Tax-benefit microsimulation and income redistribution in Ecuador

ENIGHUR is a nationally representative cross-sectional survey on income and expenditures of households in Ecuador. The aim of the survey is to provide information on the distribution and structure of income and expenditures of Ecuadorean households based on the demographic and socio-economic characteristics of the household members (INEC, 2012). The survey is conducted approximately every eight years. The latest ENIGHUR is for years 2011–12 and contains information for 39,617 households and 153,444 individuals. The survey follows a probabilistic two-stage sample design in nine self-represented cities, and a three-stage design in the rest of the country.³ The sampling unit is the dwelling defined as the person or group of people living in the same housing structure (dwelling), sharing meals, and who depend on a common budget. Information about the household and each member of the household occupying the dwelling is collected in the survey.

ENIGHUR contains very detailed income information. Gross employment and self-employment income from principal and secondary occupations are reported, including extra pay, bonuses, inkind income and self-consumption from self-employment activities. Contributions to social security and tax payments are also collected for both, employees and self-employed workers.⁴ Moreover, detailed information about affiliation to social security is reported, allowing to categorize individuals to specific social security regimes. Cash transfers from social benefits, such as public pensions, the HDT, Joaquín Gallegos Lara transfer and the housing grant are also available. Other sources of income such as income from capital and property, private transfers, remittances, and other rents, are also reported in the data. Finally, detailed data on household expenditures is also available in the survey, as well as information on demographic and socio-economic characteristics of each member of the household.

The present study takes advantage of the availability of reported data on social benefits, taxes and social insurance contributions to compare their effect on income inequality and poverty relative to that of simulated tax-benefit components calculated from ECUAMOD, which is presented in the next section.

Our study makes use of ECUAMOD, the tax-benefit microsimulation model for Ecuador. ECUAMOD has been developed as part of UNU-WIDER’s project on ‘SOUTHMOD-simulating tax and benefit policies for development’ in which tax-benefit microsimulation models have being built for selected developing countries. ECUAMOD and other country models from the SOUTHMOD project have been implemented on the EUROMOD software, which provides a harmonized setting for cross-country comparative analysis on the redistributive effect of tax-benefit policies (see Sutherland & Figari 2013). ECUAMOD combines detailed country-specific coded policy rules with representative household microdata to simulate direct and indirect taxes, social insurance contributions, as well as cash transfers for the household population of Ecuador. ECUAMOD is a static model in the sense that tax-benefit simulations abstract from behavioural reactions of individuals and no adjustments are made for changes in the population composition over time. Simulation results for ECUAMOD have been validated against external statistics (see Jara, Varela, Cuesta, & Amores, 2017).

The underlying microdata used in ECUAMOD comes from ENIGHUR 2011–2012. Adjustments to the data and variables, for the construction of ECUAMOD’s input data, are kept to a minimum. On the one hand, individuals recorded as domestic employees in a household have been dropped, as information about their own household (e.g. number of children, expenditures) is not available. In total, 103 individuals (0.07 per cent of the sample) were dropped from the original sample, leaving us with a sample of 153,341 individuals. No households were dropped and no adjustments to the weights were made as a result of dropping individual observations. On the other hand, ENIGHUR data includes a single variable covering all pension payments. This variable includes contributory pensions, as well as severance pay and alimony for divorce and children. This single variable is disaggregated into five variables in the ECUAMOD database according to the characteristics of the recipients: (i) old-age pension, (ii) disability pension, (iii) survivors’ pension, (iv) severance pay, and (v) alimony payments. In our analysis, the first three variables are considered as part of public pensions, severance pay is included in social benefits and alimony payments are included as part of market income.

ECUAMOD version 1.4 covers policy years 2011–2017 for the purpose of tax-benefit simulations, based on ENIGHUR 2011–2012. To account for time inconsistencies between the input data year and the policy year in the simulations, market incomes and non-simulated tax-benefit variables in the data are adjusted using source-specific updating factors (see Jara et al. 2017). In this study, 2011 policies (as on June 30th) in Ecuador are used for the analysis.

ECUAMOD aims to simulate the main tax and benefit components of household disposable income in Ecuador, where household disposable income is defined as the sum of market income plus government cash transfers of all household members net of income tax and social insurance contributions.⁵ However, simulation of tax-benefit instruments depends on the relevant information being available in the underlying dataset used in the simulations, in this case ENIGHUR 2011–2012. Instruments which are not simulated are taken directly from the data and included as part of disposable income. The remainder of this section describes briefly the policy instruments simulated in ECUAMOD for year 2011, as well as the underlying assumptions used in the simulations of each income component. Table A.1 in the appendix summarizes the tax and benefit components included in the model, differentiating between simulated and non-simulated components, and provides information about why simulation was not feasible.

Validation of simulated taxes and benefit is a crucial part of microsimulation modelling. In this section, we propose comparing information about the number of benefit recipients and taxpayers as well as aggregate annual expenditure in social benefits and revenue from taxes and SICs obtained with ECUAMOD and reported in ENIGHUR against external official data.

The numbers of recipients of benefits and payers of taxes and SICs obtained with ECUAMOD and reported in ENIGHUR are compared with external benchmarks in Table 1. In terms of benefits, the results show that ENIGHUR data captures around 91 per cent of beneficiaries of the Human Development Transfer compared to external statistics. The underrepresentation in the data is likely to be related to the fact that external statistics report the total number of beneficiaries over the whole year, whereas the survey asks respondents whether they have received the transfer over the past three months. ECUAMOD underestimates the number of recipients of the Human Development Transfer by around 17 per cent compared to external statistics. As mentioned in the previous section, in order to assess eligibility for the HDT in our simulations, we replicated the index of the Social Register with the input data and fixed the threshold of eligibility by identifying the same number of people below the official threshold. Note that this is different from calibrating the number of recipients of the transfer because other eligibility conditions, simulated in ECUAMOD, apply in addition to being below the threshold of the composite index, such as the presence of children in the household or non-affiliation to social security in the case of elderly adults and disabled persons. The underestimation of recipients of the HDT in ECUAMOD could therefore be related to non-fulfilment of eligibility conditions of individuals below the threshold due to differences in characteristics with actual recipients but also to a certain degree of digression from the authorities as to who is considered eligible for the transfer, which cannot be captured in our simulations.

The number of recipients of the disability carer benefit Joaquín Gallegos Lara obtained from ECUAMOD matches those of ENIGHUR because, as previously mentioned, simulations of the disability carer benefit are based on actual benefit receipt in the data. Disability carer benefit cannot be fully simulated because eligibility depends on the severity of disability and this information is unavailable in the data. ENIGHUR and ECUAMOD results underestimate the number of recipients of the disability carer benefit reflecting that this population subgroup is not properly captured by the survey.

In terms of taxes and SICs, ECUAMOD underestimates the number of payers compared with data from the Ecuadorean Social Security Institute (IESS) and the Internal Revenues Service (SRI), which might be related to income underreporting in the underlying input data. Note, however, that the underestimation of personal income tax and SICs payers is even more severe when reported payers from ENIGHUR are compared with external statistics. In particular, ENIGHUR data captures only 43 per cent of personal income taxpayers compared to 70 per cent obtained with ECUAMOD, reflecting the difficulties of collecting tax data in household surveys and highlighting the usefulness of microsimulation to calculate this type of income. The underestimation of self-employed SICs payers in ENIGHUR is related to the fact that the data does not contain information on payments to rural workers’ SICs (Seguro Campesino), which are simulated in ECUAMOD and are included in the official figures. The gap between the number of income taxpayers obtained from ECUAMOD and that of official statistics is most likely driven by income underreporting and underrepresentation of individuals at the top of the market income distribution in ENIGHUR data. Note that in Ecuador (like in many other Latin American countries) deductions from personal expenditures apply to income tax and the exemption threshold for income tax payment is rather high. In 2011, annual incomes below the USD 9,210 threshold were exempt from income tax, whereas the annualised minimum wage was equal to USD 3,168. Most individuals non-affiliated with social security have low earnings and, therefore, even under the assumption of full-compliance in our simulations they would pay no income tax.

Table 2 presents a validation of reported and simulated aggregate annual expenditure in social benefits and revenue from taxes and SICs. Aggregate expenditure in the HDT from ENIGHUR data matches well the external benchmark, which reflects the fact that survey data in Ecuador captures well the bottom of the income distribution. ECUAMOD simulations underestimate aggregate expenditure in the HDT by around six per cent in comparison to ENIGHUR data and nine per cent with respect to external statistics, which is in line with the underestimation of benefit recipients. Reported and simulated amounts of Joaquín Gallegos Lara transfer underestimate the aggregate amounts compared with external sources by around 40 per cent, reflecting the underestimation of benefit recipients in the data (see Table 1).

In terms of taxes and SICs, our simulations underestimate the aggregate amount of personal income tax and SICs compared with external statistics. ECUAMOD simulations capture around 82 per cent of revenue from personal income tax and 98 per cent of employee SICs. On the other hand, results obtained directly from ENIGHUR data provide an extremely poor match against the external benchmark. Reported personal income tax amounts from ENIGHUR capture only 22 per cent of personal income tax revenue. Reported employee SICs’ amounts provide a better fit, capturing 88 per cent of employee SICs revenue, compared to 98 per cent captured by ECUAMOD results. The poor representation of aggregate annual revenue from personal income tax provided by ENIGHUR data is most likely related to the abovementioned difficulties of reporting this type of deductions in surveys and points to the importance of using simulated taxes to assess the effect of this component on the income distribution, as its effect would be otherwise underestimated. The underestimation of income tax might also be related to the fact that, in general, surveys fail to properly capture the top of the income distribution and this is particularly the case for developing countries like Ecuador. The comparison of simulated and reported revenue from self-employed SICs with respect to official statistics is rather puzzling. Both simulated and reported results overestimate the revenue from self-employed SICs compared to the external benchmark. The results are counterintuitive given that the number of payers of self-employed SICs obtained from ECUAMOD and ENIGHUR underestimates the number of payers with respect to official statistics. The discrepancies might be related to the categories of workers for which SICs revenue has been reported by the authorities. ECUAMOD and ENIGHUR results consider as self-employed any person reporting income from self-employment, whereas external statistics might be based on people officially registered as self-employed at IESS. Finally, Table 2 provides a comparison of aggregate revenue from VAT. Compared with external statistics, ECUAMOD underestimates VAT by around 44 per cent. VAT reported by ENIGHUR represents only 19 per cent of the official aggregate amounts. However, it is worth noting that ENIGHUR reports VAT payments by the self-employed only, as part of their expenditures. VAT payment by households is not directly reported by ENIGHUR. On the other hand, ECUAMOD simulates VAT payments by households only, whereas the official statistics on VAT include payments by both households and firms.

Before comparing poverty, inequality and the redistributive effect of tax-benefit systems between our two approaches, it is worth considering the extent to which the size of different tax-benefit components varies across the income distribution depending on the use of simulated or reported incomes. Figure 1 shows the relative size of five tax-benefit components in Ecuador, where the average size of each income component is measured as a percentage of average household disposable income by household disposable income decile, and on average for the whole population. Direct and indirect taxes, and social insurance contributions are shown as negative values as they are subtracted in the calculation of disposable income.

Figure 1

Download asset Open asset

Under both approaches, our results show that social benefits play an important role at the bottom of the income distribution. Social benefits represent on average 20 per cent of household disposable income for individuals in the bottom decile of the income distribution with ECUAMOD and 22 per cent with reported benefits from ENIGHUR. The slightly larger size of social benefits with ENIGHUR data reflects the fact that ECUAMOD simulations of the Human Development Transfer slightly underestimate the number of recipients and aggregate amounts of this benefit compared to ENIGHUR, as shown in Tables 1 and 2 above.¹²

ECUAMOD and ENIGHUR results differ more in terms of the relative size of taxes and SICs. Simulated direct taxes represent an important proportion (5.4 per cent) of household disposable income only for the top decile of the income distribution. This might be due to the high exemption threshold for income tax payment and also to the presence of deductions from personal expenditures, which are applied to taxable income. Direct taxes reported in the data account only for 1.7 per cent of household disposable income for the top decile. The discrepancies between ECUAMOD and ENIGUR results might be driven by underreporting or misreporting of personal income tax in ENIGHUR and reflect the difficulties of recording this type of income in surveys. Indirect taxes in ECUAMOD are mostly proportional with respect to disposable income, although the relative size of this income component is slightly larger for the first income decile pointing to a minor regressivity. The mostly proportional nature of indirect taxes is in line with a recent paper by Rojas Baez (2017) and is driven by the presence of a zero per cent VAT rate on goods and services considered basic necessities. The relative size of indirect taxes in ENIGHUR is on average smaller than that of ECUAMOD and they appear to affect mostly the top income decile. Differences between ENIGHUR and ECUAMOD results in terms of indirect taxes are due to the fact that ENIGHUR reports VAT payments from expenditures related to self-employment activities only, whereas VAT payments for all households are calculated in ECUAMOD simulations based on expenditure information available in the survey. Finally, reported and simulated social insurance contributions are progressive in nature but represent a smaller share of household disposable income with results based on ENIGHUR data.

Table 3 confirms the pattern of tax-benefit components with respect to disposable income in our two approaches, where the progressivity of different policy instruments is assessed using the Suits index. The Suits index measures the ratio between the cumulative proportion of a policy instrument (e.g. income tax) and the cumulative proportion of income. The value of the Suits index lies within the interval [−1, 1]. A value of −1 would imply an extremely regressive policy instrument (e.g. the poorest person pays all the tax), whereas a value of 1 depicts extreme progressivity (e.g. the richest person pays all the tax). Our results show that direct taxes and SICs are progressive with respect to disposable income but to a larger extent with results obtained from ECUAMOD. Public pensions are regressive, whereas social benefits are progressive. Finally, ECUAMOD and ENIGHUR provide contrasting results in terms of progressivity of indirect taxes. Reported VAT in ENIGHUR are progressive with respect to disposable income, whereas ECUAMOD results point to a minor regressivity of indirect taxes, which is in line with the fairly proportional nature of this income component (see Figure 1).

Table 4 presents the results of our comparison of inequality and poverty indicators calculated from the reported ENIGHUR data and the simulated ECUAMOD data. The results are computed for individuals according to their household disposable income (HDI) per capita (i.e. HDI divided by household size). HDI is calculated as the sum of all income sources of all household members net of income tax and SICs. Using HDI per capita is the approach proposed by the National Institute for Statistics and Census (Instituto Nacional de Estadística y Censos, INEC) to calculate income poverty and inequality.

In terms of income inequality, Table 4 shows that the Gini coefficient obtained with the two sets of results are very similar. The Gini coefficient obtained from ENIGHUR data is slightly higher (by only three per cent) but the difference with the estimate from ECUAMOD data is small. The same applies to results for the p90/p10 ratio. Table 4 also presents the Atkinson index using three values for the parameter of ‘inequality aversion’, ε = 0.5, 1 and 2. Calculating the Atkinson index for different values of ε allows us to change the importance attached to variations at different points in the income distribution, with larger values of ε being more sensitive to variations at the lower end of the income distribution. ENIGHUR data provides slightly larger estimates for the Atkinson index. The gap between ENIGHUR and ECUAMOD data estimates is the largest for the Atkinson index with ε = 0.5, with a difference of eight per cent of the size of the ECUAMOD estimate.

The second and third sections of Table 4 compare statistics on absolute poverty and extreme poverty headcounts calculated under the two approaches. The national poverty lines of USD 72.87 per month for poverty and USD 41.06 per month for extreme poverty are used in the calculations.¹³ As it was the case for inequality, absolute poverty and extreme poverty calculated using reported ENIGHUR data are slightly higher than those based on simulated incomes. Our results further distinguish poverty estimates between households living in the urban or rural area of the country. The discrepancies between poverty estimates calculated from ENIGHUR data and ECUAMOD data are more marked in the urban area. The gap between urban extreme poverty from ENIGHUR data and ECUAMOD simulations is around 15 per cent, although this represents 0.4 percentage points given the low levels of extreme poverty in urban areas of the country.

The comparison of poverty and inequality calculated under our two approaches points to only minor differences. However, as observed in the previous section, the relative size and progressivity of particular tax-benefit components can vary considerably depending on the use of reported or simulated incomes. For instance, reported information on personal income tax payments seems to suffer from underreporting in the survey, which translates into a modest impact of reported income tax on household disposable income (see Figure 1). ECUAMOD simulations match better external statistics and translate into a larger effect of income tax at the top of the income distribution and therefore lower income inequality. From a policy perspective, it is important to have a clear idea of the effect of individual tax-benefit instruments on poverty and inequality. This is particularly relevant given the different degrees of measurement errors (e.g. underreporting or underrepresentation of top incomes) that individual income components might suffer from. The following section compares the extent to which the effect of tax-benefit components differs depending on whether reported or simulated incomes are used in the analysis.

Assessing the role of taxes and benefits in reducing poverty and inequality is essential to improve the design of existing and new policy instruments. Table 5 compares results for the Gini coefficient, the poverty headcount and extreme poverty headcount of household disposable income and household market income per capita under our two approaches. The results show that direct taxes and cash transfers reduce income inequality by 4.1 points (50.2 minus 46.1) based on ECUAMOD data. The redistributive effect of the tax-benefit system is reduced by almost half based on ENIGHUR data, accounting for 2.5 points reduction in inequality (50.2 minus 47.7). Absolute poverty is reduced by 4.1 points when simulated disposable income is compared to market income, whereas a 3.2 points decrease is observed with reported incomes from ENIGHUR data. The effect of the tax-benefit system on extreme poverty is more similar between the two approaches: extreme poverty is reduced by 3.7 (3.2) points when ECUAMOD (ENIGHUR) data is used.

The effect of specific tax-benefit components on income inequality and poverty under our two approaches is presented in Tables 6 and 7. The marginal contribution of individual tax-benefit components is obtained from the difference between inequality/poverty for disposable income with and without the tax-benefit component in question. As such, in order to assess the effect of public pensions and social benefits, we subtract each component separately from disposable income and recalculate income inequality and poverty. In order to assess the effect of direct taxes and SICs, we add back each component separately to disposable income and recalculate poverty and inequality. Finally, we also present poverty and inequality estimates for disposable income minus indirect taxes.

Under both approaches, the income component that reduces inequality the most is social benefits; accounting for a reduction of around 1.7 points, when Gini for disposable income is compared to Gini for disposable income minus social benefits (column A minus column C). The role of public pensions is minor in both cases. Table 6 points to a very different effect of direct taxes and SICs depending on the approach used in the analysis. SICs reduce income inequality by 1.3 points based on ECUAMOD data, whereas the effect of SICs is 0.9 points when reported incomes are used (column A minus column E). The differences are even more striking for direct taxes, which contribute to inequality reduction by 1 point with ECUAMOD data and only 0.3 points with ENIGHUR data (column A minus column D). Finally, indirect taxes tend to slightly increase inequality according to ECUAMOD simulations, whereas inequality decreases by 0.8 points when the Gini for disposable income is compared to the Gini for disposable income minus indirect taxes reported in ENIGHUR (column A minus column F).

Table 7 presents the effect of tax-benefit instruments on absolute poverty and extreme poverty under our two approaches. The income component that reduces poverty and extreme poverty the most is social benefits and the effect is very similar under the two approaches. Social benefits reduce poverty by 3.4 points and extreme poverty by 2.9 points with both ECUAMOD and ENIGHUR data (column A minus column C). The effect of public pensions is also similar under the two approaches, reducing poverty by around 1.4 points and extreme poverty by around 1 point (column A minus column B). The role played by direct taxes and SICs is, on the other hand, very modest under the two approaches and adding them back to disposable income results in slightly lower poverty levels. Finally, our two approaches provide a very different picture of the effect of indirect taxes. Subtracting indirect taxes from disposable income increases poverty by 2.1 points and extreme poverty by 0.9 points based on simulated data from ECUAMOD, whereas neither poverty nor extreme poverty are affected when reported VAT is deducted from disposable income calculated from ENIGHUR data (column A minus column F).

The results presented in this section highlight the extent to which individual tax-benefit instruments might be subject to different degrees of measurement error depending on the approach used for the analysis, and the effect of such differences on income inequality and poverty. On the one hand, the effect of social benefits on income inequality and poverty is similar under the two approaches. However, it is important to bear in mind that ECUAMOD results underestimate the number of HDT beneficiaries due to the complexity of simulating eligibility rules for this cash transfer. On the other hand, the larger effect of simulated direct taxes on income inequality reflects the fact that reported tax payments in the data suffer from underreporting or misreporting and therefore simulating personal income tax based on market income information available in the survey captures a larger number of taxpayers. As noted before, problems of underreporting and underrepresentation of the top of the market income distribution also result in an underestimation of tax revenue if ECUAMOD results are compared to external statistics. Finally, the use of comparable income information under both approaches has an important effect on the results. Our two approaches provide a very different picture of the effect of indirect taxes. This is due to the fact that indirect taxes reported in ENIGHUR relate only to VAT payments from expenditures linked to self-employment activities, whereas ECUAMOD results represent VAT payments by all households calculated from expenditure information available in the survey.

Differences between INEC (2016) figures and the results presented in this study are due to the use of different surveys in the analysis. INEC (2016) figures are based on the National Survey of Employment, Underemployment and Unemployment (Encuesta Nacional de Empleo, Desempleo y Subempleo, ENEMDUR).

For more information about CEQ, see http://www.commitmentoequity.org/.

The nine self-represented cities are: Cuenca, Machala, Esmeraldas, Guayaquil, Loja, Manta, Quito, Ambato, and Santo Domingo.

Respondents are asked to report how much was deducted from their monthly gross income in terms of social insurance contributions. Additionally, the survey asks employees to report how much was deducted from their monthly gross income in terms of income tax, and the self-employed to report how much income tax they paid over the last twelve months.

Market income is the sum of employment and self-employment income, bonuses, in-kind income, self-consumption from self-employment activities, capital and property income, inter-household payments, private transfers, minus alimony payments. Imputed rent is not included as part of market income.

Utilities participation is a benefit for employees, where 15 per cent of a firm’s utilities are distributed among all employees in the firm.

Deductions from personal expenditures cannot be higher than 50 per cent of taxable income or 1.3 times the basic exempted band. Additionally, there are individual limits for each type of expenditure. Expenditure in food, housing, education, and clothing cannot exceed 0.325 times the basic exempted band, individually. Expenditure in health cannot exceed 1.3 times the basic exempted band.

In 2016, the VAT rate was increased to 14 per cent in all provinces except those hit by the earthquake in April 2016.

For more details about the simulation of ICE in ECUAMOD see Jara et al. (2017).

Additionally, the conditionality of the programme extends to prenatal health controls, sexual and reproductive health consultations, eradication of child labour and mendacity, maintenance of the dwelling, and an annual update of changes in the socio-economic situation of the household.

Our pseudo index might be upward biased with respect to the official index, as the sample used to calculate the official index targets, a population in potential need of social protection.

Previous studies based on the National Survey of Employment, Underemployment and Unemployment (ENEMDUR) have focused on the percentage of HDT beneficiaries along the income distribution or concentration shares, rather than the relative size of the HDT by deciles. Burgos Dávila (2014) shows that around 26 per cent of HDT beneficiaries are in the two lowest income quintiles, whereas ECUAMOD finds 36 per cent of beneficiaries in the two lowest income quintiles and ENIGHUR 35 per cent. In terms of concentration shares, around 38.7 per cent of the total budget allocated to the BDH is concentrated among the two lowest income quintiles according to Ponce (2011), whereas results based on ECUAMOD amount to 36.3 per cent and those based on ENIGHUR data directly to 35.05 per cent.

The extreme poverty line in Ecuador is defined in terms of the minimum value of a food consumption basket to satisfy the nutritional needs for a healthy life. The poverty line is then obtained by using the inverse of the Engel coefficient (which measures the relationship between expenditure in food consumption and expenditure in total consumption) to scale the extreme poverty line (INEC 2015).

The results presented here are based on ECUAMOD v1.4. ECUAMOD is developed, maintained and managed by UNU-WIDER in collaboration with the EUROMOD team at ISER (University of Essex), SASPRI (Southern African Social Policy Research Institute) and local partners in selected developing countries (Ethiopia, Ghana, Mozambique, Tanzania, Zambia, Ecuador and Vietnam) in the scope of the SOUTHMOD project. The local partner for ECUAMOD is Instituto de Altos Estudios Nacionales (IAEN). We are indebted to the many people who have contributed to the development of SOUTHMOD and ECUAMOD. We would also like to thank two anonymous referees for valuable comments on previous versions of the paper. The results and their interpretation presented in this publication are solely the authors’ responsibility.

1. Version of Record published: April 30, 2019 (version 1)

© 2019, Jara and Varela

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.