This research was funded by the Social Security Administration, Office of Research, Evaluation and Statistics, Division of Policy Evaluation (Contract No.: 600-98-27332). We remain solely responsible for all errors and omissions. The nonpartisan Urban Institute publishes studies, reports, and books on timely topics worthy of public consideration. The views expressed are those of the authors and should not be attributed to the Urban Institute, its trustees, or its funders.
This report is available in its entirety in the Portable Document Format (PDF), which many find convenient when printing.
ABSTRACT
Long-Term Model Development for Social Security Policy Analysis
Policymakers need to understand how Social Security reforms affect income distribution. Existing models range from simple representations of career earnings of typical workers to complex general equilibrium models. Population micro-simulation models, which project the earnings, wealth, and demographic histories of a representative sample of families, are useful for simulating many reform proposals. This report evaluates one such model - the projected cohorts model - and then discusses in detail three important issues in model development: 1) representing saving behavior, 2) capturing macro-economic effects, and 3) accounting for risk and uncertainty.
PREFACE
This report was prepared by the Urban Institute for the Social Security Administration, Office of Research, Evaluation, and Statistics, Division of Policy Evaluation (SSA/ORES/DPE) under Task Order 0440-99-36156, pursuant to Contract No. 600-96-27332.
The report was drafted by a number of authors, both Urban Institute staff and project consultants. Eric Toder of the Urban Institute directed the day to day research on the project and contributed to writing several of the chapters. Principal authors of each chapter in the report were:
| Chapter 1 |
|
Melissa Favreault, Urban Institute |
| Chapter 2 |
|
Frank Sammartino and Karen Smith, Urban Institute |
| Chapter 3 |
|
John O'Hare, Urban Institute |
| Chapter 4 |
|
Diane Lim Rogers, Urban Institute and Kent Smetters, University of Pennsylvania |
| Chapter 5 |
|
John Rust, Yale University |
Joseph Cordes, Daniel Dulitzky, and Frank Sammartino also contributed to the writing of chapter 3. Robbie Howell provided research assistance.
In preparing this report, we received numerous helpful comments on earlier drafts from David Pattison and Benjamin Bridges of the Social Security Administration and from a panel of outside experts. The outside experts were Jonathan Gruber of the Massachusetts Institute of Technology, Eric Hanushek of the University of Rochester, Olivia Mitchell of the University of Pennsylvania, John Shoven of Stanford University, and Douglas Wolf of Syracuse University. We also received helpful comments and suggestions from Rudy Penner, Lawrence Thompson, and Sheila Zedlewski of the Urban Institute and Gary Burtless of the Brookings Institution.
Theresa Plummer prepared the manuscript for publication.
TABLE OF CONTENTS
PREFACE
INTRODUCTION AND SUMMARY
I. OUTLINE OF REPORT
II. MAIN FINDINGS
1. Microsimulation Models
2. The Projected Cohorts Model
3. Modeling Saving
4. Macroeconomic Growth Modeling for Microsimulation of Social Security Reform
5. Risk and Uncertainty
REFERENCES TO INTRODUCTION AND SUMMARY
CHAPTER 1: REVIEW OF SOCIAL SECURITY MICROSIMULATION ALTERNATIVES
I. INTRODUCTION
II. WHY MODELS ARE IMPORTANT AND CLASSES OF MODELS FROM WHICH ANALYSTS CAN CHOOSE
III. DESIRED CHARACTERISTICS OF A MODEL OF SOCIAL SECURITY
1. Capacity: Ability to Capture Proposed Reforms
2. Outcomes
3. Population Definition/Aging Modules
4. Time Horizon
5. Integration of Social and Economic Research on Social Security Parameters
IV. ASSESSMENT OF DIFFERENT MODEL TYPES
1. Criteria
Representativeness and Completeness of Starting Data
Cohort Size (Aside From Stochastic Replications)
Generalizability
Ability to Oversample Parts of The Population
Theoretical and Empirical Validity of Updating Algorithms (If applicable)
Autocorrelation/Serial Correlation (If applicable)
Feedbacks
Outcomes
Programmatic Detail
Projection Assumptions
Sensitivity Analyses
Stochastic Replications Per Individual/Cell
Stochastic Replications of Aggregate Demographics/Economics
Ease of Use
Accessibility
Development and Maintenance Time and Cost
2. Type 1. Representative Worker Models
Summary Of Representative Worker Models
3. Type 2: Cell-Based Models
Summary of Cell-Based Models
4. Type 3: Population Microsimulation Models
Aging Algorithms
Open Versus Closed Populations
Simulation Time Dimension
Model Combinations
Summary of Population Microsimulation Models
5. Type 4: Computable General Equilibrium Models
Summary of Computable General Equilibrium Models
V. WHICH MODEL WHEN?
VI. CONCLUSIONS
REFERENCES TO CHAPTER 1
CHAPTER 2: REVIEW AND ASSESSMENT OF THE COHORTS SIMULATION APPROACH
I. INTRODUCTION
II. AN OVERVIEW OF THE COHORTS MODEL
1. Applicability of The Cohorts Model
2. Types of Analyses Beyond The Current Scope Of The Model
Economic Well-Being of the Retired Population
Program Interaction
Post Retirement Earnings
Social Security Trust Funds Balances
Macroeconomic Effects of Social Security Reforms
III. KEY ISSUES FOR THE PROJECTED COHORTS MODEL: MISSING DATA AND EARNINGS PROJECTIONS
1. Starting Database
Incomplete and Underreported Survey Data From the 1994 CPS
Missing SSA Administrative Data
Missing Earnings Information For Deceased Workers
Missing Earnings Information For Former Spouses
Earnings Limited to Social Security Covered Earnings
Censored Earnings
2. Calculation of Expected Lifetime Social Security Benefits
Differential Mortality
Taxation of Benefits
3. Labor-Force Participation and Earnings Projections
4. Changes in Labor Force Participation and Earnings Patterns
Increased Labor Force Participation of Women
Decreased Labor Force Participation of Older Men
Increased Relative Earnings of Women
Decreased Relative Earnings of Younger Men
Increased Share of Wives Earnings
Increased Correlation of Spouses' Earnings
Increased Earnings Inequality
Changing Macroeconomic Environment
5. Alternative Methods For Projecting Earnings
6. Demographic Projections
IV. THE PROJECTED COHORTS MODEL AS A FRAMEWORK FOR A MORE COMPREHENSIVE MODEL
1. Behavioral Response at the Micro Level
2. Macroeconomic Feedback Effects
V. SUMMARY/CONCLUSIONS
REFERENCES TO CHAPTER 2
CHAPTER 3: MODELING SAVING BEHAVIOR
I. INTRODUCTION
II. ISSUES IN MODELING SAVING BEHAVIOR
1. What Questions Do We Want to Answer?
2. What Definition of Saving Should We Use?
3. Stylized Facts With Respect to Saving
The U.S. Private Saving Rate Has Declined
Saving Rates Vary By Age, But Have Declined for All Groups
Saving Rates are Higher for Higher Income Households
Many Individuals Reach Retirement with Little Savings
Social Security and Pension Assets Constitute Most of Retiree Wealth
The Elderly Spend-Down Only Some of Their Wealth
III. OVERVIEW OF THE SAVING AND CONSUMPTION LITERATURE
1. Life-Cycle Models
2. Other Models of Saving
Hyperbolic Discounting
Behavioral Life-Cycle Model
IV. APPROACHES TO MODELING SAVING BEHAVIOR
1. Dynamic Programming (DP) Methods
2. Estimation and Projection of Wealth at Retirement
3. Projection of Wealth Based on Econometric Estimates of Annual Consumption
Consumption a Fixed Fraction of Current Income
Consumption as a Function of Permanent Income
Consumption as a Function of Cash-on-Hand
4. Issues That Models of Wealth Accumulation Must Address
Quality of Data
Treatment of Uncertainty
5. Saving and Wealth Accumulation in Existing Simulation Models
MINT
CORSIM
DYNASIM
PRISM
DYNACAN
EBRI/SSASIM2
Auerbach-Kotlikoff Overlapping Generations Model
V. SAVING AND OTHER BEHAVIORAL DECISIONS
1. Retirement Decision
Estimates of the Retirement Decision
Joint Retirement Decisions of Spouses
2. Bequests
3. Other Sources of Income
4. Portfolio Choice
VI. CONCLUSIONS
REFERENCES TO CHAPTER 3
CHAPTER 4: MACROECONOMIC GROWTH MODELING FOR MICROSIMULATION OF SOCIAL SECURITY REFORM
I. INTRODUCTION
II. UTILITY-BASED MODELS (EXPLICIT HOUSEHOLD OPTIMIZATION)
1. Ramsey Model
2. The Life Cycle Overlapping-Generations Model
3. Deterministic OLG Models
4. Uncertainty in OLG Models
5. Making Utility-Based Models More Realistic
III. AD HOC MODELS (NON-EXPLICIT SPECIFICATIONS OF HOUSEHOLD BEHAVIOR)
1. Advantages and Disadvantages of Ad Hoc Specifications
2. Overview of Ad Hoc Modeling Approaches and Examples
3. Macroeconomic Modeling with Microdata
IV. HOW MODELS ADDRESS PARTICULAR ISSUES
1. Degree of Disaggregation: Household-Level
2. Degree of Disaggregation: Producer-Level
3. Consumption, Saving, and Wealth Accumulation
4. Investment Adjustment Costs
5. International Capital Flows
Reduced-Form Equations
Small Open Economy
Ignore Capital Flows Altogether
6. Labor Supply
7. Government Sector and Government Budget Constraints
Degree of Details
The Government's Budget Constraint
8. Incorporating Uncertainty in Exogenous Variables
9. Measuring Welfare and Efficiency Gains
V. SOME POLICY APPLICATIONS
1. Cut in Social Security Benefits
2. Mandatory Private Accounts
3. Trust Fund Investment
VI. SUMMARY/CONCLUSIONS
REFERENCES TO CHAPTER 4
CHAPTER 5: STRATEGIES FOR INCORPORATING RISK, UNCERTAINTY, AND PRIVATE INSURANCE MECHANISMS IN MODELS OF SOCIAL INSURANCE
I. INTRODUCTION
II. THE IMPORTANCE OF ACCOUNTING FOR RISK AND BEHAVIORAL/EQUILIBRIUM FEEDBACKS
1. Some Preliminary Concerns
Risk vs. Uncertainty
Modeling Priorities: Risk vs. General Equilibrium Effects
2. Examples Illustrating the Need to Account for Risk and General Equilibrium Feedbacks
Example 1: Projecting the Financial Impacts of an Aging Population
Example 2: The Value of Social Security in a World of Incomplete Annuity Markets
Example 3: Social Security's Effect on Intra and Intergenerational Risk-Sharing
Example 4: Social Security's Effect on Markets and Institutions
3. Examples that Illustrate the Benefits of a Comprehensive Treatment of Risk in Social Security Policy Models
Example 1: Interaction Between Old Age and Medicare Insurance
Example 2: Interactions Between Old Age and Disability Insurance
III. DIMENSIONS OF RISK AND UNCERTAINTY
1. Sources of Risk
2. How Individuals Deal With Risk
3. Risks That Social Security Models Should Address
Evidence of Risks that Matter to Americans
Subgroups of the Populuation at Risk
IV. "SHORT-CUT" APPROACHES TO INCORPORATING RISK AND INSURANCE IN POLICY MODELS
1. An Illustration of the Short-cut Approach to Accounting for Risk
2. Limitations of the Short Cut Approach to Policy Forecasting
Absence of Objective Criteria For Evaluating Welfare Gains and Losses
Uni-dimensional Analysis of Risk
Treatment of Unfunded Liability of Social Security System
3. Using Options to Calculate the Value of Benefit Guarantees
4. Pitfalls of Using Money's Worth Measures and Other Summary Statistics
5. Concluding Remarks about the Short Cut Approach
V. THE DP/SGE APPROACH TO INCORPORATING RISK AND INSURANCE IN POLICY MODELS
1. Dynamic Programming Models
2. Information Economics, Mechanism Design, and Models of Optimal Social Insurance
Optimal Disability Insurance
Optimal Unemployment Insurance
Optimal Health Insurance
3. Stochastic General Equilibrium Models
The STY Model
Other SGE Models
Models With Endogenous Policies
Concluding Remarks on SGE Models
VI. CONCLUSIONS
APPENDIX 5-A: RECOMMENDATIONS FOR AN SSA MODELING STRATEGY
REFERENCES TO CHAPTER 5
INTRODUCTION AND SUMMARY
The Social Security Administration (SSA) is undertaking a major effort to improve its capability to evaluate the distributional effects of Social Security reforms. As part of this effort, SSA has supported the development of two large-scale micro-simulation models — the Model of Income in the Near-Term (MINT) and the Projected Cohorts Model (PCM). In addition, SSA is considering for internal use the Cornell Microsimulation Model (CORSIM), a model developed by a team of economists and sociologists at Cornell University to analyze the effects of policy reforms. These new models supplement the models used by the Office of the Chief Actuary (OCT) by adding much more detail on how Social Security affects different sub-groups of the population.
This report provides a review of the major issues in developing models to analyze the effects of reforms of the Social Security system on the distribution of income within and between cohorts. We consider the ability of models both to project accurately how reforms affect economic variables and to evaluate the net social benefits of alternative outcomes. While the emphasis is on micro-simulation models that can provide detailed projections of the economic status of sub-groups of the population under current law and proposed alternatives, we also consider alternative approaches for modeling the long-term macro-economic effects of policy changes. We discuss the strengths and weaknesses of alternative modeling strategies, in relation to the information policy-makers need to evaluate alternative reforms. In doing so, the report considers in detail some of the most complex and difficult issues in using models to analyze the effect of policy reforms. These include the choice of how to model household saving and alternative ways of incorporating risk and uncertainty in Social Security policy analysis.
I. OUTLINE OF REPORT
Chapter 1 of the report provides a detailed review of alternative models used to analyze Social Security policies. We explore the relationship between the options that policymakers want to analyze and the capabilities of alternative types of models to answer different questions. We list detailed criteria for evaluating models and then assess the strengths and weaknesses according to these criteria of four general model types — representative worker models, cell-based models, population microsimulation models, and computable general equilibrium models. The chapter concludes with a general discussion of which types of model are appropriate under which circumstances.
Chapter 2 reviews and assesses one population microsimulation model currently under development at the Social Security Administration - the Projected Cohorts Model (PCM). The general approach of the cohorts model is to begin with a complete earnings history of a recent cohort of retirees (the 1930 birth cohort). It then projects characteristics of future cohorts by adjusting earnings to reflect earnings growth and changes in labor force participation rates and relative earnings of women. The chapter provides an overview of the PCM, summarizing the types of analyses it can currently perform and those it cannot perform without further enhancements. It then considers problems and issues that the cohort model approach must address. These issues include how to correct for gaps in the current database, improving the calculation of lifetime Social Security benefits, and adjusting labor-force participation rates and earnings patterns to take account of observed changes for more recent cohorts. We conclude the chapter by discussing the strengths and weaknesses of using the cohorts model approach as a platform for a more comprehensive model that would incorporate behavioral responses and macro-economic effects.
Chapter 3 discusses issues in modeling household saving in microsimulation models. We begin by reviewing some stylized facts about saving that any model must explain and then discuss the main approaches to modeling saving in the academic literature. The central approach is the life-cycle model (LCM), which offers a simple and powerful framework to explain how households allocate consumption over their lifetime. We review empirical research that contradicts the implications of the simple form of the LCM and that either suggests ways of modifying the LCM to improve its explanatory power or offers alternative modeling approaches. We then outline broad approaches for modeling saving in microsimulation models and review the treatment of saving in current models. We conclude with a brief discussion of how to integrate modeling of saving with modeling of other behavior, especially the decision of when to retire.
Chapter 4 reviews approaches for modeling the effects of Social Security reforms on the macro-economy and for linking macro-economic models to microsimulation models that can be used to analyze income distribution. We discuss two broad approaches to macro-economic modeling — utility-based models that represent household labor supply and saving as a solution to a lifetime optimization problem and ad hoc models that impose a less formal structure. These latter models rely more on econometric analysis from historical data to derive behavioral equations for labor supply and saving. We consider the strengths and weakness of the two alternative approaches. We then discuss alternative ways of linking macro and micro models. The chapter then reviews specific features of existing and potential macro models for Social Security policy analysis, including the level of aggregation at the producer and consumer levels and how the models treat household saving and labor supply, the government sector, investment adjustment costs, and international capital flows. We conclude the chapter by exploring the alternative ways that utility-based and ad hoc models would approach key policy questions, such as the effects of cuts in Social Security benefits, mandatory private accounts, and a decision to invest the trust fund surplus in equities.
Chapter 5 discusses strategies for incorporating risk, uncertainty, and private insurance mechanisms in models of Social Security reform. The chapter begins with illustrations of why it is necessary to account for the effects of policy changes on risk-bearing and how failing to do so can lead to misleading implications on the net gains and losses from policy reforms. It then discusses the types of risks people face and the mechanisms people use (private and public) to reduce risks and explains why models of Social Security should ideally be able to address a wide variety of risks in a comprehensive way.
The final sections of Chapter 5 review different methods of incorporating risk and insurance in policy models. Because formal modeling of risk is extremely complex, we provide illustrations of "short cut" approaches to modeling risk, noting their potential applicability and possible pitfalls of some simplified approaches. We then survey the recent literature on dynamic programming/stochastic general equilibrium (DP/SGE) models, showing how they have been used to generate analyses of the welfare gains and losses from the current Social Security system and from transitions to alternatives, such as a system of individual accounts.
The report is intended to provide information to the SSA on the wide range of policy models available and on the advantages and disadvantages of alternative approaches to model development. The Urban Institute has not recommended an overall strategy for model development, although we hope that the information in this report will help SSA in deciding how best to proceed with their own model development.1 We hope also that the report will inform others on the "state of the art" in this rapidly expanding area of policy analysis.
An Appendix to Chapter 5 presents the views of Professor John Rust of Yale University, a consultant to this project and the author of chapter 5, on a strategy for SSA to expand model development. While the Urban Institute takes no position on the material in this Appendix, we believe the views of Professor Rust are worth serious consideration.
II. MAIN FINDINGS
1. Microsimulation Models.
A number of different types of models of varying complexity are available for analyzing the effects of Social Security reforms. They vary along many dimensions, including the extent to which they accurately represent the composition of the population and subgroups of interest (by sex, race, education, and income level), incorporate details of the Social Security retirement program, and allow for behavioral responses to policy changes. The simplest are representative worker models that can be used to simulate how Social Security reforms would affect the lifetime taxes and benefits of a worker with an assumed "representative" lifetime pattern of earnings. The most theoretically sophisticated are stochastic general equilibrium models that permit estimation of the economic effects of policy changes by representing household labor supply and saving behavior as the solution to a complex optimization problem, given household earnings capacity and initial wealth and government policies.
Ideally, models should allow policy makers to estimate the impact of policy changes on income and lifetime welfare of different age cohorts of the population and sub-groups within them (by income level, race, sex, and level of education). They should be flexible enough to incorporate alternative assumptions about demographic changes and behavioral responses to policies, based on recent research findings. They should be sufficiently transparent so that users understand what specific assumptions are producing the model's results. At a technical level, their computing cost should be low enough and computing speed fast enough to allow users to perform numerous simulations incorporating alternative scenarios and policy changes in response to requests for information from policymakers.
Our survey finds that no one model meets all concerns; all model types have both strengths and weaknesses. Representative worker models typically incorporate substantial details about the Social Security program. This enables them to be used to simulate the effects of different changes in payroll taxes (such as a change in the rate or the maximum earnings level) and the benefit formula (such as changes in the bend points in the replacement rate formula, cost-of-living adjustments, and spousal benefits.) But representative worker models incorporate little population detail, so they cannot be used to generate estimates of how policy changes affect income distribution. They also cannot be used to estimate overall effects of the fiscal balance of the Trust Fund or the state of the economy.
Cell-based models represent the population by age, sex, and cohort subgroups and incorporate average earnings profiles, Social Security benefit replacement rates, and tax rates in each population cell or subgroup. They are useful for tracking aggregates such as Trust Fund balances. But they are inadequate for studying the effects of Social Security proposals on detailed population sub-groups because they do not generate sufficient detail on the distribution of individual earnings histories.
Population Microsimulation models simulate the lifetime earnings, wealth, and demographic profiles of a representative sample of families. They include models such as the Model of Income in the Near Term (MINT) and Projected Cohorts Model (PCM) that project retirement incomes of sample individuals based on historical relationships, but do not incorporate a full year by year evolution of economic and demographic processes. They also include models such as the Cornell Microsimulation Model (CORSIM) and the Urban Institute's Dynamic Simulation of Income (DYNASIM) model that simulate annual changes in demographic characteristics and economic outcomes for the sample population. The advantage of population microsimulation models is that they can capture the interactions of multiple program changes and social and economic processes. This enables users to simulate the effects on lifetime incomes and benefits of different cohorts and demographic groups of changes in Social Security rules. But the complexity of these models can make their results difficult to interpret and dependent on assumed future behavioral processes that may be improperly specified. While microsimulation models typically incorporate some behavioral responses based on statistical research results, they do not incorporate a theoretically consistent model of economic behavior. They also cannot be used to generate overall economic forecasts without being linked to some type of external macro-economic model that represents the production side of the economy and solves for market clearing wages and interest rates.
General equilibrium models provide a consistent behavioral framework for analyzing the effects of policy changes on labor supply, saving, total economic output, and social welfare. But these models are extremely complex and are often based on strong a priori assumptions about the structure of household preferences. To make them tractable, they typically represent the population and the parameters of taxes and benefit programs in a highly stylized fashion. This makes them less useful than dynamic microsimulation models in generating detailed distributional estimates and in analyzing changes in program details. But their advantage is they are only the modeling tool that provides new insights on the "big picture" effects of reforms on saving and work effort.
The Projected Cohorts Model.
The Projected Cohorts Model (PCM) is a population microsimulation model currently under development at SSA. As part of this contract, we were asked to evaluate this model in some detail.
The PCM uses as its base data a sample drawn from an exact match of the 1994 U.S. Census Current Population Survey (CPS) and Social Security Earnings Records (SSER) of individuals who became eligible for Social Security retirement benefits in 1992 and thus were born in 1930. Based on the experience of the 1930 birth cohort, it projects earnings histories of future cohorts by adjusting the earnings histories of the base year cohort. In its present form, it can be used to simulate the effects of changes in the Social Security benefit formula and payroll taxes on different groups of earners. Adding imputations of other sources of income (pensions and income from non-pension saving) and linking the PCM with a macro-economic model could potentially make the model usable for performing broader analyses of the impacts of policy reforms on the level and distribution of income of future cohorts of retirees.
The cohorts model is superior to representative worker models and cell-based models for analyzing the distributional effects of Social Security reform proposals, such as changes in the structure of the benefit formula. Its advantage is that it incorporates the full actual earnings and demographic history of a cohort. There are some gaps in the data because of missing observations and because earnings subject to Social Security are censored at levels that were much lower in relation to the average wage in earlier years than today. But SSA has developed a method for correcting censored earnings and imputations for missing earnings histories of deceased workers and former spouses could be added.
The main strength of the cohorts model is that it is able to simulate an entire lifetime of earnings for members of a cohort, using a simple computational process. Its main weakness it that the simple adjustment it uses will not adequately project earnings histories of future cohorts if their lifetime earnings do not resemble earnings histories from the 1930 cohort. Although the model adjusts labor force participation rates and relative earnings of women to 1990 levels, it may be missing important changes that are already evident in earnings patterns for later cohorts. These include further increases in female labor force participation rates, especially for middle-aged and older women, a more positive correlation of earnings levels between spouses, and an increase in the inequality of the earnings distribution. The best way to fix these problems is to develop methods that incorporate information available in the partial histories of later cohorts.
The PCM in its present form cannot be used to analyze complete future total incomes because it does not include all income sources of beneficiaries and does not project marriage histories. In particular, it does not project income from pensions, other assets, and "partial retirement" earnings of Social Security beneficiaries. The PCM could be enhanced by modeling saving and retirement behavior and by including macro-economic feedback effects. The general approach used by the PCM has advantages for projecting saving because it simulates the entire earnings history of a worker. But simulating other behavioral responses (such as the timing of retirement) requires more complex algorithms that depart from the simple scaling adjustments of the cohorts model.
Modeling Saving.
The economics profession is far from a consensus on a theory of saving or on how to interpret results of econometric studies on how saving incentives affect net saving. Different approaches can yield very diverse predictions about how proposed policy reforms would affect household saving. Moreover, evaluation of policy reforms, such as a shift to individual accounts, is highly sensitive to assumptions about saving responses of households.
The most powerful and widely used theory of saving is the life cycle model (LCM). The basic premise of the LCM is that individuals are forward looking in planning to save to finance the decline in earnings that accompanies old age. The simple version of the LTM ignores uncertainty and assumes that utility is time separable and depends only on consumption and leisure in different periods. It also assumes that individuals have unlimited access to credit markets at the market interest rate and neither leave nor receive bequests. The simple LCM yields strong predictions about the paths of consumption and saving, given any pattern of lifetime earnings. Key implications are that consumption in any year depends only on total wealth (including the present value of future earnings and Social Security wealth) and not on either current income or on the types of assets in which wealth is held. But a significant amount of empirical research contradicts the implications of the simple form of the LCM.
Enhancing the LCM to account for uncertainty, liquidity constraints, the effects of government transfer programs, and bequests makes the model more consistent with observed data. Modified approaches that account for these factors, and less traditional approaches such as the "behavioral" life cycle model, suggest that both current and lifetime income could affect current consumption and that the propensity to consume out of different forms of wealth may differ.
Existing large-scale microsimulation models typically contain very rudimentary treatments of household consumption. A standard approach is to project wealth of future retirees by a simple reduced form relationship that projects wealth at retirement based on wealth in some earlier period and other household characteristics.
An alternative would be to introduce a consumption function directly into the model, compute saving as the difference between income and consumption, and build up household wealth year by year based on projected saving and an assumed distribution of rates of return. Such an approach would allow for direct simulation of changes in policies (such as individual accounts) on wealth accumulation but the results from such simulations would be highly sensitive to the exact specification of the consumption function. Because the current state of research still provides only limited guidance on how best to model consumption, models should be structured to facilitate testing of the consequences of alternative representations of consumption behavior.
Macroeconomic Growth Modeling for Microsimulation of Social Security Reform.
Social Security reforms may have significant effects on labor supply and saving. As a consequence, their effects on wages, interest rates, and the growth rate of output may be too large to be disregarded. Micro-simulation approaches, on their own, cannot account for these overall economic effects. Some type of macro-economic model is needed to supplement microsimulation models.
There are a wide variety of macroeconomic models that are used for different purposes. For the purpose of modeling the long-run effects of Social Security reforms, demand side models that are useful for forecasting short-run fluctuations in output and employment are not helpful. What is required is some type of neo-classical growth model that projects long-run output as a function of labor supply, the stock of capital, and state of technical knowledge. But within this grouping, there are a number of available options.
One useful way of classifying empirical macroeconomic models is by whether or not they explicitly model optimizing behavior. We classify models in these two groups as utility-based models and ad hoc models.
The advantage of using utility-based models is that they are theoretically rigorous and consistent and, because they are based on explicit modeling of preferences, can generate measures of the effects of policy changes on economic welfare. In addition, structural models are less subject to the "Lucas critique", which states that relationships estimated under one policy regime will not be stable if the policy regime changes, than ad hoc models. Utility-based models are less subject to the Lucas critique because their parameters are arguably "deep" parameters that reflect underlying preferences and technical production relationships instead of behavioral responses observed in an environment that is likely to change.
The advantage of using ad hoc models is that they typically include more sectoral details and can be tailored to model policy reforms that have been observed in the past. If their behavioral specifications are more detailed, they may be more suitable than structural models that represent household preferences in an over-simplified fashion. The problems of some ad hoc models result from excessive aggregation (for example using a single consumption relationship for all households, regardless of age) and can be reduced by estimating separate equations for sub-groups that behave differently.
Microsimulation and macro-economic models can be linked with either a "bottom-up" or "top-down" approach. In both approaches, some type of macro model must be used to generate production relationships and demands for labor and capital and to solve for equilibrium prices. The difference between the two approaches is in the treatment of household decisions. In the "bottom-up" approach, household decisions in the microsimulation model determine aggregate factor supplies. These are then fed into the macroeconomic model, which generates factor demands and solves for prices. The two models run simultaneously until demand and supplies are equilibrated. In the "top down" approach, the calculation of equilibrium prices and outputs is done wholly within the macro model, which generates both demand and supply for factors of production. The microsimulation model is then used to allocate aggregate factor supplies from the macro model among households.
Risk and Uncertainty.
Individuals face a wide variety of risks that can adversely affect their income security in retirement. In the context of proposals to supplement or partially replace the current Social Security system with individual accounts, discussion has focused on risks associated with fluctuations in returns on equity. But there are many other potentially more important risks that people face. These include financial risks associated with wage declines, unemployment, uninsured health care costs, divorce or death of a spouse, living longer than expected (thereby using up assets), loss of pension coverage, and potential changes in Social Security benefits and taxes. It is important to model these various risks in a comprehensive way. It is also important for models to consider the interactions of different public programs in altering risks, including Social Security retirement and disability benefits, Unemployment Insurance, Medicare, and other Federal and state transfer programs. Finally, government measures to reduce risks may interact with actions the private sector takes, such as private health and life insurance, employer pensions, privately purchased annuities, and within-family transfers.
There is now a rapidly growing economics literature that examines how individuals use Social Security and other public and private mechanisms to cope with risk. In reviewing this literature, we contrast the more formal dynamic programming/stochastic overlapping generations (DP/SGE) models with "short cut" approaches.
Short cut approaches provide important advantages in simplification and ease of communication of results, but have several limitations. First, to the extent they are not based on a consistent utility maximization framework, they are unable to assess the effects of alternative policies on individual welfare. Second, they typically fail to account for potentially large behavioral and general equilibrium feedback effects. This could lead to errors in the direction as well as the size of effects of policy changes. Finally, they are usually subject to the "Lucas critique," in that the estimated relationships they use are not invariant to changes in the policy regime.
DP/SGE models address these problems and are becoming sufficiently realistic to be taken seriously in forecasting and policy analyses. They have produced valuable insights on how household responses to reforms of Social Security and the changes in economic equilibrium that the responses produce may affect the welfare of different cohorts in the population. For example, as discussed in the chapter, some DP/SGE models show that Social Security's adverse effect on household saving results in a net reduction in individual welfare, even though individuals benefit from the opportunity to purchase an "actuarially fair" annuity.
But DP/SGE models still suffer from important shortcomings. They typically lack a "unified" model of social insurance that provides a single, consistent, integrated framework of policy evaluation. The computational complexity of solving DP/SGE models makes them difficult and expensive to use. They require large amounts of data to identify and estimate their unknown parameters, which for this reason often lack sufficient empirical validation. They lack a good representation of employer behavior, such as the demand for labor in different age groups and the decision to offer pensions, health insurance and other fringe benefits. Finally, the models fail to capture the potential effects of macro-economic shocks and must assume an equity premium instead of solving for equity returns endogenously. Moreover, there is as of now no good explanation for the observed high equity premium and large volatility in equity prices.
DP/SGE models are still a long way from being empirically justified. They are unable to solve and estimate a full life cycle model with realistic treatment of labor/leisure and consumption/saving decisions. They are also unable to incorporate various additional choices, such as educational and career choices and marital and childbearing decisions.
Recent research with advanced DP/SGE models has found that the current PAYGO system reduces the welfare of most cohorts because of its adverse effects on work effort and capital accumulation. This raises the question of whether the large popularity of the current system reflects intergenerational transfers that benefited early cohorts in the system or some component of risk reduction that is not adequately captured in the models. Further advances in this type of work could produce new insights on the political economy of the current system and how it might best be reformed.
REFERENCES TO INTRODUCTION AND SUMMARY
Citro, C. and E. Hanushek (1997). Assessing Policies for Retirement Income: Needs for Data, Research and Models. National Academy Press. Washington, D.C.
Notes
1. See, for example Citro and Hanushek (1997) and the Appendix to Chapter 5 of this report by John Rust.
This report is available in its entirety in the Portable Document Format (PDF), which many find convenient when printing.
Latest Research
Browse by Topic:
PDF
Printer-friendly summary
