Pensions Social Protection & Labor Policy Note March 2017 | Number 21 Highlight What Can We Learn about This paper uses the FINDEX and administrative data to Pensions from the FINDEX estimate the gender, age and income composition Data?1,2 of workers covered by a mandated pension Zaineb Majoka and Robert Palacios scheme in a large number of developing countries. We find that the pension coverage gender gap is I. Overview around 25 percentage points; The life cycle theory of consumption is based on a consumer choice model where men and women are 63 and individuals try to maximize their utility by adjusting their consumption based on 37 percent of those covered, expected income over their lifetime. Since income and consumption needs vary over respectively. time, consumption is likely to be highest in middle age, whereas it declines around retirement. Research studies have tried to use this model to understand the savings It also provides evidence behavior of older individuals, but the findings are not consistent. Some studies have found that elderly continue to save even in retirement while others find a hump- on the determinants of shaped relationship where savings decrease when individuals retire.3 Baldacci et voluntary savings for old age. al. (2010) look at household consumption and saving patterns in relation to public Along with the expected spending on pensions and find that in emerging Asian economies, a 1 percent positive relationship with increase in public spending on pension raises consumption by 1.5 percent.4 income levels within and Hence, understanding patterns of savings and factors responsible for the variation has across countries, we find important policy implications. The objective of this note is to complement the existing that individuals are less knowledge on saving patterns and pension systems by looking at: likely to save for old age 1. Gender, age and income composition of workers potentially covered by mandatory the more the government pension schemes; and spends on public pensions. 2. Micro and macro determinants of voluntary saving for old age. Other things constant, men and individuals with an account in a formal financial institution are more likely to voluntarily save for old age. 1 This note was authored by Zaineb Majoka and Robert Palacios. The authors can be contacted at mzaineb@ worldbank.org and rpalacios@worldbank.org 2 Citation guideline: Majoka, Z. and Palacios, R. 2017. “What Can We Learn about Pensions from FINDEX Data?” World Bank. Washington D.C. 3 Crown, W. H. 2002 4 IMF 2011. 1 Policy Note: Pensions March 2018 | Number 21 II. Data and Methodology Figure 1: Pension coverage and regular wage This policy note uses the Global Financial Inclusion Database earners compared or FINDEX5 and World Bank administrative data on pension systems around the world.6 FINDEX provides a rich source % reported to be contributors to pension scheme 80 of information on a variety of topics, but two questions in y = 0.0087x2+0.6218x–2.5629 particular provide a unique opportunity to understand pension 70 R2 = 0.7892 coverage and the determinants of voluntary savings for old 60 age: whether an individual received a wage payment in the last 12 months and if an individual saved for old age in the 50 past 12 months. Binary variables (0/1) of responses to these questions are used in the analysis. Administrative data on 40 coverage, presence of social pensions, median income, and 30 pension spending are used to complement this. 20 For summary statistics, we look at only low and middle income countries7 whereas a full range of countries, depending on data 10 availability, is used in the regression analysis. We use a probit 0 model clustered at a country level, to assess the impact of micro 0 20 40 60 80 and macro variables on the probability to save for old age.8 % reporting receiving regular wage The paper is organized as follows: The next section presents Source: FINDEX data and World Bank pension database; excludes high income countries and five countries that do not have mandated national the rationale for estimates of mandated pension coverage and pension schemes. its composition. Section IV presents regression results aimed at identifying the determinants of voluntary savings for old age. Particular attention is paid to the effect of public pension spending given the importance of the subject in the pension Despite the high correlation, the 45 degree line suggests that literature and for public policy. The last section concludes and some wage earners, particularly in lower income countries, are suggests areas for further research using the FINDEX database. able to avoid payroll taxes. Nevertheless, these workers are the best proxy for contributors for whom individual level data are available at a global level. III. What Can Be Inferred from FINDEX Using wage earners as a proxy, we can begin to roughly about Who Is Covered by Mandated estimate the composition of those covered by mandated Pensions? pensions in each country. To begin with, we find that there is a significant gender difference in most countries, suggesting Mandatory pension schemes may apply to self-employed that men are far more likely to be covered by pensions than workers or others that do not receive regular wages, but the women. The figures are shown for each country in Annex 1. vast majority of those contributing to these schemes are wage By weighting them by the actual adult population we estimate earners. Therefore, we expect a strong correlation between the that globally, approximately 6 out of 10 people covered by the share of working age people in a country that reports having a pension system are men.9 The data also allow us to estimate the regular wage and the share of a working age group contributing distribution of covered population in income quintiles10 and to a mandated pension scheme. The FINDEX survey asks age groups (see Figure 2). The prime working age population whether an individual received a wage in the past 12 months. covered by pensions is on average older than those not covered. By comparing this variable for each country with the World As expected, older workers tend to be nonwage earners Bank’s database of administrative data on pension coverage in reflecting the fact that they are not eligible to receive pensions. Figure 1, we can observe that this correlation does indeed hold. Finally, we find that more than 49 percent of those likely to be covered are found in the top 40 percent of the income distribution of their respective countries.11 Only 13 percent are found in the bottom quintile. 5 In 2011, the World Bank launched the Global Financial Inclusion database or FINDEX to track progress on a range of indicators related to financial inclusion. The indicators are based on interviews with about 150,000 nationally representative and randomly selected adults age 15 and above in more than 140 economies. 9 Of the total wage earners, approximately 63 percent are male and 37 percent 6 World Bank 2014. are female, accounting for the difference of 25 percentage points. 7 We use median income to define the cutoff range. There are 76 countries left 10 Available on request. in the sample after dropping high income countries. 11 Quintile distribution of population potentially covered by pension schemes 8 Bryan, M. L. and S. Jenkins 2013. in each country is available on request. 2 Social Protection & Labor | World Bank Group old age, we use the current ratio of pension spending to GDP Figure 2: Age distribution of wage earners as a proxy for pension wealth. Other studies have shown that (potentially covered by mandated pension there is a close relationship between the unfunded pension liability of a country and its current level of pension spending, scheme) particularly in mature pension schemes.14 While this seems 50% intuitive due to the long-term nature of pension obligations, No wage Wage the direct relationship between stock and flow will be weaker in immature schemes and where there have been major 40% reforms which reduce future spending. These conditions apply to a minority of countries. 30% Median per capita income is used in the analysis to control for the effect of income level on private savings. Several cross- 20% country studies show that the impact of real per capita income on private savings is more pronounced in low income countries 10% as compared to the high income countries.15 Other macro level variables, such as pension coverage rates and presence of a social pension scheme control for whether an individual is 0% 15–24 25–39 40–49 50–59 60+ likely to have access to alternate schemes that can negatively affect the probability to save for old age. Source: FINDEX Data. Individual characteristics affecting savings behavior are well known and include income level and age. Another study by Demirguc-Kunt et al. (2016), tests the effect of a variety of IV. Pensions and Savings for Old Age independent individual and macro variables on probability to save for old age.16 They find that individuals in the 36–45 age One of the most important empirical questions in the literature group, belonging to the top income percentile with education is how a pension system affects an individual’s propensity to and employment are more likely to save for old age. Also, save for old age. Since the 1970s,12 researchers have looked having an account at a financial institution increases this for evidence that public pay-as-you-go pension schemes that probability while there is only a small overall gender gap in increased pension wealth were offset by reductions in voluntary saving for old age. The effects of these variables must be taken savings as individuals adjusted their behavior based on the into account in order to isolate the pension wealth effect. classic life cycle consumption theory. In more than two dozen studies at both the macro and individual levels, most found This note uses a more parsimonious model and controls for evidence of some offset.13 The policy implication of this finding median income per capita in explaining the savings decision. suggests that the most common form of national pension A confounding factor in the analysis relates to what was found system is likely to reduce national savings with consequent in the previous section, namely, that a subset of the individuals effects on economic growth and capital market development. tested are covered by a mandated pension scheme. When many of these individuals answer the question ‘did you save for old The FINDEX data provide the first opportunity to test this age in the last 12 months?’ their positive response may simply relationship using a global data set, albeit with a discrete rather confirm that they contributed to the mandated scheme. It is than continuous dependent variable. The data tell us only important, therefore, to try to distinguish between this group whether a person claims to have saved for old age, but not and those who are not covered by the mandate given our how much they saved. Nevertheless, given that in middle and interest in what drives voluntary savings. lower-middle income countries, only 13 percent of individuals claimed to have done so, the determinants of this dichotomous Table 1 presents the results for our main specification for both choice are revealing. types of worker, those that do and do not earn regular wages.17 In order to test the proposition that greater pension wealth is associated with a reduced tendency for individuals to save for 14 See Holzmann, Palacios and Zviniene 2004. 15 Loayza et al. 2001. 16 Demirguc-Kunt, et al. 2016. 12 The earliest empirical tests of this hypothesis include Feldstein 1974 and 17 We also ran two step regression (Bryan, and Jenkins 2013) using the same Munnell 1974. In contrast, there is little evidence supporting Barro’s hypothesis dependent and independent variables. Country fixed effects were added for around the same time of a kind of Ricardian equivalence that led individuals individual variables whereas samples were clustered at the country level in to save more in anticipation of higher taxes that would be needed to pay off regression with macro variables. Results were almost the same. Another set of unfunded pension liabilities. regressions was run by adding regional dummies to control for social norms 13 Six of these studies found no significant effect. See IMF 2011 for a summary and with and without coresidence rates. The results were almost the same and of the evidence. can be shared on request. 3 Policy Note: Pensions March 2017 | Number 21 Table 1: Multivariate analysis of the choice to save for old age, workers with and without regular wages Dependent Variable: Saved for Old Age in Past 12 Months (yes = 1) Independent Variables Margins (wage earners) Margins (nonwage earners) Male –0.001 0.014*** [0.008] [0.004] Respondent Age 0.0016*** 0.008*** [0.003] [0.002] Respondent Age Squared –0.000*** –0.000*** [0.000] [0.000] Account Holder (Yes/No) 0.183*** 0.116*** [0.020] [0.011] Household Income per Capita Quintiles: Top 0.122*** 0.068*** [0.014] [0.008] 4th 0.089*** 0.062*** [0.015] [0.008] 3rd 0.052*** 0.038*** [0.013] [0.008] 2nd 0.037*** 0.016*** [0.016] [0.006] Bottom Ref. Ref. Log of Median Income 0.120*** 0.052*** [0.019] [0.017] Pension Spending –0.017*** –0.011** [0.005] [0.004] Total Contribution Rate*Coverage –0.000 0.000 [0.000] [0.000] Social Pension (Yes/No) 0.053 0.040 [0.041] [0.047] Predicted Probability 0.2890 0.1598 No. of Observations 29,684 55,602 Pseudo R 2 0.112 0.148 Note: Marginal effects along with robust standard errors in brackets clustered at country level. Asterisks denote the following levels of significance: * p < 0.10, **p < 0.05, *** p < 0.01. The regression includes the standard explanatory variables financial sector and as expected has a positive and statistically on age but uses the income quintile to which the individual significant effect on probability to save for old age whereby belongs rather than a continuous income variable. Both individuals who have an account are 63.3 to 72.6 percent more variables show the expected positive signs and are statistically likely to save for old age. significant. The quadratic form of age is also added to the analysis to control for variation in trend as an individual grows The median income per capita variable captures the large older. It has the expected negative sign which indicates that cross-country variation and is also positive and significant. The the probability to save for old age increases with age until an main difference in the results for these conditional variables is individual hits the tipping point after which the probability to that men are more likely to save for old age only among those save decreases. not earning regular wages. This likely reflects the fact that the regular wage earners are the ones covered by the mandated Another binary variable added to the model is whether an contributory scheme. Whether the formal sector worker is a individual has an account at a financial institution. This can be man or a woman, he or she is forced to contribute so that there treated as a proxy for financial literacy or access to the formal will be no difference between the sexes. 4 Social Protection & Labor | World Bank Group The results shown in Table 1 also include two variables related pension schemes across the low and middle income countries to the design of the pension system in each country that could and to add to the existing evidence that individuals save less potentially affect savings for old age. The first is the mandated when their pension wealth is increased through public pension contribution rate. Again, this should affect only the population schemes. In both cases, it is the first time that the evidence covered by the mandate which may choose to save less for is based on such a comprehensive sample of low and middle retirement in the face of large mandated contributions. The income countries. second is the presence of a large noncontributory pension (social pension). This is a dichotomous variable based on an The database also contains other variables that would be worth estimate of social pension wealth.18 Although the signs on the considering as possible factors influencing savings for old age. coefficients on both variables are what would be expected, For example, the survey covers the propensity to save for other neither is statistically significant. purposes such as emergencies and if an individual borrowed money for medical needs. Combined with information about Meanwhile, the pension spending variable is statistically out-of-pocket health spending at the country level, further significant for both types of workers but more so for wage analysis might reveal indirect positive effects of expanding earners as would be expected. A one percentage increase in health insurance coverage on individual’s willingness to save pension spending is associated with a 5–6 percent reduction for old age. Also, as time series data become available, the in the probability that an individual saves for retirement. impact of major changes in pension policy can be studied by Although not shown here, this result held over a number looking at changes in voluntary savings behavior. of alternative specifications including those controlling for country. References The results are in line with the most prevalent finding in the empirical literature, namely, that individuals in countries Baldacci, Emanuele, Giovanni Callegari, David Coady, Ding where governments make larger pension commitments tend Ding, Manmohan Kumar, Pietro Tommasino, and Jaejoon Woo. to save less for old age. Conversely, this implies that reforms 2010. “Public Expenditures on Social Programs and Household that reduce pension wealth—increasing the retirement age, Consumption in China.” IMF Working Paper 10/69, International reducing accrual rates, taxing pension income, etc.—should Monetary Fund, Washington, DC. http://www.imf.org/external/ result in higher savings rates. pubs/ft/wp/2010/wp1069.pdf. The result for workers not earning regular wages is surprising Bryan, Mark and Stephen Jenkins. 2013. “Regression Analysis at first glance. Why should workers not covered by the pension of Country Effects Using Multilevel Data: A Cautionary Tale.” system also react to high pension spending/wealth? A possible Discussion Paper 7583, Institute for the Study of Labor, Bonn, answer is that many of these workers do have accrued pension Germany. wealth themselves or through their spouses. In middle income countries, there is much evidence that workers move in and Crown, William. 2002. “Life Cycle Theories of Savings and out of the formal sector labor force. To the extent that they Consumption.” Encyclopedia of Aging. Accessed August 30, 2016. can count on some of the future pension spending for their http://www.encyclopedia.com/doc/1G2-3402200229.html. old age, they are likely to save less. The data also confirm that Demirguc-Kunt, Asli, Leora Klapper and Georgios A. Panos. women are much more likely to work without regular wages 2016. “Saving for Old Age.” Working Paper 7693, World Bank, and therefore not to be covered by the pension system. To the Washington, DC. extent that their spouses are covered by the pension system, they are also potentially affected by pension wealth. Feldstein, Martin. 1974. “Social Security, Induced Retirement, and Aggregate Capital Accumulation.” The Journal of Political Economy, 82(5): 905–926. V. Conclusions and Direction for Future Research Holzmann, Robert, Robert Palacios and Asta Zviniene. 2004. “Implicit Pension Debt: Issues, Measurement and Scope in This paper has used the FINDEX database to shed light on the International Perspective.” Social Protection Discussion Paper likely composition of the population covered by mandatory 0403, World Bank, Washington, DC. International Monetary Fund. 2011. “The Challenge of Public Pension Reform in Advanced and Emerging Economies.” IMF, 18 This measures the present value of future spending (through 2040) on social pension in each country assuming current parameters and country specific Washington, DC. demographics. See Palacios and Knox-Vydmanov 2014. 5 Policy Note: Pensions March 2017 | Number 21 Loayza, Norman, Klaus Schmidt-Hebbel and Luis Serven. 2001. Performance Indicators.” Public Administration and Development, “Saving in Developing Countries: An Overview.” The World Bank 34: 251–264. Economic Review, 14(3): 393–414. World Bank. 2014. “Pensions: Data.” Retrieved from: http://www Munnell, Alicia. 1974. “The Effect of Social Security on Personal .worldbank.org/en/topic/socialprotectionlabor/brief/pensions- Saving.” National Tax Journal, 27(4): 553–567. data. Palacios, Robert and Charles Knox-Vydmanov. 2014. “The Growing Role of Social Pensions: History, Taxonomy and Key 6 Social Protection & Labor | World Bank Group Annex 1: Gender Distribution of Wage Earners by Country Country Female Male Country Female Male Albania 40.66% 59.34% Madagascar 49.38% 50.62% Angola 27.59% 72.41% Malawi 34.33% 65.67% Armenia 38.53% 61.47% Mali 29.92% 70.08% Azerbaijan 29.64% 70.36% Mauritania 38.92% 61.08% Belize 42.18% 57.82% Mauritius 39.01% 60.99% Benin 38.57% 61.43% Mexico 41.95% 58.05% Bhutan 27.14% 72.86% Moldova 50.41% 49.59% Bolivia 37.94% 62.06% Mongolia 46.33% 53.67% Botswana 41.73% 58.27% Montenegro 46.09% 53.91% Brazil 39.64% 60.36% Namibia 36.46% 63.54% Bulgaria 48.98% 51.02% Nepal 33.92% 66.08% Burkina Faso 40.58% 59.42% Nicaragua 34.98% 65.02% Burundi 38.83% 61.17% Niger 28.17% 71.83% Cameroon 32.48% 67.52% Nigeria 33.43% 66.57% Chad 27.43% 72.57% Pakistan 16.03% 83.97% China 40.98% 59.02% Panama 36.46% 63.54% Colombia 38.73% 61.27% Peru 35.35% 64.65% Congo, Dem. Rep. 23.70% 76.30% Philippines 36.88% 63.12% Congo, Rep. 38.43% 61.57% Romania 45.82% 54.18% Côte d’Ivoire 24.06% 75.94% Rwanda 37.88% 62.12% Dominican Republic 42.25% 57.75% Senegal 33.36% 66.64% Ecuador 33.29% 66.71% Serbia 47.34% 52.66% El Salvador 38.91% 61.09% Sierra Leone 38.53% 61.47% Ethiopia 35.76% 64.24% South Africa 50.85% 49.15% Gabon 41.43% 58.57% Sri Lanka 34.31% 65.69% Georgia 42.15% 57.85% Sudan 39.14% 60.86% Ghana 31.75% 68.25% Tajikistan 32.51% 67.49% Guatemala 36.68% 63.32% Tanzania 47.17% 52.83% Guinea 26.05% 73.95% Thailand 37.05% 62.95% Haiti 26.74% 73.26% Togo 36.66% 63.34% Honduras 25.53% 74.47% Tunisia 22.39% 77.61% India 29.49% 70.51% Turkey 24.87% 75.13% Indonesia 35.69% 64.31% Uganda 35.95% 64.05% Jamaica 44.56% 55.44% Ukraine 51.36% 48.64% Kazakhstan 51.74% 48.26% Venezuela, RB 39.45% 60.55% Kenya 46.46% 53.54% Vietnam 38.01% 61.99% Kyrgyz Republic 49.26% 50.74% West Bank and Gaza 15.70% 84.30% Macedonia, FYR 39.13% 60.87% Zambia 40.93% 59.07% 7 The findings, interpretations, and conclusions expressed herein are those of the author(s), and do not necessarily reflect the views of the International Bank for Reconstruction and Development/The World Bank and its affiliated organizations, or those of the Executive Directors of The World Bank or the governments they represent. The World Bank does not guarantee the accuracy of the data included in this work.