Universal Basic Income Programs: How Much Would Taxes Need to Rise? Evidence for Brazil, Chile, India, Russia, and South Africa

Abstract Using microsimulations this paper analyzes the poverty and tax implications of replacing current transfers and subsidies by a budget-neutral (no change in the fiscal deficit) universal basic income program (UBI) in Brazil, Chile, India, Russia, and South Africa. We consider three UBI transfers with increasing levels of generosity and identify scenarios in which the poor are no worse off than in the baseline scenario of existing social transfers. We find that for poverty levels not to increase under a UBI reform, the level of spending must increase substantially with respect to the baseline. Accordingly, the required increase in tax burdens is high throughout. We find that the increase in the average tax rate that would be consistent with not hurting the poor is almost universally above 30%, limiting the feasibility of a UBI reform due to political resistance and efficiency costs.


Introduction
A Universal Basic Income (UBI) is often seen as an attractive policy option to replace existing targeted transfer and subsidy programs fully or partially. This replacement, in theory, can provide an income floor or broad-based safety net for income related contingencies (which helps with consumption-smoothing), avoid errors of exclusion frequently observed in targeted programs, eliminate issues of stigma, entail administrative simplicity, and ensure more political buy-in because everybody could potentially receive a (net) benefit (Gentilini, Grosh, Rigolini, & Yemtsov, 2019). 1 In order to derive all these benefits and at the same time not hurt the poor by the switch, government spending may need to be higher than under the existing transfers system. However, expanding government spending may trigger macroeconomic risks (e.g. high inflation and/or unsustainable debt levels). Thus, when considering switching to a UBI it is not worse off than in the existing system turn out to be not feasible by our definition above. For most countries the required increase in average tax rates range between 38% to over 400% for the UBI scenarios in which the poor are not harmed. In only two of the fourteen cases, the average tax rate would have to be raised by less than 30%: about 27% in Chile and 29% in Brazil.
Our paper contributes to the literature that analyzes the fiscal feasibility of a poverty reducing UBI. At the theoretical level, Hanna and Olken (2018) argue that a UBI may be difficult to fund in the context of developing countries because of the narrowness of the tax base (i.e. the number of individuals paying taxes tends to be relatively low due to informality, tax exemptions, tax evasion, and tax loopholes). At the empirical level, however, the funding challenges occur in advanced countries as well. For European countries, Browne and Immervoll (2017) use EUROMOD data to simulate the effects of existing means-tested cash transfers versus a UBI. They find that in Finland, France, Italy, and the United Kingdom, a budget-neutral UBI that equals the countries' guaranteed minimum income requires 'very large tax-revenue changes' (Browne & Immervoll, 2017, p. 333). Ghatak and Jaravel (2020) evaluate the feasibility of using a flat tax rate to fund a generous UBI in the UK. Using a conventional optimal taxation model with key features of the UK's current tax and transfer system and income distribution, they find that ' … it is possible to fund a large annual UBI … but that increasing taxes on the most affluent alone would be insufficient' (Ghatak & Jaravel, 2020, p. 1). While they find that a generous UBI may be feasible despite the disincentives to work, it would require a 45% flat income tax, a large tax rate applied to everyone, including the poor. In the same vein, Magnani and Piccoli (2020) evaluate the feasibility of using income taxes to fund a generous UBI in France and find that it would require a flat tax rate of 49%.
Results for North America are similar to those found for European countries. Hoynes and Rothstein (2019) analyze the distributional and behavioral effects of a UBI in the United States applying a framework simplified to a single equation that captures the key parameters of transfers programs, including a 'canonical' (that is, genuinely universal) UBI. They find that a UBI that is generous enough to meet the basic needs of households without earnings would be very expensive (twice the cost of all existing transfer programs) andunless substantial new funds are put in-highly regressive (Hoynes & Rothstein, 2019). Kesselman (2018) carefully examines the tax implications of a number of proposals to replace existing targeted programs in Canada with a self-financed (that is, raise tax revenues to pay for the universal transfer) basic income guarantee (BIG). The author concludes that-because of the required increase in taxesthe ' … self-financing approach may even exacerbate the economic and political barriers to a feasible BIG' (Kesselman, 2018, p. 433).
A key contribution of our paper is to assess the fiscal feasibility of a budget-neutral UBI that does not harm the poor in a set of middle income countries. Our results are quite in line with those found in the empirical literature for advanced countries. While there are country-specific and cross-country analyses for middle income countries, almost all of them focus on the spending side without considering the implications or estimating the changes in the tax burden, and what the latter mean for the post-UBI poverty and inequality indicators. 8 Furthermore, existing research focuses on the impact on poverty and inequality measured with disposable income, which only includes direct taxes and transfers rather than consumable income, which includes the impact of indirect taxes (and consumption subsidies) too. As shown in Lustig (2018), indirect taxes can impoverish the poor quite frequently, a fact that is completely missed when the poverty levels are estimated with disposable income. To the best of our knowledge, ours is the first paper that does all three: imposes budget neutrality, accounts for the tax burden for the poverty analysis, and uses consumable income as the welfare indicator. In addition, our paper makes three further contributions. First, we rely on a clear set of conditions to select feasible UBI scenarios based both on poverty impacts and the required increase in tax burdens. Second, by analyzing the budgetary conditions prevailing in our selected countries, we also Universal basic income programs 1445 consider whether a UBI could be financed through a higher fiscal deficit or lower spending in other items. Third, our estimates are based on a unique dataset in which the incidence of taxes and transfers were generated using a common methodological framework discussed in Lustig (2018).
The paper is organized as follows. Section 2 discusses the microsimulation methodology of this paper and renders more details regarding the household survey data and administrative accounts of the five countries considered in this paper. Section 3 presents our results regarding the impact of each budget neutral UBI scenario on the inequality, poverty, and tax incidence in all five countries. Finally, section 4 provides our concluding remarks.

Methodology and data
Our purpose is to estimate the impact of budget neutral UBI programs on poverty, inequality and tax burdens. More precisely, we want to show the potential trade-off between the generosity of a UBIand, hence, its poverty impactand the required increase in taxes. For this, we resort to microsimulation to construct prefiscal and postfiscal household income per person for each country and for seven scenarios. The scenarios are the following: (1) baseline, (2) spending neutral UBI financed by direct taxes, (3) spending neutral UBI financed by indirect taxes, (4) equivalent benefits UBI financed by direct taxes, (5) equivalent benefits UBI financed by indirect taxes, (6) poverty gap UBI financed by direct taxes, and (7) poverty gap UBI financed by indirect taxes. The UBI scenarios are ordered from the least (spending neutral) to the most generous (poverty gap). Table 1 describes the characteristics of each scenario.
The microsimulations are carried out applying the so-called 'accounting framework:' that is, we ignore behavioral and general equilibrium effects on labor supply, consumption patterns, and taxable income. In other words, we capture the first-round effects. While this is a limitation, first-round effects are considered a reasonable approximation for the short-run. More importantly and as mentioned before, for the purposes of our analysis, they are enough to assess the fiscal and political feasibility of the options considered here. If the policy alternatives are not feasible in the absence of labor supply and other behavioral responses, they would be even less feasible if the latter were taken into account.
Our microsimulations allow us to obtain measures of poverty and inequality with prefiscal and postfiscal incomes for each scenario. They also allow us to quantify and compare the required increases in direct or indirect tax burdens. The selected prefiscal income here is market income plus contributory pensions. 9 The postfiscal income is consumable income: that is, prefiscal income plus direct transfers and subsidies net of personal income and consumption taxes. While data on inequality and poverty is usually reported for disposable income (income after direct taxes net of cash transfers), we use consumable income as the relevant welfare indicator because our simulations contemplate raising consumption taxes. Consumable income captures what people are able to consume in actual terms after one takes into account not only personal income taxes but also consumption taxes. Consumable income is constructed by adding to prefiscal incomes the transfers reported in the household surveys plus the simulated subsidies, and minus the simulated direct (personal income) and indirect (consumption) taxes. The income concepts used in the analysis are summarized in Figure 1. 10 As we shall explain below, for simulating the impact on tax burdens, we will need to make use of gross income (for robustness check of our results as reported in Supplementary Appendix 5 (available in Supplementary Materials)) and disposable income too.
Having calculated the income concepts, for each scenario we calculate the Gini coefficient, the poverty measures and the indicators of tax burdens. In our simulations, we consider two poverty measures: the poverty headcount and the squared poverty gap. The poverty headcount is widely used in policy circles but fails to capture the impacts of poverty reforms among the extremely poor. To give an example, assume that social assistance covers the extremely poor well, but coverage is not as good among households whose income or consumption lies close to the poverty line. A spending-neutral UBI reform may show greater poverty reduction when measured with the poverty headcount index (because with the UBI all households close to the poverty line would now receive a transferand hence jump over the poverty line); but the reform could come at the expense of higher extreme poverty, because the budget would be taken away from the extremely poor to be redistributed among a greater number of people.  The squared poverty gap measure, by giving a greater weight to the welfare of the extremely poor, will capture such an increase in extreme poverty. As we shall see below, in South Africa a spending neutral UBI reform would reduce the headcount ratio but increase the squared poverty gap.
In our analysis of the poverty impact, we use country-specific poverty lines which vary by countries' income levels, since in wealthier countries, higher poverty lines are more appropriate. We use US$3.20 a day in PPP for India, $5.50 for Brazil and South Africa, and $11 for Chile and the Russian Federation. Supplementary Appendix 1 (available in Supplementary Materials) provides more details regarding the source and rationale for these poverty lines.
The impact on tax burdens is assessed by the difference in the incidence of taxes (average tax rate) of each decile for each scenario and the baseline. The incidence of direct taxes here is defined as the ratio of direct taxes to prefiscal income (market income plus pensions). For robustness check and in Supplementary Appendix 5, we use gross income instead of prefiscal income to broaden the tax base and reduce the necessary increase in tax rated even further. For the incidence of indirect taxes, we use the ratio of indirect taxes to disposable income.
For each UBI budget neutral scenario, we use the CEQ Desktop Tax Simulator tool, an Excel-based tool developed by the Commitment to Equity Institute (CEQI) at Tulane University, to measure the necessary change in direct or indirect taxes to achieve budget neutrality. The CEQ Desktop Tax Simulator achieves budget neutrality for each scenario by multiplying the existing tax rates (in the baseline scenario) by a constant multiplier. The existing tax rates are the observed direct tax incidence by decile with respect to market income plus pensions (or gross income in Supplementary Appendix 5) and the indirect taxes with respect to disposable income. The advantages of using this approach to achieve budget neutrality are twofold. First, it is easy to implement in practice: everybody's taxes are increased proportionally. Second, it keeps the share of taxes paid by each income decile constant (i.e. constant concentration shares of taxes) across scenarios. The details of the microsimulations using the CEQ Desktop Tax Simulator are described in Supplementary Appendix 2 (available in Supplementary Materials).
In order to assess the desirability and feasibility of the simulated UBI programs, we first identify the scenarios in which the poor are no worse off than in the baseline with two types of poverty indicators: the conventional (anonymous) incidence and severity of poverty and the nonanonymous fiscal impoverishment (Higgins & Lustig, 2016). The latter could be interpreted as applying the Pareto criterion to the pre-reform poor. We shall see that even if the headcount and the severity of poverty are lower under some scenarios, some of the prefiscal poor are made worse off. For scenarios that pass the 'poverty test', we assess their fiscal feasibility by first looking at the increase in the tax incidence (average tax rate) by decile. While there are no general conventions on the latter, we leave out a policy scenario that implies an increase in the average tax rate above 30%. The scenarios that fulfill the poverty condition and are feasible but imply a change in average tax rates above this threshold are excluded from the feasible set.
We then look at whether the fiscal gap could be absorbed through a higher deficit (lower surplus) or a reduction in government spending on other items. Both options seem quite unfeasible too. Thus, it seems that in order to implement a feasible budget neutral UBI that is funded by taxes without incurring high efficiency costs, the size of the universal transfer may entail that some of the poor will be made worse off comparing to the existing system of transfers and subsidies.
The main sources of information for this paper are twofold. The household surveys and administrative data on transfers programs come from the World Bank's Atlas of Social Protection Indicators of Resilience and Equity (ASPIRE). 11 The household surveys (Table 2) include information on social assistance programs disaggregated by program, welfare indicator (disposable income, expenditure, or consumption), and household demographic characteristics. The ASPIRE database posts program-level information on social protection programs including spending, number of beneficiaries, and program design features from administrative sources. Administrative data were available for all but Russia; for the latter we used World Bank staff estimates. This database is used to validate/compare amounts spent on social assistance programs from administrative data by country, with the total benefit amounts retrieved from household surveys. The second main source of information are the fiscal incidence of taxes and subsidies by decile published in the Commitment to Equity Data Center on Fiscal Redistribution (CEQ Data Center). 12 An advantage of using the data in the CEQ Data Center is that the studies were produced using a common methodological framework described in Lustig (2018). The World Development Indicators database is used for some of the indicators reported in this study, such as the income groups of countries, gross domestic product (GDP), and purchasing power parity (PPP).
The cash transfer programs included in the baseline and which are replaced by a UBI in the simulated scenarios cover noncontributory programs only: that is, means-tested conditional and unconditional cash transfers, cash transfers based on categorical targeting (e.g. people with disabilities), and noncontributory pensions. 13 Contributory pensions and unemployment compensation programs are not included in the analysis because, by definition, these programs have an insurance component and reforming them involves complexities (e.g. financing for the transition period) that are beyond the scope of this study. 14 We also do not consider public works or cash-for-work programs because strictly speaking they are not transfers as people need to work to receive them. Supplementary Appendix 1 provides a detailed list of existing cash transfer programs (which together with spending on consumption studies) were replaced in our simulations by a UBI.
To calculate the inequality, poverty, and tax burdens in each scenario, we use the incidence of taxes and consumption subsidies by decile housed in the Commitment to Equity Data Center on Fiscal Redistribution. The CEQ Data Center includes information on the incidence of direct (mainly, personal income tax) and indirect taxes (VAT, excise and other consumption taxes) and subsidies by decile from fiscal incidence analyses that used a common methodological framework (Lustig, 2018). 15 Details on data sources, methodological assumptions, and results by country can be found in: Brazil (Higgins & Pereira, 2014), Chile (Martinez Aguilar, Fuchs, Ortiz-Juarez, & Del Carmen Hasbun, 2017), India (Kundu & Cabrera, 2022), the Russian Federation (Lopez-Calva, Lustig, Matytsin, & Popova, 2017), and South Africa (Inchauste, Lustig, Maboshe, Purfield, & Woolard, 2017). As indicated before, the fiscal incidence studies used here for our baseline scenario is based on the accounting framework and, thus, the derived prefiscal income cannot be presumed to equal the true counterfactual income in the absence of taxes and transfers. Therefore, the analyzed scenarios should be considered as a first-order approximation. 16 In essence, the underlying assumption is that payroll taxes and contributions (both by employee and employer) in the formal sector are borne by labor and that consumption taxes (and subsidies) are fully shifted forward to consumers. This is equivalent to assuming that the supply of labor and demand for goods and services are perfectly inelastic. 17 The baseline scenario is not, however, a mechanical application of statutory rules. We analyze the incidence of taxes by their economic rather than their statutory incidence and take into account tax evasion and avoidance. For example, individuals who do not report being registered in the social security administration are assumed not to pay personal income and payroll taxes. In the case of consumption taxes, we generate effective rates of taxationwhich we use in place of the statutory ratesby calculating the actual revenues collected by the revenue authority over the actual sales value of the taxable base.

Results
To recapitulate, our analysis is based on simulations of the first-round effects on poverty, inequality, and tax burdens that result from changing the baseline social assistance and subsidies system (baseline scenario) with three alternative budget-neutral UBI scenarios. We have called these scenarios spending-neutral, equivalent benefits, and poverty gap. In order to focus on the implications for taxes, the initial resource pool includes not only spending on transfers but also on consumption subsidies. In the spending neutral UBI scenario, the current cash transfers and consumption subsidies are replaced by a per capita UBI keeping total spending on cash transfers and subsidies unchanged and direct or indirect taxes are adjusted accordingly to ensure budget neutrality. In the poverty gap UBI scenario the universal transfer is set equal to the average poverty gap and direct or indirect taxes are adjusted accordingly to ensure budget neutrality. We treat these two scenarios as a lower and upper bound. 18 In the in-between equivalent benefits scenario, the universal transfer is set equal to total spending on transfers and subsidies divided by total beneficiaries of at least one of the direct transfer programs considered and direct or indirect taxes are adjusted accordingly to ensure budget neutrality. Considering the baseline and the three UBI scenarios and the two possible sources of financing the fiscal gapdirect and indirect taxeswe have seven scenarios in total (see Table 1 above). Table 3 shows daily transfers and subsidies per person and per poor person in the baseline and transfers per person in the UBI scenarios. The last three columns are of particular interest because they record the ratio of the UBI transfer divided by the transfers and subsidies per poor person in the baseline. As expected, a spending neutral scenario results in a reduction in average benefits for the poor when compared to the baseline situation. In the equivalent benefits UBI, the poor receive roughly the same as in the baseline in India and Russia, somewhat more in South Africa, and considerably more in Brazil and Chile. Needless to say, the amount in transfers the poor receive under the poverty gap UBI is at least three times as much as what they receive currently except for Russia where the increase is close to just 30 percent. The latter is the result of the fact that in Russia categorical targeting of existing transfer programs approximates a universal system much more than the current targeted schemes in the other countries. Universal basic income programs 1451

Poverty and inequality
In Table 4, we show the results of the UBI simulations on poverty and inequality. As expected, in general, the spending neutral scenario results in higher poverty headcount ratio and squared poverty gap index and an increase in the Gini coefficient. However, this is not true for every country. In the case of India, all three indicators are lower under the spending neutral scenario than the baseline. How can that be? It is telling us that the Public Distribution System was not benefiting the poorest individuals in comparison to a scenario in which we distribute the total benefit equally across the whole population. In the case of South Africa, while the headcount ratio falls under the spending neutral scenario, the squared poverty gap and the Gini coefficient rise. South Africa illustrates how the headcount ratio, by focusing on what happens to poor (and nonpoor) individuals near the poverty line, can lead to a misleading appreciation. We would not want to switch to a system in which the poorest of the poor are worse off. For the spending neutral scenario, whether budget neutrality is achieved through direct or indirect taxes makes little difference. Another general result that comes out as expected is that, under the poverty gap scenario, the headcount ratio, the squared poverty gap index, and the Gini coefficient fall in all the countries compared to the baseline. And, as expected, the reduction is higher when the budget neutral UBI is financed with direct taxes. 19 If poverty and inequality reduction and providing a social protection floor to all were our only goals, then clearly this is the most desirable scenario of the ones analyzed here. In fact, if one could target the amount spent in this scenario perfectly to those who are prefiscal poor, poverty could be eradicated. However, and as we are able to confirm later, the required increase in taxes to fund such a scenario may be too onerous.
Thus, a more realistic scenario from the tax implications point of view may be the equivalent benefits one. This scenario is the closest of those considered here to introducing a UBI in which the poor (measured with anonymous indicators as the ones shown on Table 4) are as well off as in the baseline. Although this is not strictly true for each individual, it is true for the poor on average. In this scenario, the poverty and inequality indicators are lower than in the baseline for all the countries. Therefore, it would seem that this scenario could be an option if the goals are to reduce poverty and provide a social protection floor for all. However, if we look at Table 5, which shows the change in consumable income per person by decile for individuals ranked by their prefiscal income for each UBI scenario versus the baseline, we observe that in India, Russia, and South Africa the poorest deciles lose under the equivalent benefits scenario (compared to the baseline, that is). In these three countries, thus, even a relatively generous budget neutral UBI would result in the poorest being worse off because they would end up paying more in taxes than what they receive in benefits. This result holds whether budget neutrality is achieved by raising more direct or indirect taxes.
This table shows a nonanonymous comparison and reveals that some of the poorest could be impoverished by a policy change that would mimic the equivalent benefits scenario. The fact that anonymous and nonanonymous indicators can yield different stories regarding the poor was first introduced by Higgins and Lustig (2016). When determining whether a certain policy change is desirable from the poor's perspective, nonanonymous indicators such as fiscal impoverishment must be considered. Otherwise, we may miss the fact that we are making the poorest in a country worse off. For instance, if we use the nonanonymous indicator shown in Table 5, the spending neutral scenario in the case of India reveals the fact that every decile that includes people in poverty, is worse off. Thus, this is another example of how we would have missed this important fact if we just relied on the anonymous poverty measures (headcount ratio and squared poverty gap) shown in Table 4 and alluded above. Universal basic income programs 1453

Tax burden
Based on the implications on consumable income (Table 5), one can conclude that a spending neutral UBI reform is undesirable because the poorest deciles are made worse off in all countries. One can also conclude that an equivalent benefits UBI reform is undesirable for India, Russia, and South Africa for the same reason. An equivalent benefits UBI reform seems 'doable' in the cases of Brazil and Chile: the poor are not made worse off while the net losses in consumable income for the top deciles are small (around 6% or less). The poverty gap scenario makes the poor better off in all the countries. However, the losses in consumable income for the top deciles can be very large (see India and South Africa, for example). The only country in which the losses for the top are small (significantly below 5%) is Russia. This is because the categorical targeting in Russia is probably the closest to a UBI and the average benefit under a UBI is the closest to the current system (Table 3).
In sum, all the countries pass the 'poverty test' under the poverty gap scenario. However, what is the change in tax burdens? Figure 2 shows the change in the incidence of taxes (average tax rate) per decile between the equivalent benefits and poverty gap scenarios and the baseline. 20 How feasible would a UBI be if budget neutrality is attained through taxes? Recall that we chose to define infeasibility as an increase in the average tax rate equal or above 30%. The poverty gap scenario would imply such a large increase in tax burdens for every country that is clearly not feasible. Not even for Russia, where the increase in the incidence of direct taxes and indirect taxes would be about 50 and 38%, respectively. For the equivalent benefits scenario, the 'poverty test' is passed by Brazil and Chile. For both countries, the simulated required increase in direct tax rates is large: about 64 and 71%, respectively. The required increase in indirect taxes is about 29% in Brazil and 27% in Chile. The latter may be borderline feasible if we assume away behavioral responses. However, the required increases in the average indirect tax rates are large enough that behavioral responses could end up limiting their feasibility.
Our results are consistent with the existing empirical analyses of budget-neutral generous UBI proposals in advanced countries as well as income floor transfer programs in sub-Saharan Africa. Browne and Immervoll (2017) find that in Finland, France, Italy, and the United Kingdom, a budget-neutral UBI that equals the countries' guaranteed minimum income requires 'very large tax-revenue changes' (Browne & Immervoll, 2017, p. 333). International Monetary Fund (2017) calculates the fiscal cost of 'a UBI set at 25% of median per capita income at about 6-7% of GDP in advanced economies … ' (International Monetary Fund, 2017, p. X). Kesselman (2018) examines proposals to replace existing targeted programs in Canada with a basic income guarantee and concludes that issues such as an effective marginal tax rate of about 50% for the bottom decile would create a significant disincentive for work that make these proposals infeasible. Hoynes and Rothstein (2019) find that funding a UBI that both replaces the existing transfer programs in the US and increases transfers to low-income families would require the federal tax revenue to increase by about 100%. Ghatak and Jaravel (2020) and Magnani and Piccoli (2020) find that a generous UBI in the UK, and France would require a flat tax rate of 45 and 49%, respectively. Lastly, Lustig, Jellema, and Martinez Pabon (2019) examine the feasibility of budget-neutral income floor transfer programs in nine countries in sub-Saharan African. Even though such policies are less resource-intensive than an unconditional UBI and even when the income floor is set equal to the poverty line of these countries, the required increase in direct or indirect taxes to fund them is too burdensome to make them deem feasible.
The required increase in direct tax rates has been calculated under the assumption that UBI transfers are not subject to taxes. Some may object to this assumption because it could exaggerate the required increase in tax rates. In low-and middle-income countries cash transfers are overwhelmingly not part of taxable income. As a robustness check, however, Supplementary Appendix 5 reproduces all results under the assumption that the direct tax rates are applicable to gross income. The hope is that the expansion of the tax base reduces the need for tax rates to Chile (2013) India (2011) Russia (2016) South Africa (2014) Spending Neutral Chile (2013) India (2011) Russia (2016) South Africa (2014) Spending Neutral Universal basic income programs 1455 increase as much as reported in Figure 2. However, this reduction is minimal and our results remain virtually unchanged: The only cases in which a budget neutral UBI would neither hurt the poor nor require an increase in tax rates beyond 30% are equivalent benefits UBI financed using indirect taxes in Brazil and Chile in which the new indirect tax rates would be 27.5 and 25.6% higher than the baseline, respectively.

Deficit financing and cutting other expenditures
The above results suggest that, in general, a budget neutral UBI financed by an increase in taxes appears not to be feasible for poverty gap scenario and not even for the equivalent benefits  (2015) Chile ( (Kundu & Cabrera, 2022); Russia (Lopez-Calva et al., 2017) and South Africa (Inchauste et al., 2017). Note: The incidence of direct taxes here is defined as the ratio of direct taxes to market income plus pensions. For the incidence of indirect taxes we use the ratio of indirect taxes to disposable income. Simulated scenarios displayed in this table are as follows: (1) Equivalent benefits: scenario in which each individual receives the average direct transfers and indirect subsidies of beneficiaries in the current system; (2) Poverty gap: scenario in which UBI transfer equals the average poverty gap for each country. See Supplementary Appendix 3 (available in Supplementary Materials) for the decile-level tax rates in the baseline and the necessary change in tax rates to achieve budget neutrality.    Universal basic income programs 1457 scenario. There are two other options: let the government deficit increase or cut spending in other items. Figure 3 shows the government deficit as a share of GDP. India, Brazil, and South Africa have relatively large government deficits so for them the first option is not really feasible.
In the case of Chile, the budget surplus would not only be wiped out but there would be a deficit. In the poverty gap scenario, it would be around 4.3%; that is, large. Thus, deficit financing does not seem to be an option either. Cutting down on other expenditures for all countries and both scenarios is high except for India and for the case of equivalent benefits: expenditures would need to be cut by 0.8%. For the other countries, current spending would have to be reduced by 4.3% in Russia (equivalent benefits scenario) to close to 26% in Chile and India (the poverty gap scenario).

Concluding remarks
Our results suggest that, even in middle income countries, it may be very challenging to design a UBI program that would leave the poor not worse off than under the existing transfers and subsidies schemes and simultaneously keeps the increased tax burden in check.
Thus, whenever reducing poverty is the primordial goal, it appears that targeted transfers (either means-tested or categorical) will be a superior option. Many decades ago, Akerlof (1978) formally showed that if beneficiaries could be selected ('tagged', using his terminology) without a cost, a welfare system that gives transfers to people with special needs or characteristics (e.g. the poor) would be superior to a uniform negative income tax that gives the same benefit to everyone.
This rationale is the basis for targeted programs. In categorical targeting, the benefits are provided to people with characteristics that are exogenous in the sense that these characteristics are more difficult to manipulate or change through behavior (e.g. social pensions to the elderly). In means-tested targeting, the benefits are provided to people with, for instance, incomes below a certain threshold. The challenge is that people can change a characteristic such as income through behavior: e.g. non-beneficiaries could decide to work less in order to become eligible. 21 The evidence for developing countries, however, does not show that transfers targeted to the poor result in a reduction in adult labor supply (see, for example, Fiszbein andSchady (2009), chapter 4 andBastagli et al. (2016)).
Our results and conclusions are sensitive to a few key assumptions and characteristics of the countries included in our analysis. First, we assume that the administrative costs of current targeted transfer programs are not large. In other words, savings derived from eliminating targeting mechanisms would not be sufficient to pay for a significant portion of a generous UBI. This assumption is in line with existing research: the administrative costs of targeted transfers programs range between 3 to 8 percent of the value of transfers and the 'targeting' component of these admin costs ranges between 18 and 75 percent of total admin costs . 22 Second, in the countries included in our analysis, the poor are significantly covered by the current transfer programs. Thus, the UBI would need to be sufficiently generous to make sure the poor are not made worse off. In countries where the coverage of the poor of existing targeted transfer programs is low, even a modest UBI would make many of the poor better off. However, if the problem of coverage is due to the lack of resources as opposed to improper targeting, those countries would be able to reduce poverty more if they continue to allocate additional resources to the targeted programs as opposed to implementing a modest UBI. Third, if the budget neutrality assumption is relaxed and a country can fund a portion of its UBI program by borrowing (without causing macroeconomic disruptions, of course), then even a relatively generous UBI may be feasible. However, a country that continuously borrows to fund its social spending would risk a debt crisis unless it also has the necessary economic growth to keep the debt-to-GDP ratio relatively constant. Fourth, political resistance may make it hard if not impossible to increase the size of current targeted transfer programs. In such cases, the only way to benefit the poor may be to benefit the whole population, i.e. to implement a UBI. In this paper we do not analyze the case of adding a UBI to the existing targeted programs, but our results indicate that a generous UBI would be harder to implement politically because the increase in taxes has to be more than before to fund both the existing transfer programs and the UBI. Lastly, our findings do not apply to UBIs with categorical criteria such as UBI for children or an unconditional old age pension (Canelas & Niño-Zaraz ua, 2022). Since the required amount of additional spending would be lower, the required tax increases may end up being manageable.

Notes
1. UBI programs are also consistent with welfarist (Fair, 1971;Mirrlees, 1971) and non-welfarist optimal tax theory (see Kanbur, Paukkeri, Pirttil€ a, & Tuomala, 2018;Kanbur, Pirttil€ a, & Tuomala, 2006) which in the derivation of the optimal tax schedule assume that redistribution will take place through lumpsum uniform transfers. For a survey of the optimal tax theory, see, for example, Tuomala (1990). 2. The reference years and data sources are in Table 2. The classification of countries by category of gross national income per capita in 2011 purchasing power parity is based on World Bank thresholds; see https://www. datahelpdesk.worldbank.org/knowledgebase/articles/906519-world-bank-country-and-lending-groups. Russia is classified as a high-income country because that is how it was classified in 2016, the data collection year of the survey for which the analysis presented here was conducted. 3. To assure the poverty remains unchanged, we use the conventional (anonymous) indicators of poverty such as the headcount ratio as well as the fiscal improvement indicator (Higgins & Lustig, 2016) which is a nonanonymous index. The nonanonymous measures of poverty compare each individual to himself/herself to determine what percentage of individuals would become poorer following the implementation of a policy. So, for example, it is possible for the poverty headcount ratio to decrease, yet some poor individuals become more impoverished. 4. Of course, to be sure that poverty is eradicated in full, the UBI would have to be set equal to the poverty line and the total fiscal cost would equal the poverty line times the population. We did not consider this scenario because the ensuing fiscal burden makes it unrealistic in all the cases we considered here. 5. We opted for calculating the equivalent benefit transfer considering the beneficiaries of at least one direct transfer program (and not both transfer and subsidy programs) because, as it is discussed in our methodology section, our simulations allocate subsidies to each household based on the decile of income distribution its members belong to so, implicitly, every individual in the decile receives the average benefit of the subsidy. In other words, the number of beneficiaries from subsidies equals the total population in the survey, which would put us in the spending neutral scenario. 6. If the existing programs are highly targeted toward the poor and only provide little benefit to the non-poor, replacing them with this UBI would hurt the poor. That is, because the average benefit to recipients is kept constant, the poor recipient would lose, and the non-poor recipient would gain from this UBI. Therefore, in the context of poverty alleviation, analyzing the equivalent benefit scenario allows us to assess whether the gains from the elimination of the error of exclusion exceeds the losses from the reduction in progressivity of the existing targeted programs. 7. In contrast, Rigolini, Lustig, Gentilini, Monsalve, & Quan (2019) did not consider subsidies as a source of financing nor impose strict budget neutrality. 8. For example, Caamal-Olvera, Huesca, and Llamas (2022) (2019) and Rigolini et al. (2019). Lustig et al. (2019) examine the feasibility of budget-neutral income floor transfer programs in nine countries in sub-Saharan African. The required increase in direct or indirect taxes is too burdensome to make them feasible. 9. We chose to treat contributory pensions as part of the prefiscal income to avoid the distortions that treating them as pure transfers introduced. See chapter 1 in Lustig (2018) for a detailed presentation on this matter. 10. Note that for the baseline scenario we start from observed disposable income in the microdata and work "backwards" to construct baseline gross income and prefiscal income and work "forward" to generate the baseline consumable income. 11. https://www.worldbank.org/en/data/datatopics/aspire. See Appendix 1 for more details. 12. To view and download information, visit http://www.ceqinstitute.org and click on the Data Center tab. 13. See Appendix 1 for more detail. The choice of programs a UBI would replace, and the way we interpret the results, deserves some explanation. The choice is not dictated by the belief that a UBI should replace specific programs, but by our intention to explore the implications on poverty, inequality, and the tax burden when certain programs are replaced. In this respect, the simulations that follow should not be viewed as an actual recommendation for the countries included in the paper-any country-specific proposal would require analyses that are better tailored to the specific context of that country. Rather, the objective is to reach a better understanding of how varying contexts affect the impacts on poverty, inequality, and tax burden of varying UBI schemes. 14. For a discussion of contributory social security systems and the challenges of reforming them see, for example, Barr and Diamond (2008). 15. For a detailed description of the fiscal incidence analysis method applied see Chapters 1, 6, and 8. It is important to note that the incidence of direct and indirect taxes is not the incidence of statutory rates but the economic incidence. Due to tax evasion or informality, which are widespread in developing countries, a significant number of self-employed and salaried workers may not pay direct taxes; and consumers in rural areas and those who purchase from informal sellers (for example, street vendors, farmers' markets, and so onwhen the survey contains a question about place of purchase) may not pay consumption taxes such as VAT or excise taxes. The studies housed in the CEQ Data Center make assumptions about informality and evasion. Typically, individuals who do not report being registered in the social security administration are assumed not to pay personal income and payroll taxes. In the case of consumption taxes, for purchases from informal sellers it is assumed that no consumption taxes are paid (at least, directly at the time of purchase although the price of the good may carry the effect of taxes on inputs). If there is no information on the place of purchase, some studies assume that households in rural areas do not pay consumption taxes. For details, we refer to the country studies cited above. The information can be found in the CEQ Data Center by clicking here. 16. A first-order approximation suffices for a reasonable impact estimate. Coady et al., for instance, state, "The first order estimate is much easier to calculate, provides a bound on the real-income effect, and is likely to closely approximate a more sophisticated estimate. Finally, since one expects that short-run substitution elasticities are smaller than long-run elasticities, the first-order estimate will be a better approximation of the short-run welfare impact" (Coady et al., 2006, p. 9). Moreover, although public spending on, for example, education, health, and infrastructure has an inherent investment element that is likely to affect long-run inequality and poverty dynamics, we do not attempt to capture these dynamic effects. 17. "The economic incidence, strictly speaking, depends on the elasticity of demand and/or supply of a factor or a good, and the ensuing general equilibrium effects. In essence, the accounting approach implicitly assumes zero demand price and labor supply elasticities, and zero elasticities of substitution among inputs, which may not be far-fetched assumptions for analyzing effects in the short-run, especially when changes are small" (Lustig, 2018, p. 18). 18. We also tried a UBI reform with the uniform transfer equal to the country-specific poverty line. The tax burden would be too high to make it feasible so we discarded this scenario at the outset. 19. This is expected because direct taxes are more progressive than indirect taxes. 20. The increase in direct tax incidence under the poverty gap scenario ranges from 42% in Russia to over 5000% in India. The equivalent benefits scenario, as expected, implies a lower increase in tax burdens. However, it is still quite large. The range is from 10% in South Africa to 119% in India. These increases are sufficiently large to conclude that such a reform would not be politically feasible and, even if it were, the efficiency costs could be high. Would a budget neutral UBI financed with indirect taxes be feasible? The poverty gap scenario would entail an increase in the incidence of indirect taxes between 35% for Russia to 263% for India. Again, too high to make it politically feasible and with likely high efficiency costs. The equivalent benefits budget neutral UBI has a smaller range in the change in the burden of indirect taxes: from 7% in India to 28% in South Africa (and for the rest of the countries the change is also above 20%). We do not show results for the spending neutral scenario because the changes are minimal, but they are available upon request. 21. Hence the amount of effort that governments place in defining eligibility criteria that are complicated and nontransparent. The more complicated and obscure, the more difficult it is for people to change their behavior to become eligible. In Mexico, for instance, the eligibility criteria of the now defunct Progresa/ Oportunidades/ Prospera program used a formula that was kept secret from the public. 22. However, if there is an outlier country in which the "targeting" costs of an existing transfer program is much higher than the estimates in the literature and if that country can assure universality of a UBI at a much lower admin costs, then a generous UBI may be funded by reducing the admin costs without the need to increase taxes.