95198 DISCUSSION PAPER NO. 1501 Defining, Measuring, and Benchmarking Administrative Expenditures of Mandatory Social Security Programs Oleksiy Sluchynsky February 2015 Defining, Measuring, and Benchmarking Administrative Expenditures of Mandatory Social Security Programs Oleksiy Sluchynsky* February 2015 * Oleksiy Sluchynsky is a Senior Economist with the World Bank. Correspondence should be sent to the World Bank, 1818 H St NW, Washington DC 20433; e-mail: osluchynskyy@worldbank.org The author is especially grateful to Raluca Golumbeanu for assistance in data collection, Robert Palacios for very valuable comments and input, and other colleagues from the World Bank for their advice and support in conducting this research. Abstract This study provides a framework for comparison and benchmarking of administrative expenditures of public and private social security programs. The paper presents the genesis of the inquiries into the subject, reviewing some of the most relevant literature on administrative expenditures and the costs of mandatory programs produced over the past two decades. The quantitative analysis builds on the extensive body of literature, but our framework evolved considerably from earlier studies. Our dataset includes over 100 observations and a broad set of explanatory variables. We developed and compared a number of standardized cost indices discussing their advantages and limitations. We also discuss major cost components and their shares in total program costs. The analysis explains over 90 percent of variation in administrative expenditures. It confirms some of the hypotheses expressed in the earlier studies and presents new evidence of driving factors for costs. We developed three different specifications for statistical analysis. The first set looks at the impact of design of a program on total costs. The second group of specifications assesses differences in costs of managing pension liabilities between the public and private mandatory pension schemes. Finally, on the basis of the third model we generate benchmarks for staffing levels and for the total administrative expenditures. We compare those to the actual indicators and develop standard performance ratios, providing insights into design variations and performance of the programs. We conclude with a discussion of data limitations and implications of our findings. JEL Classification: H55, H83, G23 Keywords: Administrative Costs; Public Pension; Social Security; Public Administration i Table of Contents Executive Summary............................................................................................................. 1 I. Introduction ................................................................................................................. 5 II. Formulating the Research Question ............................................................................ 9 III. Scope of Analysis ....................................................................................................... 17 IV. Our Data and Structure of Costs................................................................................ 22 4. 1. Institutional organization and total expenditures ............................................. 23 4. 2. Key elements of the cost structure ..................................................................... 25 4. 3. Functional analysis: contribution collection and benefit payment .................... 27 V. Cost Normalization .................................................................................................... 29 5. 1. Uses of national income, revenues, and expenditures in cost normalization .... 31 5. 2. Administrative costs and pension liabilities ....................................................... 31 5. 3. Per-member costs............................................................................................... 33 VI. Data Analysis and Cost Benchmarking ...................................................................... 37 6. 1. Administrative expenditures and program design ............................................. 37 6. 2. Administrative expenditures and pension liabilities .......................................... 40 6. 3. Administrative expenditures and institutional organization ............................. 42 6. 4. Performance against benchmarks ..................................................................... 45 6. 5. Implications for choice of cost indices ................................................................ 47 6. 6. Global benchmarks ............................................................................................. 48 VII. Quality Aspects in Cost Measurement: What is Left to Residual .............................. 52 VIII. Conclusions ................................................................................................................ 53 References ........................................................................................................................ 55 Annex 1: List of Public Pension Programs and Abbreviations Used ................................. 58 Annex 2: Key Institutional and Operational Indicators ..................................................... 60 Annex 3: Benchmarking Performance of Public Pension Programs ................................. 62 Annex 4: Benchmarking Costs Performance .................................................................... 64 ii Tables Table 1: Summary of Literature .............................................................................................. 16 Table 2: Classification of the Public Social Security Administration ....................................... 24 Table 3: Activity of Pension Accounts (Thousands of Contributing Members) ...................... 34 Table 4: Choice of Denominator in Cost Indices and Associated Biases ................................ 36 Table 5: Administrative Expenditures and Program Design ................................................... 38 Table 6: Factors Affecting the Cost of Managing Pension Liabilities ...................................... 40 Table 7: Staffing Requirements for Pension Administration .................................................. 42 Table 8: Key Factors Affecting Costs of Public Pension Programs .......................................... 44 Table 9: Choice of Denominator for Cost Index and Correlation with Cost Benchmark ........ 48 Figures Figure 1: U.S. SSA Staffing and Cost per Beneficiary (1978–1998) ........................................... 6 Figure 2: Administrative Costs as Share of the Imputed Covered Wage................................ 18 Figure 3: Agency Rank and Median Administrative Expenditures (Income Adjusted) ........... 25 Figure 4: Costs of Managing Pension Assets .......................................................................... 26 Figure 5: Share of Benefit Payments in Banks by National Social Security Agencies ............. 27 Figure 6: Allocation of Labor Resources within the Social Security Agencies ........................ 28 Figure 7: Contribution Rate and Administrative Costs ........................................................... 30 Figure 8: Costs of Managing Pension Liabilities (Percentage of Total Assets or IPDs) ........... 32 Figure 9: Economies of Scale in Administrative Expenditures................................................ 49 Figure 10: Per-Beneficiary Cost Spreads for a Midsize Operation (Nominal US$) ................. 50 Figure 11: Economies of Scale in Staffing Requirements ....................................................... 50 Figure 12: Beneficiary per Staff Ratios for a Midsize Operation ............................................ 51 Figure 13: Quality Cost Tradeoffs ........................................................................................... 52 iii Executive Summary This study was motivated by an interest toward determinants of the operating costs of public social security programs and implications of policy reform for the institutions that administer them. A simple comparison of the administrative expenditures of the different types of schemes may often be misleading, and cost differentials do not always imply inefficiencies. A comprehensive framework is needed to address various biases and make a meaningful comparison of programs of different types, sizes, and organizations. There also is significant interest toward comparing performance of publicly versus privately managed pension schemes. The wave of reforms with partial or full privatization of the national social security programs in the 1990s and early 2000s along with the perception of excessive charges imposed by the private providers generated a considerable body of literature focusing on the private defined contribution (DC) plans. Yet, that type of research generated few implications for the public schemes and institutions. While members of the publicly managed programs do not bear the costs of administration directly, such schemes have their own risks. Public programs are prone to agency problems, often resulting in overstaffing, over-resourcing, or under-provision of quality services. Policymakers and administrators often face the same operational choices and challenges under public or private management, with cost implications. By focusing on the performance of systems and institutions rather than on cost incidence, this paper offers a generic approach to the cost analysis with some emerging recommendations relevant to schemes of all types. Our analysis builds on the extensive body of literature for both public and private pension schemes. It summarizes key findings, lays out a new systematic framework for quantitative analysis, and develops program-specific performance benchmarks for both labor resources and operating costs. The framework evolved considerably from earlier studies. Our dataset includes over 100 observations and allows for greater confidence of statistical inferences. Our data has a broader set of explanatory variables and allows zooming in on functional accounting of costs. Remarkably, our analysis explains over 90 percent of variation in administrative expenditures among the observations of our sample. We confirm some of the hypotheses expressed in the earlier studies and present new evidence of driving factors for costs. Administration of mandatory social security programs is a complex operation. There are significant systemic, institutional, and operational differences among the schemes. Sometimes, the same agency operates multiple schemes that are very diverse in nature. Often, one program can be managed by multiple agencies. We discuss several important challenges in defining comparable cost measures and propose a set of guiding principles. Availability and quality of the data is a major constraint as the data differs dramatically from country to country and from institution to institution. There are significant heterogeneities in how social security agencies report their operational and expenditure information. One 1 clear recommendation emerging from this work is a need to promote standardized reporting of operating costs, including functional accounting wherever possible. This paper consists of several parts. We first perform a structural analysis of costs and review elements associated with various aspects of operations and their contribution to the overall cost function. We also review several conventional cost indices, exposing weaknesses associated with each type of cost normalization (including uses of gross domestic product (GDP), revenues, expenditures, members, and so on). We present an alternative index in which pension liabilities serve to normalize costs. We further discuss common biases of normalization and summarize their impact on ten conventional cost indices. Our key finding from this analysis is that the best normalization for comparative analysis is achieved when using the number of members (or even better, beneficiaries only) adjusted for the level of national income (for example, GDP per capita). With that measure, we observed a group of countries with exceptionally high administrative expenditures. Notably, from 21 institutions in Sub-Saharan Africa in our sample, 14 agencies were in this outlier ategor . In all specifications of our regression analysis, this category was coded as a separate qualitative variable and came out as highly significant. Among the key findings for both structural and regression analyses is unequivocal evidence that the variable cost of benefit management is much greater than the cost of contribution collection. Mere recordkeeping of the contributors does not seem to affect the staffing requirements or overall costs in a significant way. However, the operation of contribution collection and provision of additional services does, implying the importance of fixed costs over variable costs for that line of business. This may be due to the fact that agencies do not really provide direct service to contributors and mostly interact with their employers, so statistical association with the number of contributors is loose. On the other hand, the number of beneficiaries alone explains over 80 percent of the variation in staffing levels (and hence, significantly, the total costs). This has important implications for the debate over the proper institutional home for the contribution collection function. For a well- established internal collection function, the argument for outsourcing and consolidation with tax collection is weak on the basis of cost reduction alone. It does not mean that other potential systemic improvements could not be achieved by such reforms (for example, reduced administrative burden, possibly improved compliance, or overall improvements in economic efficiency due to reduced informality). Yet, where significant investments are required to establish or modernize a collection function, both tax and social security contributions systems could benefit from a well-coordinated effort. On the benefit-management side, given significant variable costs, considerable economies of scope may exist. This may argue in favor of consolidating various benefit programs under a unified administration (for example, universal basic pensions and earnings-related pensions, retirement benefits, and various short-term or other special benefits, especially where these cover mostly the same groups of beneficiaries). 2 In our quantitative analysis, we develop three different regression specifications. The first two are equivalent to cost normalization by members and by pension liabilities, respectively. We investigate the effects of various elements of program design (such as private versus public management, in-house collection, and special schemes) as well as the level of economic development and the quality of institutions on current administrative expenditures. These are the main observations:  In line with other studies, there are clear economies of scale (expressed either in terms of scheme members or pension liabilities). Yet, there are dramatic differences in how functions of managing services for beneficiaries versus contributors add to the overall cost function. The difference can be up to one order of magnitude.  The evidence of cost differentials of in-house versus outsourced collection is weak. One possible explanation is that the modes of organizing collection function vary significantly, so capturing such a variation under one categorical variable constitutes a measurement challenge. At the same time, managing special supplementary programs and benefits (such as health, unemployment, and member loans) produces notable increments in operating costs.  While the evidence of cost differentials between defined benefit (DB) and defined contribution (DC) schemes (public or private) is weak, the results show the strong effect of private management on the costs of pension plans. However, there are indications that this effect may reflect differences in the maturity and coverage of the schemes and thus fade in the longer term. We also observe that schemes that require the management of financial assets (DB or DC) produce incremental costs, indicating advanced complementary resources (both skills and systems).  The level of economic development has a strong impact on costs, suggesting that more developed countries can manage pension schemes more efficiently, possibly taking advantage of better technologies, infrastructure, and institutions. However, using the Government Effectiveness Index, we find that as technologies spread over time they may become less important in explaining cost differences, and what may ultimately matter is the quality of governance. We further find that as economies develop and as new technologies become available they lead to the substitution of capital for labor in managing social security programs. We then proceed to our third specification that is used for benchmarking operational performance. That specification consists of two steps. The first step is to benchmark optimal uses of labor resources in program operation. The second step is to benchmark levels of current administrative expenditure. Notably, the spread between low and high estimates for programs of the same size and same economic environment can be four-fold and is driven by parameters of design and operation (for example, asset management function, in- house collection, or operation of special supplementary schemes). This suggests that inferences about the level of administrative expenditures should always be done keeping in mind the institutional context for each program. 3 We also produce individual benchmarks (in both labor resources and overall operating expenditures) for each of the programs in our sample considering the nature of their operation and their institutional context. We estimate the degrees of deviation from the individual benchmarks and develop three performance categories: categor A is for the programs that perform at or close to the benchmark; categor B is for the progra s that moderately deviate from the benchmark; and ategor C is for the programs where performance coefficients are more than double the predicted levels. Most of the programs i ategor C are suspects for operational inefficiencies, especially those in which the benchmark coefficients are multiples of the predicted levels. Out of the 11 programs where the expenditure-to-benchmark ratio exceeds 5, 8 are located in Sub-Saharan Africa. For programs in Uganda, Kenya, and Ghana, the ratios are 26, 15, and 11, respectively. It may be easy to overspend when operating significant surpluses, which all three happen to have, but excessive administrative costs certainly cannot be sustained as schemes mature. We conclude with remarks on the implications of data limitations, especially in the quality of services provided by different programs. To properly interpret the results of comparative cost studies, we point to the need to look beyond our results and use special operational and beneficiary surveys to capture information on the performance and satisfaction of various stakeholders with the administration of programs (including information on processing times, compliance costs and various overheads, and overall perception of service quality). This study provides a framework for analyzing the operational efficiency of public social security programs. It also helps guide complex organizational transformations that involve the reallocation of resources between functions, adopting new technologies, developing synergies between multiple agencies, and outsourcing. Decisions on optimal investments in systems, processes, and people require clear understanding of the key factors that affect the costs of operating schemes of various types and scope. 4 I. Introduction Mandatory social security programs play an important role in the lives of millions of individuals by securing a stable income over their lifespan. The administration of these programs is a complex operation defined by their objectives and design. For example, retirement income programs that provide only flat benefits will not require the extensive machinery of contribution collection and recordkeeping. In contrast, earnings related schemes, especially individual retirement savings, will require not only elaborate mechanisms of contribution collection but also provisions for individual accounts and the management of assets. There is growing pressure to improve the efficiency of public social security programs and constrain their costs. There also is significant interest in comparing the performance of publicly and privately managed pension schemes. The objective of this study is to present a framework for comparative analysis of operational efficiency for mandatory pension programs and develop program-specific performance benchmarks. A good illustration of efficiency improvements at work is an experience of continuous administrative transformations within the Social Security Administration (SSA) of the United States. Figure 1 shows how over a period of 20 years (from 1978 to 1998), the agency was able to achieve significant unit cost reduction along with a 25 percent cut in the total staff in the context of a 30 percent expansion of the beneficiary base over the same 20-year period. The agency implemented a series of adjustments for technical efficiency and cost efficiency. That is, the agency attempted to produce greater output with the same or reduced resources and attempted to alter the combination of labor and capital in the pursuit of further cost reductions by adopting new technologies. Other examples of similar efficiency improvements include the Marshall Islands Social Security Administration and the Swaziland National Provident Fund. These agencies recently implemented dramatic reforms, resulting in efficiency gains and a 30 percent reduction in their staff. 5 Figure 1: U.S. SSA Staffing and Cost per Beneficiary (1978–1998) $55 Cost per beneficiary $50 $45 $40 65,000 70,000 75,000 80,000 85,000 90,000 Total Staff of SSA Source: Author’s al ulatio s ased o data fro the U.S. SSA. Note: Inflation adjusted. Conceptually, there are three types of questions concerned with efficiency: (1) for a given level of resources, is the output maximized (technical efficiency); (2) is the combination of resources the most optimal (cost efficiency); and (3) does the output represent the optimal product for the society (allocative efficiency or effectiveness)? Several studies have attempted to measure the technical efficiency of public pension programs (that is, if members are serviced in the most cost-efficient manner and if public transfers operate at the optimal cost). Our study builds on that analysis but also attempts at benchmarking cost efficiency in the utilization of labor resources for given types of programs and technology. The focus of allocative efficiency is on whether systems offer services that best fit the needs of society. There are several categories of studies of this sort. Some discuss alternative designs or organizational modes (for example, whether contribution collection or management of assets should be centralized, with specific focus on identifying economies of scale and scope). Others look at the tradeoffs in spending resources on improving services for current beneficiaries versus expanding coverage of existing programs to new members. Yet others investigate dynamic allocative efficiency in benchmarking the optimal packages of services over time as economies develop.1 Finally, there are aspects of equity in the allocation of the total cost of operating social security among different groups of plan members and general public. Those are studies that 1 For example, Robalino et al. (2008) and Palacios et al. (forthcoming). 6 look at the incidence of costs across participants with different incomes, demographics, or participation profiles. The focus of quantitative analysis presented in this paper is on technical and cost efficiency. Allocative efficiency or equity is outside the scope of our study (although we do develop and suggest cost estimates that could guide the allocative efficiency analysis). The fact that different countries adopted different types or mixes of products complicates the task of comparative analysis as an efficient combination of resources for one type of program may be suboptimal for another type of program. Therefore, a simple comparison of the costs of widely differing types of schemes may be misleading. Additionally, more complex programs allow for greater variation in the quality and types of services, so cost differentials should not always imply inefficiencies. Even within the same class of programs, benchmarks of operational efficiency are difficult to obtain. Box 1 raises questions on the consistency of policy advice in the absence of such a framework. Box 1. Reforms of the Croatian Pension Insurance Institute The World Bank was involved with providing support to the government of Croatia since the early stages of reform for its national pension system. In 2002, in the do u e t Croatia: Pe sio “ ste I est e t Proje t, a Wor ld Bank team noted that the [a]dministrative costs of the pension system in Croatia are unjustifiably high, and reflect significant inefficiencies and overstaffing in the Pension Institute. Currently, these costs run to 3.7 percent of total benefits, while 2 percent is a typical share based on international experience. Reducing these costs to regular levels ould sa e . per e t of GDP a uall , ithout sig ifi a t loss of effe ti e ess. However, by 2006, in a follow-up do u e t e titled Proje t Paper on Restructuring Pe sio “ ste I est e t Proje t for the Repu li of Croatia, a e tea recommended that [ ]y the end of 2007 CIPI should reduce its administrative costs from 1.8 to 1.2 percent of pension expenditures and improve its productivity by 20– per e t. The reader should note that while over the five-year period the cost index did drop below the originally suggested benchmark, the suggested additional 30 percent reduction lacks sufficient justification. While the agency was undergoing some structural changes around that time, it is exactly this lack of a consistent quantitative framework that produces such ambiguity in defining a reference point. Costs may vary over time within the same program or as the program undergoes systemic changes. They also vary across countries often for the same program types even after adjusting for the size of the schemes and other important factors. In this paper, we use tools of quantitative analysis and data on administrative expenditures and operational setup to develop a framework and assess the technical and cost efficiency of institutions in charge of public retirement programs. High administrative expenditures may be a symptom of inefficiencies in some systems, but in other systems, these high administrative expenditures may simply indicate public choice for systems of more diverse and high-quality services that come with high costs. Factors of quality are very difficult to quantify (see discussion in section 7). Those include, for example, better accessibility and greater variety 7 of services, more competent staff and higher responsiveness of administration, greater individual choice and more transparent systems, and effective enforcement and reduced fraud. Better and more complex services require investments in systems and people. For example, a substantial part of the debate around reform of Social Security in the United States has been focused on the types of services, efficiency, and costs of the reformed administration. A lot of that discussion is around benchmarking and costing of different bundles of services compared to the current set provided by the SSA.2 We provide a summary of one of such studies in box 2. At the same time, there are some fundamental institutional differences across countries that may create cost differentials for similar types of programs. As James et al. (2001) indicate with reference to experiences of setting up individual account systems, [p]robably the least-cost alternatives and trade-offs are available for industrialized rather than for developing countries. Industrialized countries have access to existing financial institutions, lo er tradi g osts, passi e i est e t opportu ities, a d ore effe ti e go er a e. […] In developing and transitional countries, particularly those with small contribution and assets bases, investment costs are likely to be higher and the opportunities for reducing fees lo er. In fact, we confirm this statement in our analysis and show that for less developed countries, a substantial institutional cost premium may be unavoidable. Hence, high costs may not represent a problem in itself but rather point in the direction of further inquiries on a case-by-case basis. Important factors responsible for cost differences are scheme coverage, benefit generosity, maturity of the program, and others. We discuss all of them and their effects in the context of the data available for this analysis. We have collected data on operational organization and various components of administrative expenditures for over 100 public programs internationally (see annexes 1 and 2 for details), which is the largest sample among similar studies to date. Our objective is to standardize presentation of administrative expenditures, ensuring consistency in comparing schemes of different types and capturing some obvious deviations from the expected performance of the various programs as projected by our simulations. We also are interested in the composition of total expenditures and cost components associated with various functions. While analysis is perhaps less conclusive here, given the differences in cost accounting and gaps in information, it is instructive in terms of the magnitude of various factors as they enter the total cost function. The analysis also suggests considerable scope for standardization of accounting and reporting of costs across these institutions. 2 See Genetski (1999) for a review of a possible decentralized model versus Hart et al. (2001) for options for a centralized system. 8 Box 2. Cost Analysis of Reform Options for the U.S. Social Security Program In a discussion of the possible centralized organization of the Individual Accounts (IA) of the reformed Social Security, Hart et al. (2001) look at two hypothetical models with basic and advanced levels of services. The higher-service program is intended to represent an IA program that would provide participants with as many features and services as those offered today by leading private providers of financial services and by employers who offer defined contribution plans, such as a 401(k). It, therefore, would require more extensive new information systems and processes. The authors note that for additional functions, the SSA would require an estimated 7,000 to 33,000 additional employees under the basic- and higher-service IA examples, respectively. The range of additional operational costs is defined between US$0.7 and US$3 billion (or the equivalent of an additional US$3 to US$15 per member, including both covered employees and active beneficiaries). These are significant increases compared to the current mode of operation, and hence, such cost analysis clearly cannot be ignored in the process of reform discussions. The remainder of the paper is organized as follows. First, we present the genesis of the inquiries into the subject, reviewing some of the most relevant literature on the administrative expenditures and costs of mandatory programs produced over the past two decades. Our primary objective is to review the methodologies used, although each paper comes with a rich set of findings and recommendations. We then present our set of administrative data on the public programs and develop a number of standardized cost indices, discussing their advantages and limitations. We also discuss various major cost components and their shares in total costs. Finally, we proceed with simulations on the basis of our data and develop three different models. The first model looks at the impact of design of a program on its total costs. Our second group of model specifications assesses differences in costs of managing pension liabilities between the public and private mandatory pension schemes. Finally, on the basis of the third model we generate benchmarks for staffing levels and for the total administrative expenditures. We compare those to the actual indicators and develop standard performance ratios, providing insights into the differences in the performance of various programs. II. Formulating the Research Question The first notable generation of comprehensive research inquiries into the subject of administrative expenditures and efficiency of mandatory pension programs internationally was produced in the early-to-mid 1990s. The focus was primarily on exploring administrative inefficiencies and on benchmarking operational performance by comparing expenditures of the public and private pension plans with centralized versus decentralized modes of organization. Two types of approaches emerged: one in which cost indices were constructed on the basis of recurrent program expenditures or revenues and another in which costs were measured as applied to individual members (either as one-off charges or as cumulative costs over the period of plan participation). The latter measure emphasized the incidence aspect, bringing analysis from the macro-level down to micro-level. 9 One of the most comprehensive early studies that adopted macro-methodology is found in Mitchell et al. (1993). Their sample of the costs of managing national social security systems, including retirement programs, dates back to 1986 and includes 25 countries of Latin America and the Caribbean and 24 countries of the Organisation for Economic Co- operation and Development (OECD). The costs were defined quite broadly as expenditures borne by the state to provide certain inputs in exchange for services of the social security system. They did not differentiate between particular types of retirement programs or cost incidences in different schemes. To explain cross-country differences, the cost function included the explanatory variables of technological and infrastructural characteristics, program organization, and the level of national income as a proxy for input prices. The studies show that administrative expenditures of social security systems exhibit considerable economies of scale and cannot be explained simply by technological differences in the production of such services across countries. James and Palacios (1995) point to fundamental difficulties in measuring and comparing the administrative costs of mandatory systems. They explain some of the differences by quality differentials, subsidized operation, and the non risk-taking nature of the public sector provisions, concluding that pu li l a aged old-age programs tend to understate their true administrative costs and overstate their efficiency relative to privately managed plans. They also indicate biases of simplified cost ratios, particularly evident in immature or small and poor systems (which we discuss in greater detail below). They propose a measure that would better reflect various internal and external factors that affect costs as administrative cost per member over income per capita, although recognizing that it is only a crude adjustment for the higher input prices and the higher-quality services. This study perhaps also is the first effort to compare individual systems to their corresponding benchmarks. Statistical analysis on the basis of a sample of 50 countries indicated that national schemes in Austria, Chile, Finland, and Kuwait cost more to administer than predicted by the model, while schemes in Canada, Denmark, and Mauritius that feature universal flat benefits cost less than predicted.3 Bebczuk and Musalem (2008) also made an attempt at benchmarking operational efficiency of public pension programs, although on the basis of simple cost ratios. They find that the group with the most inefficient programs includes Belize, Sierra Leone, Fiji, Tanzania, Philippines, Costa Rica, and South Africa. They exhibit a ratio of operating expenses to gross income between 24 percent and 5.5 percent. Their intermediate group includes Jersey, New Zealand, Ghana, Egypt, France, the United States, Japan, and Ireland. The ratio of operating expenses to gross income here ranges between 3.3 percent and 1 percent. Finally, they find that the most efficient countries are Denmark, Sweden, Ecuador, Guatemala, Singapore, 3 Our analysis that follows confirms these findings for Canada, Denmark, and Mauritius while also indicating that Finland operates close to its benchmark. For corresponding programs in Australia, Chile, and Kuwait we did not have data. 10 Korea, Finland, Sri Lanka, and Malaysia, where the operating expense ratio is below 1 percent of gross income. These findings should be interpreted with care given serious limitations in the simple cost indices. We further discuss biases of these measures. Valdes-Prieto (1994) zooms in on four national programs and compares the costs of public and privately operated schemes in Chile, the United States (including both public and voluntary private schemes), Malaysia, and Zambia, focusing specifically on different types of services offered by each system. This is one of the earliest studies that adopted a micro- approach that captured and converted all lifetime member costs (before or after retirement) to the equivalent charge ratios and subsequently to the annual absolute cost per member. It also is one of the few studies that explicitly accounted for the costs incurred to the beneficiary after retirement.4 By the late 1990s, in the United States, the discussions of privatization of the U.S. Social Security program were at their height, generating a significant body of literature on the organization and costs of various alternative provisions. Diamond (1998) was one of the most influential studies on the topic, where issues on the cost measurement of privately managed pension plans were summarized. Around the same time, the interest in comparing the operations of publicly and privately managed programs intensified. This occurred as reforms of public pension schemes unfolded in a number of countries with the shifting mandate for retirement income provisions from the public to private sector. Systemic reforms resulted in significant changes in the machinery of administration with partial or full privatization. Other countries were closely watching and contemplating similar reforms.5 This prompted the second generation of studies of the costs of mandatory pension plans. However, the focus this time noticeably shifted toward privately managed schemes and to the cost incidence with analysis of the effects of various charges levied on plan members.6 Much of the literature on the subject was generated from Latin America and transition economies of the former socialist block. The new schemes were fully funded and resulted in 4 Using the early experiences of the Chilean insurance industry providing annuity products, the study shows that under reasonable assumptions up to 50 percent of all the costs of individual pensions could be incurred after becoming a pensioner. Since the time the study was published, however, the insurance industry has significantly evolved, premiums have substantially decreased, and products have grown in diversity. Mackenzie (2002), for example, indicates that today most annuitants in many OECD countries can expect to be subject to costs between 5 and 10 percent for converting lump sums into an annuity. Furthermore, Rocha and Thorburn (2006) conclude that today Chilean annuitants have a deal that is even better than annuitants in other countries, which is in part explained by the large supply of indexed instruments in Chile. 5 Where no national mandatory schemes existed or where coverage was limited, the focus of discussions was often on reforms of the civil service pension programs as well as sustainable and cost-effective initiatives of e pa sio of the o erage to e populatio groups, like i I dia, for e a ple see The Proje t OA“I“ Report. Submitted by the Expert Committee for Devising a Pension System for India. January 2000). 6 Valdes-Prieto (1994) suggested several reasons for differences in costs and charges; for example, due to implicit subsidies of publicly managed programs and profit margins and indirect taxes by private pension providers. 11 accumulation of assets under the management of private providers. Typically, under such settings, operations are no longer subsidized and costs get passed on to members in the form of implicit or explicit charges. The charges were of different types and applied at different times throughout the accumulation and payment phases. When combined and compounded over time, such charges can consume a significant part of future benefits. Hence, the authorities recognized the need to analyze and regulate costs to protect plan members against excessive charges. In line with Valdes-Prieto (1994), a number of studies began to emphasize a lifetime approach to measuring the cost incidence of charges in both voluntary and mandatory pension schemes. For example, Murthi et al. (1999) indicate that various fees accumulated over a lifetime could consume over 40 percent of individual pension account value in the U.K. Mitchell (1999) produced standardized presentation of costs with simulations on the basis of data from the new mandatory pension program in Mexico and found that depending on the assumptions, over the long term, on average between 30 to 40 percent of contributions could go to fund commissions. (While not significantly different from aggregate commission loads in neighboring countries that introduced similar reforms, the author discusses several factors that still would work both to increase and to reduce charges in the new Mexican system in the medium-to-long run). Given the variety of types and rates of charges, such analysis requires standardization. Building on Diamond (1998) and reflecting on the experiences of private pension industries in OECD countries and approaches found in other studies, Whitehouse (2000) presents a formal framework of interrelations between various measures of charges, including those with equivalent effects on contributions (reduction in premium), on earnings (reduction in yield), and on resulting accumulations (charge ratio).7 These indices provide for aggregation of various types of charges over the lifetime of plan members and allow for consistent comparison of fees across different plans, bringing multiple forms of charges to a set of comparable indicators.8 However, as box 3 indicates, policy implications of these ratios in the multi-pillar program context are not straightforward. With that framework, Dobronogov and Murthi (2005) surveyed early experiences of reforms in Poland, Kazakhstan, Croatia, and Hungary. On the basis of available data, they observed that over a i di idual’s lifetime, the charges could result in an average of 1 percent reduction in yield or 19 percent reduction in assets of mandatory programs. Their study also made an attempt at functional accounting of costs and investigated connections between the charges and actual costs. While observing deficiencies in reporting of costs by 7 As one particularly useful observation, the analysis suggested a rule of thumb that under reasonable assumptions and continuous plan participation over the long-term, a 1 percent charge against pension assets is equivalent to a 20 percent charge on contributions. 8 A related challenge is the disclosure of fees to plan members. For a comprehensive discussion of the issues and a review of the situation in selected OECD countries, see Turneri and Witte (2008). 12 pension providers, their simulations suggest that for private individual accounts to be viable, they should be funded by a minimum contribution of 4–6 percent of wages. This is the only benchmark of that kind of which we are aware. In many countries, however, the rates of supplementary mandatory or voluntary pension contribution remain quite low, which raises questions regarding the financial viability of such policies. Palacios (2005) also attempts to determine the relationship between costs and charges and finds a high correlation between the two. Furthermore, the author uses a sample of 49 pension fund managers from 8 countries of Latin America and finds significant economies of scale (on both the cost per contributor and per affiliate basis). He discusses the implications of an alternative organization of systems by centralizing the management of funds and critiques that approach. Corvera et al. (2006) investigate the performance of the pension industry in Latin America and apply the same methodology to 67 pension managers in 10 countries that operate mandatory private pension programs in the region. They find significant dispersion of charges both across and within the countries and produce a ranking of providers according to the equivalent lifetime cost effect on members. While some of the cross-country variations can be explained by differences in the services provided, the limited competition as well as the presence of state-owned managers is blamed for the differences within a country. Tapia and Yermo (2008) adopted a simplified methodology to compare the effects of various fees and charges. Their indicator of equivalent annual reduction in yield is the sum of all annual charges in U.S. dollar terms divided by total assets, without any extrapolation.9 The small size of their sample (18 countries) did not allow for a regression analysis of various factors; hence, they looked at some of the key factors independent of each other. As a result, the discussion remains largely inconclusive. Apart from an evident relationship between the adopted measure and the maturity of the systems,10 the results did not provide clear evidence of the impact of size or concentration of the industry on charges. However, they point to the potential effects of those factors along with the nature of the collection system, composition of the investment portfolio, and others. 9 Diamond (2011) uses these estimates to debate alternative modes of institutional organization of the national DC pension plans and implications for costs to members, arguing for a centralized system with wholesale interactions with fund managers. 10 Results suggest two different groups of countries: a set of countries from Latin America and a set of countries from Central and Eastern Europe that are more recent reformers. 13 Box 3. Cost Incidence and Multi-pillar Pension Programs Application of indices proposed in Whitehouse (2000) is quite common, including in the context where new DC schemes have recently been introduced on top of legacy DB programs. Such indices are often used to scrutinize the cost efficiency of new private pension schemes compared to legacy plans. One serious limitation of such measures and of all similar indices, however, is that they do not capture differences in generosity of various plans (see discussion in section 5). Thus, for example, a low charge against high contribution (relative to wages a eat up ore of a i di idual’s resour es o pared to a high charge against a lower contribution rate. An individual will simply pay more in absolute dollar terms to administrators for the former plan. We used sample data from 21 countries in Gómez Hernández and Stewart (2008) to calculate and confirm high negative correlation between the 40-year charge ratios and corresponding contribution rates. So, small DC schemes often will be seen as expensive. Yet, this does not necessarily imply their inefficiencies if put in the context of multi-pillar programs. Let us consider three alternative scenarios of a multi-pillar program in figure B3. Each case is presented by an average absolute contribution amount to the legacy DB component I (C-I) and new DC component II (C-II). The grey areas are average costs per member in absolute terms. Case A represents the larger DC component, while cases B and C have a smaller DC component. Keeping average costs per member constant, simply by the design of the scheme and not any other intrinsic differences, plan B will show a less favorable value for the reduction in premium for the DC component. However, comparison between cases A and B would be incomplete without incorporating administrative costs of the legacy DB component. When it is added, the result indicates that in both cases the individual pays the same for the overall program administration. So, purely from a cost perspective, it is hard to argue for either option. In plan C, the legacy plan has quite high costs per member. While shares of the costs may be very similar in both the legacy DB and new DC plans, the argument from the cost perspective may well be for further expansion of a more efficient DC component. Figure B3: Contribution Space Within Multi-pillar Pension Programs A B C C-I cost contributions C-I contributions C-II C-II cost Source: Author’s desig . 14 One immediate implication from that body of research is a need for a more detailed analysis of the impact of organization of various systems on their administrative costs. Some studies specifically focus on assessing the advantages of consolidated collection of pension contributions with other social insurance contributions and income taxes.11 A report funded by the International Federation of Pension Fund Administrators (FIAP) (2006) surveyed the processes and costs of collecting pension contributions in 11 countries that implemented mandatory defined contribution pension schemes, comparing them in the context of multiple organizational modes of the collection.12 If anything, the study reflects the complex nature of cost definition and measurement, given the multiplicity of agencies associated with the collection function and varying incidence of costs. From a simplified binary framework adopted by FIAP’s report, Anusic’s (2005) work is a serious improvement in terms of organizing the knowledge about the collection function and assessing its cost impact.13 In response to inquiries into the efficiencies of unifying tax and social insurance contribution collection, the study emphasizes a continuum of options. It defines five specific modes of administration depending on the responsibilities of different agencies over managing the flows of money and information on social insurance contributions. It also captures operational costs of institutions that are directly responsible for social insurance as well as the cost of functions performed by other agencies for social insurance institutions. The author uses social insurance revenues, total social insurance expenditures, and GDP as denominators to construct simple cost indices for a set of over 30 European countries.14 There also have been a number of country-specific studies of administrative performance and efficiency of the national mandatory social insurance programs.15 We provide a summary of literature on the cross-country studies and select country-focused studies in table 1. 11 See for example, Barrand et al. (2004). 12 This study follows a framework defined in Demarco and Rofman (1999). 13 For a more detailed analysis of the contribution collection process, see Fultz and Stanovnik (2004). 14 Interestingly, the study finds no clear association between the mode of organization of the collection function and administrative costs. Furthermore, the study does not confirm the hypothesis that a consolidation of social insurance administrations implies lower administrative costs, suggesting considerable lags in administrative savings as a result of such reforms, possibly due to political factors that slow down the reform process. There are important dynamic problems, however, in measuring the cost impact of collection integration overtime. First, the denominator changes over time (as a result of compliance improvements). Second, moving functions from one organization to another implies a need for aggregation of costs pre- and post- impact from all agencies involved in such a transfer. 15 See, for example, Chlo o Pola d a d Gru išić a d Nuši o ić o Croatia. Yoo (2002) applied a very innovative approach to the Korean Public Pension Schemes with the uses of the stochastic cost frontier function model and a decade worth of panel data from the national pension agencies. The major observation is that, on average, the Korean system could produce the same outputs at half of the current costs, suggesting reforms with restructuring and management innovations, including operational integration of public pension schemes. 15 Table 1: Summary of Literature Publications Number of Type of Indices (in chronological order) Countries/Region Mitchell et al. (1993) 49/OECD & LAC AE over GDP Valdes-Prieto (1994) 4/World Lifetime charges per member James and Palacios (1995) 50/World AE per member (over national income) Mitchell (1999) Mexico Lifetime charges per member Murthi et al. (1999) U.K. Lifetime charges per member Whitehouse (2000) 13/World Multiple indicators Hart et al. (2001) U.S. AE per member James et al. (2001) LAC & U.S. Multiple indicators Szilágyi (2004) 8/ECA & LAC Annual charges per member Anusic (2005) 30/Europe AE over expenditures, revenues, GDP Dobronogov and Murthi (2005) 4/ECA Lifetime charges per member Palacios (2005) 8/LAC AE and charges per member Corvera et al. (2006) 10/LAC Lifetime charges per member FIAP (2006) 11/World Collection costs over contributions Chłoń-Do iń zak et al. (2007) n.a./World Multiple indicators Bebczuk and Musalem (2008) 24/World AE over expenditures Gómez, Hernández, and Stewart (2008) 21/World Lifetime charges per member Tapia and Yermo (2008) 18/World Annual charges per member Note: ECA = Europe and Central Asia region; LAC = Latin America and Caribbean region; AE = Administrative Expenditures (current annual). Finally, a significant body of literature exists on the operations of occupational and voluntary plans.16 Research covers the span of issues from market organization of the sector to benchmarking the performance of specific plans. A Toronto-based group CEM Benchmarking Incorporated specializes in benchmarking the cost and performance of investments and the administration of pension funds, foundations, and sovereign wealth funds. It maintains one of the most extensive global databases for the private sector and uses data to generate analysis and research with performance comparisons and insights into best practices.17 These types of studies are quite relevant to public sector programs, as the private sector is an excellent incubator for ideas and cost-optimizing solutions. Such 16 For example, some earlier studies by Caswell (1976) and Mitchell and Andrews (1981) provide evidence of economies of scale and explore the effects of other factors on the administrative costs of private pension plans. 17 Bauer et al. (2010) use that database (with observations from 463 DB and 248 DC funds over a period beginning 1990) to study the performance of the U.S. pension industry. One interesting conclusion of that study is that the DB plans may be better cost watchers compared to the DC plans, given incentives. An earlier study on the basis of the same database by Lum (2006) indicates that adjusting for asset mix, size, and style, Ca adia fu ds are the orld’s lo est ost fu ds. 16 analysis generally provides direction on more efficient modes of organization for public sector programs, especially those of small size or narrow statutory coverage. In closing this section, we again need to emphasize that it always is important to differentiate the objectives of various cost studies and the questions that they help to answer. While some studies help to access the effectiveness of a particular program, they may not be very helpful in addressing the issue of efficiency. For example, using simple cost indices may show that a program spends as much on benefits as on administration, which may raise questions on the effectiveness of that particular program; at the same time, it may operate efficiently from a technical perspective, that is, producing services or outputs with the most efficient use of available resources. Alternatively, some mature programs with broad coverage and generous benefits may look impressive in terms of the relative share of expenditures that go into administration; however, such programs may be overstaffed or overspending on other inputs.18 III. Scope of Analysis Cost esti ati g is […] diffi ult i the est of ir u sta es. It requires both science and judgment. And, since answers are seldom—if ever—pre ise, the goal is to fi d a reaso a le a s er. (U.S. Government Accountability Office [GAO], 2007) The incidence of costs of public pension programs will vary across countries. Some will be directly or indirectly borne by the members of the program (active or inactive), while others will be addressed with the general budget. Figure 2 indicates that the cost of running a public pension program can constitute a substantive share of the covered wage bill, with the median at 1 percent for a sample of 70 country observations. On the high end of the spectrum, there are CNSS of Burkina Faso, NSSF of Kenya, GIPF of Namibia, NASSIT of Sierra Leone, POPF of Botswana, SSNIT of Ghana, SNPF of Swaziland, and NPF of the Solomon Islands—all with operational costs above 3 percent of the covered wage19. 18 James and Palacios (1995) find that Indonesia and Kenya spend 30 and 72 percent of contributions, respectively, on administrative costs, which is more than they pay out in benefits. At the same time, Japan and the United States spend 1 percent or less of benefits and contributions on operating expenses. Their statistical analysis, however, indicates that both sets of countries are spending approximately what would be expected, given their per capita incomes and the numbers of covered workers and pensioners. So they note that those ratios raise questions about the overall wisdom of starting old-age security programs in small, poor countries, but they do not tell much about the internal efficiency of those schemes. 19 See annex 1 for abbreviations. 17 Figure 2: Administrative Costs as Share of the Imputed Covered Wage 8.0% 7.0% 6.0% 5.0% 4.0% 3.0% 2.0% 1.0% 0.0% Source: Author’s al ulatio s. Note: Data labels presented only for select countries. Agency abbreviations listed in annex 1. Ultimately, whether or not subsidized from public funds, most of the costs will be shared among economically active individuals. There are issues of whether such total cost allocations are regressive and whether they represent the most dynamically equitable outcomes. We do not directly deal with those issues. Keeping that in mind and unless we explicitly mention it, we refer to the costs in this paper as total expenditures associated with operating a public pension program. Clarity is required, however, on what needs to be captured in the cost measure. In what follows, we discuss several important challenges and propose a set of guiding principles. We do admit, however, that in many cases the ultimate and accurate measures will be impossible to obtain and the best we can do is rely on approximations. There are significant systemic, institutional, and operational differences in how public pension programs are designed and managed. Often the same agency operates multiple programs or the administration of one particular program can be distributed across multiple agencies. We, therefore, identify three approaches to constructing the cost measures:  Programmatic. With the programmatic approach, the focus is on the overall administration of one particular scheme. So, when administration functions are shared among multiple institutions, all such related expenses get captured, and when one agency operates multiple programs, the costs of other programs get excluded. 18  Institutional. Under the institutional approach, the measure is constructed around one institution that manages one or multiple programs (or parts of such programs).  Functional. The functional approach, in principle, allows for an across-the-border comparison of different institutional or operational elements but requires diving into the intricate details of functional organization of each program.20 Anusic (2005) is an example of a functional approach. Such an approach also is useful as it allows for a broader analytical context of the comparison exercise for pension programs, for example, by bringing in the collection side of tax administration or the benefit side of various mass payment systems. There are several expense categories of administrative systems that should be considered. All administrations have to bear regular operational expenses in the form of labor cost, office maintenance, supplies, utilities, and so on. In addition, they may incur significant capital expenditures. They sometimes bear expenses that are not directly related to the core benefit administration (for example, the SSA in the United States provides certain tax processing services to the Internal Revenue Service [IRS]; similarly, some public pension agencies have corporate mandates imposed on them by the state to manage certain publicly owned businesses or assets, for example, state-owned recreation facilities). Some in-kind benefits, such as rehabilitation services, can arguably be treated as both benefits and costs. Other expenses are never incurred directly and come in the form of implicit subsidies (for example, use of public assets such as office premises or other infrastructure) or as opportunity costs (when office buildings form part of the pension assets under the real-estate investment schemes for pension reserves). There also are expenses not borne by the public administration but incurred by the participants (for example, in the form of bank charges for contribution remittances or benefit payments). Whether it is a programmatic, institutional, or functional approach, all direct and indirect current operational expenses related to the program administration should ideally be included in constructing a consistent cost measure. Capital expenditures should ideally be averaged out (or amortized) over several years; alternatively, they could be completely excluded (which is the approach that we follow in our comparative analysis). Operational expenses of various unrelated functions (for example, rehabilitation services) should be excluded. We also recommend excluding all expenses for providing in-kind services (on both benefits and costs). Formal or informal costs—external to the program administration—are conceptually of three types. First, bank charges and the like, while easier to capture or estimate, are often a function of the overall efficiency of the national infrastructure and hence, are outside the control of the pension administration. It would be ideal, therefore, to assess the efficiency 20 Proper functional accounting of costs in public plans is very rare. The only cases we are aware of include the United States, Canada, and New Zealand. 19 of the services provided by the financial intermediaries separately and exclude the associated costs from the cross-country comparison. However, this is not always possible, and in most cases, such costs remain part of the total expenditures presented in this analysis. We further discuss the magnitude of these costs. Second, there often are numerous informal costs borne by the program members (for example, fee for benefit claim facilitation service or postman fees). Informal fees are interesting because they compensate for services that are otherwise inadequate or unavailable and hence, arguably constitute an extension of conventional administration. From this perspective, such facilitation fees transform conventional administration from a low-cost or low-services scheme to a high-cost or high-services program, although altering the incidence of cost in a less transparent manner. When there is a void in some areas of formal administration, it gets quickly filled with alternative provisions. There are plenty of examples of such arrangements. Experience shows that often a whole formal or informal industry could emerge and fill the void of services, such as increased awareness or understanding of scheme rules, contribution collection services,21 benefit facilitation, complaints processing, and legal representation.22 All those direct costs to beneficiaries often remain unaccounted. In the end, low-quality formal services often translate into additional transaction costs for the beneficiaries.23 However, as long as we treat all service altering arrangements as different and discrete packages, the issues of associated cost differences pertain to allocative efficiency and equity and are not the focus of our analysis. Box 4 discusses some other approaches to optimizing benefit packages from a cost perspective. 21 In Chile, in order to perform electronic collection and support the payment of social security contributions through the Internet, the pension fund administrators (AFPs) created Previred, an agency that has captured the large companies market (FIAP 2006). The portal allows those who employ workers, both businesses and home-helpers, and self-employed workers to make their monthly payments in an easy and secure manner. The service is free for employers. 22 In the United States, the National Organization of Social Security Claimants' Representatives is an association of over 4,000 attorneys and other advocates who represent social security and supplemental security income claimants. The members provide representation services for claimants (for example, advocating change in the disability determination and settlement process). 23 See discussion in Palacios (forthcoming). 20 Box 4. Limiting Program Coverage as a Cost-Controlling Strategy There are important tradeoffs between coverage, the quality of services, and costs. Palacios and Pallarès-Miralles (2000) indicate that coverage is a function of the level of national income. This can be explained by several factors, including infrastructure and technological constraints. The costs associated with reaching out to individuals employed in less formal jobs or living in more remote communities could be excessive. Thus, limiting coverage of mandatory contributory programs often is an efficient mechanism of cost controls. Mandate often is defined on a sectoral basis with some professions or industries excluded due to the anticipated high compliance * and enforcement costs. Alternatively, some sectors set limits on minimum earnings, so that contributions would not be required from individuals with very low incomes, often providing services informally. Finally, there often are minimum participatory requirements for firms in terms of their size. As countries develop, they reduce these floors and accommodate greater number of beneficiaries at a lower cost. Therefore, holding everything else constant, it will be inappropriate to compare the costs of delivering social security, say in India, with its current 10 percent coverage of formal contributory schemes and with a hypothetic 40 percent coverage by the same schemes. Those are two different products. * Note: In the United States, participation in social security for various groups of workers was mandated only gradually over time. It was not until 15 years after the introduction of the social security system that the non-farming self-employed could join the system; the farming self-employed joined after an additional 4 years. Third, there are conventional costs of compliance with a contributory mandate in the form of time and effort of employers spent on keeping records, preparing and filing returns, making payments, preparing for audits, getting trained in new rules and procedures, following up on various inquiries, and so on. There is a separate body of literature on organization and costs of compliance with revenue collection (usually in the context of tax payments). Contribution collection can be organized in many different ways. For example in Egypt, employers send information on wage changes only once a year and to a single agency, while in Chile, employers send detailed reports to multiple providers each month. This implies differences in burden in different countries. Furthermore, if the collection of data is not synchronized between various components of the mandatory program, it further increases compliance costs. Slemrod and Yitzhaki (2002) based on their literature review for the United States and a group of OECD countries o lude that [i] al ost all ases the private compliance costs dwarf the public ad i istrati e osts of olle ti g ta es. Thus, by minimizing public administration costs (that is, by not providing adequate services of counseling or electronic submission), the costs are simply pushed out to the domain of 21 nontransparent compliance costs perhaps producing a heavier burden for smaller businesses and the self-employed than for medium-sized and large firms.24 In closing this section, it is interesting to note that different collection and recordkeeping models may not necessarily produce differences in administrative expenditures but rather differences in the distribution of resources in the cycle of benefit administration. Some countries still have quite a weak collection and recordkeeping systems that do not even provide for centralized or electronic facilities and rather shift the burden of recordkeeping on employers who would have to support the employee claim of pension rights at the time of retirement with valid records of earnings and contributions. This implies pushing costs to the end of the administrative cycle of the program, that is, putting fewer resources in the collection effort and more resources in claims verification and processing. So, the structure of cost could be revealing of the maturity of the administrative systems. Arguably, the legacy contributory schemes will have more resources in benefits processing, while more advanced systems likely will spend more effort on contribution collection and in-house record maintenance. As systems mature administratively, they will be strengthening the collection side. The trend further is reinforced when new DC schemes get introduced with their significant upfront data quality requirements. IV. Our Data and Structure of Costs Availability and quality of data differs dramatically from country to country and from institution to institution. There are significant heterogeneities in how pension agencies report their budgets and operational information. The truth is that those reports were never meant to be standardized across countries. In collecting data for this research, we adopt a set of standard definitions of cost categories. However, reporting norms and practices often do not conform to our definitions. For example, we are not always able to differentiate between labor and non-labor costs or segregate capital expenditures from the current operating expenses. In a good number of cases, however, we are able to obtain those details, which allow us to assess potential biases in cases where such breakdown is not possible. We collected data from over 100 public social security programs around the world (see annexes 1 and 2 for data description). Those programs vary in size. The smallest in our sample is the Falkland Islands Pension Scheme with 600 contributing members. The largest scheme is the Old-Age, Survivors, and Disability Insurance (OASDI) program in the United States that covers over 160 million active contributors and some 50 million beneficiaries. The nature of the operation and institutional organization of the programs in our sample 24 As a cost-control measure in such circumstances, various simplified contributory regimes could be a good alternative where compliance and collection costs are unreasonably high, if such are seen as a major deterrent to coverage expansion. Again, in our study, we view these issues as pertaining to allocative efficiency and simply control for them in our model without analyzing them in detail. 22 also vary. We were able to obtain separate data on at least 10 noncontributory pension schemes. On the other end of the spectrum, there are a significant number of social security institutions that offer a broad range of benefits, including maternity, child allowance, unemployment, sickness insurance, and others. We used a combination of methods to obtain data. In most cases, the information was obtained directly from annual statements and reports of the social security agencies or other offi ial statisti s a aila le o the age ies’ e portals. We also obtained data from around 25 agencies using a detailed questionnaire. Our objective was to collect information on each agency as a whole with all its multiple programs. In some cases, we were able to obtain data on separate programs or on institutions operating at the subnational levels (for separate states, autonomous regions, or overseas territories), which increased the number of observations for our analysis. The data includes the nature of the schemes operated by each agency, coverage in terms of the contributors and beneficiaries, financial flows, accumulated assets, staffing and number of offices, and the level and details of administrative expenditures. Our main data set is presented in annex 1 and it includes both raw data expressed in the national currencies and various indices constructed on the basis of that data (which we discuss in the following section). To make our data set consistent, we report data on labor resources and administrative costs related to the associated functions. Thus, where certain functions of an agency were excluded from the analysis, we had to prorate both staffing numbers and cost allocations associated with such functions (as we did, for example, with Medicare-related and reimbursable services provided by the SSA in the United States). Conversely, if we combined functions performed by various agencies under one observation pertaining to a particular program, we combined both labor resources and total costs (like in the case of the Sri Lankan Employees’ Provident Fund that uses collection and other services provided by the Department of Labor). 4. 1. Institutional organization and total expenditures To better visualize our data, we developed a typology of the administration of public schemes reflecting varying degrees of institutional complexity, assigning higher ranks to more complex organizational structures, as summarized in table 2. 23 Table 2: Classification of the Public Social Security Administration Agency No. of Observations Rank Benefit Programs Managed by Social Security Agencies in Our Sample 1 Plain (universal) Basic Pension (BP) 3 Means tested BP and/or disability BP, or earnings related DB 2 schemes with no in-house service records (for example, some civil 9 service schemes) Earnings related DB/DC/Provident Fund schemes (possibly with an 3 49 associated small non-pension scheme, like housing loans) Multiple in-house pension schemes and/or additional extensive 4 58 non-pension schemes (assistance, health, etc.) Source: Author’s desig . Figure 3 shows median levels of administrative expenditure for each type of program in our sample. We normalized expenditures by the total members25 and by beneficiaries only. We further adjusted expenditures by the ratio of national incomes per capita in each country and in the United States, which implies equivalent costs of running the same operation in the United States. In constructing median values, we excluded a group of schemes that we considered as outliers in terms of excessive costs per member adjusted for income differences.26 From the 21 Sub-Saharan African institutions in our sample, 14 are in this category. There is no particular pattern emerging from this analysis. For schemes that do not involve recordkeeping of earnings, programs with DB (rank 2) require more resources compared to flat-benefit schemes. The result for ranks 3 and 4 depends on the denominator used in normalization. However, overall it is consistent: there is a significant increase in the per- beneficiary measure, indicating extra costs associated with the functions of employee record management and contribution collection. The drop in per-member costs for rank 3 indicates that average recordkeeping costs per member is much lower than the average benefits management costs per beneficiary. This is entirely consistent with all other findings in this paper. The drop in average costs of the most complex programs (rank 4) is a challenge to explain. We associate it with important biases of this simple measure we use, including program size, generosity, level of economic development, and so on. We discuss them in greater detail in section V. In the regression analysis that follows, we did not find categorical variables for different ranks of the programs consistently significant, except for basic pension. However, certain elements associated with the nature of operational organization of the programs did play a role in explaining cost differences. 25 Members are defined as the total number of beneficiaries of all cash programs and contributors/insured for whom records are kept in-house. 26 Those institutions include BWA-POPF, BFA-CNSS, GHA-SSNIT, KEN-NSSF, KEN-LAPT, MLI-NSII, NAM-GIPF, PHL-GSIS, RWA-RSSB, SEN-SII, SLE-NASSIT, SWZ-PSPF, TZA-GEPF, TZA-PPF, UGA-NSSF (see annex 1 for definitions). 24 Figure 3: Agency Rank and Median Administrative Expenditures (Income Adjusted) 1,800 Current admin exp per member (US eqv), US$ 1,600 1,400 Members 1,200 Beneficiaries 1,000 800 600 400 200 0 1 2 3 4 Agency rank Source: Author’s al ulatio s. 4. 2. Key elements of the cost structure In what follows, we discuss key observations from the cost structures of various institutions. For 71 observations in our sample, in which we separately provided information on expenditures on capital investments and depreciation, we found that the median for such costs is only around 5 percent of the total administrative expenses. In some exceptional cases, however, capital expenditures are up to one-third of the total operational budget (for example, in the case of Maldives, which was in the process of establishing a new agency to run a new national contributory program at the time of collecting this data). For 74 observations, the median share of the direct labor cost in current expenditures is 57 percent, although the variation is extremely broad from 6 to 90 percent, in part due to reporting differences, with the lowest share of labor cost reported for the Swedish national DC program and the U.S. Thrift Savings Plan for public sector employees (two programs similar in nature of operation). We also note that for the 27 countries in which data on both direct labor costs and asset management expenses are available, the correlation between the sizes of those two cost components (as a share of total current expenditures) is negative 81 percent, implying that as systems accumulate and actively begin to manage considerable financial assets, direct labor costs become insignificant in explaining total cost differentials. 25 For 39 programs with available data on pension asset management expenses, such expenses constitute 25 percent, as a median, of the total current expenses. (Only one-third of the programs in this subset were DC schemes.) We find no direct association between the volume of assets and the share of reported asset management expenses in total current expenditures. At the same time, larger pools of assets are clearly less expensive to manage on per unit basis, while for the smaller portfolios there is a significant dispersion in asset management costs.27 Figure 4: Costs of Managing Pension Assets 1.6% Asset management expenses (% total assets) 1.4% 1.2% 1.0% 0.8% 0.6% 0.4% 0.2% 0.0% 0 50,000 100,000 150,000 200,000 250,000 300,000 Total assets (USD, millions) Source: Author’s al ulatio s. With 31 available observations on office rent expenditures, the median reported amount constitutes 1.3 percent of the total current expenses. In the Netherlands, Northern Ireland, Kosovo, and the Maldives, however, where accounting recognizes these costs more explicitly, office rent expenditures reach a 10 percent share, which perhaps is closer to the actual situation with such costs borne by the pension agencies.28 The median for benefit delivery costs for 30 countries in which such data is available stands at around 5 percent of the total current administrative expenditures of the agency. While bank charges seem to be part of those costs, more analysis is required on the classification 27 We note, however, that such direct comparison does not take into account the composition of the portfolios and the nature of the management practices. 28 In Kosovo, the new pension agency that was established to manage a national DC scheme selected and leased premises for its main office on a commercial basis. In Northern Ireland, assessment of the rent which would be payable on an open market basis for most buildings occupied by the Social Security Agency is charged to the age ’s operati g ost, as part of otio al osts. 26 of expenditures in that category on a case-by-case basis. In Romania, for example, where such costs are reported at 54 percent of the total administrative expenditures, the benefit delivery services have been outsourced. Another interesting observation is that the next three most expensive delivery services are in neighboring Georgia, Azerbaijan, and Armenia (in the range of 40 to 50 percent of total administrative expenses). At the same time, in all four countries, those costs as a share of the total benefit expenditures still are relatively small (between 1 and 2 percent). We do not have data on the cost differences associated with various types of benefit delivery, but as evidenced by our data presented in figure 5, as countries develop, they tend to process a greater share of social security benefit payments through the banking system. This perhaps is a function of the efficiency of the financial sector, costs of these services, and program coverage. Figure 5: Share of Benefit Payments in Banks by National Social Security Agencies 120% 100% 80% Bank payments 60% 40% 20% 0% $- $10,000 $20,000 $30,000 $40,000 $50,000 $60,000 $70,000 GDP per capita Source: Author’s al ulatio s. Finally, for five countries where we have information on the delivery of individual account statements, the associated costs are small and range from less than one to five percent of the total current administrative expenditures. 4. 3. Functional analysis: contribution collection and benefit payment We also investigated resource allocation between the key administrative functions of contribution collection and benefit payments. Explicit functional cost accounting has been adopted so far only in a handful of countries, including the United States, Canada, and New 27 Zealand. We conducted a small survey with select public social security agencies and requested, through special questionnaires, information on staff associated with the functions of contribution collection, benefit payments, and all other staff (IT support and maintenance and general management). For the nine agencies that operate in-house contribution collection and for which we received such data, figure 6 (a) and (b) suggest that benefit administration is much more resource-intensive than the process of contribution collection on a per head basis. The difference between the presentations in (a) and (b) is that in the former, one observation depicts one institution (with its collection and benefit functions) while the latter expands the sample, incorporating additional agencies that do not operate in-house collection function. Based on these observations, the beneficiary payment side may require between 3 and 10 times more staff per member serviced compared to the contribution collection side. We will further validate these findings in our regression analysis. Figure 6: Allocation of Labor Resources within the Social Security Agencies Contribution Collection Benefit Payments 70 LN (Members over Staff**) 10 60 (a) (b) 50 Members ratio* 8 40 30 6 20 10 0 4 0 10 20 30 40 50 60 70 4 6 8 10 12 14 16 18 20 Staff ratio* LN (Members**) Source: Author’s al ulatio s. Note: * Members ratio: active contributors over beneficiaries; Staff ratio: estimated staff involved in contribution collection over estimated staff involved in benefit payments. **Members: active contributors and beneficiaries; Members over Staff: ratio of total members over staff of the agency. This finding has important implications for understanding member accounting in the cost analyses. Specifically, for the same membership size, the programs that do only benefit payments will always look more expensive on a per-member basis compared to the programs that collect contributions and pay benefits. Therefore, for the same 100,000 members, it may seem to be more expensive to run a basic pension-type program relative to a contributory scheme as measured on a per-member basis. Hence, it is important to recognize that bias in any comparative analysis. Finally, some agencies operating multiple programs produce programmatic accounting of costs (for example, in Canada, St. Kitts and Nevis, Sweden, and the United States). In fact, in 28 some of those cases, given the different nature of the programs, we included separate observations on those programs in our data set. V. Cost Normalization Having all current administrative costs aggregated and converted to a common currency does not make them directly comparable given significant underlying heterogeneities in program design, size, and institutional and operational setup. Some normalization is required. For the purposes of reporting operational efficiency, we favor the measure suggested by James and Palacios (1995), which is income-adjusted annual current cost per member,29 and in this paper, we offer a more intuitive variation of it (see annex 2). At the same time, justice needs to be done to other measures, and we now discuss the advantages and limitations of various alternative indices broadly used in the literature. 30 As important caveats of such analysis, we need first to consider several common biases:  Maturity bias. This is evident when newer earnings-related schemes with very few beneficiaries and payouts turn out more expensive as compared to older and stabilized schemes.  Financing bias. This implies that noncontributory schemes or contributory schemes with significant budget subsidies cannot be compared with financially balanced contributory schemes, if contribution revenues are used as the denominator. (To address this, the sum of contributions and benefits can be used instead.)  Generosity bias. This bias reveals itself when the schemes of the same organization and coverage do not look the same when differences in the rules of accruals (or in contribution rates) are significant. Figure 7 indicates that the share of program revenues that finance administrative costs generally increases as the contributory mandate shrinks; however, the variation remains quite significant. (To remedy this problem, some use GDP as the denominator.)  Coverage bias. This bias will disqualify GDP as a useful denominator if various sector- specific schemes or schemes with very narrow coverage (by design or implementation outcomes) need to be compared.  Technology bias. Several studies point to the fact that more advanced technologies and better infrastructure are available in more developed countries, hence availing more cost-efficient solutions. This information is not easy to reflect in any of the cost indices. 29 Based on their regression analysis, the authors note that the administrative costs rise at a much lower rate than per capita income, owing to higher productivity of more developed countries. So, such income adjustment overcorrects. This is in interesting contrast to Mitchell et al. (1993) that constructed proxies to capture cross-country differences in production technologies but found them insignificant explanatory variables in explaining cost differences of Social Security. 30 Valdes-Prieto (1994) offers a related discussion of alternative indices with their limitations. 29  Operational bias. Biases of all sorts exist when resources are shared with other programs or functions to sustain operational synergies (for example, in contribution collection or benefit administration).31  Size and membership biases. These biases are quite a few and diverse. First, fixed costs imply that smaller schemes will be costlier to manage. Second, we point above to the implications of the composition of the membership bases (beneficiaries versus contributors) for per-member cost accounting. Third, we present empirical evidence that the number of inactive members may pose important implications for any measure used. Fourth, some programs provide service to special groups of beneficiaries, such as widows and the disabled, where additional administrative resources presumably would be required to assess eligibility. More generally, some programs require complex categorical or resource eligibility checks for potential beneficiaries. Figure 7: Contribution Rate and Administrative Costs 70% Admin costs over contribution revenues 60% 50% 40% 30% 20% 10% 0% 0% 5% 10% 15% 20% 25% 30% 35% 40% Statutory contribution rate (% of wage) Sources: Author’s al ulatio s. Note: The sample includes only programs with the collection function operated largely in-house. The costs of managing pension assets are excluded. The choice of the denominator for comparable cost measures is important. It needs to relate to something that the social security system produces, such as delivery of a general public good, administration of contributions and benefits, management of pension liabilities, or servicing various members and beneficiaries. The following sections discuss corresponding candidates for the denominator. 31 For example, it is common for the civil service pension schemes to rely on the personnel and payroll units of various government branches and agencies to conduct public information and contribution collection. 30 5. 1. Uses of national income, revenues, and expenditures in cost normalization Some studies use GDP as a denominator to normalize costs, which for the same coverage and program type is sufficient. The problem, however, is that coverage is not the same across countries or even across schemes within the same country. So, the coverage bias is a major deficiency of this index. When the focus is on financial flows, conventional indices are composed on the basis of contributions collected or benefits paid. We do report those indices in our table in annex 2. However, these measures have the greatest number of biases associated with them, which we report in table 4. At the same time, just as with other indices, comparison across programs of similar size and design can be fair. 5. 2. Administrative costs and pension liabilities All retirement schemes, whether funded or not, are in the business of liability management. If such liabilities can be clearly defined and measured, one could compare the cost efficiency of various types of schemes on that basis. In figure 8, we use the estimated Implicit Pension Debt (IPD) of selected unfunded or partially funded mandatory DB schemes and reported assets of DC schemes (including provident funds) to normalize administrative expenditures. For comparison, we also present data on combined equivalent annual charges from private mandatory pension programs.32 Out of 52 observations for which we had information on either total pension assets or IPD, 30 schemes are DC, including 12 schemes that are privately managed (see data in annex 2). 32 The numerator is the actual current expenditures of public programs and combined annual member charges in private schemes. Hence, we ignore the difference between the costs and expenditures of operating private programs assuming that profit margins at the national level are insignificant. In fact, Palacios (2005) reports high correlation between the costs and charges in private plans. 31 Figure 8: Costs of Managing Pension Liabilities (Percentage of Total Assets or IPDs) Piblic schemes Private schemes 4.50% 4.00% DB external DB in-house Publicly Privately collection collection managed DC managed DC 3.50% 3.00% 2.50% 2.00% 1.50% 1.00% 0.50% 0.00% “WE-““IA/NPFs OA… Uruguay-PRIVATE LTU-SSIFB SEN-SII IND-ESS MFD-PDIF BRA-NSSI KOS-KPST UKR-PFU PHL-SSS PRT-SSI URY-SIB MAR-NSSF IND-EPFO TON-RFB KEN-NSSF Poland-PRIVATE UGA-NSSF UK-(GBR)PS ROM-NPSIF USA-SSA/OASDI HRV-PII KGZ-KRSF USA-TSP SGP-CPF LKA-EPF THA-GPF UK-(FIS)FIPS Chile-PRIVATE CAN-CPP KOR-NPS SLB-NPF SWZ-SNPF Mexico-PRIVATE EST-SIB Estonia-PRIVATE Sweden-PPA HUN-CANPI POL-ZUS+KRUS MDA-NOSI VUT-VNPF FJI-NPF WSM-SNPF NZL-(CI) NSF Bolivia-PRIVATE Peru-PRIVATE Costa Rica-PRIVATE Hungary-PRIVATE BWA-POPF El Salvador-PRIVATE Argentina-PRIVATE Sources: Author’s al ulatio s. For private schemes, data from Tapia and Yermo (2008). Note: Agency abbreviations listed in annex 1. The impressions that we get from figure 8 are as follows: i. There is significant dispersion in simple cost indices among DC schemes. It also appears that centralized publicly managed programs are necessarily the least costly arrangements compared to privately managed decentralized DC schemes, perhaps in part due to the following attributes of the public DC schemes in our sample: (a) some of them operate in less developed countries with relatively more expensive financial sector infrastructure, (b) additional costs of in-house contribution collection, and (c) smaller size of some of those schemes. ii. Unfunded DB programs seem much less expensive to manage. This is not a reflection of their greater efficiency but rather an indication that management of individual funded liabilities requires much greater effort, providing a public good of a completely different nature. For five countries that operate two parallel mandatory programs, the contrast in costs of managing funded versus unfunded liabilities is particularly striking (although given the discussion in box 3, not necessarily conclusive). We also note that the DB schemes in the sample are larger and more mature. iii. Contribution collection seems to be associated with some additional costs of program management. Nevertheless, the impact does not seem highly significant. These observations are not without shortfalls, as the simple measure we use is subject to biases of maturity, generosity, and size. We further validate these findings with regression analysis in section 6. 32 5. 3. Per-member costs Using the number of program members as a means of normalization of costs seems obvious. Unfortunately, the methods are not that straightforward and deserve a detailed discussion. Our approach to membership accounting is based on the principle of liability management. In broad terms, the business of pension administration is to (i) accumulate, (ii) manage, and (iii) discharge the member pension liabilities. Hence, all those to whom such liabilities relate should be generally considered members of the program. We discuss these three components of pension administration and how we propose to count the associated members under each component in greater detail. Under contributory regimes, the business of accumulating pension rights is encapsulated in the contribution collection and processing function. Contributing individuals become direct beneficiaries of administration services, and we consider them as members. There are inactive members, or those without contributions, who also accumulate pension liabilities. For example, in the context of operation of most European pension programs, certain inactive member groups can constitute part of the category referred to as insured (for example, students, military servicemen, or recipients of certain public benefits). In line with our proposed definition of pension administration services, our approach is to count and include all members who accrue new rights in a given period, including all members who earn such rights without contributions.33 Furthermore, we count all members who are directly supported by the mechanisms of the system, that is, who accumulate pension rights and for whom the records are kept in-house.34 33 Our definition, however, provides for a minor complication of vesting, or minimum period of contributory participation in the scheme to earn and in the future benefit from pension rights. For all practical reasons, we suggest ignoring that issue, as in most cases some form of compensation, usually a lump sum payment, will be provided to members who fail to earn the required minimum. 34 An alternative approach to counting members would be to include all those who accrue benefit on a noncontributory basis. While there is no contribution, the service of rights accumulation still is provided. This applies to both noncontributory earnings-related civil service schemes and universal pension programs. In some cases, such an accumulation process is literal and gradual and is related to the years of residency (for example, Denmark) or taxable work (for example, Netherlands and U.K.). This approach helps answer an important policy question regarding the cost differentials of alternative design options for the same group of potentially eligible individuals. Hence, such an approach would help make inferences about allocative efficiency. A pure mechanical measure (focusing only on benefit recipients of the noncontributory schemes) will lack such insight. Furthermore, there are important caveats in using such indexes for inferences about allocative efficiency and in the choices between different design options on the basis of costs. One could argue that direct current beneficiaries of earnings-related schemes that provide survivorship benefit are both contributing members and their families. Note that some DC schemes price such services separately and collect a special premium to service survivors of current members. Following this logic, the DB schemes have an implicit premium for survivorship benefit. If there is a premium, there must be well-defined beneficiaries. At the same time, plain universal basic pension programs in most cases do not offer survivorship benefits. So, while the latter schemes could be seen as providing coverage to all working age individuals (without contributing), the earnings-related schemes with an implicit or explicit survivorship premium cover a smaller 33 The liability management side of pension administration covers all those members of the program who conceptually have legal claims against the total outstanding stock of pension liabilities. That includes (i) all those currently contributing (or who otherwise accumulate new pension rights); (ii) all inactive members who do not accumulate new rights in any form but whose implicit or explicit rights still constitute part of the total liabilities managed by the pension agency (often referred to as members with dormant accounts); and (iii) all those currently in receipt of the regular benefits. While categories (i) and (iii) will be counted as members under other components of pension administration services, our approach is to exclude the second category (dormant accounts) from the definition of members. Thus, this side of the pension business administration should not produce any additional members for our count.35 Not all workers, however, regularly contribute,36 and he e, defi itio s of a ti e orker ary from program to program. Valdes-Prieto (1994) defi es the o ept of effe ti el o ered o tri utors as registered perso s aki g at least one contribution in the last 12 months. Variations of this approach seem common in how pension agencies report data on their active members. To provide the idea of differences in the reported numbers of active contributing workers versus total accounts, we present data from our survey in table 3. Table 3: Activity of Pension Accounts (Thousands of Contributing Members) Burkina Faso Philippines Singapore Indonesia Denmark Sri Lanka Malaysia Malaysia Vanuatu Pakistan Uganda Guyana Jordan Ghana Kenya USA Country Agency CNSS ATP SSNIT NIS ESS SSC NSSF EPF SSO EOBI SSS CPF EPF NSSF TSP VNPF Year 2004 2007 2006 2002 2007 2006 2007 2007 2006 2007 2007 2007 2005 2005 2008 2006 Registered 150 3,800 1,200 590 23,700 1,500 3,600 12,000 11,800 2,700 27,000 3,100 11,900 248 3,900 40 Active 73 3,100 850 130 7,900 660 900 5,400 5,500 1,800 7,900 1,500 2,100 138 2,600 28 Activity rate 49% 82% 71% 22% 33% 44% 25% 45% 47% 67% 29% 48% 18% 56% 67% 70% Source: Author’s al ulatio s. Note: Agency abbreviations listed in annex 1. In many cases, more than half of all accounts remain passive. The reasons for such differences are few. In less operationally efficient systems, in which identification means are weak, some workers end up having multiple accounts when moving from one employer to group of workers but also indirectly provide services to a broader group (those who can claim the survivorship benefit). This observation may tend to equalize allocative efficiencies of the earnings-related and universal schemes. One way of dealing with this issue is to exclude the survivorship element from the contribution rate and to use adjusted revenues in the denominator for proper comparison. 35 This approach, however, may constitute problems with incorporating closed earnings-related schemes into the cross-program analysis, but that is a very small group of programs that do not attract significant research interest. 36 For analysis of contribution density, see Arenas de Mesa et al. (2004) who use household survey data linked with the Chilean social security records for over 20 years. 34 another.37 Other reasons include inactivity of accounts due to prolonged periods of unemployment, temporary migrants, and members of specialized schemes (for example, for civil servants) who left their employer and opted for deferred retirement. On discharge of pension liabilities or the benefit side of pension administration, the approach may seem straightforward: all those in receipt of pension payments should be counted as members. However, there are three types of issues here. First, there are lump sum payments (for example, provident fund type schemes or new DC schemes usually pay only lump sums or make arrangements with external annuity providers). Our approach is to include lump sum payments along with regular payments in the count of beneficiaries.38 The second group of issues is associated with common benefits, such as survivor pensions: some institutions report only family cases, others only beneficiaries (family members), yet others report both. For our purposes, we recommend using the count of all beneficiaries who are the actual collectors of the benefits. Finally, many pension agencies administer benefits other than pensions. Those typically include various short-term benefits (sickness, maternity, child, family, and unemployment allowances); special schemes (industrial injury pensions); multi-pillar pension schemes; and several other assistance programs. Conceptually, the institutional approach will dictate incorporating all the schemes managed by the same agency and all those benefiting from such schemes into the institutional cost measure. A programmatic approach focusing on the pension program only would require some form of cost proration. In either case, on the practical side, while capturing the count of beneficiaries of special, multi-pillar pension, or assistance schemes is relatively straightforward, there often are problems with the availability or interpretability of the count of beneficiaries of certain short-term benefits and other schemes.39 The provision of such benefits, therefore, should be controlled in the cross-country comparison. In our analysis, we account for beneficiaries in two different ways. We combine the counts of regular maternity, children, and family allowances with the numbers of pension recipients, which constitutes our totals for the beneficiary numbers in all our results and regressions. We also account for the provision of other benefits, such as sickness, unemployment, health insurance, and personal loans. However, given the complexity of measurement and interpretation of these types of benefits, we use a categorical variable and activate it each time when at least one of these programs is available. 37 A number of countries explicitly have recognized that problem and as part of the system modernization effort implemented projects of record cleanup and consolidation. 38 Arguably, the level of effort across these cases is comparable. One-off lump sum payments require extensive regular certification as all first-time claims. Regular (monthly) payments while less resource intensive are numerous. 39 For example, as is often the case with short-term health benefits, there are risks of double counting when the same members are covered and benefit from multiple insurance programs. 35 These factors should be carefully considered in the choice of the denominator in universal indices. Table 4 summarizes various options, including associated biases. Table 4: Choice of Denominator in Cost Indices and Associated Biases Size/membership Operational Bias Technology Bias Generosity Bias Financing Bias Coverage Bas Maturity Bias Bias Cost Index Denominator 1. Contributions + + + + + 2. Benefits + + + + + 3. Contributions + Benefits + + + + 4. GDP + + + + 5. Pensions Assets or IPD + + + + + 6. Covered wages + + + + 7. Contributors (+) + + + 8. Beneficiaries + + + + 9. All members + + + 10. All members (income adjusted) (+) + + Source: Author’s desig . Let us illustrate some of these biases at work. There is a maturity bias associated with (2): younger earnings-related schemes that cannot be compared to some older stabilized schemes (see the example of Kosovo in the table in the annex). Financing bias is present in (1): noncontributory schemes or contributory schemes with significant budget subsidies that cannot be put on the same line of comparison with fiscally balanced contributory schemes (see the cases of New Zealand and Netherlands). To address these problems, often, we find that a composite measure (3) is used. However, there is another generic problem—generosity bias—as schemes with the same organization, coverage, and operational costs will not look the same when differences in the contribution and/or benefit rates are significant (for that reason programs in Poland and St. Kitts and Nevis, while very similar in relative coverage and institutional organization, cannot be directly compared). To remedy this problem, some would use the administrative costs per GDP ratio (4). This measure, however, has its own problem, a coverage bias. Consider, for example, the provident fund for the formal sector employees in India (EPFO), civil service pension plan in the United States (TSP), and the national pension program in Estonia (managed by SIB). In these cases, the differences in total cost per GDP measure for SIB and TSP are dramatic (almost 80 times), reflecting differences in the relative size of those programs in the economy (and perhaps additional institutional subsidies that TSP receives). However, income adjusted per member costs are almost the same for SIB and TSP. At the same time, while the cost per GDP is 3 times smaller in EPFO than in SIB (with labor coverage of that program 10 times smaller in India), the income adjusted per member cost of EPFO is 7 times larger, indicating potential inefficiencies. 36 In summary, all these measures are quite informative but should be used only with subsets of comparable programs. We also note that there may be a progression in the usefulness and applicability of various indices. Indices in (1), (2), and (3) may be more relevant when a system is stable and mature. However, when administrative costs devour most revenues or investment profits, such indices will not reveal much about the health of the program. Rather, indices in (4), (9), or (10) would be more revealing. VI. Data Analysis and Cost Benchmarking We provide three alternative quantitative models for our analysis. We first investigate the effects of program design on administrative expenditures (for example, private versus public management). We then study factors explaining the differences in administrative expenditures in managing pension liabilities. Finally, we proceed with benchmarking analysis for individual programs. We account for the program size by using the information on the total number of members. However, we track separately the active contributing or effectively insured members (where the pension agency keeps corresponding records of such members in-house) and the beneficiaries (including recipients of old-age, disability, survivors, work injury, and other pensions as well as cash benefits like maternity, family, and child allowances). We find significant differences in how total counts in these two groups affect total costs. To control for heteroscedasticity, all our quantitative variables are in natural logs. Corresponding coefficients, therefore, indicate percentage change in the endogenous variable as a result of the 100 percent increase in the explanatory variables. 6. 1. Administrative expenditures and program design The first data set is the most comprehensive and includes all observations of our sample, including 116 publicly managed programs (see annex 1 and annex 2) and 12 additional observations of the combined administrative charges of privately managed pension programs.40 The cost function is constructed as follows: ln EXP = a0 + a1 ln BEN + a2 ln INS + a3 DCSCHEME + a4 PRIVATEMGT + a5 COLLECTION + a6 SHUL + a7 BP + a8 ln GDPpc + a9 OUTLIERS + e, where EXP is the total operating expenses;41 BEN is the total number of beneficiaries serviced by the program or agency;42 and INS is the total number of active contributors (or 40 Sources include Tapia and Yermo (2008) and FIAP statistics (http://www.fiap.cl). 41 This definition treats expenditures of public programs and total charges imposed under mandatory private schemes equally. It also includes all costs associated with asset management (in-house or outsourced). 37 insured). The five categorical variables are correspondingly for DC scheme; DC scheme managed by private agency(ies); schemes where contribution collection is operated largely in-house; schemes with additional benefits (sickness, health insurance, unemployment insurance, or loans to members); and basic pension schemes (rank 1 in table 2). GDPpc is the national income per capita to account for differences in technology and quality of institutions. Finally, we define a group of programs that are outliers in terms of excessive costs per member adjusted for income differences (see section 4.1); in all our regressions this variable was highly significant. Table 5 contains the results. Table 5: Administrative Expenditures and Program Design Independent Variables Ln Total Operating Expenses (a) (b) ln BEN 0.45 0.60 (11.97) (17.15) ln INS 0.15 0.06 (6.07) (2.59) DCSCHEME – 0.44 (1.87) PRIVATEMGT – 3.33 (8.49) COLLECTION – 0.05 (0.21) SHUL – 0.22 (1.16) BP – -1.44 (-2.46) ln GDPpc 0.65 0.59 (8.08) (8.86) OUTLIERS 1.47 1.84 (3.88) (6.40) CONSTANT 4.29 3.69 (5.35) (4.96) Observations 128 128 2 Adjusted R 0.72 0.84 Note: t-statistic in parentheses. Our first specification is simply to account for program size and income differences, while the second specification introduces design elements and generates considerable additional explanatory power. These are the key observations: 42 This generally includes old-age pensions, disability pensions, survivor’s pensions, work-injury pensions, other pension benefits, maternity (parental) allowances, family allowances, child allowances, and other assistance and compensations. 38  In line with other studies, we identify economies of scale as the coefficients by both the beneficiaries and insured are less than one. At the same time, confirming the observations of section 4.3, there are significant differences in the effects produced by beneficiaries versus contributors. This is consistent across all our specifications and subsets of data. There are significant differences in variable costs between two different lines of social security operations. We discuss this finding later in the text.  The results show the robust and significant effect of private management of pension plans. However, as we move to our long-term specifications, we find that the effect is not robust and may, in part, reflect differences in the maturity and coverage of the schemes in the shorter term.  The evidence of cost differentials between the DB and DC schemes is weak. In fact, substituting this variable with the categorical variable for fund management produces somewhat stronger results. This indicates that the mere fact of managing financial assets (in either DC or DB schemes) is associated with some additional costs, thereby reflecting a need for advanced skills and systems.  We expected that bringing the collection function in-house would increase costs. However, the results do not show robust evidence for such an increase. One possible explanation is that the modes of organizing the collection function vary significantly (see Anusic 2005), so capturing such a variation under one categorical variable constitutes a measurement challenge. We also note that as far as variable costs are concerned, the function of managing active contributors is not that impactful. Perhaps the collection function, however loosely defined, is a relatively small add-on in terms of variable costs. The conclusion is that the argument for consolidation of the collection function cannot be supported on the basis of cost reduction alone but rather on the basis of other systemic improvements (for example, reduced administrative burden, improved compliance, and overall improvements in economic efficiency due to reduced informality). This finding does not extend to start-up or other fixed costs. Where significant investments are required to establish or modernize a collection function, both tax and social security contributions systems could benefit from a well-coordinated effort.  Administrative costs increase less than proportionately with increases in income per capita. One possible explanation suggested by earlier studies is that more developed countries can manage pension schemes more efficiently, taking advantage of better technologies, infrastructure, and institutions (see Mitchell et al. 1993 and James and Palacios 1995).  In line with figure 3, plain basic pension schemes that do not require a history of contributions to establish eligibility are less expensive to manage. At the same time, we did not find robust evidence that schemes that serve the public sector only or means- tested pension programs are systematically less expensive on average. This can be in line with our earlier observation of the disproportional importance of the benefit- management function relative to contribution management. As long as the agency does 39 benefit calculations and payments, the costs of recordkeeping do not differ much across different program types. 6. 2. Administrative expenditures and pension liabilities We now construct a specification that would reveal the long-term effects of various design factors on costs. We use a measure of pension liabilities and validate preliminary findings from section 5. 2. To control for maturity and generosity biases, we introduce a measure of the average member account value.43 The cost function is constructed as follows: ln EXP = a0 + a1 ln LIABILITY+ a2 DCSCHEME + a3 PRIVATEMGT + a4 COLLECTION + a5 ln GDPpc + a6 ln ACCOUNT + a7 OUTLIERS + a8 GOV + a9 SHUL + e where EXP is the total operating expenses defined as in the previous equation; LIABILITY equals the total reported assets of the DC schemes or an estimated Implicit Pension Debt (IPD) for the DB schemes;44 the categorical variables are the same as in the previous equation; ACCOUNT is the ratio of total DC assets or DB IPD over total members (to approximate maturity); GOV is the government effectiveness index45. We use the same observations as in the previous regression; however, the sample is smaller due to the limited data on the implicit pension debt of public programs. These are the same observations as figure 6 illustrates. Table 6 contains the results. Table 6: Factors Affecting the Cost of Managing Pension Liabilities Independent Variables Ln Total Operating Expenses (a) (b) (c) ln LIABILITY 0.84 0.90 0.90 (16.97) (20.08) (21.09) DCSCHEME 1.19 0.40 0.39 (3.46) (1.30) (1.40) PRIVATEMGT 0.98 0.50 0.51 (2.83) (1.67) (1.82) COLLECTION 0.55 0.98 0.87 (2.09) (4.27) (3.91) 43 We define members as all contributors and beneficiaries for the DB schemes and as contributors only for most DC schemes in our sample (recognizing that retiring members in most cases liquidate their balances at the point of separation). 44 Source of data is Holzmann et al. (2004). 45 Government effectiveness captures perceptions of the quality of public services, public administration, public infrastructure, the quality of the civil service and the degree of its independence from political pressures, the quality of policy formulation and implementation, and the credibility of the government's commitment to such policies (World Bank 2012). 40 Independent Variables Ln Total Operating Expenses (a) (b) (c) ln GDPpc – 0.63 1.11 (4.39) (5.57) ln ACCOUNT – -0.75 -0.80 (-5.22) (-6.09) OUTLIERS – 2.01 2.71 (4.75) (6.47) GOV – – -0.72 (-3.22) SHUL – – 0.51 (2.30) CONSTANT -3.07 -3.06 -6.63 (-2.34) (-2.83) (-4.20) Observations 52 52 51 2 Adjusted R 0.88 0.93 0.94 Note: t-statistic in parentheses. At first look, our baseline specification reveals costs differences associated with our core set of design elements. However, as we introduce the adjustment for maturity, the marginal effects on costs of the scheme design become smaller and less significant. Here are some more specific observations:  The value of the coefficient by pension liabilities (<1) indicates economies of scale in management of pension liabilities. The larger, older, and more generous schemes tend to be less expensive to manage per unit of liability. However, effects of these three factors are not distinguishable here.  The DC schemes, private sector schemes (all DC in our sample), and schemes that operate a fund management function (not shown among these results) are all associated with a cost increment. However, the results are not robust and their significance depends on choice of specification. As the maturity indicator is added, the scheme design becomes a less significant explanation of the cost differentials. This may suggest that in the long run, design largely does not matter and the costs should not be the key driving factor in policy choices over a particular design type.  The sign of the coefficient of the average account size is as expected and suggests that unit costs decline as schemes mature. In the long run, this factor may compensate for the possibly higher overall management costs of the DC schemes, if the same account over time can generate greater value under the DC arrangements compared to implicit wealth generated under a DB scheme. So, the policy focus should be on the wealth- generating patterns of different types of schemes and not just on cost ratios.  This specification indicates that in-house collection of contributions is associated with higher administrative expenditures.  We find again that the value of the coefficient for national income per capita is positive but less than one, indicating that more developed countries may have better institutions and access to better technologies, and so can manage pension liabilities more 41 efficiently. We also experimented with several governance indices and found that index that captures government effectiveness produces the most robust effects (specification [c] in table 6).46 Higher levels in that index are associated with lower administrative costs. Interestingly, as this index captures most of institutional factors, the response to changes in the levels of national income becomes close to one. These results seem to suggest that technologies as they spread over time become less important in explaining cost differences, and what ultimately matters is the quality of governance.  We also note that managing special supplementary programs and benefits produces increments in operating costs. 6. 3. Administrative expenditures and institutional organization We now proceed with the third and main model of our analysis, which we also use for performance benchmarking of mandatory programs. The sample of programs used in the regression contains only publicly operated programs and only those programs for which we have information on staffing levels. We adopted a two-step approach. In the first step, we assess and benchmark technical efficiency using the data on staffing levels. We then obtain residuals from this step as an indication for over- or under-staffing and use them in the regression of the second step in which we look at the cost efficiency of the same programs. In the final results, we can then distinguish sources of deviations from the benchmarks. Step I regression equation is constructed as follows: ln STAFF = a0 + a1 ln BEN [+ a2 (ln BEN)2] + a3 ln GDPpc + a4 COLLECTION + a5 SHUL + e, where the new variable is STAFF as the total number of staff in the agency. Table 7: Staffing Requirements for Pension Administration Independent Variables Ln Total Number of Staff (a) (b) (c) ln BEN 0.66 0.72 – (20.57) (25.32) (ln BEN)2 – – 0.03 (27.58) ln GDPpc – -0.11 -0.13 (-1.98) (-2.57) COLLECTION – 0.70 0.67 (3.52) (3.64) SHUL – 0.52 0.52 (3.26) (3.52) 46 The governance indices were not found to be significant in other regressions. 42 Independent Variables Ln Total Number of Staff (a) (b) (c) CONSTANT -1.31 -1.76 2.41 (-3.26) (-3.00) (4.85) Observations 99 99 99 2 Adjusted R 0.81 0.88 0.90 Note: t-statistic in parentheses. Remarkably, the number of beneficiaries (recipients of retirement benefits) alone explains over 80 percent of the variation in the staffing levels. The level of the coefficient indicates economies of scale. We experimented with several alternative specifications and found that the number of the insured or contributors for whom records are kept in-house produced only a small additional power, and the significance of its coefficient drops to almost zero when we add various categorical variables. This may be due to the fact that the agencies do not really provide direct service to contributors and largely interact with their employers, so association with the contributor numbers is loose. Mere recordkeeping of the contributors does not seem to significantly affect staffing requirements but contribution collection and special additional services does, implying the importance of fixed costs over variable costs for that line of business. We also found that design of the scheme (DC versus DB) or sectoral affiliation (public only versus private sector) did not produce any systematic differences in staffing requirements. In most regressions, we obtained slightly better fitting results when using a squared function for beneficiaries. It may reflect the fact that the variable costs on top of significant fixed costs are relatively indistinguishable for smaller plans, so the quadratic function captures that aspect better. In our benchmarking we used quadratic functions in both staffing requirements and administrative costs regressions.47 An interesting observation can be made regarding the negative sign of the income per capita coefficient in the staffing regression. It indicates that as economies develop and new technologies become available, they tend to substitute capital for labor. This particularly conforms to the case for the U.S. Social Security Administration depicted in figure 1. We obtained residuals from (c) and used them as STAFF_RES in the Step II regression: ln EXP_NAMC = a0 + a1 (ln BEN)2 + a2 (ln INS)2 + a3 ln GDPpc + a4 STAFF_RES + a5 COLLECTION + a6 SHUL + a7 FUNDSMNGMT + a8 OUTLIERS + e. The principal difference in this regression is that here we use the progra ’s total operati g expenses net of explicit direct costs associated with asset management, often external to the administration (EXP_NAMC). Our reasoning was that practices of managing pension 47 We also observed that the coefficients of linear terms become insignificant in the quadratic specifications. 43 assets vary substantially and so do the associated costs and norms of reporting those costs. By taking out those costs, we focused on benchmarking only the core operation mechanisms. We do admit, however, that total and clear segregation of those costs was not possible in all cases.48 To capture various related costs, we added a categorical variable FUNDMGT associated with the management of financial assets in either DC or DB schemes. Table 8: Key Factors Affecting Costs of Public Pension Programs Independent Variables Ln Total Operating Expenses Net of Asset Management Costs (a) (b) (ln BEN)2 0.03 0.03 (20.52) (27.01) (ln INS)2 0.003 – (2.70) ln GDPpc 0.46 0.49 (9.18) (10.15) STAFF_RES 0.69 0.72 (6.43) (8.08) COLLECTION – 0.42 (2.54) SHUL – 0.45 (3.52) FUNDSMNGMT – 0.50 (3.21) OUTLIERS 1.47 1.53 (6.08) (7.54) CONSTANT 8.43 7.15 (19.39) (15.53) Observations 99 99 2 Adjusted R 0.91 0.94 Note: t-statistic in parentheses. One of the most important and already familiar observations is a striking difference in the coefficients by beneficiaries versus insured (specification [a]), which consistently reflects across all our models and specifications (including linear and quadratic). This indicates that the variable costs are much more important in the management of beneficiaries relative to the management of contributor accounts. In other words, there are significant economies of scale in managing active member accounts. Furthermore, variable costs become insignificant when the categorical variable of in-house collection is added. These findings have interesting implications for the policy of collection administration. Specifically, as significant capital investments have been made in the collection function, transferring that function to an external agency, such as tax administration, may not be a significant cost 48 In most such cases, investment management remains largely in-house and is often restricted to passive, limited, or illiquid portfolios, including public debt and real estate. 44 saver. Hence, if the capacity of an external collection agency is weak, the advantage of such outsourcing may be questionable. At the same time, significant up-front cost-saving advantages could be seized if new schemes can rely on the existing collection infrastructure. On the beneficiary side, given significant variable costs, there may be considerable economies of scope, which may argue in favor of consolidating various benefit programs under a unified administration (for example, universal basic pensions and earnings-related pensions or retirement benefits and various short-term or other special benefits, especially where these mostly cover the same groups of beneficiaries). The significance of the coefficients by GDPpc and by the residual from the staffing regression again suggests that more developed countries can operate their schemes more efficiently and that a significant part of variation in costs can be explained by the deviation of staffing levels from projected benchmarks. Additional services to active members (benefits of sickness, health, unemployment, or individual loans) increase the total costs. The significance of the coefficient by funds management may capture two aspects: (a) the additional costs of managing pension assets that cannot be distinguished from the total costs are important, and (b) even if costed out separately, such functions may be associated with a cost premium, for example, for higher skilled staff. 6. 4. Performance against benchmarks Annex 3 contains the results of our benchmarking analysis, providing actual and fitted estimates for both staffing and total administrative expenditures. It also provides benchmark coefficients: where such coefficients are close to 1, the program operates at its benchmark. While we ran regressions only for 99 observations where we had full information on staffing levels, our methodology allows for producing expenditure benchmarks for the programs outside of that sample. We treated two specific parameters as anomalies: (a) deviations from projected staffing levels and (b) outliers with excessive administrative expenditures. To benchmark the expenditures, we turned both parameters to zero. Effectively, our expenditure benchmarks assume the benchmark performance on the staffing side. Where we have information on the actual staffing levels for particular programs, such information provides additional insight into program performance. Where such additional information is not available, the benchmark ratio simply indicates how the program performs relative to the cost benchmark and given staffing performance just at the benchmark. We will illustrate further how to interpret these results. We further group programs in three categories. Category A is for progra s that perfor at the benchmark. Expenditure coefficients for this group are within the range of plus and minus 25 percent of the benchmark to allow for measurement errors and minor 45 inefficiencies. Around 30 percent of the programs in our sample are in this category for the total expenditure performance. Categor B is for progra s that de iate fro the e h ark e o d per e t ut ot exceeding 100 percent. This may indicate moderate inefficiencies of under- or over- resourcing in either staff or other expenditures. Around 50 percent of the programs in our sample fall under this category and may merit further inquiries into potential operational optimization. Programs where performance coefficients are more than double the predicted levels are in ategor C . Most of them are suspects for operational inefficiencies, especially those where benchmark coefficients are multiples of the predicted levels. Out of the eleven programs where the expenditure-to-benchmark ratio exceeds 5, eight are located in Sub- Saharan Africa. In the cases of Uganda, Kenya, and Ghana the expenditure coefficients are 26, 15, and 11, respectively. It may be easy to overspend when operating significant surpluses, which all three happen to have, but excessive costs certainly cannot be affordable as schemes mature. Where we have information on the staffing levels, we can define several categories in terms of how staffing affects overall expenditure outcomes.  There is a AA ategor of progra s i hi h oth staffi g a d e pe ditures are at the benchmark (13 programs in our sample).  Where both staffing and expenditure benchmarks are on the higher side, it indicates that osts are ai l dri e e essi e staffi g. There are progra s i the CC category. In the cases of Norway and U.K.-Northern Ireland, both staffing and expenditure ratios are of the same magnitude (around 5 and 3, respectively), indicating that staffing is a key driving factor behind the higher levels of administrative expenditures. We do note, however, that findings of excessive costs for some European agencies, including in Norway, Ireland, and U.K.-Northern Ireland, should be taken in the context of operating multiple programs, including active labor market programs and various targeted welfare schemes, which our analysis cannot fully capture. The same is true for the U.S. SSA that has targeting costs to consider. The Canadian public sector plan falls into this category perhaps due to the additional services it provides to employers outside of the government, requiring a more extensive compliance enforcement effort, even though we have classified its collection effort as insignificant compared to conventional private sector plans. Uganda is an interesting case; in 2008, the NSSF embarked on the mission to decentralize its operation, and by 2009 the agency had 4 times the predicted requirements for staffing. The decentralization never took full effect, but it subsequently was found that many staff were underutilized and roles duplicated. Thus, a decision was made to downsize and release close to 40 percent of 46 the staff,49 which would be in line with recommendations of our analysis but still far from the benchmark performance.  At the lower e d of resour e o it e t is the BB ategor . We fi d, for e a ple, that the DC schemes in Denmark, the Maldives, and Singapore, operate with very low demand on both labor and other resources when the costs of asset management are excluded, which may reflect efficient operation. At the same time, there are low-end resource-commitment programs like the public sector pension scheme in Afghanistan, which is a candidate for suboptimal operation. While serving beneficiaries across Afghanistan, the operation is concentrated in the capital city, where 90 percent of beneficiaries have to travel in order to get service. Additionally, payments are made only once a year. Clearly, low costs come with poor service here.  There is a group of programs in which expenditures are excessive while staffing levels are at or below the benchmark. For example, Burkina Faso and Tanzania (PPF) are in the C ategor for e pe diture, a d those also happe to e flagged a s outliers in our regressions. An important point to note is that their staffing levels are on the low side. We made additional investigation into the overall costs of labor. For the observations where we had separately reported labor costs, we calculated the ratios of those costs per staff numbers and adjusted by GDP per capita. We then calculated the median alues of that i de for e pe diture oeffi ie t ategories B-lo , A , B-high , and C a d those stood at . , . , . , a d . , respectively. This implies that those programs that overspend have a tendency to pay premiums relative to the rest of the economy to their staff.  Finally, there are programs in which total expenses against the benchmark are on the low side while staffing is at or above the benchmark. The most notable cases are the Sri Lankan Farmers Pension Scheme and the Syrian GOSI. It seems that those programs retain inexpensive labor, possibly lack investments in technology and infrastructure, and operate suboptimally in terms of resource allocation between labor and capital. 6. 5. Implications for choice of cost indices Our analysis leaves many programs outside the scope of the benchmarking exercise. However, it can offer guidance on the ultimate choice of a simple index that makes cost comparisons meaningful. The following table provides estimates of correlation between our expenditure benchmark ratios and a set of conventional cost indices that are discussed in section 5. 49 NSSF Business Reorganization Exercise: Rightsizing for Efficiency. 2010. http://www.nssfug.org/uploads/NSSF%20Business%20reorganisation%20exercise.pdf . 47 Table 9: Choice of Denominator for Cost Index and Correlation with Cost Benchmark Contribution Benefit Insured + Beneficiaries Beneficiaries GDP Revenues Expenditure (GDPpc adjusted) (GDPpc adjusted) 11% 43% 56% 69% 80% Source: Author’s al ulatio s. The results broadly reflect our earlier discussion of biases associated with various indices. Expenditure over GDP is not a good predictor of program health. Using program revenues or benefit expenditures does not produce sufficient explanatory power either. Using the combined total number of insured and beneficiaries and adjusting for differences in incomes provides better results among the commonly used indices. Apparently, even better results can be achieved by using just the number of beneficiaries in the denominator. This should not be surprising as our regressions show disproportionate importance of the numbers of beneficiaries compared to the insured in explaining the cost differences. 6. 6. Global benchmarks To further guide inquiries into the cost indices analysis, we generated a set of benchmarks on the basis of our model. Annex 4 provides a set of just-on-benchmark estimates of the index of Administrative Expenditures over Beneficiaries for different program sizes and different levels of national income. The direct costs of asset management, if any, are excluded. We provide estimates for five stylized country income categories (GDP per capita in current US$ indicated): low income (LI: US$500); lower-middle income (LMI: US$2,500); upper-middle income (UMI: US$8,000); lower-high income (LHI: US$15,000); and upper- high income (UHI: US$40,000). In each case, we provide 5 schedules: a baseline low estimate; three estimates which correspondingly are the functions of in-house collection, fund management, and special supplementary schemes assessed individually; and a high cost estimate in which all three functions are combined. Results indicate that with differences in functional organization and services, the spread between the high and low estimates is almost fourfold. With such a broad benchmark, we are cautioned against providing advice on the appropriate level of expenditures for a particular program without detailed inquiry into its operational organization. We also observe the same schedule of economies of scale across all stylized cases. Figure 9 depicts such a schedule as a proportion of a midsize plan with 500,000 beneficiaries. 48 Figure 9: Economies of Scale in Administrative Expenditures 4.00 3.75 3.50 3.25 3.00 2.75 2.50 2.25 2.00 1.75 1.50 1.25 1.00 0.75 0.50 0.25 - Beneficiaries Source: Author’s calculations. Note: Per-beneficiary costs relative to per-beneficiary costs of a plan with 500,000 beneficiaries. For plans with 100,000 beneficiaries, the premium for their smaller size is 50 percent over the costs of similar plans with 500,000 beneficiaries. Similarly, larger-sized plans could be 25 percent less expensive (per beneficiary) to manage. One particular application of this schedule could be the planning of consolidation programs. This schedule can also be helpful in more flexible benchmarking of program costs. Figure 10 depicts low and high scenarios for per-beneficiary costs of a program with 500,000 beneficiaries for different levels of GDP per capita. As our regression coefficients indicate, costs increase less then proportionally with increases in income levels. To illustrate applicability of this method, let us assess the benchmark spread of expenditures for a plan with 100,000 members in an economy of US$15,000 GDP per capita. First, we use from figure 10 the spread for the 500,000 plan, which is US$50–200. Then from figure 9, we find that the size premium for a 100,000 plan is approximately 50 percent. This means that the benchmark spread we are looking for is US$75 –300. 49 Figure 10: Per-Beneficiary Cost Spreads for a Midsize Operation (Nominal US$) 550 500 450 400 350 300 250 200 150 100 50 - GDP per capita Source: Author’s al ulatio s. We also developed similar benchmark schedules for staffing levels. For the convenience of application, we provided workload schedules in which beneficiaries are in the numerator. Figure 11 indicates economies of scale in staffing requirements to manage programs of different sizes. Figure 11: Economies of Scale in Staffing Requirements 1.40 1.20 1.00 0.80 0.60 0.40 0.20 - Beneficiaries Source: Author’s al ulatio s. Note: Workload relative to workload of a plan with 500,000 beneficiaries. 50 Interpretation of results is similar to the one provided for the costs schedule. Plans with 100,000 beneficiaries require approximately 30 percent more staff per beneficiary compared to the same type of plan with 500,000 beneficiaries. At the same time, the larger- sized schemes require 30 percent less of human resources per beneficiary. Figure 12 depicts low and high scenarios for the workload ratios of a program with 500,000 beneficiaries for different levels of GDP per capita. As our regression coefficients indicate in table 7, staffing requirements per beneficiary actually decline with increases in income levels (the reverse of that relationship is presented in the workload schedule). The variation between the low and high scenarios is defined by adding functions of in-house contribution collection and provision of special schemes, which we found statistically significant in explaining cost differences. Note that the curve on top indicates the results of the low resource requirement, and the curve at the bottom is the high resource estimate (with both of these additional functions factored in). It is important to emphasize again the wide spread in the benchmark, which is threefold between the low and high estimates. Hence, any recommendations on staffing levels should be carefully crafted with incorporating information on the nature of operational organization or services a particular agency provides. Figure 12: Beneficiary per Staff Ratios for a Midsize Operation 1,100 1,000 900 800 700 600 500 400 300 200 100 - GDP per capita Source: Author’s al ulatio s. 51 VII. Quality Aspects in Cost Measurement: What is Left to Residual In this section we discuss the implications of not having full data on the nature and quality of services for our results. Service quality comes at a cost. Higher operational expenses may reflect better services (for example, more frequent and direct communications with clients, faster processing of benefit claims, and more inclusive payment methods). Unfortunately, differences in the nature of particular bundles of service or variation in their quality remain uncaptured. Hence, we need to interpret our results and recommendations on benchmarking with caution. Figure 13 illustrates the limitations of our analysis. Figure 13: Quality Cost Tradeoffs Cost H B M A L C Quality Source: Author’s desig . This diagram introduces an additional dimension to our multivariate regression and effectively collapses our regression line into one single point, A . Assuming that there is a relationship between the costs and quality of services offered, our regression generates an average cost estimate around an average bundle of services that are of average quality. Thus, if our assessment finds that a particular scheme operates just at its benchmark, it corresponds to line M in the diagram. However, it does not necessarily imply that the scheme offers services at an efficient level. In fact, given the cost, the scheme can offer a service bundle that would place it to the left of poi t A , which is a suboptimal cost and quality combination. Or it can offer exceptional services, which would put it on the far right of poi t A , implying an efficiency premium in such a system. So, on-the-benchmark results of our simulations can only be taken to imply equal probabilities of over- and under- efficient operation. However, if the osts e eed our e h ark, it puts su h a s ste o li e H, hi h ea s that ith the opti al operatio arked poi t B there o is a greater pro a ilit that the system operates sub-efficiently. Finally, with the estimate below the benchmark, we are 52 o o li e L, a d so, there is a higher pro a ilit that the s ste osts less tha the average cost estimate of the same bundle of services would suggest. In summary, the tools of conventional analysis with narrow sets of explanatory variables can only produce very limited inferences about the performance of various programs. In each case, we need to look beyond our results. Special operational and beneficiary surveys could help capture information on the performance and satisfaction of various stakeholders with program administration, including processing times, compliance costs and various overheads, and overall perception of service quality. At the same time, our methodology and findings help point in the direction of such additional inquiries. Furthermore, it is difficult to establish a clear association between the costs and value of service, even in theory, to support any assumption about the shape of the cost line in figure 13. Even if we could define distinct bundles of services, ranking of their social preferences and hence, value for society would not be straightforward. While it is beyond the scope of this paper, we can take a brief look at this challenge. For example, in the context of a DC plan where individuals may have a choice of an investment option, let us consider three alternative operational arrangements: (1) there are three investment options with switching allowed once a month, (2) there are five investment options but with switching allowed no more than twice a year, and (3) there is only one default investment portfolio. Perhaps both (1) and (2) are superior to (3) but choosing a superior solution between (1) and (2) is not clear-cut. In fact, on a society utility plateau, (1) and (2) may be equally preferable but the costs may considerably and systematically differ. So, it would be difficult to establish a clear association between the costs and values (quality) of services in this case.50 Thus, the gap in accounting for the variation in service value and quality in the operation of mandatory social security programs limits conclusiveness of our benchmarking analysis. VIII. Conclusions As countries develop and seek to provide coverage to greater segments of their population, administrative costs become an important aspect of reforms, especially where new mandate extends to low-income or informal-sector workers. New technologies pave the way for effective outreach, monitoring, and recordkeeping, while infrastructure improvements (including financial services) provide for better access. Comparing and benchmarking administrative expenditures helps assess the efficiency of different modes of operational organization of public social security programs. It provides guidance on reform strategy, choice of alternative organizational models, and trade-offs in 50 We also note that all quality improvements with their additional investment costs exhibit diminishing returns. While (3) is one extreme of the spectrum, at the other end of the same spectrum, we may have an unlimited number of options with unlimited switching allowed and all instantly available. 53 instituting various new operational elements. Inquiries into operational efficiency often prompt complex organizational transformations. Among trend-setting practices are cutting redundant staff, employing more advanced technologies, sharing certain functions with other public entities, and outsourcing select tasks to other agencies. To decide on optimal investments in systems, processes, and people, it is important to understand the key factors that affect the costs of operating various schemes. First, it is important to recognize economies of scale and scope in managing social security programs, including their magnitudes in contribution collection and benefit management. Synergies with existing mechanisms should always be sought. Private management of the schemes will be more expensive compared to the public option but differences may disappear over the long term. At the same time, funding of pension liabilities (in either DB or DC schemes) will always involve cost premium, given advanced complementary resource requirements. These considerations will involve important policy decisions. Finally, the level of economic development has a strong impact on costs, suggesting that more developed countries can manage social security schemes more efficiently, possibly taking advantage of better technologies, infrastructure, and institutions. As technologies spread over time, they may become less important in explaining cost differentials. Yet, quality of governance seems to remain an important indicator of the financial health of any program in both the short and long run. 54 References Anusic, Zoran. 2005. International Experience in Consolidated Social Contributions and Tax Collection, Reporting and Administration. Mimeo, World Bank. Arenas de Mesa, Alberto, Jere Behrman, and David Bravo. 2004. Chara teristi s of a d determinants of the density of contributions in a Pri ate “o ial “e urit “ ste . Michigan Retirement Research Center, University of Michigan. Barrand, Peter, Stanford Ross, and Graham Harrison. 2004. Integrating a Unified Revenue Administration for Tax and Social Contribution Collections: Experiences of Central and Eastern European Countries. IMF Working Paper WP/04/237, International Monetary Fund. Bauer, Rob M. M. J., K. J. Martijn Cremers, and Rik G. P. Frehen. 2010. Pension Fund Performance and Costs: Small is Beautiful. Working Paper 10-04, Yale International Center for Finance. Bebczuk, Ricardo N., and Alberto R. Musalem. 2008. Public Pension Funds around the World: Governance, Investment Policies, and Performance. Mimeo. Bikker, Jacob A., and Jan de Dreu. 2007. Operating Costs of Pension Schemes. In Costs and Benefits of Collective Pension Systems, edited by Onno W. Steenbeek and S. G. van der Lecq. Caswell, Jerry W. 1976. Economic Efficiency in Pension Plan Administration: A Study of The Construction Industry. Journal of Risk and Insurance 43: 257–273. Chlon, Agnieszka. 2002. Administrative Costs of Pension Funds in Poland in International Perspective. Regional Meeting for the Eastern and Central European Countries, Tallin, Estonia. Chłoń-Do iń zak, Ag ieszka, Piotr De derski, Mariusz Kubzdyl, and Dariusz “tańko. 2007. Costs and Charges of Pension Funds: International Comparison and Methods of Approximation. Warsaw: Institute for Structural Research. Corvera, F. Javier, J. Mateo Lartigue, David Madero. 2006. Comparative Analysis of Administrative Fees of Pension Funds in Latin American Countries. Mimeo. Demarco, Gustavo, and Rafael Rofman. 1999. Collecting and Transferring Pension Contributions. Social Protection Discussion Paper 9907, World Bank. Diamond, Peter. 1998. Administrative Costs and Equilibrium Charges with Individual Accounts. Working Paper 7050, National Bureau of Economic Research, Cambridge, Massachusetts. ———. . E o o i Theor a d Ta a d Pe sio Poli ies. The Economic Record 87 (Special Issue): 2–22. Dobronogov, Anton, and Mamta Murthi. 2005. Administrative Fees and Costs of Mandatory Private Pensions in Transition Economies. PEF 4 (1): 31–55. FIAP (International Federation of Pension Funds Administrators). 2006. Collection costs in Pension Fund Systems. Santiago: FIAP. Fultz, Elaine, and Tine Stanovnik, eds. 2004. Collection of Pension Contributions: Trends, Issues, and Problems in Central and Eastern Europe. Budapest: International Labour Organization. 55 GAO (United States Government Accountability Office). 2007. Cost Assessment Guide: Best Practices for Estimating and Managing Program Costs. Exposure Draft, GAO. Genetski, Robert. 1999. Administration Costs and the Relative Efficiency of Public and Private Social Security Systems. The Cato Project on Social Security Privatization 15. Gru išić, Mihaela, a d Mustafa Nuši o ić. 2006. Fee Redistribution and Cost Rationalization in the Second Pension Pillar of the Croatian Pension System. Zagreb: The Institute of Economics. Hart, Lawrence E., Mark Kearney, Carol Musil, and Kelly Olsen. 2001. ““A’s Estimates of Administrative Costs under a Centralized Program of Individual Accounts . Social Security Administration. Holzmann, Robert, Robert Palacios, and Asta Zviniene. 2004. Implicit Pension Debt: Issues, Measurement and Scope in International Perspective. Social Protection Discussion Paper 0403, World Bank. Hernández, Denise Gómez, and Fiona Stewart. 2008. Comparison of Costs and Fees in Countries with Private Defined Contribution Pension Systems. Working Paper 6, International Organization of Pension Supervisors. James, Estelle, James Smalhout, and Dimitri Vittas. 2001. Administrative Costs and the Organization of Individual Account Systems: A Comparative Perspective. In New Ideas About Old Age Security, edited by Robert Holzmann and Joseph Stiglitz. Washington, DC: World Bank. James, Estelle, and Robert Palacios. 1995. Costs of Administering Public and Private Pension Plans. Finance and Development 32 (2): 12–16. Lum, Hubert. 2006. The World’s Lo est Cost Fu ds. CEM Benchmarking Inc. Mackenzie, George A. 2002. The Role of Private Sector Annuities Markets in an Individual Accounts Reform of a Public Pension Plan. IMF Working Papers 02/161, International Monetary Fund. Mitchell, Olivia S. 1999. E aluati g Ad i istrati e Costs i Me i o’s AFORE“ Pe sio “ ste . Pension Research Council, Wharton School, University of Pennsylvania. Mitchell, Olivia S., and Emily S. Andrews. 1981. Scale Economies in Private Multi-Employer Pension Systems. Industrial and Labor Relations Review 34 (4). Mitchell, Olivia S., Ping-Lung Hsin, and Annika E. Sunden. 1993. An Appraisal of Social Security Administration Costs. World Bank. Murthi, Mamta, J. Michael Orszag, and Peter Orszag. 1999. Administrative Costs Under A Decentralized Approach to Individual Accounts: Lessons from the United Kingdom . In New Ideas about Old Age Security, edited by Robert Holzmann and Joseph E. Stiglitz. World Bank Palacios, Robert. 2005. Systemic Pension Reform in Latin America: Design and Early Experiences. In Workable Pension Systems: Reforms in the Caribbean, edited by P. Desmond Brunton and Pierto Masci. Inter-American Development Bank; Caribbean Development Bank. Palacios, Robert, and Montserrat Pallarès-Miralles. 2000. International Patterns of Pension Provision. World Bank. 56 Palacios, Robert. Forthcoming. Framework for Implementing Social Programs (FISP). World Bank. Palacios, Robert, Oleksiy Sluchynsky, and Sergiy Biletsky. Forth o i g. “o ial Pe sio s Part II: Their Design and Implementation. In Pension Reform Primer. Washington, DC: World Bank. Rea, John, and Brian Reid. 1998. Trends in Ownership Cost of Equity Mutual Funds. Perspective 4 (3). Investment Company Institute. Robalino, David, Gudivada Venkateswara Rao, and Oleksiy Sluchynsky. . Pre e ti g Poverty among the Elderly in MENA Countries: Role and Optimal Design of Old-Age Subsidies. Unpublished Manuscript. Rocha, Roberto, and Craig Thorburn. 2006. Developing Annuities Markets: The Experience of Chile. World Bank. Slemrod, Joel, and Shlomo Yitzhaki. 2002. Tax Avoidance, Evasion, and Administration. I Handbook of Public Economics, edited by A. J. Auerbach and M. Feldstein. Szilágyi, Imre. 2004. Survey of Pension Fund Costs and Fees. OECD Working Party on Private Pensions. WD(2004)22. Tapia, Waldo, and Juan Yermo. 2008. Fees in Individual Account Pension Systems: A Cross- Country Comparison. OECD Working Papers on Insurance and Private Pensions 27. Turneri, John A., and Hazel A. Witte. 2008. Fee Disclosure to Pension Participants: Establishing Minimum Requirements. Rotman International Centre for Pension Studies. Valdes-Prieto, Salvador. 1994. Administrative Charges in Pensions in Chile, Malaysia, Zambia, and the United States. Policy Research Working Paper 1372, Policy Research Department, World Bank. Whitehouse, Edward. 2000. Administrative Charges for Funded Pensions: An International Comparison and Assessment. Human Development Network. Washington, DC: World Bank. ———. 2001. Administrative Charges for Funded Pensions: Comparison and Assessment of 13 countries. In Private Pension Systems: Administrative Costs and Reforms. Private Pensions Series. Paris: Organisation for Economic Co-operation and Development. World Bank. 2012. Worldwide Governance Indicators. World Bank. Yoo, Keum-Rok. 2002. Evaluating the Operational Efficiency of Korean Public Pension Schemes: A Stochastic Cost Frontier Approach. Korean Social Science Journal 29 (1): 137–162. 57 Annex 1: List of Public Pension Programs and Abbreviations Used Country Year Population GDP pc, US$ Institution / Program Abbreviation Afghanistan 2006 27,518,809 300 MOLSA Pension Department AFG-PD Armenia 2007 3,072,450 3,000 State Service of Social Security ARM-SSSS Australia 2007-2008 21,072,500 40,660 ComSuper+All other expenses AUS-ComSuper Azerbaijan 2007 8,581,300 3,850 State Social Protection Fund AZE-SSPF Bahamas 2007 333,609 21,680 National Insurance Board BHS-NIB Bahrain 2007 759,560 24,320 General Organisation for Social Insurance BHR-GOSI Bahrain 2006 743,522 21,320 Pension Fund Commission BHR-PFC Belize 2005 291,800 3,820 Social Security Board BLZ-SSB Bhutan 2007/2008 676,040 1,770 National Pension and Provident Fund BTN-NPPF Bosnia Herzegovina (Federation 2003 3,783,067 2,210 Federal Fund for Pension and Disability Insurance BIH-(F)FPDI Bosnia Herzegovina (Republika 2003 3,783,067 2,210 Fund for Pension and Disability Insurance of the Republika Srpska BIH-(RS)FPDI Botswana 1999 1,692,814 3,470 Old Age Pension -- Department of Labor and Social Security BWA-OAP Botswana 2006/2007 1,864,831 6,040 Public Officers Pension Fund BWA-POPF Brazil 2008 191,971,506 8,540 National Social Security Institute BRA-NSSI Burkina Faso 2004 13,290,189 380 National Social Security Fund BFA-CNSS Canada 2007 32,976,000 43,180 Old Age Security Program CAN-OAS Canada 2007 32,976,000 43,180 Canada Pension Plan CAN-CPP Canada (Alberta) 2008 3,500,000 82,490 Alberta Public Service Pension Plan CAN-(AB)PSPP Canada 2007 32,976,000 43,180 Public Service Pension Plan CAN-PSPP Canada 2007 32,976,000 43,180 Canadian Forces Pension Plan CAN-CFPP Croatia 2007 4,436,000 13,200 Pension Insurance Institute HRV-PII Cyprus 2007 853,814 25,120 Social Insurance Services CYP-SIS Czech Republic 2007 10,334,160 16,860 Social Security Administration CZE-CSSA Denmark 2007 5,461,438 56,890 ATP (Danish Labor Market Supplementary Pension) DNK-ATP Denmark 2007 5,461,438 56,890 Special Pension Savings (SP) Scheme DNK-SP Dominica 2005 72,000 4,160 Social Security Board DMA-SSB Egypt, Arab Rep. 2004 75,718,360 1,040 Public and Private Entreprises Employees Pension Fund EGY-PPPF Egypt, Arab Rep. 2004 75,718,360 1,040 Government Employee Pension Fund EGY-GEPF Estonia 2007 1,341,672 15,940 Social Insurance Board EST-SIB Fiji 2006 833,330 3,720 National Profident Fund FJI-NPF Finland 2008 5,313,399 50,780 Social Insurance Institution of Finland (KELA) FIN-SIIF France (New Caledonia) 2007 242,400 36,390 Social Welfare Fund of New Caledonia (CAFAT) FRA-(NCL)CAFAT Georgia 2008 4,307,011 2,970 Social Service Agency GEO-SSA Ghana 2006 22,393,338 570 Social Security and National Insurance Trust GHA-SSNIT Grenada 2006 102,823 5,490 National Insurance Board GRD-NIB Guyana 2002 758,834 950 National Insurance Scheme GUY-NIS Hungary 2007 10,055,780 13,800 Central Administration of National Pension Insurance HUN-CANPI India 2006-2007 1,109,811,147 860 Employees' Provident Fund Organization IND-EPFO Indonesia 2007 224,669,595 1,920 Employees’ Social Security (JAMSOSTEK) IDN-ESS Ireland 2007 4,356,931 59,610 Department of Social and Family Affairs IRL-DSFA Jordan 2006 5,542,000 2,680 Social Security Corporation JOR-SSC Kenya 2006/2007 36,771,613 610 National Social Security Fund KEN-NSSF Kenya 2006 36,771,613 610 Civil Service Pension Scheme KEN-CSPS Kenya 2007 37,754,701 720 Local Authorities Pension Trust KEN-LAPT Korea, Rep. 2006 48,297,000 19,710 National Pension Service KOR-NPS Kosovo 2007 1,785,000 2,620 Kosovo Pension Administration KOS-KPA Kosovo 2007 1,785,000 2,620 Kosovo Pension Savings Trust KOS-KPST Kyrgyz Republic 2006 5,192,100 550 Kyrgyz Republic Social Fund KGZ-KRSF Latvia 2007 2,276,100 12,640 State Social Insurance Fund (VSAA) LVA-SSIF Lithuania 2007 3,375,618 11,580 State Social Insurance Fund Board (SODRA) LTU-SSIFB Macedonia, FYR 2007 2,039,838 3,880 Pendion and Disability Insurance Fund MFD-PDIF Malaysia 2007 26,555,654 7,000 Employees Provident Fund (KWSP) MYS-EPF Malaysia 2006 26,094,742 6,000 Social Security Organisation (PERKESO) MYS-SSO Maldives 2010 315,900 4,690 Maldives Pension Administration Office MDV-MPAO Mali 2007 12,408,824 580 National Social Insurance Institute (INPS) MLI-NSII Marshal islands 2005 55,792 2,480 Marshall Islands Social Security Administration MHL-SSA Mauritius 2005 1,243,253 5,050 Mauritius National Pensions Fund MUS-MNPS Micronesia, Fed. Sts. 2007 110,123 2,300 Federated States of Micronesia Social Security Administration FSM-SSA Moldova 2008 3,633,369 1,670 National Office of Social Insurance (CANS) MDA-NOSI 58 List of Public Pension Programs and Abbreviations Used (Continued) Country Year Population GDP pc, US$ Institution / Program Abbreviation Morocco 2007 31,224,136 2,410 National Social Security Fund (CNSS) MAR-NSSF Morocco 2007 31,224,136 2,410 Moroccan Pension Fund (CMR) MAR-MPF Namibia 2001 1,861,828 1,910 Social Pension NAM-SP Namibia 2007 2,088,671 4,230 Government Institutions Pension Fund NAM-GIPF Netherlands 2007 16,381,696 47,510 Social Insurance Bank NLD-SIB New Zealand 2007 4,228,300 31,850 Superannuation NZL-Super New Zealand (Cook Islands) 2009 21,300 14,080 National Superannuation Fund NZL-(CI) NSF Norway 2008 4,768,212 94,570 Labour and Welfare Administration (NAV) NOR-NAV Pakistan 2006-2007 159,144,934 800 Employees' Old-Age Benefits Institution PAK-EOBI Philippines 2007 88,718,185 1,620 Social Security System PHL-SSS Philippines 2007 88,718,185 1,620 Government Service Insurance System PHL-GSIS Poland 2007 38,120,560 11,160 Social Insurance Institute POL-ZUS Poland 2007 38,120,560 11,160 Agricultural Social Security Fund POL-KRUS Portugal 2007 10,608,335 21,040 Social Security Institute PRT-SSI Romania 2007 21,546,873 7,860 National Pension and Social Insurance Fund ROM-NPSIF Rwanda 2007 9,454,534 360 Rwanda Social Security Board RWA-RSSB Samoa 2006 179,004 2,470 Samoa National Provident Fund WSM-SNPF Senegal 2001 10,164,729 480 Social Insurance Institute for Old-Age Pensions (IPRES) SEN-SII Sierra Leone 2006 5,270,799 270 National Social Security and Insurance Trust SLE-NASSIT Singapore 2007 4,588,600 38,520 Central Provident Fund SGP-CPF Slovak Republic 2006 5,391,409 12,810 Social Insurance Agency SVK-SIA Solomon Islands 2008 510,672 1,290 National Provident Fund (NPF) SLB-NPF South Africa 2007/2008 48,257,282 5,930 South African Social Security Agency ZAF-SSA South Africa 2007/2008 48,257,282 5,930 Government Employees Pension Fund ZAF-GEPF Spain 2007 44,878,945 32,100 National Institute of Social Security ESP-NISS Sri Lanka 2005 19,668,000 1,240 Employees’ Provident Fund LKA-EPF Sri Lanka 2001 18,797,000 840 Farmers Pension Scheme LKA-FPS St. Kitts and Nevis 2007 48,790 10,520 Social Security Board KNA-SSB St. Lucia 2000 155,996 4,540 National Insurance Corporation LCA-NIC St. Vincent and the Grenadines 2004 108,531 3,880 National Insurance Services VCT-NIS Swaziland 2007 1,151,399 2,560 Swaziland National Provident Fund SWZ-SNPF Swaziland 2008 1,167,834 2,430 PSPF - Public Service Pension Fund SWZ-PSPF Sweden 2007 9,148,092 50,560 Swedish Social Insurance Agency (SSIA+NPFs):1st Pillar OA Pensions SWE-SSIA/NPFs (OA Pen) Sweden 2007 9,148,092 50,560 Premium Pension Authority (incl PP+Funds): 2d Pillar SWE-PPA Syrian Arab Republic 2007 20,082,697 2,020 General Organization for Social Insurance SYR-GOSI Tanzania 2007/2008 42,267,667 490 Government Employee Pension Fund TZA-GEPF Tanzania 2007 41,276,209 410 Parastatal Pension Fund TZA-PPF Tanzania (Zanzibar) 2004/2005 1,040,659 300 Zanzibar Social Security Fund TZA-(ZZB)ZSSF Thailand 2007 66,979,359 3,690 Government Pension Fund THA-GPF Thailand 2005 65,945,675 2,670 Social Security Office THA-SSO Tonga 2009 103,967 2,990 Retirement Fund Board TON-RFB Trinidad and Tobago 2007 1,328,216 15,740 National Insurance Board TTO-NIB Turkey 2007 73,003,736 8,860 Social Security Institution TUR-SSI Uganda 2009 32,368,000 490 National Social Security Fund UGA-NSSF Ukraine 2007 46,509,350 3,070 Pension Fund of Ukraine UKR-PFU United Kingdom (Great Britain) 2007/08 60,980,304 45,900 UK Pension Service UK-(GBR)PS United Kingdom (Anguilla) 2006 13,600 11,900 Social Security Board UK-(ANG)SSB United Kingdom (Falkland Islan 2007 3,000 69,330 Falkland Islands Pension Scheme UK-(FIS)FIPS United Kingdom (Jersey) 2006 90,000 58,890 Social Security Department UK-(JER)SSD United Kingdom (Northern Irela 2007/2008 1,720,000 43,390 Northern Ireland Social Security Agency UK-(NI)NISSA United States 2007 301,580,000 46,460 SSA Supplementary Security Income USA-SSA/SSI United States 2007 301,580,000 46,460 SSA Old Age Survivor Disability Insurance USA-SSA/OASDI United States (Alaska) 2007 676,987 65,730 Alaska Permanent Fund Dividend (PFD) Division USA-(AK)APFDD United States 2008 304,375,000 47,210 Thrift Savings Plan USA-TSP Uruguay 2007 3,323,906 7,210 Social Insurance Bank (BPS) URY-SIB Vanuatu 2006 222,200 2,020 Vanuatu National Provident Fund VUT-VNPF Vietnam 2008 86,210,781 1,050 Vietnam Social Security VNM-VSS Yemen 2006 21,637,666 880 General Corporation for Social Security YEM-GCSS Yemen 2006 21,637,666 880 General Agency for Pensions and Social Security YEM-GAPSS Zambia 2007 12,313,942 940 Workers Compensation Fund ZMB-WCF 59 Annex 2: Key Institutional and Operational Indicators Contribution Revenues, Total Assets, 000 US$ Admin Exp / ConRev Admin Exp / BenExp Admin Exp, 000 US$ Benefit Expenditure, Admin Exp / InsBen, Admin Exp / InsBen, Admin Exp / InsBen, No InsBen / No Staff No of Beneficiaries PrivSectorScheme InHouseCollection FundManagement Admin Exp / GDP InHouseRecords SHUL Schemes No of Insured No of Offices over GDPpc US eqv US$ DCScheme No of Staff 000 US$ 000 US$ Agency US$ AFG-PD No No No No No No 1 125 320,000 54,370 435 22,213 .. 166 0.002% $3.06 \ 1.02% 0.75% .. .. ARM-SSSS No Yes Yes No No Yes 70 904 586,486 536,183 1,242 249,681 .. 8,013 0.09% $7.14 $111 0.24% 3.21% .. .. AUS-ComSuper No No Yes Yes Yes No n.a. 610 271,929 193,867 763 4,873,345 1,399,915 212,202 0.02% $455.57 $521 1.12% 4.35% 15.16% $17,572,156 AZE-SSPF No Yes Yes No No Yes 81 2,616 1,700,200 1,312,900 1,152 1,113,197 .. 37,131 0.11% $12.32 $149 0.32% 3.34% .. .. BHS-NIB No Yes Yes Yes Yes Yes 27 488 146,752 31,894 366 139,499 155,500 29,830 0.41% $166.98 $358 0.77% 21.38% 19.18% $1,492,000 BHR-GOSI No Yes Yes Yes Yes Yes 4 385 374,466 14,580 1,011 130,104 289,754 17,330 0.09% $44.54 $85 0.18% 13.32% 5.98% $3,930,751 BHR-PFC No No Yes Yes Yes Yes n.a. n.a. 54,000 10,020 n.a. 252,356 278,167 8,596 0.05% $134.27 $293 0.63% 3.41% 3.09% $4,000,654 BLZ-SSB No Yes Yes Yes Yes Yes 13 250 71,719 14,148 343 15,646 25,164 6,850 0.61% $79.78 $970 2.09% 43.78% 27.22% $150,000 BTN-NPPF Yes No Yes Yes Yes Yes n.a. n.a. 38,210 2,855 n.a. 3,657 17,461 868 0.07% $21.14 $555 1.19% 23.74% 4.97% $157,490 BIH-(F)FPDI No Yes Yes Yes No No n.a. n.a. 209,699 288,613 n.a. 443,410 445,690 21,120 0.25% $42.38 $891 1.92% 4.76% 4.74% .. BIH-(RS)FPDI No Yes Yes Yes No No n.a. n.a. 132,320 173,692 n.a. 171,732 147,017 8,310 0.10% $27.15 $571 1.23% 4.84% 5.65% .. BWA-OAP No Yes No No No No n.a. n.a. 774,417 77,200 n.a. 22,706 .. 1,022 0.02% $13.24 $177 0.38% 4.50% .. .. BWA-POPF Yes No Yes Yes Yes No n.a. 35 83,329 3,719 2,487 192,714 148,363 32,125 0.29% $369.05 $2,839 6.11% 16.67% 21.65% $4,112,767 BRA-NSSI No Yes Yes Yes Yes Yes n.a. 39,559 53,741,233 24,950,929 1,989 118,335,666 n.a. 5,086,237 0.31% $64.63 $352 0.76% 4.30% n.a. .. BFA-CNSS No Yes Yes Yes No No n.a. 800 73,362 297,061 463 15,435 43,661 27,798 0.54% $75.04 $9,175 19.75% 180.09% 63.67% .. CAN-OAS No Yes No No No No 587 n.a. 18,356,909 4,447,602 n.a. 31,228,029 .. 106,135 0.01% $23.86 $26 0.06% 0.34% .. .. CAN-CPP No Yes Yes No Yes No 587 n.a. 12,280,000 4,758,774 n.a. 24,509,841 .. 407,784 0.03% $23.93 $26 0.06% 1.66% .. $108,556,085 CAN-(AB)PSPP No No Yes No Yes No n.a. 57 48,075 19,290 1,189 202,898 .. 25,134 0.01% $373.10 $210 0.45% 12.39% .. $4,470,310 CAN-PSPP No No Yes No Yes No 3 700 294,979 231,913 753 4,266,831 .. 176,892 0.01% $335.73 $361 0.78% 4.15% .. $26,314,144 CAN-CFPP No No Yes No Yes No n.a. 139 87,532 108,798 1,412 2,105,020 .. 40,778 0.00% $207.70 $223 0.48% 1.94% .. $7,330,794 HRV-PII No Yes Yes No No No 112 3,399 1,579,463 1,577,301 929 5,917,113 .. 112,370 0.19% $35.60 $125 0.27% 1.90% .. .. CYP-SIS No Yes Yes Yes No Yes n.a. n.a. 421,352 132,265 n.a. 650,626 619,328 6,097 0.03% $11.01 $20 0.04% 0.94% 0.98% $4,605,660 CZE-CSSA No Yes Yes Yes No Yes 92 8,578 4,880,187 3,347,121 959 15,647,591 17,553,596 265,670 0.15% $32.29 $89 0.19% 1.70% 1.51% .. DNK-ATP No Yes Yes No Yes No 1 226 3,116,000 716,000 16,956 1,408,049 .. 132,998 0.04% $34.71 $28 0.06% 9.45% .. $71,458,740 DNK-SP Yes Yes Yes No Yes No 1 110 2,547,500 158,450 24,600 307,879 .. 38,209 0.01% $14.12 $12 0.02% 12.41% .. $10,121,791 DMA-SSB No Yes Yes Yes Yes Yes 2 45 17,169 4,489 481 10,344 10,242 1,567 0.52% $72.37 $808 1.74% 15.15% 15.30% $100,664 EGY-PPPF No Yes Yes Yes No Yes 453 23,000 13,910,000 5,184,350 830 1,613,882 952,190 41,477 0.05% $2.17 $97 0.21% 2.57% 4.36% $15,213,743 EGY-GEPF No No Yes Yes No Yes 62 8,000 4,792,000 1,999,000 849 1,258,828 1,517,049 21,465 0.03% $3.16 $141 0.30% 1.71% 1.41% $22,043,040 EST-SIB No Yes Yes No No No 4 609 650,000 675,770 2,177 1,623,212 .. 11,974 0.06% $9.03 $26 0.06% 0.74% .. .. FJI-NPF Yes Yes Yes Yes Yes No 6 242 331,050 85,577 1,722 178,850 154,605 11,259 0.36% $27.03 $338 0.73% 6.30% 7.28% $1,848,371 FIN-SIIF No Yes Yes No Yes Yes 287 5,864 2,670,000 2,730,600 921 16,282,279 .. 441,206 0.16% $81.70 $75 0.16% 2.71% .. $1,325,668 FRA-(NCL)CAFAT No Yes Yes Yes Yes Yes n.a. 413 97,563 122,441 533 865,633 736,338 36,466 0.41% $165.75 $212 0.46% 4.21% 4.95% $266,571 GEO-SSA No Yes No No No No 77 2,206 2,274,709 1,362,227 618 555,743 .. 18,955 0.15% $13.91 $218 0.47% 3.41% .. .. GHA-SSNIT No Yes Yes Yes Yes Yes 73 n.a. 854,761 73,311 n.a. 87,152 312,937 54,167 0.43% $58.36 $4,757 10.24% 62.15% 17.31% $1,289,747 GRD-NIB No Yes Yes Yes Yes Yes 3 75 36,715 3,889 541 9,111 16,519 2,289 0.41% $56.37 $477 1.03% 25.12% 13.86% $193,704 GUY-NIS No Yes Yes Yes Yes Yes n.a. 545 130,533 43,352 319 23,472 29,079 4,143 0.57% $23.82 $1,165 2.51% 17.65% 14.25% $111,940 HUN-CANPI No Yes Yes No No No 9 3,872 4,356,500 3,680,304 2,076 15,128,590 .. 158,309 0.11% $19.70 $66 0.14% 1.05% .. .. IND-EPFO Yes Yes Yes Yes Yes No 242 19,510 44,404,000 5,100,230 2,537 2,671,993 6,127,836 211,634 0.02% $4.28 $231 0.50% 7.92% 3.45% $56,547,543 IDN-ESS Yes Yes Yes Yes Yes Yes 128 2,997 7,941,017 802,504 2,917 442,886 949,153 182,171 0.04% $20.83 $504 1.09% 41.13% 19.19% $6,715,130 IRL-DSFA No Yes Yes No No Yes 132 4,840 3,002,276 1,500,504 930 20,469,579 .. 770,649 0.30% $171.15 $133 0.29% 3.76% .. .. JOR-SSC No Yes Yes Yes Yes No 19 1,276 689,176 213,548 707 313,962 509,931 27,373 0.18% $30.32 $526 1.13% 8.72% 5.37% $5,218,618 KEN-NSSF Yes Yes Yes Yes Yes No 39 1,800 900,000 50,000 528 33,566 75,529 37,528 0.17% $39.50 $3,009 6.48% 111.80% 49.69% $1,109,557 KEN-CSPS No No No No No No n.a. 198 393,000 158,700 802 238,541 .. 2,692 0.01% $16.96 $1,292 2.78% 1.13% .. .. KEN-LAPT No No Yes Yes Yes No n.a. n.a. 22,862 4,720 n.a. 10,562 20,828 2,460 0.01% $89.17 $5,754 12.38% 23.29% 11.81% $93,179 KOR-NPS No Yes Yes Yes Yes No 97 4,833 17,740,000 1,830,600 4,049 4,566,447 21,106,200 484,538 0.05% $24.76 $58 0.13% 10.61% 2.30% $190,827,199 KOS-KPA No Yes No No No No 33 158 550,000 151,077 956 100,659 .. 2,806 0.06% $18.57 $329 0.71% 2.79% .. .. KOS-KPST Yes Yes Yes No Yes No 1 24 238,000 1,592 9,983 2,506 .. 2,748 0.06% $11.47 $203 0.44% 109.67% .. $380,822 KGZ-KRSF No Yes Yes Yes No No 54 1,900 1,029,300 524,000 818 141,766 192,210 6,376 0.22% $4.10 $347 0.75% 4.50% 3.32% $62,262 LVA-SSIF No Yes Yes No No Yes 43 n.a. 1,202,400 1,066,609 n.a. 3,628,299 .. 31,044 0.11% $13.68 $50 0.11% 0.86% .. .. LTU-SSIFB No Yes Yes Yes No Yes 50 3,970 1,467,000 1,319,807 702 3,093,047 3,712,766 89,471 0.23% $32.11 $129 0.28% 2.89% 2.41% .. MFD-PDIF No Yes Yes Yes No No 31 733 419,347 280,249 954 590,020 489,924 10,993 0.14% $15.71 $188 0.40% 1.86% 2.24% .. MYS-EPF Yes Yes Yes Yes Yes Yes 62 5,176 5,400,000 1,075,742 1,251 6,172,165 8,414,667 153,686 0.08% $23.73 $158 0.34% 2.49% 1.83% $92,591,586 MYS-SSO No Yes Yes Yes Yes Yes 45 1,142 5,454,799 259,081 5,003 264,430 432,408 35,943 0.02% $6.29 $49 0.10% 13.59% 8.31% $3,856,127 MDV-MPAO Yes Yes Yes Yes Yes No 1 27 35,000 14,500 1,833 24,438 14,250 569 0.04% $11.49 $114 0.25% 2.33% 3.99% $15,625 MLI-NSII No Yes Yes Yes Yes Yes n.a. 1,717 193,185 43,809 138 53,878 109,193 16,172 0.23% $68.24 $5,466 11.77% 30.02% 14.81% $63,772 MHL-SSA No Yes Yes Yes Yes No 2 26 10,486 3,240 528 10,674 17,621 928 0.67% $67.63 $1,267 2.73% 8.70% 5.27% $56,623 MUS-MNPS No Yes Yes Yes Yes No n.a. n.a. 300,000 253,600 n.a. 190,295 46,345 3,763 0.06% $6.80 $63 0.13% 1.98% 8.12% $1,293,589 FSM-SSA No Yes Yes Yes Yes No 5 34 21,590 6,363 822 13,664 12,784 1,232 0.49% $44.06 $890 1.92% 9.01% 9.63% $47,322 MDA-NOSI No Yes Yes Yes No Yes n.a. 1,316 900,000 1,029,500 1,466 589,240 507,686 14,485 0.24% $7.51 $209 0.45% 2.46% 2.85% .. 60 Key Institutional and Operational Indicators (Continued) Benefit Expenditure, Admin Exp / InsBen, Admin Exp / InsBen, Admin Exp / InsBen, No InsBen / No Staff Revenues, 000 US$ No of Beneficiaries PrivSectorScheme InHouseCollection FundManagement Admin Exp / GDP Total Assets, 000 InHouseRecords Admin Exp, 000 SHUL Schemes No of Insured No of Offices Admin Exp / Admin Exp / Contribution over GDPpc US eqv US$ DCScheme No of Staff 000 US$ ConRev BenExp Agency US$ US$ US$ MAR-NSSF No Yes Yes Yes Yes Yes 71 4,750 1,970,000 818,745 587 880,091 1,352,728 79,999 0.11% $28.69 $553 1.19% 9.09% 5.91% $2,481,588 MAR-MPF No No Yes No Yes No 12 419 855,837 528,368 3,304 1,705,011 .. 21,007 0.03% $15.18 $293 0.63% 1.23% .. .. NAM-SP No Yes No No No No n.a. n.a. 620,338 107,822 n.a. 33,931 .. 3,049 0.09% $28.28 $688 1.48% 8.99% .. .. NAM-GIPF No No Yes Yes Yes No n.a. 143 72,370 38,439 775 130,950 152,214 31,949 0.36% $288.32 $3,167 6.82% 24.40% 20.99% $4,984,554 NLD-SIB No Yes No No No No 13 3,247 8,000,000 4,720,092 1,454 40,786,300 .. 281,653 0.04% $59.67 $58 0.13% 0.69% .. .. NZL-Super No Yes No No No No 20 560 2,280,000 891,575 1,592 6,514,560 .. 29,322 0.02% $32.89 $48 0.10% 0.45% .. .. NZL-(CI) NSF Yes Yes Yes Yes Yes No 1 n.a. 5,631 254 n.a. 406 4,488 577 0.19% $98.10 $324 0.70% 142.30% 12.87% $19,685 NOR-NAV No Yes Yes No No Yes 293 n.a. 2,500,000 2,121,423 n.a. 44,166,667 .. 1,648,936 0.37% $356.80 $175 0.38% 3.73% .. .. PAK-EOBI No Yes Yes Yes Yes No 63 938 1,820,000 302,677 2,263 57,170 80,482 9,902 0.01% $4.66 $271 0.58% 17.32% 12.30% $1,824,236 PHL-SSS No Yes Yes Yes Yes Yes 178 6,074 7,863,340 1,729,399 1,579 1,315,229 1,339,788 128,391 0.09% $13.38 $384 0.83% 9.76% 9.58% $5,368,274 PHL-GSIS No No Yes Yes Yes Yes 58 3,100 1,355,558 233,778 513 699,981 884,213 137,946 0.10% $86.79 $2,489 5.36% 19.71% 15.60% $8,886,117 POL-ZUS No Yes Yes Yes No Yes 327 47,588 14,074,500 8,396,800 472 42,610,298 49,017,251 1,133,460 0.27% $50.44 $210 0.45% 2.66% 2.31% $1,253,816 POL-KRUS No Yes Yes Yes No Yes 273 6,390 1,610,000 1,586,800 500 4,949,616 429,740 173,875 0.04% $54.39 $226 0.49% 3.51% 40.46% .. PRT-SSI No Yes Yes Yes No Yes n.a. n.a. 4,480,804 4,572,059 n.a. 22,365,400 #VALUE! 766,454 0.34% $84.66 $187 0.40% 3.43% n.a. .. ROM-NPSIF No Yes Yes No No No 43 4,737 5,479,432 4,677,128 2,144 9,383,599 .. 161,583 0.10% $15.91 $94 0.20% 1.72% .. .. RWA-RSSB No Yes Yes Yes Yes No 30 227 216,304 29,121 1,081 6,726 28,951 7,932 0.23% $32.32 $4,171 8.98% 117.93% 27.40% $206,608 WSM-SNPF Yes Yes Yes Yes Yes Yes n.a. 165 22,526 4,224 162 8,371 11,388 1,666 0.38% $62.28 $1,171 2.52% 19.90% 14.63% $113,450 SEN-SII No Yes Yes Yes Yes No 9 227 170,929 92,175 1,159 28,262 38,089 5,457 0.11% $20.74 $2,007 4.32% 19.31% 14.33% $68,209 SLE-NASSIT No Yes Yes Yes Yes No 6 227 126,749 19,559 645 4,488 16,759 4,874 0.34% $33.32 $5,733 12.34% 108.60% 29.08% $51,176 SGP-CPF Yes Yes Yes Yes Yes Yes n.a. 1,200 1,545,000 962,317 2,089 7,671,600 11,735,056 77,670 0.04% $30.98 $37 0.08% 1.01% 0.66% $93,708,995 SVK-SIA No Yes Yes Yes No Yes 39 5,839 2,000,000 1,419,470 586 4,383,401 3,939,048 102,449 0.15% $29.96 $109 0.23% 2.34% 2.60% .. SLB-NPF Yes Yes Yes Yes Yes Yes n.a. 104 135,960 3,708 1,343 5,369 13,410 3,473 0.53% $24.87 $896 1.93% 64.70% 25.90% $109,409 ZAF-SSA No Yes No No No No n.a. 7,528 18,036,174 12,386,396 1,645 8,817,002 .. 630,303 0.22% $50.89 $399 0.86% 7.15% .. .. ZAF-GEPF No No Yes Yes Yes No 4 705 1,160,000 403,280 2,217 3,406,495 3,678,878 146,379 0.05% $93.64 $734 1.58% 4.30% 3.98% $102,904,534 ESP-NISS No Yes Yes Yes No Yes n.a. 13,000 19,151,400 9,660,574 2,216 118,060,711 124,256,584 782,302 0.05% $27.15 $39 0.08% 0.66% 0.63% .. LKA-EPF Yes Yes Yes Yes Yes No n.a. 600 2,100,000 93,841 3,656 169,396 271,795 3,545 0.01% $1.62 $61 0.13% 2.09% 1.30% $4,027,362 LKA-FPS No Yes Yes Yes Yes No n.a. 400 388,800 10,366 998 1,300 n.a. 688 0.00% $1.72 $95 0.21% 52.89% n.a. $26,515 KNA-SSB No Yes Yes Yes Yes Yes 2 104 30,953 4,525 341 10,744 22,222 3,538 0.69% $99.73 $440 0.95% 32.93% 15.92% $337,037 LCA-NIC No Yes Yes Yes Yes Yes n.a. 95 41,004 5,556 491 7,360 17,849 2,259 0.32% $48.52 $497 1.07% 30.70% 12.66% $187,922 VCT-NIS No Yes Yes Yes Yes Yes n.a. 53 33,782 4,373 720 5,533 7,987 1,461 0.35% $38.30 $459 0.99% 26.41% 18.30% $100,000 SWZ-SNPF Yes Yes Yes Yes Yes No n.a. 121 125,000 10,970 1,124 10,007 13,243 4,583 0.16% $33.71 $612 1.32% 45.80% 34.61% $161,405 SWZ-PSPF No No Yes Yes Yes No n.a. n.a. 33,000 21,000 n.a. 36,185 51,134 6,232 0.22% $115.40 $2,206 4.75% 17.22% 12.19% $969,075 SWE-SSIA/NPFs (O No Yes Yes No Yes No n.a. 1,300 7,557,655 2,120,000 7,444 27,468,459 .. 398,321 0.09% $41.16 $38 0.08% 1.45% .. $129,975,129 SWE-PPA Yes Yes Yes No Yes No 2 210 5,838,802 449,000 29,942 67,468 .. 277,717 0.06% $44.17 $41 0.09% 411.63% .. $42,019,480 SYR-GOSI No Yes Yes Yes Yes No 23 1,986 1,730,448 180,142 962 274,413 n.a. 8,519 0.02% $4.46 $103 0.22% 3.10% n.a. $151,396 TZA-GEPF No No Yes Yes Yes Yes n.a. n.a. 28,200 743 n.a. 1,136 8,690 1,031 0.00% $35.63 $3,379 7.27% 90.78% 11.87% $46,736 TZA-PPF No Yes Yes Yes Yes Yes 6 170 84,186 16,813 594 19,954 67,127 9,022 0.05% $89.33 $10,123 21.79% 45.22% 13.44% $325,292 TZA-(ZZB)ZSSF No Yes Yes Yes Yes Yes n.a. 42 40,000 816 972 493 3,968 363 0.12% $8.90 $1,379 2.97% 73.78% 9.16% $18,403 THA-GPF Yes No Yes Yes Yes Yes 1 277 1,177,586 20,158 4,324 220,419 657,545 20,087 0.01% $16.77 $211 0.45% 9.11% 3.05% $10,926,651 THA-SSO No Yes Yes Yes Yes Yes n.a. 5,931 8,467,410 1,388,210 1,662 598,332 2,178,012 48,549 0.03% $4.93 $86 0.18% 8.11% 2.23% $9,659,074 TON-RFB Yes No Yes Yes Yes No n.a. 20 3,726 189 196 1,811 3,436 306 0.10% $78.27 $1,216 2.62% 16.92% 8.92% $35,009 TTO-NIB No Yes Yes Yes Yes Yes 15 614 501,450 120,615 1,013 163,811 237,357 15,393 0.07% $24.74 $73 0.16% 9.40% 6.49% $2,259,786 TUR-SSI No Yes Yes Yes No Yes n.a. 24,779 15,059,964 9,566,647 994 55,481,612 33,809,680 752,150 0.12% $30.54 $160 0.34% 1.36% 2.22% .. UGA-NSSF Yes Yes Yes Yes Yes No 24 600 230,000 10,000 400 20,225 129,556 22,314 0.14% $92.97 $8,816 18.97% 110.33% 17.22% $640,394 UKR-PFU No Yes Yes Yes No No 760 36,000 15,350,000 13,321,042 796 19,493,069 15,002,234 262,694 0.18% $9.16 $139 0.30% 1.35% 1.75% .. UK-(GBR)PS No Yes No No No No n.a. 11,890 32,105,000 12,000,000 1,009 146,066,704 .. 1,254,955 0.04% $104.58 $106 0.23% 0.86% .. .. UK-(ANG)SSB No Yes Yes Yes Yes Yes n.a. 25 7,526 771 332 1,988 7,513 1,644 1.02% $198.17 $774 1.67% 82.69% 21.89% $59,444 UK-(FIS)FIPS Yes Yes Yes Yes Yes No 1 n.a. 600 17 n.a. 1,727 3,380 214 0.10% $346.47 $232 0.50% 12.38% 6.32% $47,085 UK-(JER)SSD No Yes Yes Yes Yes Yes n.a. n.a. 87,458 28,165 n.a. 272,730 271,514 10,142 0.19% $87.72 $69 0.15% 3.72% 3.74% $1,072,703 UK-(NI)NISSA No Yes Yes No No Yes 35 5,336 880,000 986,027 350 7,309,112 .. 377,720 0.51% $202.42 $217 0.47% 5.17% .. .. USA-SSA/SSI No Yes No No No No 1526 23,000 156,677,557 7,359,525 320 39,927,000 .. 2,700,000 0.02% $366.87 $367 0.79% 6.76% .. .. USA-SSA/OASDI No Yes Yes No Yes No 1526 44,000 163,177,000 52,464,978 4,901 584,967,000 .. 5,800,000 0.04% $26.90 $27 0.06% 0.99% .. $2,239,438,000 USA-(AK)APFDD No Yes No No No No n.a. n.a. 352,000 595,000 n.a. 984,522 .. 7,066 0.02% $11.87 $8 0.02% 0.72% .. .. USA-TSP Yes No Yes No Yes Yes 1 380 2,600,000 353,000 7,771 11,660,000 .. 104,300 0.001% $35.32 $35 0.07% 0.89% .. $260,000,000 URY-SIB No Yes Yes Yes Yes Yes n.a. 4,338 1,004,886 1,276,927 526 2,421,305 1,578,122 157,960 0.66% $69.23 $446 0.96% 6.52% 10.01% $623,698 VUT-VNPF Yes Yes Yes Yes Yes No 2 39 27,922 710 734 1,802 8,438 1,417 0.32% $49.48 $1,138 2.45% 78.61% 16.79% $68,691 VNM-VSS No Yes Yes Yes Yes Yes 714 16,500 8,800,000 3,000,000 715 n.a. n.a. 57,300 0.06% $4.86 $215 0.46% n.a. n.a. $4,232,545 YEM-GCSS No Yes Yes Yes Yes No 8 395 98,900 4,163 261 3,664 27,217 2,573 0.01% $24.96 $1,318 2.84% 70.22% 9.45% $183,908 YEM-GAPSS No No Yes No Yes Yes 22 917 473,500 71,022 594 76,636 .. 6,014 0.03% $11.04 $583 1.26% 7.85% .. $1,119,903 ZMB-WCF No Yes Yes Yes Yes No 21 277 168,000 39,367 749 6,903 24,970 4,743 0.04% $22.87 $1,130 2.43% 68.71% 18.99% $32,000 61 Annex 3: Benchmarking Performance of Public Pension Programs Country-Agency Staffing Total Admin Expenditures, US$ Benchmark Coefficient Performance category Actual Projected Actual Projected Staffing Expenditures Staffing Expenditures AFG-PD 125 202 166,000 747,000 0.62 0.22 B B ARM-SSSS 904 1,358 8,013,000 19,000,000 0.67 0.42 B B AUS-ComSuper 610 505 82,824,000 50,100,000 1.21 1.65 A B AZE-SSPF 2,616 2,783 37,131,000 44,800,000 0.94 0.83 A A BHS-NIB 488 263 29,830,000 16,900,000 1.85 1.77 B B BHR-GOSI 385 160 12,011,000 11,100,000 2.40 1.08 C A BHR-PFC n.a. 132 8,596,000 8,438,000 n.a. 1.02 n.a. A BLZ-SSB 250 202 6,850,000 4,421,000 1.24 1.55 A B BTN-NPPF n.a. 94 868,000 1,303,000 n.a. 0.67 n.a. B BIH-(F)FPDI n.a. 1,005 21,120,000 9,871,000 n.a. 2.14 n.a. C BIH-(RS)FPDI n.a. 684 8,310,000 6,769,000 n.a. 1.23 n.a. A BWA-OAP n.a. 185 1,022,000 3,131,000 n.a. 0.33 n.a. B BWA-POPF 35 28 5,808,000 1,136,000 1.25 5.11 A C BRA-NSSI 39,559 81,442 5,086,000,000 2,630,000,000 0.49 1.93 B B BFA-CNSS 800 1,298 27,798,000 4,262,000 0.62 6.52 B C CAN-OAS n.a. 3,627 106,100,000 277,000,000 n.a. 0.38 n.a. B CAN-CPP n.a. 3,866 301,600,000 483,000,000 n.a. 0.62 n.a. B CAN-(AB)PSPP 57 49 7,084,000 10,000,000 1.15 0.71 A B CAN-PSPP 700 294 82,860,000 38,600,000 2.38 2.15 C C CAN-CFPP 139 169 15,548,000 22,400,000 0.82 0.69 A B HRV-PII 3,399 1,654 112,400,000 61,400,000 2.05 1.83 C B CYP-SIS n.a. 681 6,097,000 28,600,000 n.a. 0.21 n.a. B CZE-CSSA 8,578 10,285 265,700,000 320,000,000 0.83 0.83 A A DNK-ATP 226 695 22,228,000 106,000,000 0.33 0.21 B B DNK-SP 110 214 15,063,000 33,400,000 0.52 0.45 B B DMA-SSB 45 106 1,567,000 2,475,000 0.43 0.63 B B EGY-PPPF 23,000 22,421 41,477,000 122,000,000 1.03 0.34 A B EGY-GEPF 8,000 9,321 21,465,000 51,800,000 0.86 0.41 A B EST-SIB 609 784 11,974,000 33,200,000 0.78 0.36 A B FJI-NPF 242 384 8,084,000 8,696,000 0.63 0.93 B A FIN-SIIF 5,864 3,788 441,200,000 493,000,000 1.55 0.89 B A FRA-(NCL)CAFAT 413 613 36,466,000 53,200,000 0.67 0.69 B B GEO-SSA 2,206 1,774 18,955,000 26,100,000 1.24 0.73 A B GHA-SSNIT n.a. 741 54,167,000 4,879,000 n.a. 11.10 n.a. C GRD-NIB 75 95 2,289,000 2,637,000 0.79 0.87 A A GUY-NIS 545 486 4,143,000 4,430,000 1.12 0.94 A A HUN-CANPI 3,872 3,535 158,300,000 133,000,000 1.10 1.19 A A IND-EPFO 19,510 13,506 200,800,000 116,000,000 1.44 1.73 B B IDN-ESS 2,997 3,907 91,238,000 52,900,000 0.77 1.72 A B IRL-DSFA 4,840 2,173 770,600,000 192,000,000 2.23 4.01 C C JOR-SSC 1,276 778 27,373,000 14,200,000 1.64 1.93 B B KEN-NSSF 1,800 338 37,528,000 2,505,000 5.32 14.98 C C KEN-CSPS 198 390 2,692,000 2,215,000 0.51 1.22 B A KEN-LAPT n.a. 82 2,366,000 689,000 n.a. 3.43 n.a. C KOR-NPS 4,833 3,481 415,000,000 213,000,000 1.39 1.95 B B KOS-KPA 158 310 2,806,000 4,363,000 0.51 0.64 B B KOS-KPST 24 21 1,248,000 506,000 1.15 2.47 A C KGZ-KRSF 1,900 1,937 6,376,000 7,938,000 0.98 0.80 A A LVA-SSIF n.a. 1,988 31,044,000 67,300,000 n.a. 0.46 n.a. B LTU-SSIFB 3,970 4,701 89,471,000 118,000,000 0.84 0.76 A A MFD-PDIF 733 912 10,993,000 12,700,000 0.80 0.87 A A MYS-EPF 5,176 4,215 128,800,000 127,000,000 1.23 1.01 A A MYS-SSO 1,142 1,358 35,123,000 38,000,000 0.84 0.92 A A MDV-MPAO 27 119 569,000 3,175,000 0.23 0.18 B B MLI-NSII 1,717 523 16,172,000 3,503,000 3.28 4.62 C C MHL-SSA 26 57 827,000 1,046,000 0.45 0.79 B A MUS-MNPS n.a. 816 2,068,000 22,000,000 n.a. 0.09 n.a. B FSM-SSA 34 82 969,000 1,421,000 0.41 0.68 B B MDA-NOSI 1,316 4,910 14,485,000 37,000,000 0.27 0.39 B B 62 Benchmarking Performance of Public Pension Programs (Continued) Country-Agency Staffing Total Admin Expenditures, US$ Benchmark Coefficient Performance category Actual Projected Actual Projected Staffing Expenditures Staffing Expenditures MAR-NSSF 4,750 3,855 79,999,000 60,100,000 1.23 1.33 A B MAR-MPF 419 824 15,933,000 17,700,000 0.51 0.90 B A NAM-SP n.a. 253 3,049,000 2,942,000 n.a. 1.04 n.a. A NAM-GIPF 143 220 9,977,000 5,457,000 0.65 1.83 B B NLD-SIB 3,247 3,788 281,700,000 306,000,000 0.86 0.92 A A NZL-Super 560 901 29,322,000 58,400,000 0.62 0.50 B B NZL-(CI) NSF n.a. 16 438,000 860,000 n.a. 0.51 n.a. B NOR-NAV 14,523 2,777 1,649,000,000 325,000,000 5.23 5.07 C C PAK-EOBI 938 1,193 9,902,000 10,200,000 0.79 0.97 A A PHL-SSS 6,074 7,729 119,700,000 93,000,000 0.79 1.29 A B PHL-GSIS 3,100 1,494 121,000,000 18,500,000 2.08 6.54 C C POL-ZUS 47,588 26,069 1,133,000,000 618,000,000 1.83 1.83 B B POL-KRUS 6,390 5,547 173,900,000 135,000,000 1.15 1.29 A B PRT-SSI n.a. 13,361 766,500,000 475,000,000 n.a. 1.61 n.a. B ROM-NPSIF 4,737 4,768 161,600,000 126,000,000 0.99 1.28 A B RWA-RSSB 227 255 6,891,000 1,372,000 0.89 5.02 A C WSM-SNPF 165 110 1,666,000 1,859,000 1.50 0.90 B A SEN-SII 227 530 5,457,000 3,358,000 0.43 1.63 B B SLE-NASSIT 227 207 4,874,000 936,000 1.09 5.21 A C SGP-CPF 1,200 3,058 77,670,000 267,000,000 0.39 0.29 B B SVK-SIA 5,839 4,941 102,400,000 132,000,000 1.18 0.78 A A SLB-NPF 104 112 3,473,000 1,268,000 0.93 2.74 A C ZAF-SSA 7,528 12,778 630,300,000 278,000,000 0.59 2.27 B C ZAF-GEPF 705 1,146 39,035,000 34,000,000 0.62 1.15 B A ESP-NISS 13,000 26,014 782,300,000 1,190,000,000 0.50 0.66 B B LKA-EPF 600 473 3,545,000 5,411,000 1.27 0.66 B B LKA-FPS 400 123 688,000 1,131,000 3.26 0.61 C B KNA-SSB 104 94 3,512,000 3,913,000 1.11 0.90 A A LCA-NIC 95 117 2,259,000 2,882,000 0.81 0.78 A A VCT-NIS 53 105 1,461,000 2,360,000 0.50 0.62 B B SWZ-SNPF 121 109 4,139,000 2,014,000 1.11 2.06 A C SWZ-PSPF n.a. 162 6,232,000 2,862,000 n.a. 2.18 n.a. C SWE-SSIA/NPFs (OA 1,300 1,799 71,486,000 252,000,000 0.72 0.28 B B SWE-PPA 210 483 35,513,000 69,200,000 0.44 0.51 B B SYR-GOSI 1,986 711 8,519,000 10,900,000 2.79 0.78 C A TZA-GEPF n.a. 61 1,031,000 385,000 n.a. 2.68 n.a. C TZA-PPF 170 300 9,022,000 1,639,000 0.57 5.50 B C TZA-(ZZB)ZSSF 42 68 363,000 314,000 0.62 1.16 B A THA-GPF 277 250 15,086,000 5,351,000 1.11 2.82 A C THA-SSO 5,931 5,967 48,188,000 98,400,000 0.99 0.49 A B TON-RFB 20 17 262,000 366,000 1.15 0.72 A B TTO-NIB 614 677 15,393,000 34,900,000 0.91 0.44 A B TUR-SSI 24,779 30,560 752,200,000 626,000,000 0.81 1.20 A A UGA-NSSF 600 129 22,314,000 851,000 4.64 26.22 C C UKR-PFU 36,000 29,198 262,700,000 330,000,000 1.23 0.80 A A UK-(GBR)PS 11,890 9,436 1,255,000,000 734,000,000 1.26 1.71 B B UK-(ANG)SSB 25 41 1,547,000 1,860,000 0.61 0.83 B A UK-(FIS)FIPS n.a. 6 214,000 949,000 n.a. 0.23 n.a. B UK-(JER)SSD n.a. 213 7,426,000 25,400,000 n.a. 0.29 n.a. B UK-(NI)NISSA 5,336 1,579 377,700,000 115,000,000 3.38 3.28 C C USA-SSA/SSI 23,000 5,815 2,700,000,000 460,000,000 3.96 5.87 C C USA-SSA/OASDI 44,000 44,162 5,100,000,000 5,520,000,000 1.00 0.92 A A USA-(AK)APFDD n.a. 585 7,066,000 59,900,000 n.a. 0.12 n.a. B USA-TSP 380 675 94,300,000 86,600,000 0.56 1.09 B A URY-SIB 4,338 4,865 158,000,000 149,000,000 0.89 1.06 A A VUT-VNPF 39 30 1,417,000 484,000 1.31 2.93 B C VNM-VSS 16,500 13,438 57,300,000 122,000,000 1.23 0.47 A B YEM-GCSS 395 75 2,573,000 713,000 5.30 3.61 C C YEM-GAPSS 917 684 6,014,000 5,907,000 1.34 1.02 B A ZMB-WCF 277 273 4,583,000 2,653,000 1.02 1.73 A B 63 Annex 4: Benchmarking Costs Performance (Administrative Costs Net of Asset Management Expenditures over Beneficiaries, Nominal US$) 150 320 560 140 300 520 130 LI: US$500 280 LMI: US$2,500 480 UMI: US$8,000 120 260 440 Baseline 240 110 400 100 Collection in-house 220 200 360 90 320 Special schemes 180 80 160 280 70 Funds management 140 240 60 120 200 50 High estimate 100 40 160 80 30 60 120 20 40 80 10 20 40 - - - Beneficiaries Beneficiaries Beneficiaries 750 1,300 700 650 LHI: US$15,000 1,200 UHI: US$40,000 1,100 600 1,000 550 500 900 450 800 400 700 350 600 300 500 250 400 200 300 150 100 200 50 100 - - Beneficiaries Beneficiaries Source: Author’s al ulatio s. 64 Social Protection & Labor Discussion Paper Series Titles 2012-2014 No. Title 1501 Defining, Measuring, and Benchmarking Administrative Expenditures of Mandatory Social Security Programs by Oleksiy Sluchynsky, February 2015 1425 Old-Age Financial Protection in Malaysia: Challenges and Options by Robert Holzmann, November 2014 1424 Profiling the Unemployed: A Review of OECD Experiences and Implications for Emerging Economies by Artan Loxha and Matteo Morgandi, August 2014 1423 Any Guarantees? An Analysis of China’s Rural Minimum Living Standard Guarantee Program by Jennifer Golan, Terry Sicular and Nithin Umapathi, August 2014 1422 World Bank Support for Social Safety Nets 2007-2013: A Review of Financing, Knowledge Services and Results by Colin Andrews, Adea Kryeziu and Dahye Seo, June 2014 1421 STEP Skills Measurement Surveys: Innovative Tools for Assessing Skills by Gaëlle Pierre, Maria Laura Sanchez Puerta, Alexandria Valerio and Tania Rajadel, July 2014 1420 Our Daily Bread: What is the Evidence on Comparing Cash versus Food Transfers? by Ugo Gentilini, July 2014 1419 Rwanda: Social Safety Net Assessment by Alex Kamurase, Emily Wylde, Stephen Hitimana and Anka Kitunzi, July 2012 1418 Niger: Food Security and Safety Nets by Jenny C. Aker, Carlo del Ninno, Paul A. Dorosh, Menno Mulder-Sibanda and Setareh Razmara, February 2009 1417 Benin: Les Filets Sociaux au Bénin Outil de Réduction de la Pauvreté par Andrea Borgarello et Damien Mededji, Mai 2011 1416 Madagascar Three Years into the Crisis: An Assessment of Vulnerability and Social Policies and Prospects for the Future by Philippe Auffret, May 2012 1415 Sudan Social Safety Net Assessment by Annika Kjellgren, Christina Jones-Pauly, Hadyiat El-Tayeb Alyn, Endashaw Tadesse and Andrea Vermehren, May 2014 1414 Tanzania Poverty, Growth, and Public Transfers: Options for a National Productive Safety Net Program by W. James Smith, September 2011 1413 Zambia: Using Social Safety Nets to Accelerate Poverty Reduction and Share Prosperity by Cornelia Tesliuc, W. James Smith and Musonda Rosemary Sunkutu, March 2013 1412 Mali Social Safety Nets by Cécile Cherrier, Carlo del Ninno and Setareh Razmara, January 2011 1411 Swaziland: Using Public Transfers to Reduce Extreme Poverty by Lorraine Blank, Emma Mistiaen and Jeanine Braithwaite, November 2012 1410 Togo: Towards a National Social Protection Policy and Strategy by Julie van Domelen, June 2012 1409 Lesotho: A Safety Net to End Extreme Poverty by W. James Smith, Emma Mistiaen, Melis Guven and Morabo Morojele, June 2013 1408 Mozambique Social Protection Assessment: Review of Social Assistance Programs and Social Protection Expenditures by Jose Silveiro Marques, October 2012 1407 Liberia: A Diagnostic of Social Protection by Andrea Borgarello, Laura Figazzolo and Emily Weedon, December 2011 1406 Sierra Leone Social Protection Assessment by José Silvério Marques, John Van Dyck, Suleiman Namara, Rita Costa and Sybil Bailor, June 2013 1405 Botswana Social Protection by Cornelia Tesliuc, José Silvério Marques, Lillian Mookodi, Jeanine Braithwaite, Siddarth Sharma and Dolly Ntseane, December 2013 1404 Cameroon Social Safety Nets by Carlo del Ninno and Kaleb Tamiru, June 2012 1403 Burkina Faso Social Safety Nets by Cécile Cherrier, Carlo del Ninno and Setareh Razmara, January 2011 1402 Social Insurance Reform in Jordan: Awareness and Perceptions of Employment Opportunities for Women by Stefanie Brodmann, Irene Jillson and Nahla Hassan, June 2014 1401 Social Assistance and Labor Market Programs in Latin America: Methodology and Key Findings from the Social Protection Database by Paula Cerutti, Anna Fruttero, Margaret Grosh, Silvana Kostenbaum, Maria Laura Oliveri, Claudia Rodriguez-Alas, Victoria Strokova, June 2014 1308 Youth Employment: A Human Development Agenda for the Next Decade by David Robalino, David Margolis, Friederike Rother, David Newhouse and Mattias Lundberg, June 2013 1307 Eligibility Thresholds for Minimum Living Guarantee Programs: International Practices and Implications for China by Nithin Umapathi, Dewen Wang and Philip O’Keefe, November 2013 1306 Tailoring Social Protection to Small Island Developing States: Lessons Learned from the Caribbean by Asha Williams, Timothy Cheston, Aline Coudouela and Ludovic Subran, August 2013 1305 Improving Payment Mechanisms in Cash-Based Safety Net Programs by Carlo del Ninno, Kalanidhi Subbarao, Annika Kjellgren and Rodrigo Quintana, August 2013 1304 The Nuts and Bolts of Designing and Implementing Training Programs in Developing Countries by Maddalena Honorati and Thomas P. McArdle, June 2013 1303 Designing and Implementing Unemployment Benefit Systems in Middle and Low Income Countries: Key Choices between Insurance and Savings Accounts by David A. Robalino and Michael Weber, May 2013 1302 Entrepreneurship Programs in Developing Countries: A Meta Regression Analysis by Yoonyoung Cho and Maddalena Honorati, April 2013 1301 Skilled Labor Flows: Lessons from the European Union by Martin Kahanec, February 2013 To view Social Protection & Labor Discussion papers published prior to 2013, please visit www.worldbank.org/spl. Abstract This study provides a framework for comparison and benchmarking of administrative expenditures of public and private social security programs. The paper presents the genesis of the inquiries into the subject, reviewing some of the most relevant literature on administrative expenditures and the costs of mandatory programs produced over the past two decades. The quantitative analysis builds on the extensive body of literature, but our framework evolved considerably from earlier studies. Our dataset includes over 100 observations and a broad set of explanatory variables. We developed and compared a number of standardized cost indices discussing their advantages and limitations. We also discuss major cost components and their shares in total program costs. The analysis explains over 90 percent of variation in administrative expenditures. It confirms some of the hypotheses expressed in the earlier studies and presents new evidence of driving factors for costs. We developed three different specifications for statistical analysis. The first set looks at the impact of design of a program on total costs. The second group of specifications assesses differences in costs of managing pension liabilities between the public and private mandatory pension schemes. Finally, on the basis of the third model we generate benchmarks for staffing levels and for the total administrative expenditures. We compare those to the actual indicators and develop standard performance ratios, providing insights into design variations and performance of the programs. We conclude with a discussion of data limitations and implications of our findings. About this series... Social Protection & Labor Discussion Papers are published to communicate the results of The World Bank’s work to the development community with the least possible delay. This paper therefore has not been prepared in accordance with the procedures appropriate for formally edited texts. 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. The boundaries, colors, denominations, and other information shown on any map in this work do not imply any judgement on the part of The World Bank concerning the legal status of any territory or the endorsement or acceptance of such boundaries. For more information, please contact the Social Protection Advisory Service, The World Bank, 1818 H Street, N.W., Room G7-803, Washington, DC 20433 USA. Telephone: (202) 458-5267, Fax: (202) 614-0471, E-mail: socialprotection@worldbank.org or visit us on-line at www.worldbank.org/spl. © 2015 International Bank for Reconstruction and Development / The World Bank