Targeting versus Universality: Is There a Middle Ground?1,2

A global average of 45 percent of the bottom quintile is enrolled in social safety net programs. However, this figure is significantly lower in poor countries. In the last twenty years, coverage of these programs has been increasing steadily in Sub-Saharan Africa and Asia where a growing number of flagship programs have been introduced. For example, in Tanzania, the coverage of Productive Safety Nets Program increased from 0.4 to 10 percent of population between 2013 and 2016. In Ethiopia, a similar program covers more than eight percent of the population. In the Philippines, close to 25 percent of the population is covered. At least 142 countries now have such programs with

low and middle-income countries spending an average of around 1.5 percent of GDP with median spending at 1.1 percent of GDP.But there is a high degree of variation in spending where Kenya spends 0.37 percent while Liberia spends 2.64 percent. 3spite this historic expansion, limited budgets and imperfect implementation have led to mixed results.In low income countries, less than one in five people in the poorest quintile receive any transfer.This is partly because there are many households clustered around the poverty line and distinguishing between them at a given point in time often leads to arbitrary outcomes.Also, information may not be updated frequently enough.
These problems are well known to those administering and supporting targeted programs.Low coverage of many social assistance schemes is due to budget constraints which often leads to rationing.A program that is not so constrained and pays benefits through the middle part of the distribution will reduce these errors.
Recently, the idea of a Universal Basic Income (UBI) has been resurrected, this time as an answer to the job losses anticipated in the wake of automation and artificial intelligence.Proponents of universal benefits also point out that it helps overcome the exclusion errors inherent in targeted programs.In the few examples of existing UBI programs such as Mongolia and Iran, a significant impact on poverty has been observed.In Mongolia, the rather large UBI reduced poverty by 33.7 percent and inequality by 21 percent. 4World Bank (2018).
4 Yeung and Howes (2015).Poverty rates were 32.6 percent without the UBI, and 21.6 percent with the transfer; inequality as measured by the Palma ratio declined from 1.76 to 1.39.
In Iran, the quasi-UBI led to a 5-percentage point decline in poverty. 5In both cases, budgets were tied to revenue generated from natural resources. 6,7  cost of a UBI depends on the level of benefit chosen.
A recent IMF study estimated that it would cost around 6-7 percent of GDP in higher income countries and 3-4 percent in developing countries if the benefit were to be set at 25 percent of median per capita income. 8In Mexico, it will cost 7 percent of GDP if the benefit is set at the minimum rural poverty line. 9In order to have a significant impact on poverty, a UBI would require an expenditure close to five percent of GDP, as was the case in Iran and as has been proposed in India. 10This is more than three times the average spending on social assistance in low and middle income countries.As Luke Marinelli, an economist at the University of Bath said, "An affordable UBI is inadequate, and an adequate UBI is unaffordable." 11 An alternative to UBI that also seeks to ensure a minimum income level is the negative income tax or NIT.Milton Friedman proposed to give people below a certain level of income a transfer effectively extending the progressive income tax schedule into negative territory.Tondani (2009) shows that both UBI and NIT can result in the same outcome in terms of net income distribution.However, the two options differ substantially in terms of implementation.First, a UBI requires a larger fiscal footprint -both spending and taxes -than the NIT.Most 5 Salehi-Isfahani (2014).
6 Yeung and Howes (2015).Poverty rates were 32.6 percent without the UBI, and 21.6 percent with the transfer; inequality as measured by the Palma ratio declined from 1.76 to 1.39.
11 https://www.nature.com/articles/d41586-018-05259-xrelevant for developing countries, however, is the fact that a prerequisite for the NIT is the ability to accurately assess the income of an individual or family.This is simply not possible in low and middle-income countries where the informal sector is dominant and most income at the bottom of distribution is unobserved.
On the other hand, availability of data that is correlated with income or consumption is increasing rapidly with the emergence of digital government.It has made it possible to use multiple administrative databases and social registries that cover most of the population to rank households based on proxies and mimic a negative income tax.By using proxies that are normally correlated with consumption based on household survey data, the government could 'claw back' benefits for more affluent households based on consumption estimates.
This tapering of a cash transfer with broad but not necessarily universal coverage is similar to what is done with the social pension in Chile.Households are ranked from poorest to richest based on administrative data and those in the bottom 60 percent of the distribution and with elderly members are eligible for the noncontributory pension.This is reduced however, for each peso of contributory pension income.The result is a tapered benefit, which tries to alleviate old age poverty while increasing contributory savings.
Here "taper" refers to decrease in benefits with increase in income or consumption.A tapered UBI or TUBI could take an infinite number of shapes depending on the parameters of the taper chosen.The next section simulates the poverty impact of a UBI and two particular forms of TUBI using data from 52 low and middleincome countries.

II. Data and Methodology
The simulations which follow are based on data found in the Global Micro Database (GMD), a collection of household surveys that have been is harmonized across countries. 12The countries were selected based on data availability from 2008 or later.There are 52 countries with around 4.2 million observations in total.To estimate poverty rates, per capita income or per capita consumption variable in LCU is used depending on which concept is used to measure national poverty in a particular country. 13lative poverty rates are measured since the objective here is to compare the poverty impact of two types of cash transfers -UBI and TUBI -in both low and middleincome countries.The poverty line is set at 50 percent of median per capita income or consumption in each country.The poverty impact of a Universal Basic Income (UBI) scheme is simulated where the amount of transfer to each individual is 5 percent of average per capita income or consumption.The total budget required for such a universal scheme varies across countries costing between 1.8 and 5.3 percent of GDP.Keeping this budget constant, a TUBI is designed where bottom quintiles receive a higher transfer amount, which then gradually decreases to zero (Figure 1).
Since the budget required for these schemes is much higher than what LMICs currently spend on their safety nets programs, an alternative TUBI design is also considered where the total budget is reduced to half while coverage is also until sixth decile (Figure 1).
The TUBI can have both an income effect and a substitution effect whereas UBI generates only an income effect. 14There is little empirical evidence on any significant impact on labor market behavior for the UBI. 15The impact of the TUBI will depend on, among other things, the marginal tax rate implied by the taper.
14 The income effect is caused by recipients' overall income increase.The substitution effect takes place when work earnings are partially offset by reduced TUBI cash transfer.

Figure 1: Comparison of UBI and TUBI Transfer Schedule
Source: Author's calculations.Note: TUBI1 uses the same budget as UBI while TUBI2 uses half the budget.The simulations also ignore the differences between each type of transfer in terms of inclusion and exclusion errors and differences in the cost of administration.
Finally, this analysis has thus far ignored the impact of financing the transfers.This is done in the next section.

III. Results and Implications 16
The average impact of a flat UBI of 5 percent of average per capita consumption or income on the head count poverty rate is around 4.6 percentage points.Figure 2 shows that this hides significant variation across countries -between 1.9 and 10.3 percentage points.
The impact is higher in countries with a greater degree of inequality, probably because average per capita consumption is higher relative to that in lower quintile. 17or example, in Botswana, there is a 6.6 percentage point reduction in headcount poverty.Similarly, in South Africa where the poverty gap index is 7.4 percent before the 16 Results by country are available upon request.17 IMF 2017 transfers, the reduction in poverty rate due to UBI is 10.3 percentage points.The average reduction in inequality is 1.9 Gini points. 18xt, we compare the poverty impact of UBI with TUBI using the budget as described in Figure 1.All else equal, TUBI has a greater poverty impact than UBI ranging between 3.1 and 19.1 percentage points. 19However, if the budget is kept constant, targeted programs, even with high exclusion errors, lead to greater improvement in welfare as compared to universal programs (Hanna  and Olken 2018).It is primarily because of a higher per beneficiary transfer to the poor.The impact on headcount poverty rates of a TUBI with a lower budget is similar as shown in Figure 3.This is due to much steeper taper, where payments end by the sixth decile rather than significantly lowering the average amount of transfer to bottom deciles.This helps illustrate a key trade-off, namely, that maintaining the poverty impact with a lower budget comes at the cost of increasing the 18 If a flat 5 percent UBI is implemented along with a 5 percent VAT on individual welfare, the average reduction in headcount poverty rate and poverty gap index is 4.3 and 2.3 percentage points respectively.
19 Country level results available on request.

Note:
The budget for TUBI1 is the same as the UBI while TUBI2 uses half the budget.
probability that work effort is reduced due to the higher marginal tax.At the same time, it shows that a more realistic level of spending on an TUBI can have the same impact on poverty as a UBI that requires a budget twice as large.
To this point, we have ignored the impact of the revenue side of the equation.While it is possible that existing spending could be reallocated to these programs (including blanket subsidies in some countries), in most cases, it would require an increase in total social assistance spending.This could be financed by a variety of new taxes some of which would offset the poverty reduction observed in Figures 2 and 3. Consumption taxes are more regressive than other taxes 20 such as land or wealth taxes and would undo some of the gains for the poor thus reducing the poverty impact.In this sense, simulating the use of a value added tax to finance the above UBIs or TUBIs provides a kind of lower bound estimate of the net poverty impact.
Figure 4 shows the impact of levying a flat 5 percent VAT to finance UBI or TUBI. 21Post-taxes, the transfer to the bottom quintiles shown by the solid line remains positive with only minor reductions to the amounts received by the poorest deciles.The net transfer becomes slightly negative by the 7 th decile and sizably negative for the top quintile.The top quintile, on average, pays twice as much as the bottom decile receives.
Figure 5 shows the largest, median, and smallest net poverty impact of the combined tax and transfer in a VAT-financed, UBI or TUBI.The latter option yields twice the impact for the same budget.The results mirror the 20 Recent evidence suggests that consumption taxes are less regressive than previously thought, see Bachas et al., (2018).
21 LAC countries have been excluded from these simulations because of lack of household consumption estimates in household survey data.findings of Harris et al. (2018) where a UBI funded by eliminating VAT exemptions leads to larger net gains to the poorer households.Here a more progressive TUBI is financed by the VAT with even stronger results.

IV. Implementation Challenges
As discussed, these simulations ignore administrative costs.These will be higher for the TUBI due to the need to rank households by level of welfare. 22Would these costs significantly alter the results?Data on the cost of targeting are rarely published and difficult to disentangle in most cases.
The figures that are available suggest that the huge advantage of the tapered benefit over the UBI will not change significantly.For example, the massive enrollment process used to score households in Pakistan for its cash transfer scheme cost US$2.2per household yielding a ratio of about 6.5 percent of annual spending.
Assuming that this is repeated every three or four years, the reduction in funds available compared to a UBI would be less than 2-3 percent of total spending.A similar exercise in Bangladesh yielded a of about 5 percent if the cost was incurred once every three years.
These are examples of 'census sweeps' or surveys that are conducted periodically and apply to most or all of the population.Middle-income countries tend to use an application process that includes a determination of eligibility.Table 1 shows the total administrative 22 Other costs are common to both UBI and TUBI.Ideally, there would be no need for an application or enrolment process.A population registry with uniquely identified individual members of households would be harnessed and payments would be made directly into bank accounts.These were essentially the conditions that allowed for the almost universal transfers in Iran and Mongolia but would represent a major challenge for most developing countries where identification systems are often rudimentary and do not cover the entire population and financial inclusion is limited.
costs and the share of the cost related to eligibility determination, i.e., targeting.The figures show an average targeting cost of 3.5 percent of total spending.However, the cost of targeting will be significantly higher if extensive data is collected periodically to determine or update poverty status.
Increasingly, countries can use administrative databases that contain data on individuals and households that allow them to assign scores based on proxy indicators such as owning vehicles, land or property, energy consumption, and other variables correlated to consumption and wealth.The shift away from cash and towards digital transactions will add to the amount of data available.Turkey already uses 28 databases in its integrated social information system to determine relative need for the purposes of health insurance and social assistance.Egypt does something similar using 34 linked databases.The ability to link so much personal data obviously implies the need for safeguards of the kind recently adopted by the European Union.23  approaches allow policymakers to mimic the kind of progressive income tax system found in high-income countries and to implement a TUBI.However, to the extent that errors are introduced relative to a universal benefit, the simulated reduction in poverty rates will be lower. 24Having moved beyond the narrow targeting that currently prevails however, the exclusion errors should be greatly reduced although the error will now take the form of a greater or lesser amount of transfer than appropriate.Nevertheless, arbitrary differences between similar households should be reduced compared to a dichotomous world where many receive nothing.This will cost more in most countries.Some countries could finance UBIs or TUBIs by reforming universal subsidy programs as in the case of Iran's energy subsidy reforms.A budget neutral carbon tax of the type being rolled out in Canada is another environmentally friendly option. 25Countries with natural resource revenues could choose to distribute it in the form of cash transfers, as in the case of the state of Alaska in the United States.A significant number of countries already spend the share of GDP simulated here and could replace those programs, as was proposed in India.Finally, some countries might choose to increase taxes to raise the required funds; in those cases, the incidence of the tax would need to be taken into account as above. 26

VI. Conclusions
Proponents of universal basic income argue that it avoids exclusion errors and minimizes potential negative behavioral effects that arise in targeted programs.Skeptics focus on its high cost relative to its impact on poverty.
Financing the UBI with a progressive tax could yield the same impact as a negative income tax in countries where incomes can be observed.But incomes cannot be observed in most developing countries.Instead, extending the logic of the proxy means test combined with increasingly available administrative data makes it possible to rank households and gradually reduce the transfer in a way that mimics the progressive income tax.The simulations presented above showed that on average the most gradual taper rate reduced poverty by twice as much as the UBI.The higher the taper rate the lower the cost but the greater the marginal tax rate on consumption/income.Either type of transfer will require additional resources, but our simulations suggest that even if financed by an earmarked VAT, the net effect is highly progressive.If governments are concerned with exclusion error among the poor, they should expand the budget and coverage of the program significantly above the poverty line.Gradually reducing the transfer to those with higher levels of consumption will reduce poverty rates much more than a UBI for the same level of spending.Ultimately, there is no cheap alternative to a cash transfer with broad coverage if significant poverty reduction with minimal exclusion is to be achieved.

Figure 2 :
Figure 2: Reduction in Headcount Poverty Rate with UBI

Figure 3 :
Figure 3: Change in Poverty Rates after TUBI vs. UBI

Figure 4 :Figure 5 :
Figure 4: Net Transfer of TUBI after VAT Financing

Table 1 : Administrative Cost of Last-Resort Income Support Program
Source:Tesliuc et al. 2014.Note: The administrative cost in Bulgaria is much higher than other countries because the program has a very stringent eligibility criterion, which makes it costly to implement.