Social Security’s Goals and Criteria for Assessing Reforms

Statement of Dallas L. Salisbury, President & CEO

Employee Benefit Research Institute

Committee on Ways and Means

Subcommittee on Social Security

U.S. House of Representatives

March 25, 1999

The views expressed in this statement are solely those of the author and should not be attributed to the Employee Benefit Research Institute, or the EBRI Education and Research Fund, its officers, trustees, sponsors, or staff, or to the EBRI-ERF American Savings Education Council. The Employee Benefit Research Institute is a nonprofit, nonpartisan public policy research organization that does not lobby or take positions on legislative proposals.

Mr. Chairman and Members of the Committee. I appreciate this opportunity to discuss Social Security’s goals and criteria for assessing reforms.

I appear today as President of the Employee Benefit Research Institute (EBRI), a non-profit research organization located here in Washington DC. EBRI does not lobby or advocate specific actions, but has worked for over 20 years to provide objective data and analysis that allows policy proposals to be evaluated.

Our first book on Social Security was published in 1982i, and we have conducted much workii since then that has documented the critical role Social Security plays in providing income to the retired population, as well as to the disabled and survivors. The primary goals of Social Security have been to provide:

  • A foundation of income for all Americans – which it has done.
  • A nearly adequate income for the lowest-income Americans – which it does.
  • Income protection against the "risk" of living much longer than one plans, or the actuarial tables suggest, by paying a life annuity that is indexed for inflation – which it does.
  • Dignity for retirees, by having the income paid through a government transfer rather than requiring family members to ask other family members for direct assistance.
  • A level of taxation that permits a pay-as-you go program with a small reserve.
  • A benefit that grows in real value by passing on productivity increases and life-style increases to retirees with a benefit formula that targets replacement of final income, as opposed to a constant level of purchasing power. (This increase in real purchasing power is shown in Chart 1. The chart also shows that increases in real purchasing power would remain, even if benefits were cut to allow for full funding under the present level of the FICA tax.)

t118c1.gif (6158 bytes)

Many reform proposals now being discussed would change these goals to some degree. The criteria for assessing reforms should be to:

  • Determine whether a reform proposal supports these goals, or changes them, and whether proposed changes are acceptable.
  • Determine whether the reform fully utilizes the present administrative and recordkeeping structure, and if not, whether the reform proposal is feasible for implementation.
  • Determine whether the reform proposal changes outcomes in terms of tax levels, benefit/income levels, and life income streams.
  • Determine whether the reform proposal reduces risk in the system¾ or increases risks.
  • Determine whether the reform proposal strengthens, weakens, or has no affect on the existing system.
  • Determine whether the affected public supports any fundamental reforms.

Our 1997 book, "Assessing Social Security Reform Alternatives,"iii contains a first chapter which provides a detailed list of sub-questions in each of these areas. Our just-released 1999 book, "Beyond Ideology: Are Individual Social Security Accounts Feasible?"iv applies this methodology to individual account proposals.

Also, the EBRI-SSASIM2 model that we have developed allows comparisons making use of analytic methods that are more complete and dynamic than what is being used by many advocates. Our model allows for the use of equity market return assumptions that are consistent with economic growth assumptions, and utilizes a thousand economic scenarios that introduce a full range of possible economic outcomes, as opposed to doing static straight-line projections. Our model uses earnings projections that reflect actual lifetime income patterns (based upon Bureau of Labor Statistics data), as opposed to the "unisex flat earnings for typical households" used by most analyses. This permits assessment of effects on more than just the "average" worker, and allows more accurate measurement of "rates of return" on individual accounts. Our model allows analysis of alternative forms of transition costs, and alternative payment periods for these costs (40 year versus 70 years, etc.).

Equity Investment of the Annual FICA Surplus

One use of the EBRI model is to assess the program finance outcome of investing the annual FICA surplus – the annual amount of FICA taxes above the costs of the program for a particular year – into the equity markets or into individual accounts for workers. We modeled (first) the collective investment in equities of just the FICA surplus, and (second) the FICA surplus plus interest payments on the existing trust fund balance (which would remain in special-issue U.S. Treasury bonds). Using the actuarial assumptions from the1998 Social Security Trustees Report, when the model is in deterministic mode, the FICA-surplus-only investment has a 75-year actuarial balance of –0.08 percent of taxable payroll; but if the FICA surplus plus the interest from the bond investment is added, the actuarial balance becomes positive at 0.46 percent of taxable payroll (see Chart 2).

t118c2.gif (6134 bytes)

But the real world is not deterministic, it is dynamic. Therefore, the model was also run in stochastic (or random) mode to evaluate what would happen if the future were allowed to have ups and downs in the economy, as it has in the past. In this mode, both levels of investment yielded an average negative 75-year actuarial balance of –0.97 percent of taxable payroll for FICA surplus only, and –0.70 percent of taxable payroll for FICA surplus plus interest. When examining the various scenarios, 91 percent of the scenarios resulted in a negative 75-year actuarial balance for the FICA-surplus-only case, while 85 percent of the scenarios had a negative 75-year actuarial balance for the FICA-surplus-plus-interest case (see Charts 3 and 4).

t118c3.gif (21121 bytes)

t118c4.gif (22073 bytes)

This would be a best-case scenario because all of the surplus would be placed into equities, whereas if individuals were allowed to invest in individual accounts as they saw fit, some of the surplus would be invested in bonds. Furthermore, in our modeling we forced the equity percentage to increase by more than the annual FICA tax surplus regardless of the performance of equities over the 12-year period. This is realistic, since over 12 years it is highly improbable that bonds would outperform equities. In addition, once the program began to have expenses greater than revenue, funds were drawn down in a manner that left the equity percentage the same. Therefore, the investment in equities had a chance for long-term returns. Lastly, collective investment does not account for all the additional start-up and administrative costs that would result from creating individual Social Security accounts, and these costs would be significant.

Regarding administration, our analysis finds that a system of personal accounts would involve a number of distinct operations:

  • First, employer deduction of payroll taxes and transmittal to a third party, with ultimate transmittal of records on whom each dollar belongs to.
  • Second, receipt of the funds by a trust company or financial intermediary.
  • Third, receipt of full information on the employee and the amount of money that went to the financial intermediary.
  • Fourth, notice to the recordkeeper and the financial intermediary of how the money is to be invested.
  • Fifth, either investment by the intermediary or the transmittal to an investment manager.
  • Sixth, regular reporting on investment results to the recordkeeper so that account balances can be maintained.
  • Seventh, a system for servicing the worker’s account and providing information on the account, the investments, and details on choice.
  • Eighth, education of the worker on the personal account system, what "investments" are, what a bond, a stock, and cash are, and on what actions they can or must take regarding his or her individual account.
  • Ninth, a system for communicating ultimate annuity options and then paying the annuities. Each of these steps involves costs, with most estimates provided to date leaving out many of these costs, or providing ranges based upon frequency of choice designed into the system. Those studies make clear that the more responsive the system, the more expensive.

Administrative Conclusions

A system of personal accounts that applies to all who now pay Social Security taxes can only function at reasonable administrative cost if it takes full advantage of the present system of payroll tax deposits. Over 5 million employers still file all records on paper, and many make deposits only once each year (see Chart 5). Other approaches could be implemented¾ but only at much higher cost to employers, workers, and the government.

t118c5.gif (28878 bytes)

An individual accounts system that seeks to use the income tax system would be more difficult to make universal and would be more difficult to enforce, as it would be tied to over 140 million individuals rather than six million employers, as is currently the case.

Basing an individual accounts system on Federal Thrift Plan (TSP) or private 401(k) plans as a model is not an accurate comparison, since the covered populations are very different (see Chart 6) and thus the costs of recordkeeping and administration also would be very different. This is most true of wage levels, and thus the expected amount of annual contributions per account. Again, using the TSP or 401(k) model for individual Social Security accounts would involve much higher costs for employers, workers, and the government than is incurred under the present Social Security program.

t118c6.gif (10145 bytes)

A major cost of any retirement program is the ultimate cost of paying benefits. Only with an ultimate annuity form of payment can a personal account be compared to the present system in terms of economic security. Social Security currently spends over 90 percent of its total administrative expense on annuitization and benefit payments.

Most analyses to date of individual accounts do not include an estimate of this cost. As shown on Chart 7, administrative cost can substantially reduce benefit levels. Even without annuity cost, a recent analysis by the CATO Institute suggested costs of $55 to $115 per worker per year for just the cost of account administration and funds investment. This did not include any expense for (1) education, or (2) compliance. The CATO analysis notes that any frequent reporting to workers, frequent investment changes, loans, or other features could substantially increase costs. A recent report from the Heritage Foundation notes that the system could make use of Electronic Funds Transfers to hold down employer costs, and suggested credit bureaus as the model for individual Social Security account administration. Most employers in the United States currently report their payroll taxes on paper, and the static records of credit bureaus have little of the dynamic nature of a personal investment account.

t118c7.gif (8844 bytes)

Benefit Effects of Individual Accounts

When assessing reform proposals against the current Social Security system, it is important to note that an individual account provides a proportional benefit, meaning an equal percentage of pay contribution at each income level. As a result, the redistribution in the current system, or the "non-proportional" delivery of benefits, is not reinforced by individual accounts. This is shown graphically in Chart 8, as the higher an individual’s income, the higher the proportion of total benefits that derive from the individual account.

t118c8.gif (8022 bytes)

USA Accounts Impact On Employer Plans

Whenever public or private employers want to create a retirement program, they must make a number of decisions. Since most employers have more than one retirement plan, part of the reason for careful analysis is to avoid causing harm to other programs. Depending on how it is designed, a Universal Savings Account (USA) could be designed to avoid adverse consequences for employer plans, or it could potentially cause nondiscrimination problems for a significant percentage of employer plans. It is premature to predict what the eventual outcome of these "testing problems" may be on a plan-specific basis; however, most sponsors would need to consider a redesign – perhaps a drastic one – of their plans, and undoubtedly some plan sponsors would seriously consider the elimination of their plans. Termination of plans could reduce retirement savings, the opposite of the intended result of USA account creation.

The primary employer design factors for retirement programs are as follows:

Possible USA Features Option I Option II
1. Employer Contributions Taxable Now Tax Deferred
2. Worker Contributions Taxable Now Tax Deferred
3. Investment Earnings Taxable Now Tax Deferred
4. Matching Contributions Amount of Match (50%; 100%) Amount of Contribution Matched (x% of pay or some flat amount)
5. Timing of Contributions End of Year Each Pay Period
6. Investment Options One/Few Many

The characteristics in bold are the most common in employer-sponsored individual account plans today. To the degree that a new mandatory universal government program were less generous than employer plans, one would expect that employees would continue to participate in the employer plan. For example, if the employer’s 401(k) plan offers no match but includes payroll deduction each pay period and several investment options, while the government plan offers no match, few investment options, and a contribution once each year, the worker would likely remain in the employer plan.

However, to the extent that a new mandatory universal government program were more generous than employer plans, one would expect that employees would stop or reduce their participation in the employer plan. For example, if the employer plan offers no match, payroll deduction each pay period and several investment options, while the government plan offers a 50 percent match up to $600 in worker contributions, few investment options, and a contribution once each year, the worker would likely drop out of the employer plan and move funds to the government plan in order to get the match.

To the extent that the new plan has the worker's contribution come out of taxable income, so that taxes are not deferred, then the balance shifts back in favor of the employer plan, all other things being equal. For example, if an employer has no matching contribution, a government program with a match could affect participation in the employer plan.

Why Would It Matter if Workers Reduce Their Participation In Employer Plans?

The tax laws that apply to employer plans are extensive and complex. Most relevant here are the "nondiscrimination rules," which, put simply, "test" 401(k) plans for relatively equal deferrals (expressed as a percentage of compensation) between lower-paid and higher-paid workers. In general, for "highly compensated employees" (known as HCEs, or those paid $80,000 a year or more) to contribute to a retirement plan, the "non-highly compensated employees" (NHCEs, or those paid less than $80,000 a year) also must do so.v Further, what the HCEs can contribute to the plan is a direct function of what the NHCEs contribute. This means that if the lower-paid workers choose not to participate or contribute, the higher-paid workers would be substantially frozen out— and there would be no reason for the employer to sponsor the retirement plan.

As a matter of public policy, Congress could also make this issue irrelevant by repeal of the nondiscrimination rules that apply specifically to 401(k) plans.vi If this happened, the level of participation by the NHCE group would not matter to the HCE group. But since repeal is unlikely, we provide analysis.

The Clinton administration, it must be noted, has been meeting with many groups in an effort to complete the design of its USA program in a form that would not have an adverse impact on employer plans. It is our hope that this good-faith effort by the administration will lead to design decisions that minimize or avoid any adverse impact on employer-based plans.vii This analysis is provided to assist in those efforts. For example, a USA-type plan design that:

  • Provides an automatic (non-matched) employer/government contribution that is tax-deferred (the same as in a 401(k) plan);
  • Requires employee contributions to be taken from taxable income (while 401(k) contributions are tax-deferred); and
  • Provides no matching contribution;

would clearly provide a set of limited incentives which would be insufficient to cause employees to leave an employer plan, and most employees would consider themselves to be better off contributing to the employer plan (where available) than to the government plan.

Were there to be a government-provided match, however, then workers in an employer plan with no match might be better off in the USA, depending on the level of the match and whether the match contribution is treated as taxable income or is tax-deferred.

For a number of design and administrative reasons, the Clinton administration is unlikely to propose what was mentioned in the State of the Union Message and described through examples in a White House fact sheet. However, we have used that plan for this analysis to show how much of a difference plan design can make and why the administration is wise to work hard on the design issue.

USA Analysis

Recently the Employee Benefit Research Institute and the Investment Company Institute have completed a two-year study of the 401(k) marketviii which has yielded detailed individual participant records (including demographic information and contribution behavior) from more than 27,000 plans. Due to strict confidentiality standards, no information on the plan sponsor’s identity was included. However, the database does break out source of contributions (e.g., employee before-tax, employee after-tax, employer matching, qualified non-elective contributions (QNECs), etc.) and we are currently working on a set of computer algorithms to classify each plan by the types of incentives provided to employees at various contribution levels (e.g., a 2 percent QNEC plus 100 percent match for the first 3 percent of compensation and a 50 percent match for the next 3 percent of compensation).

When completed, this analysis will provide unique insight into how participating employees at various compensation levels may be expected to react to various formulae adopted by the employer. It will also provide the basic framework for sensitivity analysis into the likely impact of modifications in the 415(c) and/or 402(g) limits.

Given the political timeline, we do not have the luxury of completing the pattern recognition algorithms necessary to identify the contribution formulae of 27,000 plans. However, we have taken a random sample of 6,700 plans to provide some initial insights into this policy. While this is just a small fraction of the year-end 1996 information we have collected, we believe it is still much more comprehensive than any other research database in existence.ix

It is important to note that for the preliminary analysis we are substituting the participant-specific average employer match for the marginal match.x The analysis conducted by Yakoboski and VanDerhei (1996) and Kusko, Poterba and Wilcox (1994) both demonstrate the need to consider the incentive effects of the employer’s matching formulae. Our final analysis of this proposal will provide a general framework for each plan in which the total participant’s contribution is modeled as follows:xi

Total contribution = employee deferral + employer match + QNEC, where:

  • Employee deferral will be subject to Internal Revenue Service Sections 402(g) and 415(c), and potential ADP/ACP restrictions,
  • Employer match will attempt to replicate the contribution formula in place for plan x in year t (e.g., 100 percent match on employee contributions up to the first 3 percent of employee compensation plus 50 percent match on additional employee contributions up to the next 3 percent of employee compensation),xii and
  • QNECs are determined as the amount of employer contribution that is provided regardless of employee deferral (e.g., 2 percent of compensation).

This substitution of variables would be expected to bias the results if we were attempting to analyze contribution behavior at the margin for the types of formulae seen in actual 401(k) plans where there is expected to be a significant decrease in contribution incentives after approximately the first 6 percent of compensation and a complete ban on deferrals after the first $10,000.xiii However, as long as our analysis reflects only the relatively small level of employee contributions discussed thus far (i.e., no more than $600 per year), this substitution is not likely to be significant.

For purposes of passing a nondiscrimination test unique to 401(k) plans (the so-called ADP tests), it is of utmost importance that non-highly compensated employees choose to participate in the sponsor’s plan. It is logical to assume that if any employee with limited investible funds finds an alternative arrangement with a higher match rate that they may choose to reallocate some or all of their future contributions from the 401(k) plan to the USA plan. To what extent is this likely to happen in the existing plan population? Our findings are summarized below.


The analysis consisted of the following steps:

A representative random sample of approximately 6,700 401(k) plans was taken from the EBRI/ICI 401(k) database in which there was sufficient information to determine employee deferral percentages and employer match rates for at least 90 percent of the participants in the plan.

Average match rates for each participant with the requisite information were computed.

Each participant was categorized as to whether they were a highly compensated employee (HCE) or non-highly compensated employee (NHCE).xiv

Each participant was categorized as being "at risk" or not. We defined a participant to be in the former category if the employer average match rate was less than 50 percent.

Average deferral percentages were computed for each plan for the HCEs (ADPH) and the NHCEs (ADPN).

Each plan was tested to see if it passed the basic ADP test: ADPH £ ADPN * 1.25.

Each plan was tested to see if it passed the alternative ADP test: ADPH £ min(ADPN * 2, ADPN + 2%).

Any plan that did not pass either of the above two tests was excluded from further analysis.xv

At this stage of the analysis, there are several potential methods of modeling the likely impact from a competing plan with matching contributions. Two methods were chosen to illustrate the importance of behavioral assumptions in quantifying the likely impact.xvi

  • Method One: Assume any NHCE that is "at risk" drops out of the employer’s 401(k) plan while HCEs continue their current contribution.

This "all or nothing" response to a governmental competing matching plans could be justified on several grounds. First, HCEs may not be eligible to benefit from a government match due to potential constraints on adjusted gross income (AGI). Second, it is highly unlikely that employees with salaries of at least $80,000 would leave the employer plan for a 50 percent match on only $600 (at most 0.75 percent of compensation).

The ADPs are recomputed and the percentage of plans that would be in violation of both the basic and alternative tests (assuming no corrective measures were taken) are tabulated and shown in Chart 9.

t118c9.gif (8539 bytes)

  • Method Two: Allow the substitution to be quantified.

Given that a significant percentage of NHCEs are deferring more than $600, a problem with method one is that if a NHCE were already putting in $1,000 for a 25 percent match with the employer’s current plan, why not assume that they would put in $600 to a 50 percent match for the government’s plan and leave the other $400 in the employer’s 401(k) plan? This method subtracts $600 (or the participant’s current deferral, if less) from each participant and recomputes their ADP’s.

The percentage of plans that would be in violation of both the basic and alternative ADP tests (assuming no corrective measures were taken) are tabulated and shown in Chart 10. It should be noted that this estimate of the number of plans impacted would need to be re-estimated if AGI thresholds were imposed for eligibility in the government’s matching program.

t118c10.gif (8608 bytes)

Preliminary results

Chart 9 illustrates the estimated percentage of 401(k) plans that would be in violation of the ADP tests assuming any NHCE that is "at risk" drops out of the employer’s 401(k) plan while HCEs continue their current contribution. Overall, 26 percent of all private 401(k) plans are expected to be impacted under this assumption. The percentage of plans is obviously a function of plan size, with only 15 percent of plans with 1-9 participants being impacted, increasing to 35 percent of the plans with 50-99 participants. The impact decreases for larger plans; slightly less than 25 percent of the plans with more than 500 participants were estimated to be impacted.xvii

Chart 10 illustrates the estimated percentage of 401(k) plans that would be in violation of the ADP tests assuming all employees with a match rate of less than 50 percent transfer up to $600 of their contributions from the employer’s 401(k) plan to the government’s plan. The estimated impact is obviously much smaller, since some NHCEs that are assumed to be making zero contributions in method one would still have some contributions in method two (leading to a higher ADPN) and all HCEs considered to be "at risk" in the second method would have a smaller deferral than in method one (leading to a smaller ADPH).

Approximately 13 percent of all 401(k) plans are estimated to be impacted under method two. This varies from a low of 7 percent for plans with less than 10 participants, to a high of 21 percent for plans with 50-99 participants.

In terms of the number of participants impacted, the plans estimated to be impacted under method one covered 9.7 million participants at the end of 1996 (approximately 26 percent of the universe). Method two suggests that the impacted plans had 3.9 million participants at the end of 1996 (approximately 11 percent of all 401(k) participants).

I have made available to the Committee our studies completed to date, and offer our assistance in carrying out additional studies. I have also attached to this statement a set of slides intended to add detail to some of the points contained in my statement, including the results of a 1998 survey of small employers to determine attitudes on personal accounts.

I thank the Committee for this opportunity to appear before you today and wish you the best as you seek to assure future retirement income security.


Allen, Jr., Everett T., Joseph J. Melone, Jerry S. Rosenbloom and Jack L. VanDerhei, Pension Planning: Pensions, Profit Sharing, and Other Deferred Compensation Plans, Eighth edition, Homewood, Illinois: Richard D. Irwin, Inc., 1997.

Clark, Robert L. and Sylvester J. Schieber, "Factors Affecting Participation Rates and Contribution Levels in 401(k) Plans," in Olivia S. Mitchell and Sylvester J. Schieber, eds. Living with Defined Contribution Pensions: Remaking Responsibility for Retirement (Philadelphia: University of Pennsylvania Press, 1998).

Kusko Andrea L., James M. Poterba, David W. Wilcox. Employee Decisions with Respect to 401(k) Plans: Evidence From Individual-Level Data. NBER Working Paper No. 4635, 1994

VanDerhei, Jack L., Russell Galer, Carol Quick and John D. Rea. "401(k) Plan Asset Allocation, Account Balances, and Loan Activity," EBRI Issue Brief, January 1999.

Yakoboski, Paul J. and Jack L. VanDerhei, "Contribution Rates and Plan Features: An Analysis of Large 401(k) Plan Data." EBRI Issue Brief, June 1996.


i Social Security: Perspectives on Preserving the System, ISBN 0-86643-028-8

ii For example, "Retirement in the 21st Century..Ready or Not" (1994, ISBN 0-86643-081-4), which deals with the preparation of the baby boom for retirement.

iii  ISBN7125.A783

iv ISBN 0-86643-092-X

v Technically, it is possible for plan sponsors to also use non-elective contributions to satisfy these tests as long as these contributions satisfy special vesting and withdrawal restrictions.

vi While the analysis below focuses exclusively on the ADP tests, a complete analysis of the public policy implications would require similar analysis on employee after-tax and employer matching contributions as well as the multiple use test and the potential for 401(k) sponsors to adopt the newly implemented safe harbors. See Chapter 11 of Allen, Melone, Rosenbloom and VanDerhei (1997) for an explanation.

vii It is important to note that we are referring exclusively to private 401(k) plans in the analysis below and that Section 403(b), Section 457 and public plans may or may not have similar consequences based on this proposal.

viii See VanDerhei, et. al. (1999). Analysis of account balances and loan information was provided in addition to asset allocation information.

ix Both Yakoboski and VanDerhei (1996) and Clark and Schieber (1998) had a large number of participants but what matters in this type of analysis is that a sufficient number of plans be available to determine the distribution of match rates by salary.

x This variable has been used extensively in the academic literature on an aggregate basis; however, our data provides more powerful analysis since we are able to look at the average match for a particular individual and observe their specific deferrals.

xi We will also control for the impact of both before-tax and after-tax employee contributions. Note this is likely to be extremely important if the final government plan design involves after-tax employee contributions.

xii N.B.: some plans involve a more complex function, possibly incorporating the sponsor’s profits.

xiii For purposes of the 1996 data used in this analysis, the 402(g) limit was $9,500.

xiv See Internal Revenue Code Section 414(q) for a definition. Technically, the simplified definition of HCEs implemented by the Small Business Job Protection Act of 1996 had not taken effect as of year end 1996 however we used the new definition to be more relevant for predictions of impact on plans in the post 1996 environment. Moreover, no information on the 5% owner classification is available in this database.

xv This accounted for less than 1 percent of all plans in the sample. Technically, 401(k) plans now have the flexibility to use the ADP generated by NHCEs in the previous year to test whether the current year’s ADP for HCEs is too high. This modification was not included in the current analysis since the database is temporarily limited to 1996 contribution information.

xvi A third method will be attempted as soon as the database is expanded to include information on eligible employees that choose not to contribute. Although the two methods used in these illustrations produce conservative estimates of the impact, they are not precise in that we are currently unable to observe non-participant eligibles that drag the ADP for NHCEs down further than HCEs.

xvii Technically, the smaller plans are more likely to have no HCEs among their participants and therefore relatively immune to the impact of a competing governmental plan on their ADP test. This influence gives way to the fact that larger plans appear to have more generous matches and thus are less likely to have NHCEs considered to be "at risk."

Contacting EBRI | Copyright & Disclaimer