20 minute read

6.2 Treatment of Assets in Means-Tested Programs

358 | Revisiting Targeting in Social Assistance

BOX 6.2

Treatment of Assets in Means-Tested Programs

In many Organisation for Economic Co-operation and Development countries, means-tested social assistance programs use measures of income and assets. Wealth may be measured as the sum of all assets of the assistance unit (the wealth index approach), which is compared to a threshold. Alternatively, the ownership of individual assets, regardless of value, can be used as grounds for declaring a household ineligible (the asset filter approach). Such treatments of wealth are used to exclude asset-rich households from accessing the program regardless of their current income level. To be eligible for a program, households should have both a wealth level below a given threshold and an income below a given threshold.

As with income, this creates disincentives to accumulate and perverse incentives to misreport the level of wealth. To counter these effects, programs may disregard certain assets from the asset test but not others. For example, in Greece,a households in the Social Solidarity Income program do not have to sell their houses and vehicles before applying to the program. A household can participate in the program as a homeowner if the total taxable value of the property does not exceed €150,000 (US$170,000), and/or the value of the household’s vehicles is estimated at a maximum of €6,000 (US$6,741), calculated according to Article 31 of Law 4172/2013.b However, owning other assets, such as a private recreational boat (exceeding 5 meters in length and with engine power exceeding 50 cubic centimeters), aircraft, helicopters, gliders, swimming pools, as well as financial assets over a certain threshold, automatically excludes households from the program, regardless of their income. The same applies to the current value of shares, bonds, and so forth owned over the six months prior to the application, compared with a threshold as a signal of well-being. This implies that households are expected to draw down at least some liquid assets before receiving public support. In the United States,c a few states use “broad-based categorical eligibility,” which allows lowincome Supplemental Nutrition Assistance Program beneficiary households to keep some assets, such as a car if it is used to find and keep a job; a house if the demand to the program is due to a short-term shock (for example, loss of job, divorce, or unexpected temporary disability); and a car and house if the household has seniors or people with disabilities. In Lithuania, households are excluded if the value of their property exceeds the average property value set for the residential

continued next page

How to Harness the Power of Data and Inference | 359

BOX 6.2 (continued)

area, which is determined by the government according to the norms set for housing and land, as well as other assets including vehicles, livestock, durables, savings, and shares (Tesliuc et al. 2014).

Selecting the right combination of exclusionary asset filters and their thresholds should be done after empirical analysis. Otherwise, the filters could exclude many of the poor in the target group, thus reducing the effectiveness of the program. Programs should use a representative household survey with information on household welfare as well as asset ownership or value. Failing to calibrate the asset filters, without proper ex-ante simulation of their impact, could lead to a high level of exclusion error.

The same consideration applies to programs that use the hybrid means test, which is covered in the next section. Tesliuc et al. (2014) report cases in which setting up such filters relied only on common sense or the beliefs of the social assistance administrators, instead of an empirical assessment, which resulted in large exclusion errors. In Albania’s Ndihma Ekonomike, circa 2008, the list of asset filters and conditions was estimated to exclude 90 percent of the poor from the target group in urban areas, especially working-age poor families. This was later corrected; determination of eligibility for the program was switched to a proxy means test in 2018. In Uzbekistan, the imputation coefficients for farming income were set too high and resulted in the exclusion of many poor households. In the Guaranteed Minimum Income program and the Family Allowance program in Romania in 2012, a list of 20 asset filters, which individually excluded no more than 1 to 9 percent of the households in the income target group, collectively excluded one-third of the intended beneficiaries. This situation was subsequently corrected by collecting data for simulations and calibrating the vector of asset filters to minimize the exclusion error.

Thus, the use of asset filters has both pros and cons. The use of exclusionary asset filters may prove effective for improving the focus on extreme or chronic poverty by reducing the inclusion errors and at the same time ensuring the legitimacy of the program. However, if filters are not well set, this approach may lead to exclusion errors that reduce the effectiveness of the program in reducing poverty.

a. World Bank (2019b). b. Annual objective expense for a passenger car for private use adjusted according to its age: no reduction for a period up to 5 years; a rate of 30 percent for a period of more than 5 years and up to 10 years; and a rate of 50 percent for a period of more than 10 years. c. Rosenbaum (2019).

360 | Revisiting Targeting in Social Assistance

hard times to become eligible for public support quickly. This feature will make the program faster in responding to negative income shocks. To maintain a program’s adaptability, the recertification or updating of the income of the household should also be frequent. The degree of adaptability will also depend on the frequency of income reporting in the other administrative systems.

There are some practical limits to how quickly information may be updated depending on the concepts and systems. What is practical in a given country would depend on the strength of the program’s delivery system (a topic covered in chapter 4), as well as the broader information ecosystem in the country. In general, wages are reported at least monthly or quarterly to the social security agency, but income tax records are filed only annually. In Chile, for example, labor income, pensions, vehicle ownership, and health insurance fees can be verified monthly, while capital income, school fees, and household ownership can be verified annually. In Turkey, the self-reported data included in a citizen’s application form are verified both internally and externally. The information in the system is real-time information. It is updated every second. However, when there is an evaluation for the eligibility of social assistance, before each evaluation, the system conducts an online search automatically. This process is completed once a month since social assistance payments are mostly monthly. Nevertheless, the ministry can see the most recent data related to the registered people through active inquiry of the system. Therefore, the means test must be designed considering the many sources of volatility of income and requires significant investments for data management.

By tracking the changes in the income of the assistance unit relatively frequently, means-tested programs can also calibrate the benefit level of the program and operate a large array of benefit formulae. Means-tested programs typically fall into two categories: programs that provide a flat benefit or service, and those whose benefits differ with the income level of the applicant. However, most means-tested programs offer differentiated benefits.

The most frequent means-tested programs that differentiate benefits as well as eligibility are the guaranteed minimum income programs, whose benefits equal the difference between the income level of the applicant and a guaranteed minimum. The guaranteed minimum is often linked to a measure of the subsistence level, which incorporates a range of relevant expenditures and is adjusted to reflect the composition of the assistance unit, per capita or per adult equivalent. For example, most guaranteed minimum income schemes operating in the European Union consider a broad notion of needs, including food, nonfood, and services; in contrast, the US SNAP program—which aims to ensure a minimum level of food consumption for low-income households—aims to fill only the food gap.

How to Harness the Power of Data and Inference | 361

Some countries offer one comprehensive benefit, while others use a guaranteed minimum income program as a revolving door to offer additional benefits (for example, heating or housing assistance).

By tracking changes in formal incomes frequently, means-tested programs can operate dynamic benefit formulae, for example, withdrawing benefits as a household’s earnings increase. The steepness of the taper is an important design variable. Steeper tapers target more narrowly, and flatter tapers allow more people above the eligibility or poverty line to receive some (reduced) benefits. Steeper tapers (also known as higher marginal tax rates) are expected to reduce work effort more than flatter ones. Such adjustments also have some practical implications for dynamics. In countries where formality is not widespread, individuals in the informal sector may avoid entering the formal sector and losing social benefits when formal labor income is considered. In Brazil, where means testing is used despite informality, government communications campaigns reinforced that the criterion is having an income below the threshold regardless of whether the individual is a formal or informal worker, as many people still believed that formality was an exclusion factor.17

Some higher-income countries disregard a certain proportion of the earned income from the program definition of income (they do not include it in the income test) to reduce work disincentives. Beneficiaries whose income hovers around the eligibility line face an implicit 100 percent marginal tax on earnings if benefits are determined as the difference between their income and a guaranteed minimum. For this type of program, a small increase in earnings results in an equal reduction of benefits; hence, the work disincentive may trigger a reduction in the beneficiary’s work effort. Partial disregard of the earnings of last-resort income support recipients is allowed by the social assistance legislation in many countries in Europe, such as Cyprus, the Czech Republic, Estonia, Germany, the Netherlands, Portugal, Romania, the Slovak Republic, Slovenia, and Sweden (World Bank 2019a).18 In Estonia, social assistance beneficiaries’ earned income is not considered for the first two months. After two months, 50 percent of earned income is not considered. Portugal disregards a higher share of income (50 percent) for 12 months if a new job was obtained through activation measures. If a new job was found in a different way, 20 percent of the earned income is not considered. Slovenia does not count certain income from informal work, as well as casual and nonrecurring income, when determining eligibility for lastresort income support. Another example is Serbia, where, according to the 2011 Social Welfare Law, for last-resort income support beneficiaries who are able to work, the income they receive from participating in training activities organized by the National Employment Service is not counted as additional income and does not require reassessment of

362 | Revisiting Targeting in Social Assistance

eligibility for last-resort income support. A large scope of income disregards is found in the Slovak Republic, where 25 percent of wage income, 25 percent of income from occasional work, 25 percent of the activation allowance for voluntary service, and all allowances related to participation in active labor market policies are disregarded in the income test for last-resort income support (MISSOC 2020).

Another parameter of means-tested programs that is used to reduce work disincentives is the gradual phase-out or tapered withdrawal of benefits over time (World Bank 2019a). Beneficiaries can increase their take-home income by combining their earned income with the benefits. Many EU member states allow gradual phase-outs. For example, in Croatia, social assistance beneficiaries receive the full benefit in the first month, 75 percent of the benefit in the second month, and 50 percent of the benefit in the third month. Lithuania continues to pay 50 percent of the previously paid benefit for six months to those graduating from the last-resort income support scheme, to encourage their labor market participation. Hungary modified the design of its social assistance system so that last-resort income support beneficiaries continue to receive some benefits for up to six months after gaining employment. Along similar lines, Latvia’s guaranteed minimum income benefit can be received in reduced amounts for a limited duration after securing a salaried job. In Croatia, as of September 2015, last-resort income support beneficiaries who find work can continue receiving benefits in decreasing amounts during the first three months of employment. As of 2013, the Slovak Republic pays part of the last-resort income support benefit together with the wage for 12 months. In France, the minimum income benefit is received when the beneficiary is employed for up to 750 hours per year (for a maximum of 12 months).

Important factors for the success of means testing include the following: • Databases exist that provide reasonably complete information on the income and assets of the target group and cover the part of the population that is pertinent to the eligibility decision to be made. If the program is meant to be nearly universal and screen out only the top, say, tenth or quarter of the population, this may be possible even where informality is high and data systems have low coverage. Conversely, if the eligibility threshold is set at the bottom 10 percent of the population, a verified means test would require complete information for the entire population. • For automated cross-verification, the ID numbers of individuals must be collected and available in all information systems, or a business intelligence algorithm to match individual characteristics, such as name, age, gender, address, or GPS coordinates must be used.

How to Harness the Power of Data and Inference | 363

• Ensuring that different administrative databases can communicate, which includes the harmonization of technology, language and coding, and a data field dictionary, is essential when the goal is to create interoperability19 and data integration20 to improve the quality of means testing. The level of interoperability and data integration can be different according to the organizational needs, but countries must also have proper data-sharing protocols and good data protection in place. This harmonization allows different systems to interact, to validate self-declared information and generate other measurements that program administrators need to improve accuracy in the selection of beneficiaries. • Regulatory frameworks and protocols for data exchange, privacy, and confidentiality of information must be in place. • Convenient on-demand application—virtually (online or by phone) or at local centers with dedicated and trained staff to collect information from the applicant or to trigger the interoperable system of data exchange in centers must be established at least at the municipality level.

Due to their data intensity and requirements of interoperability, means tests are common in high-income countries where the economies are largely formal and extensive government databases allow for verification of incomes. The following are some examples of how they work: • The United States defines eligibility for most social programs through an assessment of household income and selected assets, with rigorous verification to improve targeting accuracy. The United States uses social security numbers and state-level IDs to integrate information systems.

Most of the registration, database management, and eligibility decisions are decentralized to the state and/or municipal (county) level, with federal oversight and fraud control for federally funded programs. For federal programs, the main program rules are set by the federal government, but there can be some room for flexibility at the state level. For example, rules set by the federal government restrict eligibility for SNAP to those with gross incomes up to 130 percent of the federal poverty line.

However, in Alaska and Hawaii, benefit levels and income eligibility requirements are higher, and states can have different eligibility levels due to the exception of broad-based categorical eligibility.21 • The Portuguese cash transfer program for the extreme poor, Rendimento

Social de Inserção, is offered to families with total monthly income below a threshold that varies according to the household size and demographic characteristics.22 The total income is defined as the sum of verifiable income from salaries, pensions, housing subsidies, and other social programs, as well as an estimate of property income and capital income.

For property income, two estimates are added to the total income: (1) 5 percent of the difference between the property value and €193,005

364 | Revisiting Targeting in Social Assistance

(US$216,860) if the property value is greater than €193,005, and (2) 5 percent of any other property. For capital income, the estimation adds a twelfth of the maximum amount between the value of income from earned capital (interest on bank deposits, stock dividends, or income from other financial assets) and 5 percent of the total value of capital income (as of December 31 of the previous year; bank account balances, shares, bonds, or other financial assets). Moreover, the program is not offered to applicants with a total value of capital income from bank account balances, shares, investments, and other financial assets that is greater than €26,145 (US$29,050). • The Greek Social Solidarity Income program uses an income assessment based on declared income during the application stage but verified through interoperability of the systems used by different government agencies, which is made possible by the uniqueness of the applicant’s social security number. Once completed, the registration form automatically pre-fills the fields in the form using the information stored with the tax authority. This information includes the name of the applicant, date of birth, family composition, and previous year’s household income based on the applicant’s tax declaration. Moreover, the Unified System of Social Insurance allows cross-verification of the employment status of all working-age adults in the household and any pension and/or benefits received, and an asset filter is applied.23 Eligibility for the Social Solidarity

Income program also includes an asset test.

Even in countries with significant informality and limited possibilities for verification of declared incomes, means tests are sometimes used, sometimes in combination with other methods.24 In China, the Dibao program operates as a means-tested guaranteed minimum income. Early in the program’s history, verification was not done with the sort of national income, social security, or property records that are commonly used in Europe, but rather through recourse to the information available to local government cadres and community members. More recently, information systems and accountability measures are being developed with more central frameworks (Golan, Sicular, and Umapathi 2014; World Bank 2021a).

In Chile, the launch of the Social Household Registry has marked a shift in the targeting method across all social protection programs, from PMT and HMT to a means test. The reform has not required collecting new information from applicants; it involves only accessing and using existing information from administrative databases (tax records, wages, social security contributions, health insurance [public and private] contributions, unemployment insurance, pensions [contributory and noncontributory], education records,25 real estate, and vehicles). The Social Household Registry uses administrative data on formal sector incomes, complemented by self-reported informal incomes, to construct an indicator of

How to Harness the Power of Data and Inference | 365

household income. The composition of the household is self-declared but interoperable with the national ID to guard against false or duplicate IDs.26 The household income is then transformed into per capita equivalent form using a set of normative needs indices for different household members. Then, households are classified into seven socioeconomic groups, corresponding to the poorest 40 percent of the population (group 1) and six other groups corresponding to deciles five to ten. The first group was left deliberately as the largest, given that formal income information alone was not sufficient to rank households into deciles one to four. If a large share of household income is derived from informal sources and self-reported, the Social Household Registry validates the income information with an assessment of household means. The means test measures the possessions of five categories of means or expenses: cars, real estate, school tuition and fees, cost of health insurance, and the balances in pension accounts. If the household is ranked high on two or more means, its socioeconomic group is increased.

The Social Household Registry’s new means test is used to prioritize all the social assistance benefits and services in Chile. A total of 80 programs use the means test/socioeconomic groups for eligibility. Some of these programs use only socioeconomic groups. This is the case, for example, of the noncontributory child grant, which is a child allowance for children of informal sector families in the poorest six deciles. Should a program target a group that is smaller than the 40 percent poorest, additional selection criteria are considered. For example, the Securities and Opportunities program is targeted toward extremely poor families and serves only a predefined quota of beneficiaries in each district or commune. For the selection of the potential beneficiaries, the households in the Social Household Registry are ranked by their specific income per adult equivalent, which is then compared with an income threshold equal to the extreme poverty line. If the quota is smaller than the resulting caseload, the program further prioritizes families with children.

Along the same lines but with more detailed verification, Brazil combines its means-test approach with geographic targeting and uses the social registry and related mechanisms of the delivery system for beneficiary selection for a large range for social programs. The Brazilian system was initially based on self-reported income with little verification when it was established in 2001 (the Bolsa Escola, Bolsa Alimentação, Cartão Alimentação, and Vale Gás programs were later merged under the Bolsa Familia program). Geographical targeting was used to distribute quotas of caseloads of participants to municipalities for budget rationed programs. Since 2005, the country has invested heavily in the Unified Social Assistance System, which created the Social Assistance Reference Centers at the local level. The main functions of the Social Assistance Reference

366 | Revisiting Targeting in Social Assistance

Centers are to assist people benefiting from Unified Social Assistance System services and to help them access other services provided under other sectoral policies. A Social Assistance Reference Center is equipped with the human resources and infrastructure needed to establish local partnerships, as well as to mobilize and inform communities around the importance of investing in human capital to increase their productivity and how to access services, programs, projects, and social assistance benefits. At a Social Assistance Reference Center, trained staff enter people’s initial contact details into the Social Assistance System. During the faceto-face sessions, which collect the self-declaration of socioeconomic conditions in the Cadastro Único, people are informed about verification measures such that false declarations will lead households to a suspension of benefits as well as penalties that would block any member of the family from receiving assistance. Moreover, Brazil has always used cadastral surveys, spot checks, and audits financed at the federal level to assess the quality of self-declared data in the Cadastro Único; people’s trust and confidence in the staff at a Social Assistance Reference Center is also believed to reduce bias and underdeclaration of income.

Since 2014/15, Brazil has performed partial verifications of income declarations in the Cadastro Único through data sharing with several government agencies. This requires different identification numbers for each household member, including the tax ID, national ID, social security number, and labor card number,27 as Brazil still does not have an official unique digital national ID. After data entry, a cross-verification process involving different information systems runs simultaneously to compare declared income with tax records or the Relação Anual de Informações Sociais (RAIS) or formal employers information system (Cadastro Geral de Empregados e Desempregados [CAGED]) and the National Social Security system. This interoperability is recent and still partial due to the high informality levels in Brazil. However, several other processes, such as spot checks and random home visits, are also part of regular monitoring. The 2019 audit of social programs done by the Federal Court of Accounts (Tribunal de Contas da União) found R$2.2 billion (US$556 million)28 in suspicious transactions of R$55.6 billion (US$14.1 billion) benefits paid, or less than 5 percent.29 A total of 449,000 benefits were considered suspicious and 65 percent of those were receiving social assistance, which required the administrators to investigate the cases. If all those on social assistance were related to Bolsa Familia, the inclusion errors would be equivalent to 2 percent, as the program reaches more than 13.5 million households.

Means testing benefits from technology and government measures that are increasing digitization and e-governance in many countries. As technology improves and systems integration and interoperability allow the provision of better service delivery, better data quality, and better

This article is from: