose from less than 25,000 declarations in 2003 to over 130,000 in 2018 – the decrease in 2019 is likely driven by the effects of the Covid-19 pandemic.7 Furthermore, 2011 marks a change in the avail- ability of withholding information: before that period, withholding declarations were 5 In 2019, for example, taxpayers with annual income below L 158,995 were exempt, whereas the average annual minimum wage was approximately L 113,000. The exemption threshold in Honduras is particularly high when compared to other countries in the region, as documented by Bergolo et al. (2023). 6 Higher-income Latin American countries such as Chile and Uruguay have much lower informality rates and consequently tax data covers a vast swath of the total population (see Flores et al. (2020) and Burdin et al. (2014)). 7 Since the lockdown in the country started in March 2020 and the deadline to file income taxes was postponed several times, from April initially to July and finally August. 6 annual and filed on paper, while after that they became monthly and digital. For those reasons, we do not have access to withholding declarations before 2011. This limits our ability to observe important sources of income at the top, including all capital in- come and wage income for individuals whose only source of labor or mixed income is wages (and therefore do not file income tax declaration). Below we discuss in more detail how we proceed to impute incomes at the top of the administrative data for those years. 3.2.2 Corporate Income Tax & Shareholder Registry One key conceptual contribution of the Distributional National Accounts (DINA) pro- gram is to use a broader measure of income to compute inequality, considering not only income flowing directly to households but also to corporations and the govern- ment, under the assumption that eventually households are the final beneficiaries of all income generated in an economy (Alvaredo et al., 2020, Zwijnenburg, 2019). One large component of national income not flowing to households is undistributed cor- porate income. The importance of considering income retained at corporations at the household level is illustrated by the significant changes in measurement of inequal- ity and income share in the US around the 1986 tax reform, when incentives to retain income in corporations substantially changed (Nelson, 2016, Smith et al., 2022). In close collaboration with the tax authority in Honduras, we have created a com- prehensive shareholder registry for all active firms in the country. We provide more detail in Appendix B and explain briefly here the available data. We use a series of administrative records to recover shareholders and their participation in each corpo- ration, and then process the data to deal with inconsistencies and produce a final dataset at the firm-by-shareholder level, including the participation of each share- holder. Whenever direct shareholders are corporations, we apply an iterative process to recover individual shareholders and their shares through up to five levels of own- ership. This is a static dataset, reflecting the most updated data available to the tax authority as of early 2023. We then use information on corporate income tax (CIT) declarations to compute firms’ profits every year and assign the resulting profits to their shareholders using our static shareholder registry. In recent years, we observe approximately 30,000 cor- porations filing income taxes and over 50,000 identifiable domestic shareholders to whom we assign their share in corporate profits. 3.3 Macroeconomic aggregates from National Accounts Finally, we use macroeconomic aggregates from the World Inequality Database (WID) to match our aggregates from micro-data to National Accounts (NA) – such that, in 7 every year, the total amount of income observed in our dataset used to compute in- equality measures matches the totals in the national accounts.8 We use data from WID instead of Honduras’ national accounts estimates since the latter are not disaggregated enough to match the components used by the DINA approach. We provide more de- tail about these aggregates in Appendix C. 4 Building the income distribution 4.1 Unit of observation We define the unit of observation in all the analysis as individual adults (20+ years). Therefore, our definition of individual income departs from the one commonly used in household surveys where the sum of all household income is divided among all members, either equally or using some kind of weighting (the OECD guidelines, for example, weight every child as 0.3 adult). Since Honduras is an individual taxation country and there is no joint-filing by couples or households, computation of income at the top of the distribution is only possible at the individual level (“individualistic adult” approach per DINA guidelines) . 4.2 Income definition Our main set of results refers to estimates of the full distribution of pre-tax income, along the lines of the DINA guidelines (Alvaredo et al., 2020): income accruing to the owners of factors (labor and capital), before taxes but after distribution of social secu- rity benefits. We now provide a brief overview of the construction of pre-tax income distributions in the survey and tax administrative records, and include more details in the appendices. 4.2.1 Income in survey data We use survey micro-data to compute individual income as the sum of four compo- nents: wages, mixed income, capital income and imputed rent. Wages include salaried work from all activities, including additional benefits for formal employees and wages from informal employment. We also include in the wage component all pensions and retirement income. Mixed income includes all income from self-employment, includ- ing employment in the agricultural sector.9 Capital income is defined as the sum of 8 For a detailed discussion on how the WID approach differs from another initiative to bridge the gap between micro- and macro-data on inequality, the OECD-Eurostat Expert Group on Disparities in National Accounts (EG DNA), see Zwijnenburg (2019). 9 As we document in Table A.1, in the period 2004-2013 self-employment income from agricultural activities was measured using a separate module. The microdata for these specific modules is not 8 rental and interest income, when these are available - not all years in the survey data include those components. Finally, imputed rent is estimated at the household level for the entire period using different methodologies, and we divide it equally among all adults in the household. Notably, our measure of pre-tax income excludes several income components that are particularly important for low-income households – di- rect government transfers, transfers from other household and remittances are all ex- cluded from our income definition, despite being usually included in broader income measures used to calculate official poverty and inequality estimates. 4.2.2 Income in tax records We use micro-data on individual, non-incorporated taxpayers (Personas Naturales) to complement the income distribution from household surveys. We provide a thor- ough explanation of the process of generating pre-tax income from tax records in Ap- pendix D; here we summarize the key decisions taken in building the income aggre- gate. Total individual income in tax records is comprised of four components: wages, mixed income; capital income and undistributed corporate profits. Whenever we ob- serve both self-declared and third-party information for a given income source, we take the maximum value available. Wages are computed as the maximum value ob- served for each taxpayer between the third-party withholding information and the self-declared PIT (when individuals file the yearly declaration). It is adjusted to in- clude additional components (13th and 14th salaries) that are non-taxable up to cer- tain amounts. Mixed income is similarly obtained by comparing self-declared values in PIT forms and third-party withholding, and taking the maximum value.10 Income from capital is defined as the sum of dividends, interests and rental income, all avail- able through third-party withholding forms.11 Finally, we assign to each individual- year observation the sum of undistributed yearly profits in the corporations where they are shareholders, in proportion of their shares. available, only the total income measured in them. In the period when these modules were in place, income from agricultural self-employment were extremely large and inconsistent with other periods. In Appendix F we explain how we winsorize observations and adjust income from these modules to replicate aggregates similar to other years. 10 We also compare values of declared revenue from commercial activities with third-party informa- tion including VAT withholding and exports. Whenever third-party values are larger, we add these additional undeclared revenues to the individual. Since these are not net income but gross revenue, in these cases we impute implicit costs by using median profit margins declared by taxpayers with activity in the same economic sector. 11 Following most papers in the literature, we do not include income from capital gains in our in- come aggregate, particularly since also assigning undistributed corporate profits might result in double counting. 9 4.2.3 Some caveats When considering the pre-tax income aggregate used to measure inequality in our study, some caveats should be kept in mind. First, similarly to all studies that com- bine survey data and tax records, our income aggregates are not fully comparable between the two sources. Survey data does not capture all income sources (dividends, for example) nor captures the same income sources every year (interest income is only included in the survey instrument for some periods). On the other hand, tax records do not include pension income or imputed rent, and we do not try to impute those for tax record observations. Since capital income is very small at the bottom of the distri- bution and imputed rent very small at the top, we do not expect our results to be very sensitive to those choices. The extent to which pension income is relevant at the top of the distribution is less clear – it is certainly much less relevant than in high-income countries with stronger pension systems and an older population, but we are unable to precisely assess the relevance in Honduras in the absence of administrative data on pension distribution. In Appendix B, we also discuss how assigned undistributed profits are computed from a taxable profits measure. Due to meaningful changes in CIT forms over the period we study, we opt to use a measure that is more comparable over the period instead of using a broader measure of total profits. 5 Methods & Diagnostics Once we obtained distributions of income in both the survey and tax administrative data, our next step is to combine these two sources to obtain a comprehensive distribu- tion of income. In this section we describe the approach used to obtain this combined distribution and provide a series of diagnostics on the steps taken. 5.1 Merging survey and tax records Similar to other countries where labor informality is predominant and where the thresh- old to file PIT is high, tax records will only accurately capture the income of a small share of the total population. Our main goal is thus to complement the top of the income distribution observed in the survey with tax record information – where we often observe much higher incomes than those reported in the survey data. In order to do so, we follow the method of replacing survey income at the top with income from tax records. Whereas some previous work has chosen to replace specific top quantiles such as the top 1% or 10% (Burkhauser et al., 2016), we follow the same approach as Czajka (2020): for each year, we replace the survey with tax data 10 starting from the point where the quantile functions from both sources cross.12 We first collapse the tax records in up to 37 quantiles - nine quantiles between the top 10% and top 1%; nine quantiles for the top 1% to top 0.1%; 9 quantiles between the top 0.1% and 0.01%; and the highest ten quantiles up to the top 0.001%.13 As mentioned before, for the period 2003 - 2010 we have very limited information on third-party reporting, and are thus unable to capture a large share of income from individuals who do not file PIT. We adjust the level of income in top quantiles in those years to match the ratio between income declared in PIT forms and total income for the 2011-2019 period, where we observe all sources of income.14 We then compute, for a range of income levels, the share of individuals with incomes above that level in tax and survey data – as a rule, for low income levels we observe more individuals with higher incomes in the survey data, so this ratio is below unit, and at very high income levels that ratio often far exceeds unit. We choose as the merging threshold the lowest point when that ratio equals one, and replace survey data with tax records above that income level.15 We present a diagnostic assessment of our merging procedure in Table 2. In column (1) we illustrate the total reference adult population in each year, computed using the household survey. In 2019 there were approximately 5.4 million adults over 20 years- old in Honduras out of a population of over 9 million individuals. For that year, the merging point between the two sources was L465,000, the equivalent of the income of the top 0.6% highest-earning individuals in both datasets. Despite the fact that admin- istrative data includes over 600,000 individuals for that year, our merging process led us to use fewer than 40,000 taxpayers to complement the income distribution obtained from survey data – this share ranges from 0.4% - 1% in the 2012-2019 period, and is significantly lower at 0.1 - 0.2% before that, when the availability of tax records is more limited. 5.2 Matching aggregates to national accounts After the previous step, we have a comprehensive distribution of income for adults in Honduras, for each year in the period 2003-2019, constructed from microdata. It has 12 We recognize the limitations of this procedure as laid out in Blanchet et al. (2018). In future versions of this working paper we expect to also present results using alternative methods to complement survey data with tax records. 13 The top decile of the adult population in Honduras represents 300,000 - 450,000 individuals, de- pending on the year. In some years, our entire sample of individuals in tax records is lower than those numbers, so we are only able to construct a smaller share of quantiles. In practice that does not impact our results, since we document that we never replace the survey below the top 1% level. 14 We provide more information on this adjustment in Appendix G. 15 In Figure A.4, we illustrate these ratios across income levels for three years. We note that the ratio is not monotonic – since we collapse the tax data in quantiles and the survey data is sparse at high income levels, sometimes the ratio decreases as income increases and even dips below one. But the trend of more income in the tax data at higher levels is clear in all years presented, so we replace survey with tax records at the first income level at which the ratio crosses the value of one. 11 been extensively documented, nonetheless, that these income aggregates built from microdata often fail to capture the same levels of income registered in national ac- counts: De Rosa et al. (2022), for example, show that income in survey data comple- mented with administrative tax records is often only 40-60% of the total pre-tax net national income in selected Latin American countries. We show the same gap exists for Honduras: in Figure 1a, we plot the ratio between total Net National Income (NNI) in national accounts and in our microdata. The gap is relatively stable over the years: each year, income in national accounts is 30-50% larger than what we observe using microdata.16 Furthermore, we also document that this gap is vastly different across components of total income. In Figure 1b, we plot for each year the same ratio between national accounts and microdata aggregates for four components of income: wages, mixed income, imputed rent and capital income. While aggregate wages are systematically larger by approximately 20-30% each year, mixed income is often very similar in both sources, particularly for recent years when deviations are at the order of 1 - 5%, and imputed rents are often smaller in the national accounts by 10-20%. The largest driver of differences are income from capital – we observe three to five times more income from capital in the national accounts compared to microdata, even when accounting for undistributed corporate profits assigned to individuals.17 The last step to obtain the full income distribution is to multiply each of the four in- come components described above by these adjustment factors, such that total income in the microdata, as well as each of its components, matches aggregates in national accounts.18 We note that one key implication of the particularly large gap between capital income in the microdata and national account is that, since capital income is predominant at the very top of the distribution, our adjustment to national accounts will substantially increase measurements of concentration of income at the top when compared to measurements that consider only microdata. 16 We provide levels of income in national accounts data in Appendix C. 17 When adjusting microdata to national accounts, we sum distributed capital income and corporate profits in the microdata, and match it to the sum of household property income and corporate profits in the national accounts. We discuss in more details these decisions in Appendix C; the main reason for making the joint adjustment is that, when compared to other countries in the region and worldwide, the share of corporate profits in total income in the National Accounts is very large in Honduras and while property income is often negative. The sum of the two components is more in line with other countries. 18 Following the DINA guidelines, we also distribute production taxes proportionally to income, so it has no bearing on inequality measures but it does affect income levels. 12 6 Main Results 6.1 Levels and composition of income at the top We start this section characterizing the levels and composition of total income, ad- justed to national accounts aggregates, at the top of the distribution. In Figure 2, we present the average income for individuals within the top 10% highest earners in 2019, where we separate them in 37 bins – 9 bins for those between the top 10% but below the top 1% (≈ 50,000 individuals per bin); 9 bins for those in the top 1% but below the top 0.1% (≈ 5,000/bin); 9 bins for those in the top 0.1% but below the top 0.01% (≈ 500/bin); and 10 bins for those at the top 0.01% (≈ 50/bin). We present income levels both in local currency (Lempiras) as well as in USD PPP. In 2019, the average income of individuals just in the top 10% of the distribution was L213,000 or USD 20,000 per year. This level is rather close to the exemption thresh- old for Personal Income Tax (PIT) filing in 2019, suggesting that 90% of the adults in Honduras receive incomes below the exemption level.19 Average incomes increase fast once we move towards those at the very top – those just above the top 1% re- ceive L590,000 or USD 56,000 per year, those at the top 0.1% receive almost five times as much at L3 million or USD 285,000; and the 500 highest earners at the top 0.01% receive more than L30 million or USD 3.1 million in 2019.20 Consistent with find- ings from other countries in Latin America, the highest earners in Honduras receive incomes that are not too different from top earners in high-income countries: in Ta- ble A.3 we document the average income for the top 10%, top 1% and top 0.01% in several countries across the globe. While the average income of the top 10% in Hon- duras is substantially below most countries in Latin America, the average income of those at the top 1% and 0.01% are much more in line with other countries in LAC. The composition of income is also substantially different within the top 10% high- est earners in the country. In Figure 3a, we decompose total income into income from wages, mixed income, distributed capital income and undistributed corporate prof- its.21 We first note that within the top 10% highest earners but below the top 1%, income composition is mostly stable at 80% from wages and 20% from mixed income. Within the top 1%, income from capital steadily increases as the wage share decreases – and within the top 0.1% the entire increase in capital income is driven by undis- tributed profits, which steadily rises to represent close to 80% of total income within the top 0.01%. It is noteworthy that the most important source of income at the very top of the distribution is usually completely absent from personal income tax decla- 19 Notethis is total income, not taxable income, including imputed rent and likely also informal incomes not observed by the tax authority. 20 We document these income levels for selected quantiles in Table A.2 21 We present the average shares for the period 2011-2019, to reduce noise from individual years. 13 rations - because this is income realized within corporations and not distributed to individuals, it is only observable when connecting corporations to their shareholders. In the next section, we describe how our inequality measures change when we cannot observe direct ownership and instead infer it from distribution of dividends. In Figure 3b, we present a simplified version of the decomposition, splitting income between labor and capital using the convention of assigning 70% of mixed income as labor and 30% as capital. That provides a symmetric view of income composition within the top 10% – labor income represents 90% of of total income for those at the top 10%; at the top 0.1% labor and capital are equally split; and at the very top labor income tends towards zero.22 6.2 Top income shares Our key results relate to the level and evolution of top income shares in Honduras since the early 2000s. We summarize the share of income flowing to different groups of top earners in Figure 4. We highlight two features of those findings. First, the remarkable stability of those shares over the years. The share of income flowing to the top 1% highest earners, for example, starts at just below 29% in 2003, falls to approximately 25% in the period 2008 - 2013 and returns to approximately 29-30% at the most recent period. Similarly, the share of the top 0.1% stays at the range of 18% between 2003 and 2015, before climbing slightly to just above 20% in 2017-2019. These are particularly striking given all the economic and political changes faced in the country in the last two decades, including a substantial move of labor force from agriculture to the service sector; a 40% increase in the minimum wages in 2008; and the recession and political crisis that followed the 2009 military coup that deposed the president. In the following section, we document that not only the top income shares in Honduras remained stable, but the identity of individuals at the top of the distribution is highly persistent – suggesting the existence of deep roots of the inequality scenario in the country. Second, we note these numbers place Honduras among the countries with highest inequality in the world. In Table 3, we provide top shares for a series of countries for which the same distributional national accounts approach was undertaken, so results are broadly comparable with our measurements. With almost 30% of income flowing to the top 1%, Honduras ranks among the highest inequalities in the world - slightly above Mexico (28%) and the Dominican Republic (29%). Countries with higher top 22 In Figure A.2a, we show how the labor share of total income in National Accounts fell by approx- imately 6 p.p. between 2011 and 2019, from 70% to 64%. This change in aggregate labor share is fully explained by developments within the top 1%, as we show in Figure A.2b. The labor share for those at the bottom 99% was stable at approximately 80% throughout the period, while it fell by over 10 p.p. for those at the top 1%, as the participation of capital income increases from about 50% to over 60%. 14 shares include Mozambique and Central African Republic, both with 31% of income flowing to the top 1% – but in both countries these measurements are obtained from adjusted survey data, so are arguably less precise. It is crucial to note, nonetheless, that this type of detailed inequality accounting is not available for the majority of countries, so it is not clear how many of them would fare in comparison with Honduras. Its re- gional peers like Guatemala, El Salvador and Nicaragua do not have similar estimates – the only country in the region with similar estimates is Costa Rica which registers a much lower 21% share flowing to the top 1%. We also include in the table the esti- mated top shares for Honduras currently available on the WID platform – the share of the top 10%, at 52%, is not too dissimilar to our results; the share of the top 1%, how- ever, is only 18% or 12 p.p. below our estimates. This further suggests that inequality in other countries in the region, for which detailed accounting exercises have not been performed, is likely much higher than current estimates. To illustrate the importance of the series of adjustments that leads us to our main estimates using the DINA guidelines, in Figure 5 we present the share of income flow- ing to the top 1% highest earners measured using three different distributions. First, we show the share of the top 1% highest earners when we use only survey data, with- out any further adjustments. In that series, the share fell from approximately 15% in the early 2000s to less than 10% by 2019, showing a steady decline in the period and a level of top share three times smaller than our estimates using national account adjustments. We also present the top 1% share using the microdata complementing survey data with tax records, but before any adjustments to national accounts. For recent years, this series roughly doubles the top 1% share from 10% to 20%, but still falls 5 - 10 p.p. short of the estimates using National accounts information. This gap is explained in large part by the discrepancy between microdata and national accounts’ aggregates of capital income. 7 Additional Results - Tax panel data In this section we explore more in depth the top of the income distribution using tax records. We focus on three dimensions – the importance of undistributed corporate profits at the top; the dynamics and persistence of individuals at the top; and gender gaps among highest earners. For all the analyses in this section, we use the income distribution observed in the tax records, without any adjustments for macroeconomic aggregates or merging with survey microdata. For a more detailed description of of income levels and composition in tax records see Appendix E. 15 7.1 The role of undistributed corporate profits Here we document the key role played by undistributed corporate profits at the top of the income distribution, and how using microdata on the identity of shareholders matters for our inequality estimates. We start by documenting the fact that, within the top 1% of earners when consid- ering only distributed income (i.e. excluding undistributed profits), there is a strong positive correlation between income and the probability of being a shareholder. In Fig- ure 6a, we show that around 10% of individuals just within the top 1% of distributed income are large shareholders of a firm filing CIT in 2019 - we define large sharehold- ers as those owning at least 5% of a firm. This probability increases to close to 30% for those in the top 0.1% and reach approximately 60% for those within the top 0.01%. These figures show that individuals that have high distributed incomes are also more likely to be shareholders in corporations - and thus be ”assigned” undistributed profits in our comprehensive income measure. Nonetheless, when we rerank all individuals considering their total income inclu- sive of undistributed profits (orange line in Figure 6a), corporate ownership is much more prevalent at the top. The share of corporate owners is 20 p.p. higher, at 40%, for those just within the top 0.1%, and reaches more than 80% for those within the top 0.01%. This is not only driven by individuals already declaring high personal income and being assigned additional undistributed profits. In Figure 6b, we compute for each bin at the top the share of taxpayers with distributed income below the annual minimum wage - these are individuals who would not be classified as top earners otherwise, but only make the cut because we observe their ownership of corporations and assign them profits. Between 10 - 20% of individuals within the top 0.1% have distributed income below the MW. This is one indicator that directly assigning indi- viduals undistributed profits from the corporations they own generates meaningful re-ranking of the identity of top income earners, and also that a meaningful share of these earners do not receive large distributed incomes. While we have established that observing corporate ownership data matters for the identity of those on the top, an important remaining question is whether it mat- ters for estimates of top income shares using the DINA methodology. Clearly, adding undistributed corporate profits to total realized income leads to large increases in mea- sured income concentration - Alstadsæter et al. (2016), for example, show that it more than doubles the share of top 0.1% income in Norway. But the usual DINA approach also adjusts total income to include national account measures of corporate profits – it usually does so by distributing these profits in proportion to realized dividends, in- stead of directly measuring corporate ownership. To assess whether these matter, we replicate our results ignoring the corporate ownership data and instead distributing 16 profits from NA in proportion to dividends.23 We present results in Figure 7, which documents the share of income flowing to the top 1% and 0.1%, using our baseline and alternative methodologies. For the top 1% share, the gap is approximately 3 - 4 p.p. across the period - in recent years our baseline measure is approximately 28 - 30% vs. 25 - 26% when we do not use the corporate ownership data. However, the gap is much larger when we consider the top 0.1% share. Our baseline estimates suggest the top 0.1% earners receive between 18 - 20% of total income in recent years, whereas those figures are in the range 10 - 12% using the alternative methodology. These results sug- gest that using realized dividend income to adjust for undistributed corporate profits might be an incomplete exercise, and detailed data on corporate ownership might pro- vide a important lens for measurement of top income inequality. 7.2 Persistence at the top Our key results show a substantial persistence of the share of total income flowing to the highest earners individuals across time. A distinct but equally important question relates to the identity of those individuals at the top of the income distribution (Acciari erault et al., 2022). In one scenario, even if the share of et al., 2022, Auten et al., 2013, H´ income flowing to the top is stable, the identity of those highest earners changes sub- stantially as top firms are challenged by incumbents; as the demand for skills change in the job markets; and as innovators gain access to new income generating activities. Alternatively, the same individuals occupy the top of the income distribution over time – either because the majority of their income are returns to wealth that persist over time, because their successful firms are unchallenged in the market, or because opportunities for other to achieve high incomes are hampered by other structural fac- tors. Our panel data of individual taxpayers allow us not only to measure income levels and concentration at the top, but also to evaluate dynamics at the top – given that we observe an individual at the top in a given period, how likely it is they will persist at the top in future years? Evaluating the stability of individuals at the top of the distribution is also an indirect way to assess the quality of the administrative data - while a high degree of economic mobility might lead to large changes in the identity of top earners over time, on a shorter horizon we would expect to observe some degree of stability. Our main exercise, presented in Figure 8, is to compute the probability that indi- 23 To be precise, we first merge survey and tax microdata without including undistributed profits in the income definition of tax records. When expanding the joint distribution from microdata to match national accounts, we follow the same procedure as in our baseline measures, matching total distributed capital income in the microdata to the sum of household property income and corporate profits in the national accounts. 17 viduals in the top 1% or top 0.1% of the income distribution in year t will still be in the same top group in year t+1 (panel a) and t+3 (panel b). We take the same approach as Alstadsæter et al. (2016) for Norway and present results using two different income distributions: including and excluding undistributed corporate profits from the calcu- lation of individual incomes. In Figure 8a, we first show that 70-80% of individuals in the top 1% in any given year will stay in the top 1% in the following year, regardless of the income aggregate we use. This is consistent with the fact that undistributed profits represent a relatively small share of total income from most of the individuals at the top 1%. The gap between the two series is much larger when we consider per- sistence at the top 0.1% – the probability of staying in the group is close to 70% when we consider the series including undistributed profits and falls to 60-65% when only distributed income is considered. When considering persistence over three years in Figure 8b, close to 60% of individuals stay in the top 1% while that figure falls to 55% in the top 1% when considering undistributed profits (and is below 50% when only distributed income is included).24 The level of persistence in the top 1% is quite similar to other settings, while for those at the top 0.1% is significantly higher. Alstadsæter et al. (2016) show that in Norway the persistence at the top 1% is approximately 60% over one year and 50- 55% over three years; H´ erault et al. (2022) document similar levels of 65-75% over one year and 40-50% over three years in Australia. For those at the top 0.1%, where undistributed corporate profits matter much more in our settings, we measure much higher persistence levels – the probability of observing the same individuals at the top 0.1% over three years for example is 55-60% in Honduras vs. 30-40% in Norway and Australia. 7.3 Female participation at the top An additional dimension of income inequality we are able to observe using adminis- trative tax records is the gender gap: what is the representation of women at the very top of the income distribution? In Figure 9, we present the share of female taxpayers in the top quantiles of the distribution, pooled for the 2014-2019 period. We show the female share for two in- come definitions, excluding or including undistributed corporate profits. The overall 24 While the previous exercises use a discrete measure of persistence at the top, asking if individuals remained at the top 1% or 0.1% earners over time, an alternatively measure is to ask how the rank of an individual in the entire distribution changes over time, given their ranking at some initial period – that is to compute the rank-rank correlation of individual income over time. We present that exercise in Figure A.3, where we show the median rank in 2014 and 2018 for each bin of income ranking in 201125 . What we document is a very strong, positive correlation between initial ranking and median ranking four and seven years after the initial measurement – individuals at any given position in the initial period are likely to stay in a similar position over the years. 18 findings are quite similar regardless of the distribution: women are 40% of taxpayers in the top 1% – the share is stable between the top 1% - 0.1%, and then starts to decline at higher incomes until the range 20-30% within the top 0.01%. These shares of female presence among the top 1% of earners are high by interna- tional comparison. Atkinson et al. (2018) document that the share of women in top 1% are consistently in the range of 15-25% in several high-income countries. In Sweden, Boschini et al. (2017) document that the share of women in the top 1% in 2013 was 23% when capital gains were included but only 16% when it was not included. Neef & Robilliard (2021) document that the share of women in the top 1% of wage earners is in the range of 20-30% in Brazil, US, France and Costa Rica. Our estimates of 40% female participation within the top 1% are therefore higher than in all those settings. These differences between female participation among top earners in Honduras and other countries could be driven by several reasons. First, most of the countries for which granular data on top earners exists are much richer than Honduras – the threshold to be at the top 1% in Sweden in 2013 was approximately USD 140,000 PPP, for example, five times more than in Honduras where that value is stable around USD 30,000 since 2014. An yearly income of USD 140,000 PPP would be equivalent to be- ing at the top 0.05% of the distribution (using tax records income) in Honduras. In that sense, it is not strictly comparable to consider gender gaps among the top 1% in countries with vastly different income levels. Boschini et al. (2017) also document that women showing up at the top of the dis- tribution in Sweden are particularly likely to have received large capital gain incomes, which are often one-offs - they are also less likely to persist at the very top of the distri- bution than men. We first show that women at the top 1%, conditional on income, have a larger share of wage and capital income, and a lower share of mixed income, when compared to men. In Table 4, we compute gender gaps in the share of wage, mixed and capital income for those individuals at the top 1%, conditional on log-income. We document that, on average, women have a larger share of wage (3.5 percentage points) and capital income (1.1 p.p.), and a lower share of mixed income (4.6 p.p.) - the latter is usually driven by individual entrepreneurs, suggesting that the gender gap might be more relevant in those activities. We can also provide a deeper dive into the gender gaps in income composition within the very top earners. In Figure 10, we investigate whether the composition of income at top quantiles differ between women and men. For those below the top 0.1%, the composition between labor and capital income is remarkably similar: both groups receive approximately 85-90% of total income from labor and less than 10% from capital. That figure changes, nonetheless, for those within the top 0.1% - the increase in the share of capital income vs. labor income is more pronounced for women than for men. At the top 0.01%, the share of capital for women is above 50% vs 40% for 19 men, and the gap stays at those levels at higher income quantiles. Therefore, while for the majority of those at the top 1% women and men have similar income composition in terms of labor vs. capital, women’s income is more dependent on capital income at the very top (Atkinson et al., 2018, Bobilev et al., 2019). 8 Conclusion In this paper we have used rich administrative tax records, including detailed share- holder list of all corporations, to offer new measures of top income shares in Honduras. Our main results can be summarized as follows. First, we document very large top income shares for Honduras – the top 1% receive approximately 30% of total pre- tax net national income. We provide evidence not only that the inequality levels in Honduras are among the highest in the world (for which detailed measurements are available), but also that they persist with little change over two decades. Second, we also show the predominance of capital income at the top of the distribution, and partic- ularly of undistributed corporate profits. The ability to assign corporate profits to in- dividual shareholders is only possible due to the construction of a shareholder dataset for the majority of corporations in the country. We also show that this information matters: using standard methods in the literature, where undistributed profits are dis- tributed according to other distributed income, substantially underestimates income concentration within the top 0.1%. Finally, the panel dataset nature of our tax records also allows us to document a high persistence of the same individuals at the top of the distribution over time. Our estimates are obtained following the most recent literature on measuring top incomes, rendering them broadly comparable to several similar efforts. Throughout the paper, we carefully document the methodological decisions we take and caveat the limitations of our approach - more generally, all estimates of top incomes combin- ing survey data, tax records and macroeconomic national accounts carry a significant degree of uncertainty. Nonetheless, all our estimates are consistent in showing that real inequality levels are much higher and persistent over time than what survey data alone would indicate. The debate over the inequality levels and trends in Latin Amer- ica would benefit from additional similar efforts throughout the region. 20 References Acciari, Paolo, Polo, Alberto, & Violante, Giovanni L. 2022. And Yet It Moves: Inter- generational Mobility in Italy. American Economic Journal: Applied Economics, 14(3), 118–163. Alstadsæter, Annette, Jacob, Martin, Kopczuk, Wojciech, & Telle, Kjetil. 2016 (Dec.). Accounting for Business Income in Measuring Top Income Shares: Integrated Accrual Ap- proach Using Individual and Firm Data from Norway. Alvaredo, Facundo, & Gasparini, Leonardo. 2015. Recent Trends in Inequality and Poverty in Developing Countries. Handbook of Income Distribution, 2(Jan.), 697–805. Alvaredo, Facundo, Assouad, Lydia, & Piketty, Thomas. 2019. Measuring lnequality in the Middle East 1990–2016: The World’s Most Unequal Region? Review of Income and Wealth, 65(4), 685–711. Alvaredo, Facundo, Atkinson, Anthony, Chancel, Lucas, Piketty, Thomas, Saez, Em- manuel, & Zucman, Gabriel. 2020. Distributional National Accounts (DINA) Guidelines : Concepts and Methods used in WID.world. Tech. rept. 201602. World Inequality Lab. Alvaredo, Facundo, Bourguignon, Franc ¸ ois, Ferreira, Francisco H. G., & Lustig, Nora. 2023 (Oct.). Seventy-five Years of Measuring Income Inequality in Latin America. Tech. rept. Inter-American Development Bank. Assouad, Lydia, Chancel, Lucas, & Morgan, Marc. 2018. Extreme Inequality: Evidence from Brazil, India, the Middle East, and South Africa. AEA Papers and Proceedings, 108(May), 119–123. Atkinson, Anthony, Casarico, Alessandra, & Voitchovsky, Sarah. 2018. Top incomes and the gender divide. The Journal of Economic Inequality, 16(2), 225–256. Publisher: Springer. Auten, Gerald, & Splinter, David. 2023. Income Inequality in the United States: Using Tax Data to Measure Long-Term Trends. Journal of Political Economy, Nov. Publisher: The University of Chicago Press. Auten, Gerald, Gee, Geoffrey, & Turner, Nicholas. 2013. Income Inequality, Mobility, and Turnover at the Top in the US, 1987-2010. American Economic Review, 103(3), 168–172. Azevedo, Joao Pedro, Inchaust, Gabriela, & Sanfelice, Viviane. 2013. Decomposing the Recent Inequality Decline in Latin America. Policy Research Working Papers. The World Bank. 21 ˜ Bergolo, Marcelo, Londono-V´ ıo. 2023. Tax progressivity elez, Juliana, & Tortarolo, Dar´ and taxing the rich in developing countries: lessons from Latin America. Oxford Review of Economic Policy, 39(3), 530–549. Birdsall, Nancy, De La Torre, Augusto, & Caicedo, Felipe Valencia. 2011. The Washing- ton Consensus: Assessing A “damaged Brand”. Page 0 of: Ocampo, Jos´ e Antonio, & Ros, Jaime (eds), The Oxford Handbook of Latin American Economics. Oxford University Press. Blanchet, Thomas, Flores, Ignacio, & Morgan, Marc. 2018. The Weight of the Rich: Im- proving Surveys Using Tax Data. Tech. rept. hal-02878315. HAL. Bobilev, Roman, Boschini, Anne, & Roine, Jesper. 2019. Women in the Top of the Income Distribution – What Can We Learn From LIS-Data? LIS Working papers, Sept. Number: 773 Publisher: LIS Cross-National Data Center in Luxembourg. Boschini, Anne, Gunnarsson, Kristin, & Roine, Jesper. 2017. Women in Top Incomes – Evidence from Sweden 1974-2013. Working Paper Series, Aug. Number: 5/2017 Publisher: Stockholm University, Swedish Institute for Social Research. Bruil, Arjan, Essen, C´ ¨ eline van, Leenders, Wouter, Lejour, Arjan, Mohlmann, Jan, & Rabat´e, Simon. 2022. Inequality and Redistribution in the Netherlands. CPB Dis- cussion Paper, Mar. Number: 436 Publisher: CPB Netherlands Bureau for Economic Policy Analysis. Burdin, Gabriel, Esponda, Fernando, & Vigorito, Andrea. 2014. Desigualdad y sectores de altos ingresos en Uruguay : un an´ alisis en base a registros tributarios y encuestas de ıodo 2009-2011. Tech. rept. hogares para el per´ Burkhauser, Richard V., H´erault, Nicolas, Jenkins, Stephen P., & Wilkins, Roger. 2016 (Feb.). What has Been Happening to UK Income Inequality Since the Mid-1990s? Answers from Reconciled and Combined Household Survey and Tax Return Data. Tech. rept. 21991. National Bureau of Economic Research, Inc. ote d’Ivoire: 1985-2014. World Inequality eo. 2020 (May). Income Inequality in Cˆ Czajka, L´ Lab Working Paper. HAL. a, Joan. 2023. Beyond tax-survey combination: inequality and De Rosa, Mauricio, & Vil´ the blurry household-firm border. The Journal of Economic Inequality, Apr. De Rosa, Mauricio, Flores, Ignacio, & Morgan, Marc. 2022 (Aug.). More Unequal or Not as Rich? Revisiting the Latin American Exception. 22 Fairfield, Tasha, & Jorratt De Luis, Michel. 2016. Top Income Shares, Business Profits, and Effective Tax Rates in Contemporary Chile. Review of Income and Wealth, 62(S1), S120–S144. eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1111/roiw.12196. Flores, Ignacio, Sanhueza, Claudia, Atria, Jorge, & Mayer, Ricardo. 2020. Top Incomes in Chile: A Historical Perspective on Income Inequality, 1964–2017. Review of Income and Wealth, 66(4), 850–874. Gasparini, L., & Tornarolli, L. 2009. Labor informality in Latin America and the Caribbean: Patterns and trends from household survey microdata. Revista Desar- rollo y Sociedad, 63. Gasparini, Leonardo, & Lustig, Nora. 2011. The Rise and Fall of Income Inequality in Latin America. Page 0 of: Ocampo, Jos´ e Antonio, & Ros, Jaime (eds), The Oxford Handbook of Latin American Economics. Oxford University Press. Ham, Andr´ ´ del ingreso en Honduras: Un an´ es. 2011. La distribucion alisis de las prin- ˜ 1990. Revista Centroamericana de Ciencias Sociales ( cipales tendencias desde los anos RCCS ), 8(2), 123–165. Hernandez Ore, Marco Antonio, Sousa, Liliana D., & Lopez, J. Humberto. 2015. Hon- duras: Unlocking Economic Potential for Greater Opportunities. Systematic Country Diagnostic. World Bank. erault, Nicolas, Hyslop, Dean, Jenkins, Stephen P., & Wilkins, Roger. 2022. Rising H´ top-income persistence in Australia: Evidence from income tax data. Review of In- come and Wealth, Dec., roiw.12628. Lustig, Nora, Lopez-Calva, Luis F., & Ortiz-Juarez, Eduardo. 2013. Deconstructing the Decline in Inequality in Latin America. July. Publisher: World Bank, Washington, D.C. Medeiros, Marcelo, Souza, Pedro Herculano Guimar˜ abio aes Ferreira de, & Castro, F´ ´ Avila de. 2015. The stability of income inequality in Brazil, 2006-2012: an estimate using income tax data and household surveys. Ciˆ ´ Coletiva, 20(Apr.), encia & Saude 971–986. Publisher: ABRASCO - Associac ´ Coletiva. ˜ o Brasileira de Saude ¸a Michel, Veronica, & Walker, Ian. 2020. Honduras Jobs Diagnostic. Feb. Publisher: World Bank, Washington, DC. Neef, Theresa, & Robilliard, Anne-Sophie. 2021. Half the Sky? The Female Labor In- come Share in a Global Perspective. World Inequality Lab – Working Paper N° 2022/22. 23 Nelson, Susan. 2016. Paying Themselves: S Corporation Owners and Trends in S Cor- poration Income, 1980-2013. ecle. 0 edition edn. Paris: Piketty, Thomas. 2001. Les Hauts revenus en France au XXe si` Grasset. Piketty, Thomas. 2003. Income Inequality in France, 1901–1998. Journal of Political Economy, 111(5), 1004–1042. Piketty, Thomas, & Saez, Emmanuel. 2003. Income Inequality in the United States, 1913–1998. The Quarterly Journal of Economics, 118(1), 1–41. Piketty, Thomas, & Zucman, Gabriel. 2014. Capital is Back: Wealth-Income Ratios in Rich Countries 1700–2010. The Quarterly Journal of Economics, 129(3), 1255–1310. Piketty, Thomas, Saez, Emmanuel, & Zucman, Gabriel. 2022. Rethinking Capital and Wealth Taxation. World Inequality Lab – Working Paper N° 2022/18. Smith, Matthew, Yagan, Danny, Zidar, Owen, & Zwick, Eric. 2022. The Rise of Pass- Throughs and the Decline of the Labor Share. American Economic Review: Insights, 4(3), 323–340. S´ anchez, Jos´ ´ e Carlo Bermudez, & Urrea, Eduardo Enrique Pineda. 2019. La Dis- ´ sectorial del ingreso regional en Honduras para el per´ tribucion ıodo 2001 - 2018. Econom´ on (E&A), 10(2), 30–59. ıa y Administraci´ World Bank. 2020. The Gradual Rise and Rapid Decline of the Middle Class in Latin America and the Caribbean (English). Report. Zwijnenburg, Jorrit. 2019. Unequal Distributions: EG DNA versus DINA Approach. AEA Papers and Proceedings, 109(May), 296–301. 24 Figures and tables Table 1: Data availability on tax records Observations Individual yearly Witholding forms Other sources Undistributed Tax data tax filings Wages Services Dividends Interest Capital Gains Rental Third-party info Corporate income (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) 2003 38,778 24,260 0 0 0 0 44 0 0 13,639 2004 50,126 25,445 0 0 0 0 165 0 175 23,354 2005 58,136 26,556 0 0 0 0 229 0 302 29,676 2006 62,439 28,159 0 0 0 0 200 0 290 21,957 2007 67,808 31,695 0 0 0 0 281 0 356 34,440 2008 71,471 33,342 0 0 0 0 371 0 255 35,395 2009 73,549 34,024 0 0 0 0 301 0 2,384 34,800 2010 80,770 37,363 464 399 16 68 355 103 5,783 38,046 2011 252,833 41,138 61,823 31,036 8,861 94,305 427 124 32,027 37,655 2012 324,752 43,454 81,640 42,022 52,095 102,308 551 130 39,054 39,252 2013 399,716 46,568 93,508 44,360 88,386 144,197 550 145 48,241 37,625 2014 396,242 52,631 90,443 50,063 84,094 142,983 10,131 179 50,248 41,292 2015 481,806 82,378 112,976 56,726 96,209 185,636 17,917 324 51,332 52,227 2016 519,660 104,144 100,765 57,120 119,188 209,695 19,806 374 57,098 49,834 2017 510,873 124,176 105,944 57,421 76,893 214,265 21,302 409 60,524 52,048 2018 606,656 132,134 116,777 60,464 144,161 253,841 30,729 444 107,240 54,611 2019 652,459 115,923 118,158 56,089 182,194 296,584 34,497 443 105,061 55,031 Note: This table presents counts of observations in tax records, by year and source of income in- formation. Column (1) presents the total number of records in each year. Column (2) presents the number of individuals filing the yearly income tax declaration. Columns (3) through (8) present the number of individuals with withholding records for income from wages, services, dividends, interest, capital gain and rental, respectively. Column (9) presents the number of individual with some other record of third-party information, including exports, sales to the government and sales to firms declaring detailed VAT information. Finally, column (10) presents the number of individ- uals being assigned undistributed corporate profits. Since records might be available for the same individual from several different sources, the numbers of each source do not sum to the total obser- vations in one year. Before 2011, all withholding forms were presented on paper and have not been digitized, so we do not observe those records in our sample. 25 Table 2: Diagnostic of merging process between survey and tax records 20+ population (survey) Tax data observations Share obs from tax data (%) Merging point (L) (1) (2) (3) (4) 2003 3,165,874 3,169 0.10 595,000 2004 3,347,641 6,708 0.20 445,000 2005 3,489,479 6,992 0.20 505,000 2006 3,583,144 7,180 0.20 580,000 2007 3,715,919 7,446 0.20 670,000 2008 3,917,356 1,959 0.05 1,440,000 2009 4,037,917 4,041 0.10 755,000 2010 4,209,107 3,369 0.08 1,110,000 2011 4,377,335 13,171 0.30 520,000 2012 4,464,598 17,931 0.40 480,000 2013 4,588,901 41,667 0.90 320,000 2014 4,494,971 22,587 0.50 475,000 2015 4,671,707 28,198 0.60 475,000 2016 4,814,880 38,834 0.80 425,000 2017 4,942,283 44,886 0.90 395,000 2018 5,199,944 52,520 1.00 365,000 2019 5,389,481 32,532 0.60 465,000 Note: This table summarizes the merging process between household survey microdata and admin- istrative tax recordes. Column (1) presents the number of adults (20+) represented in the survey data, our underlying population of interest. Column (2) presents the number of individual observa- tions from tax data that are used in the final microdataset, i.e. that replace high-income individuals from the survey, while column (3) presents the same number as a share of the reference population. Finally, Column (4) presents the merging point between the two surveys - below that income level the final datasets includes only survey information, while above it includes only tax records. 26 Table 3: International comparison of top incomes shares (2019) Top 0.1% Top 1% Top 10% Pivotal benchmark countries China 7% 15% 42% France 4% 11% 34% India 8% 22% 57% South Africa 6% 19% 65% United States 9% 19% 46% High income share Angola 11% 26% 58% Bahrain 9% 24% 56% Central African Rep 13% 31% 65% Mozambique 13% 31% 65% Yemen 8% 24% 58% LAC (good data quality) Argentina 4% 14% 46% Brazil 5% 20% 57% Chile 8% 25% 61% Colombia 4% 19% 59% Costa Rica 4% 20% 55% Dominican Rep 6% 21% 52% Ecuador 3% 12% 42% El Salvador 3% 14% 44% Honduras (WID data) 5% 19% 52% Mexico 6% 28% 65% Peru 6% 21% 55% Uruguay 3% 13% 43% Note: This table presents the income share of the top 0.1%, top 1% and top 10% highest earners in selected countries in 2019. Data is from the World Inequality Database (WID). The figures for Honduras in this table are from the WID dataset, not the estimates in this paper. 27 Table 4: Differences in income composition at top 1% by gender (1) (2) (3) (4) Share wage Share mixed Share dist. capital Share Corp. Female taxpayer 0.048*** -0.047*** 0.003*** -0.005*** (0.00) (0.00) (0.00) (0.00) Constant 0.147*** -0.419*** -0.188*** 1.445*** (0.03) (0.01) (0.01) (0.02) Observations 285,190 285,186 285,190 285,190 R-Squared 0.01 0.02 0.01 0.21 Mean Dep Var 0.77 0.16 0.04 0.03 Note: This table presents results from regressions restricted to the individuals at the top 1% of the income distribution, pooled for the 2014-2019 period. The dependent variables are the share of total income from wages (column 1), from mixed income (column 2) and from distributed capital income (column 3) and from undistributed corporate profits (column 4), and we report the coefficient on the dummy for female taxpayers. All regression include year fixed-effects and standard errors are clustered at the individual level. Shares are winsorized between 0 and 1, since negative income can create extreme values. P-value significance levels: * 0.10 ** 0.05 *** 0.01 28 Figure 1: Adjustment factors to National Account (a) Adjustment factor to Net National Income (NNI) (b) Adjustment factors by component of NNI Note: This figures present the adjustment factors applied to the combined household survey + tax record microdata, in order for income aggregates and its components to match national accounts (NA). Panel (a) shows that, each year, net national income (NNI) in NA are 40-60% higher than in the combined microdataset. Panel (b) shows that the gaps between microdata and NA aggregates are vastly different for each income source – we adjust each income source to match the equivalent NA definition, so that aggregates in the final micro dataset match both total NNI as well as its main components. 29 Figure 2: Income levels at the top (2019) - Adjusted for National Accounts Note: This figure presents the average income for each quantile of the distribution after adjusting to national account aggregates in 2019, both in local currency (Lempiras) as well as USD PPP. It includes 37 quantiles for the top 10% highest earners - 9 quantiles for those at the top 10% but below the top 1%; 9 quantiles for those at the top 1% but below 0.1%; 9 quantiles for those between 0.1% and 0.01%, and 10 quantiles for those at the top 0.01%. In Table A.2 we provide income levels for other years in our sample. 30 Figure 3: Composition of top incomes (2011 - 2019) (a) Detailed income composition (b) Labor vs. Capital income Note: These figures present the composition of total income for each quantile of the distribution after adjusting to national account aggregates. It pools data for 2014-2019 to reduce noise, and omits two data points between top 0.03 - 0.02% due to outliers. It includes 37 quantiles for the top 10% highest earners - 9 quantiles for those at the top 10% but below the top 1%; 9 quantiles for those at the top 1% but below 0.1%; 9 quantiles for those between 0.1% and 0.01%, and 10 quantiles for those at the top 0.01%. Panel (a) decomposes total income in four sources - wages, mixed income, distributed capital income and undistributed profits. Panel (b) decomposes total income between labor and capital income, using the approach of attributing 70% of mixed income to labor and 30% to capital. 31 Figure 4: Evolution of total income shares (2003-2019) - Adjusted for National Ac- counts Note: This figure presents the share of total income accruing to the top 0.1%, 1%, 5% and 10% of individuals with highest income, each year. Income components are adjusted National Accounts, such that the sum of total income in the microdata matches total national income. 32 Figure 5: Evolution of total income shares (2003-2019) - Alternative measurements Note: This figure presents the share of total income accruing to the top 1% each year, using differ- ent measures. The ”National Accounts Adjusted” estimates are our baseline inequality measures. ”Survey + tax data” refer to inequality using micro-data only, after we merge tax records to the top of the survey distribution but before adjustments to national accounts’ aggregates. Finally ”survey data” refers to inequality measured using only survey microdata, before replace top earners with tax records. When using survey only data, we extend our top income estimates back to 2001. 33 Figure 6: The presence of shareholders at the top (a) Share of large shareholders (b) Share of taxpayers with distributed income below Minimum Wage (MW) Note: Panel (a) in this figure presents the share of individuals in top quantiles who are large share- holders of corporations, defined as having a participation of at least 5% in total shares. We present this participation for the income distribution before assigning undistributed corporate profits (in blue) and after doing so (in orange). Panel (b) presents the share of taxpayers in each top quantile of the income distribution (including undistributed corporate profits) whose distributed income (that is, excluding undistributed profits) is below the yearly minimum wage. Both figures pool data for the 2014 - 2019 period. In panel (b) we exclude one top quantile (top 0.04%) since a concentration of shareholders from one large company creates an extreme outlier (close to 30% of taxpayers in that bin are large shareholders). 34 Figure 7: Top income shares - absent corporate shareholder microdata Note: This figure presents the share of income flowing to the top 1% and the top 0.1% using two different approaches. The first one is our baseline (orange and green lines), where we impute undis- tributed corporate profits to individuals according to their ownership in corporations, merge tax records to survey data and then scale capital income (undistributed corporate profits + distributed capital income) to match the sum of corporate profits and net operating surplus of households in national accounts. The second approach is to ignore the shareholder information, which is usually unavailable information in similar studies, and instead rescale distributed capital income to match the sum of the components described above in national accounts. 35 Figure 8: Stability at the top of the distribution (a) Over one year (b) Over three years Note: These figures report the probability that an individual observed in top quantiles of the dis- tribution is still observed in the same quantiles in the following year (panel a) and three years af- terwards (panel b). We report these measures for two top quantiles (top 1% and top 0.1%) and also for two different income measurements (including corporate profits, which is our baseline income measure, and excluding corporate profits). These figures refer to tax records only, and do not incor- porate survey data or adjustment to national accounts. 36 Figure 9: Female participation at the top of the income distribution Notes: This figure presents the share of female taxpayers in each bin within the top 1% highest earners, using two different income definitions. The blue line refers to income before the assignment of undis- tributed corporate profits for shareholders, while the orange line refers to the comprehensive income definition including corporate profits. Data is pooled for 2014-2019 period. 37 Figure 10: Income composition at the top - by gender Note: This figure presents the share of labor and capital income for each quantile, separately for women (dark colors) and men (transparent). Individuals are ranked by income including undis- tributed corporate profits. The labor vs. capital definition uses the approach of attributing 70% of mixed income to labor and 30% to capital. Data is pooled for 2014-2019 period. 38 A Appendix Figures and Tables Figure A.1: Growth, poverty and inequality trends (2011-2019) 0.70 10 0.20 0.30 0.40 0.50 0.60 Honduras Poverty headcount ($6.85/day) Honduras Nicaragua 5 Nicaragua GDP growth (%) El Salvador LAC Brazil El Salvador 0 Brazil 0.10 0.00 -5 2000 2005 2010 2015 2020 2000 2005 2010 2015 2020 Year Year (a) GDP growth (b) Poverty ($6.85/day) 0.60 0.55 Brazil Honduras Gini index 0.50 Nicaragua 0.45 0.40 El Salvador 2000 2005 2010 2015 2020 Year (c) Gini coefficient Note: These figures present trends for GDP growth, poverty rate (measured using the poverty line of $6.85/day) and inequality measured by the Gini coefficient (using household survey income) for Honduras and selected other Latin American countries. Data is from the World Bank Poverty and Inequality Platform (PIP). 39 Table A.1: Relevant features of household survey over time (2001-2019) Year Self-emp. Ag. Imputed 13/14vo Formality Self-emp. Census Module rent identifier (contributes reference weights to IHSS) period 2001 Estim ptile 3 months 2001 2002 Estim ptile 3 months 2001 2003 Estim ptile 3 months 2001 2004 ✓ Estim ptile 3 months 2001 2005 ✓ Estim ptile ✓ 3 months 2001 2006 ✓ Estim ptile ✓ ✓ 3 months 2001 2007 ✓ Estim ptile ✓ ✓ 3 months 2001 2008 ✓ Estim ptile ✓ ✓ 3 months 2001 2009 ✓ Estim ptile ✓ ✓ 3 months 2001 2010 ✓ Estim ptile ✓ ✓ 3 months 2001 2011 ✓ Model ✓ ✓ 3 months 2001 2012 ✓ Model ✓ ✓ 3 months 2001 2013 ✓ Model ✓ ✓ 3 months 2013* 2014 Model ✓ ✓ 6 months 2013 2015 Model ✓ ✓ 6 months 2013 2016 Model ✓ ✓ 6 months 2013 2017 Model ✓ ✓ 6 months 2013 2018 Model ✓ ✓ 6 months 2013 2019 Model ✓ ✓ 6 months 2013 Note: Own elaboration based on yearly questionnaires and micro-data from EPHPM. ”Self-emp. Ag. module” refers to the existence of additional module on income from agricultural activities from self-employment. ”Imputed Rent” refers to the method used for estimating imputed rent: for the period 2011-2019 it is estimated from a model that imputes rent for owners using characteristics from households that rent and declare rental value; for 2001 - 2010 imputed rent is estimated using average shares of imputed rent, by percentile, from 2011 - 2019. ”13/14vo identifier” refers to the existence of question to workers whether they receive 13th and 14th additional salaries. ”Formal- ity” refers to the existence of a question asking for formality status of workers. ”Self-employment reference period” presents the differences in recall period for income from self-employment. Finally, ”Census weights” show the changes in Census weights used for the survey (* pre-census weights for 2013). 40 Table A.2: Income levels - adjusted to National Accounts 2012 2014 2016 2019 Top 10% 167 178 187 213 [19] [19] [19] [20] Top 1% 504 526 554 590 [57] [57] [55] [56] Top 0.1% 2,031 2,166 2,905 3,004 [232] [235] [288] [285] Top 0.01% 16,357 19,357 26,423 33,582 [1,866] [2,100] [2,616] [3,182] Top 0.001% 385,428 397,163 567,116 642,270 [43,961] [43,090] [56,147] [60,862] Note: This table presents the average income for each quantile, across years, in local currency (1,000s Lempiras) in the main line and in 1,000s USD ppp in brackets. These averages are for small bins, the size of which vary (top 10% refers to those between the top 10% and 9%; top 1% between the top 1% - 0.9%; top 0.1% between the top 0.1% - 0.09%; top 0.01% between the top 0.01% - 0.009%; and top 0.001% for the highest 0.01% earners. These figures refer to the final distribution of income, after adjusting for national accounts’ aggregates. 41 Figure A.2: Labor and Capital shares over time (2011-2019) (a) Labor vs. Capital income (b) Labor vs. Capital income - Top 1% vs. Bottom 99% Note: These figures present the evolution of participation of labor and capital in total income, for the 2011-2019 period. Panel (a) presents the total share of labor and capital income, using the final dataset adjusted for national accounts. Panel (b) presents these shares separately for the bottom 99% and the top 1%. 42 Figure A.3: Rank-rank correlation 2011 vs. 2015/2018 Note: This figure presents, for each quantile at the top of the distribution in 2011, the median quantile that the same individuals were observed in 2015 (orange line) and 2018 (blue line). The quantiles are 500-individual wide (meaning that the quantile 0 includes the 500 highest earners in 2011, all the way to the 500 with rank 49,500 - 50,000.). The median rank in 2015 and 2018 is calculated conditional on individual appearing on tax records. 43 Figure A.4: Survey-Tax merging methodology Note: This figure presents the ratio between the number of observations in the admin tax records and in the survey data with income above each level, for years 2015, 2017 and 2019. A ratio of one means that in both sources, the number of individuals with income above that level is the same. Survey observations with income above that level are replaced with tax record observations when constructing our comprehensive income distribution. 44 Table A.3: International comparison of top average incomes (2019 PPP USD 000’s) Top 0.1% Top 1% Top 10% Pivotal benchmark countries China 1,687 358 103 France 2,429 637 190 India 1,047 273 72 South Africa 1,365 457 155 United States 7,050 1,581 379 High income share Angola 2,614 605 135 Bahrain 5,919 1,567 359 Central African Rep 314 74 15 Mozambique 396 92 19 Yemen 324 100 24 LAC (good data quality) Argentina 3,697 1,313 427 Brazil 1,308 509 143 Chile 2,927 946 232 Colombia 1,033 447 138 Costa Rica 1,247 580 157 Dominican Rep 2,086 700 176 Ecuador 483 208 70 El Salvador 421 207 66 Honduras (WID data) 514 199 56 Mexico 1,862 828 192 Peru 1,205 448 115 Uruguay 1,282 519 173 Note: This table presents the average income of the top 0.1%, top 1% and top 10% highest earners in selected countries in 2019 (PPP USD 1,000s). Data is from the World Inequality Database (WID). The figures for Honduras in this table are from the WID dataset, not the estimates in this paper. 45 Table A.4: World Income Inequality data (2019) Country Data quality Imputation Pivotal benchmark countries China 3 Surveys and tax data France 5 Surveys and tax microdata India 3 Surveys and tax data South Africa 3 Rescaled fiscal income United States 5 Surveys and tax microdata High income share Angola 1 Adjusted surveys Bahrain 1 Adjusted surveys Central African Rep 1 Adjusted surveys Mozambique 1 Adjusted surveys Yemen 1 Adjusted surveys LAC (good data quality) Argentina 4 Surveys and tax data Brazil 4 Surveys and tax data Chile 4 Surveys and tax data Colombia 4 Surveys and tax data Costa Rica 4 Surveys and tax data Dominican Rep 4 Surveys and tax data Ecuador 4 Surveys and tax data El Salvador 4 Surveys and tax data Honduras (WID data) 0 Regional imputation Mexico 4 Surveys and tax data Peru 4 Surveys and tax data Uruguay 4 Surveys and tax data Note: This table presents WID ”data quality index” and the method of calculating inequality mea- sures for selected countries. Data is from the World Inequality Database (WID). B Corporate Income Tax & Shareholder Registry In this section we explain the process of assigning undistributed profits to individuals. 46 B.1 Computing corporate profits We start by creating a dataset of corporate income tax (CIT) declarations for every year in the period 2003 - 2019. We compute firms’ total profit as the difference be- tween total revenue and total allowable costs. We include non-taxable revenue in that computation, so our measure of profit is broader than simply taxable profits. It also includes profits for firms that are exempt from income tax or have tax credits - these are separately documented after firms’ compute their tax liability. In Table B1 we present some key descriptive statistics of that dataset. The number of corporations filing income taxes increases substantially in the period, from less than 11,000 in 2003 to over 33,000 in 2019. In Column (2), we present the aggregate profits declared by these firms. In the period 2016-2019, aggregate corporate profits fluctu- ate in the range of L45 - L55 billion. Noteworthy changes in the levels of aggregate profits happen in 2015, when profits increase by 60%, and then in 2018, when they fall by 20% from 2017. These large changes in aggregate declared profits are likely driven by the introduction and then phasing-out of a corporate minimum tax that damp- ened the incentives of large corporations to report low profit margins (cite minimum tax paper). Note that when we compare these values, computed from CIT declara- tions, with macroeconomic aggregate of property income flowing to corporations in Table C1, our numbers are much lower - in 2019, for example, corporate income is L113 billion in macro aggregates vs. L56 billion in our dataset. These discrepancies are likely driven both by the precise definition of these aggregates in each dataset, as well as the inherent challenge of estimating these quantities. We also highlight three other features of our data. First, a non-negligible share of corporations declare losses – 20-30% of total CIT filings every year. That implies that when we assign undistributed corporate profits to individuals, some of them will be assigned negative incomes - in some cases, these losses are large enough to completely compensate for higher incomes and ”exclude” individuals with very large distributed incomes from the top of our distribution. Second, before distributing corporate profits to shareholders, we net out the total amount of dividends distributed by the corpo- ration in the same fiscal period. In Column (4) we report the aggregates of corporate profits net of distributed dividends. On average, corporations distribute 15-25% of declared profits as dividends each year. Note that we do not observe distributed divi- dends before 2011, so pre- and post-dividend profits are equal. For that reason, macro adjustments for corporate profits will look differently in both periods, but incomes should be consistent over time after matching macroeconomic aggregates. 47 Table B1: Aggregates of corporate profits from tax records Number Aggregate profits Share with Profits net Profits matched Profits matched of Corporations negative profits of dividends to individuals to individuals w/ valid ID (L billion) (L billion) (L billion) (L billion) (1) (2) (3) (4) (5) (6) 2003 10,972 8,037 0.30 8,037 5,758 4,433 2004 11,774 11,909 0.30 11,909 8,497 5,465 2005 12,399 17,359 0.29 17,359 13,338 7,369 2006 13,247 23,431 0.29 23,431 19,192 8,548 2007 14,628 26,171 0.27 26,171 23,322 8,507 2008 15,674 21,221 0.29 21,221 18,479 4,485 2009 16,304 18,831 0.34 18,831 13,991 2,941 2010 17,370 22,988 0.27 22,976 15,486 5,064 2011 18,293 24,991 0.25 20,059 19,979 6,048 2012 19,147 24,015 0.23 14,397 18,657 6,534 2013 20,258 26,440 0.22 18,825 23,572 8,263 2014 21,594 27,655 0.23 23,102 26,729 8,871 2015 24,935 43,335 0.22 37,766 34,424 11,884 2016 27,146 52,673 0.20 45,407 40,541 16,371 2017 29,648 56,270 0.21 43,822 48,091 19,476 2018 32,183 45,868 0.24 32,242 41,520 11,299 2019 33,704 56,362 0.24 46,312 47,958 17,347 Note: This table presents information on corporate profits from tax records. Number of corporations refers to the number of firms filing CIT each year. Aggregate profits is the sum of taxable profits (in billion Lempiras) from all CIT filers. Share with negative profits is the percentage of filers declaring negative income. Profits net of dividends is the total amount of profits minus total amount of dis- tributed dividends. Profits matched to individuals refers to the amount of profits net of dividends that can be attributed to some shareholders using the shareholder database - this number can be larger than column (4) since profits can be negative. Finally, ”profits matched to individuals with valid ID” refers to the total profits net of dividends that are attributed to a shareholder with valid national ID - that excludes foreign corporations and foreign shareholders, for example, which do not have a national ID. B.2 Creation of shareholder list Once we compute net-of-dividends corporate profits for each period, we need to as- sign them to each shareholder in proportion to their equity in firms. We do so by using a detailed list of corporate shareholders and their shares. No consolidated and updated shareholder list of corporations exists in Honduras, so we worked with the Tax Authority to build one. We use two main datasources: Annex B of the ”professional relations table”, a document firms file when they are in- corporated, listing all relevant parties in the firm and including participation of share- holders; and the withholding form for dividends DMR 113, under the assumption that individuals receiving dividends are shareholders even if not listed in the Annex B. We emphasize that the quality of Annex B information is far from perfect: there is no for- mal validation of the information imputed, including validating the tax id (RTN) of 48 shareholders; amounts of shares sometimes are missing and/or do not sum to 100%; and the information is mostly static, with few firms ever updating their shareholder list. Under these constraints, we build the best approximation for a comprehensive list of shareholders in all corporations in Honduras. This list of shareholders include not only individual shareholders, but also corpo- rations that own shares in other corporations. In order to assign profits to individuals, we perform a recursive exercise of identifying corporations that are owned by other corporations; identifying the individual shareholders of these other corporations; and then assigning their participation in the original firm. Our resulting dataset includes approximately 800,000 relationships at the shareholder- corporation level, representing over 115,000 corporations and 200,000 individual share- holders. As presented in Table B1, in recent years only ≈ 30,000 corporations file in- come taxes every year – our shareholder list includes many firms that no longer exist, but that does not affect the distribution of income in any given year since that requires a firm to file a CIT declaration. We then proceed to merge the shareholder list to yearly CIT declarations from cor- porations, creating a database at the individual-corporation-year level, restricting our sample to corporation-years where firms filed a CIT declaration. If we focus on 2019, the most recent year in our data, we observe 31,879 firms (95% of those filing CIT) with their respective 103,000 shareholders. In terms of aggregate profits, we document in column (5) that in most year we are able to distribute 80 - 90% of total profits to some shareholder in our database (ratio between columns (5) and (4))1 . However, a much smaller share of total profits can be distributed to shareholders with a valid domestic identification number, as we document in column (6) – in 2019, only L14 billion in profits can be assigned to individuals with a national ID, out of L40 billion assigned to some shareholder. The main reason for that discrepancy (which can be massive in some years like 2018) is that some of the most profitable corporations in the country are owned by foreign shareholders, many of them corporations themselves. This is in- come that is not flowing to national taxpayers that can be linked to our other datasets compiling income from other sources, and therefore is not included in our study. Finally, we highlight that these aggregate of assigned corporate profits refer to the entire dataset of taxpayers in Honduras, only a small share of which is actually used when we combine our tax records with household survey. In every year, as docu- mented in Table C2, the total amount of corporate profits in the final microdata is much larger than the aggregate in Table B1, since individuals being assigned large losses are as a rule not at the top of the distribution, so by construction we end up 1 Insome years the amount of profits assigned to individuals is larger than the total profits. This can happen since profits are often negative – being unable to distribute negative profits from a firm, everything else equal, will increase the total amount of assigned profit in comparison with total profits. 49 with positive and large assigned profits in the final sample. C National Accounts Aggregates from WID dataset Here we discuss the national accounts aggregate we use in more detail and provide additional information on the adjustment of the microdata to macroeconomic aggre- gates. We start by using World Inequality Database (WID) aggregates for nominal Na- tional Net Income (NNI) in local currency for the period 2003 - 2019. We provide all numbers used in the analysis in Table C1. For 2019, for example, the Net National Income of reference is L 541 billion - in real terms, NNI grew by 80% in the period 2003-2019.2 We then decompose NNI in four main components – Compensation of Employees (L 260 billion in 2019); Capital Income (L148 billion); Mixed Income (L 87 billion) and Taxes on Production (L 47 billion). In Figure C2, we plot the share of main components in total NNI over time - compensation of employees represent approximately half of total income over the period; total capital income ranges from 20-30%; mixed income approximately 15% and total taxes on production 10%.3 The main change in aggregates’ shares we observe is a decrease in compensation of em- ployees from a peak of 54% in 2013 to 48% by 2019, accompanied by an equivalent 6 p.p. increase in the participation of capital income from 21% to 27%. That increase is largely explained by a substantial expansion of corporate profits in total income - the level of corporate profits increases nominally by 150% between 2013 and 2019, from L46 billion to L116 billion. As discussed in the text, we match wages in the microdata to Compensation of Employees; we match Mixed Income in the microdata to its equivalent in national ac- counts; and distribute taxes on production proportionally to total income, so it has no bearing on inequality measures. For distribution of capital income, we use a more detailed disaggregation. Using WID decomposition, capital income can be separated in property income of households; net operating surplus (NOS) of the households; corporate profits; and property income of the government. Similarly to taxes on pro- duction, we distribute the (small) property income of the government proportionally to households. Following Alvaredo et al. (2020), we match NOS to total imputed rent 2 In Figure C1, we show that the growth of NNI closely matches the growth of Gross Domestic Product in the period. The levels of GDP available at WID match those in other sources such as the World Bank’s World Development Indicators. 3 We note that not all subcomponents are available in the WID data in recent years, so we impute some of the aggregates. Specifically, no information for the four components of NNI are available for 2019, so we use the same shares observed in 2018 to recover levels in 2019, multiplying imputed shares by the NNI value in that year. Similarly, the subcomponents of capital income are not available between 2016-2019, so we impute the shares observed in 2015 and use them to recover levels of each of them in the latter period. 50 in the microdata. We are then left with corporate profits and property income of the households. Here, we note two features of WID National Account aggregates related to corpo- rate profits and property income of households, summarized in Figure C3. First, at close to 18% in more recent periods, the share of corporate profits over NNI in Hon- duras is much higher than most countries in LAC, with the exception of Nicaragua, and also much higher than high-income countries such as Spain, France and the United Kingdom. In Latin America we see a very large variation, with shares below 5% for Mexico for example, and in the range of 5 - 15% for Peru and Costa Rica. At the same time, the share of property income of households, which is mostly slightly negative in Honduras, is much lower than in other countries in the region and in high-income countries, where it ranges from 5 - 20% in recent years. While we are unable to as- sert the precise method to arrive at these figures for Honduras, these differences are substantial. Given the negative amounts of property income of households we observe, adjust- ing directly our measure of capital income in our microdata to these aggregates would mean subtracting income to match these negative amounts – we lack more granular measures on property income on national accounts, so we could not adjust income from interests and dividends separately, for example. Instead, and informed by these large gaps in NA measures between Honduras and other countries, we decide to sum up distributed property income with undistributed corporate profits, and adjust the total amount to national accounts. Therefore, when we adjust our microdata to na- tional accounts we multiply the total amount of distributed capital income and undis- tributed corporate profits in the microdata by a yearly factor that matches the total to national account aggregates.4 4 In panel (e) of Figure C3, we show that the sum of corporate profits and household property income for Honduras is less of an outlier when compared to their LAC countries. 51 Table C1: National Accounts Aggregates Year Net National Net National Compensation Capital Mixed Taxes on Property Property Net Operating Surplus Corporate Income (2021 prices) Income (nominal) Employees Income Income Production Income HH Income Govt Household Profits 2003 328.3 129.8 64.4 27.6 22.5 15.3 -0.7 0.0 7.7 20.6 2004 345.2 145.3 72.1 30.5 25.7 17.0 4.5 -0.1 8.3 17.8 2005 365.7 165.1 82.1 35.7 29.1 18.2 1.9 -0.2 9.6 24.3 2006 390.0 185.4 91.9 40.5 31.6 21.5 1.8 0.1 10.5 28.1 2007 422.1 214.0 105.6 50.1 35.8 22.6 0.6 0.1 12.2 37.2 2008 436.9 238.8 119.2 56.8 40.2 22.5 -0.2 0.9 13.2 42.9 2009 421.6 248.0 126.0 64.1 39.6 18.3 -2.2 0.0 15.1 51.1 2010 432.2 266.1 139.5 58.6 44.3 23.7 -1.6 0.0 15.4 44.7 2011 450.1 298.8 154.9 63.7 54.1 26.2 -1.9 -0.2 16.1 49.7 2012 461.9 317.6 164.9 67.6 56.2 28.9 -0.8 -0.7 18.5 50.6 2013 471.9 329.0 176.1 68.0 55.8 29.1 1.6 -0.8 20.7 46.4 2014 482.3 359.3 187.2 72.7 62.1 37.2 -1.9 -2.6 22.2 55.0 2015 509.1 405.6 200.5 95.5 66.6 42.9 -2.6 -1.9 24.8 75.2 2016 525.0 433.6 216.1 110.6 71.5 35.3 -3.0 -2.2 28.7 87.1 2017 558.2 481.8 227.4 138.0 77.6 38.8 -3.7 -2.8 35.8 108.7 2018 568.3 500.1 239.9 136.8 80.1 43.3 -3.7 -2.7 35.4 107.7 2019 591.4 541.9 260.0 148.2 86.8 47.0 -4.0 -3.0 38.4 116.7 Note: This table provides total amounts of income in national accounts of Honduras. All values are in billion Lempiras. Data is from the World Inequality Database (WID). Figure C1: Gross Domestic Product vs. Net National Income growth Note: This figure presents yearly growth of real net national income and of gross domestic product for the period 2000 - 2019. Data is from the World Inequality Database (WID). 52 Figure C2: Composition of macro incomes (WDI data) Note: This figure presents the composition of total income in national accounts over time. Data is from the World Inequality Database (WID). 53 Figure C3: Corporate Profits and Property Income of Households in comparison (a) Corporate Income - HND vs. LAC (b) Corporate Income - HND vs. High-Income (c) Property Income - HND vs. LAC (d) Property Income - HND vs. High-Income (e) Sum Corporate + Property Income - HND vs. LAC Note: These figures presents the share of different components of Net National Income for Honduras and other selected countries over time, in national accounts. Panel (a) and (b) present the partici- pation of corporate profits for Latin America Countries and High-income countries, respectively; while panel (c) and (d) present participation of property income. Panel (e) presents the participation of the sum of these two components. 54 Table C2: Total income in microdata (survey + tax records) Year Final Income Wages Mixed Capital (total) Capital (distributed) Capital (Tax data) Imputed rent Corporate 2003 89.6 41.7 31.5 7.9 2.2 0.5 7.2 6.5 2004 99.0 51.7 29.0 8.9 2.5 0.6 8.0 7.3 2005 106.3 52.7 32.2 11.6 3.2 0.9 8.6 9.5 2006 130.4 68.3 38.3 13.0 3.7 1.0 10.3 10.3 2007 155.3 79.6 47.9 14.2 3.7 1.4 13.1 11.7 2008 182.8 93.4 52.2 20.3 10.9 0.9 16.4 10.3 2009 192.8 102.0 60.6 12.0 4.3 0.7 17.1 8.7 2010 207.6 109.1 64.0 14.9 6.1 0.9 18.0 9.9 2011 208.4 113.1 58.7 15.3 3.6 1.1 21.3 11.7 2012 205.1 111.5 55.6 15.1 3.2 1.4 22.9 11.9 2013 233.6 127.1 63.9 16.8 4.6 2.3 25.8 12.2 2014 251.9 140.3 64.8 18.9 4.5 2.1 28.0 14.4 2015 271.1 148.9 69.7 22.5 5.1 2.9 30.0 17.4 2016 299.9 168.2 71.6 27.1 5.9 3.5 33.0 21.2 2017 312.1 173.6 78.1 28.1 5.7 3.6 32.2 22.4 2018 331.1 189.4 80.1 29.1 6.9 4.1 32.5 22.2 2019 360.0 200.1 88.1 29.9 6.6 5.0 42.8 23.2 Note: This table presented total amounts of income components for each year in the final microdata. It includes survey and tax records after the merging of the two sources, but before scaling up com- ponents to macroeconomic aggregates. Values are in nominal billions of Lempiras. D Income definition using tax data In this appendix we provide a detailed description of how we construct income vari- ables in the tax records. We start by presenting the composition of total income, then discussing in detail each of its components: Total individual income = + Capital Income ++ Dividend income (DMR 113) ++ Interest income (DMR 115) ++ Rental income (DMR 106) + Wage income ++ Wages (reported on yearly tax filings and/or DMR 111 ) + Mixed income ++ Income from service provision (yearly tax filings and/or DMR 112) ++ Income from commercial activities (yearly tax filings and/or other third- party sources) ++ Income from non-banking interest, commercial rent and other income (yearly tax filings) + Undistributed corporate profits 55 D.0.1 Capital income Capital income is defined as the sum of yearly gross income from dividends (DMR 113), earned interest (DMR 115), and rental income (DMR 106). This information is provided by withholding agents, which report both the total amount of gross income earned and the tax withheld. Often we observe inconsistencies in this reporting (e.g. in the case of dividends withholding should be equal to 10% of gross dividend income). Whenever that happens, we consider the withholding information to be correct (since that must match the amount remitted to the tax authority) and adjust the reported income accordingly (e.g. a common case is an agent reporting withholding equal to gross revenue, which we interpret as incorrect filing of the gross revenue, i.e. the tax base.). Even though we do have information on income from capital gains, we do not include those in the measure of total individual income, particularly since we also attribute to individuals undistributed corporate profits. D.0.2 Wage income For individuals who do not file an income tax declaration, we recover yearly gross wages from wage withholding (DMR 111). In addition to monthly wages, formal workers receive the 13th and 14th month salaries as benefits. These additional ben- efits are exempt from income taxes up to 10 monthly minimum wages each, so the observed taxable wage only includes regular wages and any benefits above the ex- emption threshold. To recover the exempt additional salaries, we complement taxable wages with the additional benefits defined as follows: AdditionalBene f it = Min( ReportedSalary ∗ (2/12), 20 ∗ MW ) For individuals who do file an income tax declaration, we potentially have an ad- ditional source of wage information, namely the declared wage on their tax filing. We perform the same adjustment to include 13th and 14th salaries. To obtain final wages for those filing taxes, nonetheless, we perform an additional exercise to take into ac- count that tax filings seem to often mislabel wage and service provision (honorary, etc.), discussed below. D.0.3 Mixed income For those not filing income taxes, we compute gross revenue from commercial activ- ities as the maximum between withholding from honorary and fees (DMR 112); and the sum of five third-party informed sources of income: sales through debit and credit cards (ATC); sales to the government (SIAFI); sales to other firms used as VAT credit (DMC); exports and sales to large firms (DMR 135). This is gross revenue, nonetheless, 56 and we are interested in profits (gross revenue net of production costs). We go back below to how we impute costs to individuals who do not file taxes. For those filing income taxes we take the following steps: • As mentioned before, taxpayers often mislabel wages and service income in their tax declaration: if we take the withholding reports as the ground truth, we can of- ten observe taxpayers declaring the correct amount of income but labeling wage income as honorary, for example (which does not have impact on tax liabilities and therefore can be seen as low stakes). In order to deal with that, we adopt the following rule: to assure that our measure of income is as encompassing as pos- sible, we define the total income from wage and services as the maximum of the self-declared income in the yearly filing and that in third-party reports. We then use the values from the source with largest total amount as the correct amounts for both wages and service income. • We then proceed to apply the allowable deductions for all self-declared revenue (wages, service, commercial, interest, rent and other income) to obtain income net of expenses. • Finally, we consider the possibility that total revenue informed by third-parties is larger than the total revenue claimed by taxpayers. Once again, since taxpay- ers often mislabel sources of income in their filings, we do not attempt to make comparisons by specific income sources: we simply compute the difference be- tween total self-reported income (already incorporating the previous adjustment for wages and honorary) and the total third-party revenue amount. When the third-party amount is larger, we impute that difference as additional revenue to the taxpayers and apply deductions as discussed below to obtain net income. For individuals who do not file an income tax declaration but are reported as hav- ing revenue from commercial activities, we need to impute deductions in order to obtain net income. We do so by using deductions from individuals who file taxes. For each year, we compute the median profit margin rate by department (19 regions) and 2-digit industry. When information at that level is missing, we simply impute the median margin at the 1-digit industry. We then estimate the net income from commercial and service provision activities for individuals not filing taxes by multiplying revenue by the imputed profit margin rate. We use the same imputed profit margin rate to recover the additional unreported net income, using third-party reports. 57 E Key descriptive statistics using tax records In this section we present a set of descriptive statistics related to our tax record only – meaning, before merging the distributions between survey and tax data, and before any adjustments to national account aggregates. Counts of observations in our tax administrative records, separately presented by income source, are presented above in Table 1: in more recent years we observe half- million individuals receiving some form of income and/or listed as shareholders of a corporation filing income taxes in that year. Approximately 110,00 individuals each year file an income tax tax declaration, which is not mandatory for individuals whose sole income sources are withheld at source. In terms of individuals withheld at source, close to 120,000 are wage workers whose salaries are withheld by their employers; 45 - 50,000 are withheld for service provision activities; 200,000 for receiving some form of dividend5 and 300,000 for receiving interest. Approximately 20 - 30,000 individ- uals each year declare capital gains, with the majority of them declaring very small amounts, and a few hundred file a separate declaration of rental income for individ- ual properties. Between 90 - 100,000 individuals in recent years are also included in other types of third-party information for commercial activities, such as exports, sales through credit cards or sales to the government. As we discuss above, we use this data to compare self-reported amounts with third-party ones, and increase the imputed in- come to individuals according to the largest amount among the two sources. Finally, in most recent years we observe between 70,000 and 80,000 shareholders in firms that filed a CIT declaration, and are thus assigned non-zero income based on undistributed profits (profits net of dividend distribution). In Figure E1a, we present the distribution of total income in the tax records, exclud- ing the imputed undistributed profits. As we discuss in section 3, one key limitation of the administrative tax records we observe is the lack of information on wages and other labor income below the minimum threshold for filing income taxes - not only in- dividuals receiving wages below that level are not required to file, but employers also do not inform or withhold annualized wages below that threshold. In the figure, we observe a peak of tax records precisely at the level of the filing threshold (right-most dashed line). We also illustrate, in the left-most dashed line, that the annualized mini- mum wage was significantly below that threshold, meaning that there is a large mass of workers receiving wages between these two levels that are not observed in the tax records. The smallest peak in the income distribution to the left of the minimum wage is comprised mostly of individuals receiving small amounts of income from dividends 5 The large number of individuals receiving dividends is mostly driven by a few large corporations, often in the financial sector, distributing very small amounts of dividends to thousands of taxpayers. In that sense, the majority of these individuals are not active shareholders with meaningful participation in corporations. 58 and interest – we illustrate that fact in Figure E1b, where we show that incomes below the minimum wage are almost exclusively comprised of capital income. The share of capital income decreases to almost zero at the level of wages start being declared and then steadily increases again at higher levels of income. Figure E1: Income distribution and composition in tax records - 2019 (a) Histogram (b) Share of capital in total income Note: Panel (a) in this figure presents a histogram of (log) total yearly income. The two dashed lines mark the value of the mininum wage (left) and the exemption threshold for PIT declaration (right). In panel (b) we present the share of capital income in total income, again marking the levels of minimum wage and PIT exemption threshold Data for 2019. We interpret the previous results as indicative that the tax records are not reliable below the minimum income level to file income taxes each year – approximately 70- 75% of tax records every year are below that level, or the equivalent of ≈ 150,000 individuals in recent years. That represents less than 5% of the reference adult popu- lation from household surveys, meaning that our ability to complement survey data with tax records is restricted to the very top of the income distribution. E.1 Levels and composition of top incomes at tax records We now describe the levels and composition of total income among the highest earners in the tax records. We focus on the top 1% highest earners.6 In Figure E2, we show the average income in each g-percentile within the top 1% for 2019.7 Individuals just inside the top 1% have an average yearly income of L350,000 or approximately USD 30,000 PPP; those at the top 0.1% receive on average L1,000; those at the top 0.01% 6 Wedefine the top 1% in relation to the total number of adults represented in the household survey and not the top 1% of tax records – e.g. since in 2019 the reference population is 5.4 million adults in the household survey, the top 1% will be the highest 54,000 earners. 7 We present the thresholds, in USD PPP, to be in selected percentiles for several years in Table E1. 59 receive L4.4 million while the top 0.001%, less than 60 individuals, receive on average L19 million or approximately USD 2 million. Figure E2: Income levels in tax records - 2019 Note: This figure presents the level of income in tax records, excluding undistributed corporate profits. Data for 2019. We also document that, within the top 1% of individuals with highest incomes, the composition of total income varies widely. In Figure E3 we present the share of total income for each g-percentile, taking the average across the 2014-2019 period. First, in Figure E3a we divide income between wages, mixed income and capital income. For those just within the top 1%, capital income is negligible - their total income is com- posed of 85% wages and 15% mixed income from self-employment and other commer- cial activities. That figure only starts to change within the top 0.1% – approximately the 5,000 individuals with highest earning each year. At that level, the share of wages in total income falls to close to 70%, mixed income represents 20% and capital income 10%. At higher levels of income, the share of capital income steadily increases all the way to 40-50% of total income at the very top, with a similar share for mixed income – within the top 0.01% highest earners, income from wages are on average always below 20% of total income. We present an alternative decomposition of total income between labor and capital in Figure E3b, under the assumption that mixed income reflects a 70% labor return and 30% capital return (Alvaredo et al., 2020). Under those assumptions, the share of income from labor steadily declines from over 90% for those just within the top 1% to 50% at the very top - within the top 0.001%, the highest quantile, the composition is actually 60% capital vs. 40% labor. 60 Figure E3: Total income in tax administrative records - 2019 (a) Histogram (b) Share of capital in total income Note: This figure presents the composition of income across top quantiles in the tax records, ex- cluding undistributed corporate profits. Panel (a) presents the breakdown for wage, mixed income and (distributed) capital income, while panel (b) presents the breakdown between labor and capital income, using the assumption that 70% of mixed income is labor income. Data for 2019. 61 Figure E4: Composition of income in tax administrative records - 2019 (a) Detailed breakdown of income composition (b) Labor vs. capital income Note: This figure presents the composition of income across top quantiles in the tax records. Panel (a) presents the breakdown for wage, mixed income, distributed capital income and undistributed corporate profits, while panel (b) presents the breakdown between labor and capital income, using the assumption that 70% of mixed income is labor income. Data for 2019. 62 Table E1: Thresholds to belong to each top quantile (USD 1,000s) Lowest income by g-percentile and year 2014 2016 2018 2019 Top 1% 283 341 369 349 [31] [34] [36] [33] Top 0.5% 441 511 550 486 [48] [51] [53] [46] Top 0.1% 998 1,166 1,248 1,009 [108] [115] [121] [96] Top 0.05% 1,379 1,641 1,741 1,509 [150] [162] [169] [143] Top 0.01% 3,448 4,243 4,477 4,346 [374] [420] [434] [412] Top 0.005% 5,554 6,676 7,256 7,203 [603] [661] [703] [683] Top 0.004% 6,797 7,602 8,794 8,203 [737] [753] [852] [777] Top 0.003% 8,655 10,118 10,866 9,585 [939] [1,002] [1,053] [908] Top 0.002% 12,224 12,973 13,766 12,744 [1,326] [1,284] [1,335] [1,208] Top 0.001% 17,731 19,167 19,764 19,951 [1,924] [1,898] [1,916] [1,891] Note: This table reports the thresholds to belong to each quantile in the tax records, for selected years, both in Lempiras and USD PPP (1,000s). F Adjustment of survey agricultural income In this section we provide more details on the adjustments we make on the survey data related to self-employed agricultural income. As we document in Table A.1, in the period 2004 - 2013 the household survey we use included an additional module on agricultural income from self-employed activities. What we observe in the data is that reported income from those activities is extremely high in that period. Further- more, the microdata available only includes total income from that module, and not a breakdown of all variables included in the survey, which limits our ability to further 63 investigate the reason for those extreme values. We take two decisions to limit the influence of income from this module. First, we topcode individual income from that module, for each year in that period, at the av- erage income + 2 standard deviations - this limits the influence of a few outliers with extreme values each year - in Table F1, we show that less than 1% of individuals each year have agriculture income topcoded in the survey, and also presents the thresh- old used for topcoding. Nonetheless, the large incomes in those periods are not only driven by a few outliers, but by the entire distribution of agricultural income being shifted upwards. So we include a second adjustment: we first compute the average real agricultural income in the period 2014-2019, after the agricultural module was discontinued, and then crate an adjustment factor for each year in the period 2003 - 2013, such that the amount of real income in each of those years match the average for the 2014 - 2019 period. These adjustment factors, which are applied to each individual agricultural income, vary from .5 in 2011 to .8 in 2005. Table F1: Topcoding agricultural module Share topcoded (%) Topcode threshold (L) 2003 0.210 14,476 2004 0.241 15,651 2005 0.277 17,030 2006 0.200 17,980 2007 0.239 19,227 2008 0.375 21,419 2009 0.267 22,597 2010 0.281 23,658 2011 0.564 25,258 2012 0.364 26,571 2013 0.143 27,942 Note: This table presents, for each year in the period 2003-2013, the share of observations for which agricultural income is topcoded and the threshold for topcoding. G Adjusting tax record data for 2003 - 2011 In Table 1, we document that our dataset includes almost no information on with- holding forms for the period 2003 -2010 - in particular, it includes no information on withheld wages, which represent a large share of income observed in the tax data. The reason for that is that withholding forms before 2011 were not electronic and have not 64 been digitized, so we do not have access to that information. For that reason, the level and composition of income in tax records for that period is very different from the most recent 2011 - 2019 period. Nonetheless, we do observe PIT filings and undistributed corporate profits for those years, and therefore can use the ratio of income in PIT filings + corporate in- come to total income in recent years to adjust total income the 2003-2010 period. We do not change directly the microdata for the 2003-2010 period. Instead, we adjust the g-percentiles used when merging tax records to survey data at the top. We proceed as follows: 1. We first create the g-percentiles used for the merging process with survey data, and compute for all years and g-percentiles the ratio between the sum of income in PIT filings plus undistributed corporate profits to total income. For 2011-2019, total income will include all forms of withholding that we do not observe in 2003-2010. 2. We then take the average value of these ratios across all periods, for each g- percentile. In Figure G1, we show the average adjustment factor, which we use to scale total income over 2003-2010, together with 95% CI of the mean to illus- trate dispersion of the adjustment factor across years. What we observe is that for individuals just on the top 1%, the adjustment factor is on the order of 2 - 2.5 - since most of their income, as we observe in other years, is composed of wage income, we miss a lot of income in 2003-2010 when withholding of wages is not observed. As we move towards higher incomes, the adjustment factor slowly converges towards unit, consistent with the fact that at the very top most of the income is from undistributed corporate profits, which we observe across the entire period. 3. We then use these adjustment factors to scale up total income in each g-percentile in the period 2003-2010. Note that we do not scale them to the average level of income observed in 2003-2010 - we simply scale it considering the ratio of PIT + corporate profits to total income, so the underlying assumption to recover the correct income level is that this ratio is stable over time. 4. Finally, when adjusting our microdata to national account aggregates, we need the composition of total income, since we perform the adjustment component- by-component. We compute the share of each component, by g-percentile, in the period 2013-2019, then apply those shares to g-percentiles in 2003-2010, recover- ing estimated income components that are then scaled-up to national accounts. 65 Figure G1: Adjustment factors for period 2003 - 2010 3 2.5 Adjustment Factor 1.5 2 1 Top [1%,0.9%] Top [0.1%,0.09%] Top 0.01% Top 0.001% n Note: This figure presents the adjustment factors used to scale total income by g-percentile, in the period 2003-2010. In the figure we plot the mean (95% CI) ratio between income from PIT filings + undistributed corporate profits and total income, by g-percentile, across years in the period 2011- 2019. 66