House Committee on Ways and Means, Subcommittee on Social Security

Trustees Reports on Social Security: The Dangers of Legislative Micro-Management of Statistical Reports

Mr. Chairman:

Thank you for the invitation to testify on the quality of information provided to workers and citizens through the annual Social Security statements and the reports of the Trustees of the Social Security program.

The initiation of annual information reports is a major and constructive innovation, as you stressed in your announcement of these hearings. The Trustees Reports, which have been available since the inception of Social Security, are a remarkable source of information, unmatched as far as I know by analogous releases in any other country.

Although these reports are very good indeed, it is important and constructive to ask whether they could be better. There has never been a data source that analysts did not believe they could improve in some way. And I shall not break that chain. Improvements are possible. For that reason, you are to be congratulated on scheduling hearings to consider possible improvements.

Having said that, I think that legislative micromanagement of these reports is more likely to reduce their quality than to improve it. Congress showed great wisdom in mandating these reports. But with even greater wisdom it left the design of these reports to non-political professionals. These civil servants—in my view among the most dedicated and capable in the federal government—have made and continue to make annual improvements and modifications based on consultation with governmental and nongovernmental professionals.

Before turning to the specifics of these reports, I want to highlight two facts. First, the debate on Social Security reform has benefited from the agreement by most participants to work from the same set of estimates about the long-term financial status of the program. These estimates have driven home the two key statistical facts about the status of Social Security. First, the system is currently running large cash flow surpluses and will continue to do so for many years. Second, the program faces a projected long-term deficit.

We all realize that any changes in the program will be phased in gradually. The proper mind-set for Social Security reform—to borrow a term immortalized by the Supreme Court in another context—is that we should move with all deliberate speed to enact the reforms on which Americans and their representatives can agree.

My second point is that agreement on the two key facts I just mentioned has permitted debate to focus on matters that are more important than finances and have little to do with them—whether the Social Security system as currently designed is the best way to serve the purposes of social insurance—to assure basic income to American retirees, the disabled, and survivors. My own view is that it is well designed for that purpose. Others disagree. But the key point is that we can focus on that debate. For the most part, we have avoided a wonkish numbers squabble. The prospects for resolution of the debate on structural reform of Social Security will diminish sharply if we become occupied by squabbles about the numbers. If the public begins to think that the experts cannot even agree on what the numbers are, they will either get diverted from the more important issues or they will get bored and tune out. We should avoid a situation in which dueling estimates contend for public attention. We should make sure that the numbers used in public debate are done carefully, according to reasonable assumptions and are presented in a manner that the public can readily understand.

So much for homilies. Now I should like to apply these principles to specific suggestions that have been advanced by various critics of the current estimates as contained in the Trustees Reports or the annual statements sent to workers. Because many assumptions go into these projections, there are many points of potential debate. I shall focus my remarks on just two. If you wish to go into others, I should be glad to respond to your questions.

Mortality Rates

The Technical Panel on Assumptions and Methods that reported last November urged the Trustees approximately to double the rate of improvement in mortality rates assumed in projecting long term costs. The 1999 Trustees Report assumed faster declines in mortality rates than have been observed since the early 1980s, but slower than rates of improvement measured over longer periods. The Technical Panel's recommendation was based on analyses by reputable demographers and others who served on the panel. Not all outside experts share the Panel's views, but the projection of sharply improved longevity can certainly be defended. Great medical advances lie in the future. They could greatly extend life expectancy. Unfortunately, as I have noted, the most recent trends in mortality rates are not so encouraging—or discouraging, from the standpoint of Social Security's finances.

For experts to miss a turning point or to expect one that hasn't occurred yet is not unusual. Nor is it unusual for them to change their minds. The problem is that experts often guess wrong.

  • In the 1930s, few foresaw the baby boom. In the 1950s and 1960s few foresaw its end.
  • Five years ago, budget experts foresaw large and explosively growing budget deficits. Today they foresee virtually permanent budget surpluses.
  • Three years ago, economists believed that unemployment much below 6 percent would trigger explosively growing wage and price inflation. Many now maintain that 4-percent unemployment without inflation can go on indefinitely.
  • Some now believe that productivity growth will remain so high that little of the projected long-term deficit in Social Security will remain. Other's think current rates cannot be sustained or even believe that they will relapse to the dismal rates that prevailed from 1973 through 1996.

The Social Security trustees often have to decide what to do when expert judgment is not yet confirmed by the facts. The prudent course of action, I believe, is to take account of the experts' views, modifying assumptions a bit at first. Then, if evidence confirms the experts' views, further adjustments are in order.

With the amply documented history of real howlers by experts in mind, I believe that we should admire the prudence with which the Trustees handled the Technical Panel's recommendations on assumptions regarding mortality rates. In the 2000 report, the Trustees increased the assumed rate of improvement in mortality by one-third—a sizeable shift, but much less than the huge shift suggested by the Technical Panel. Presumably, they found in the promise of the biological revolution sufficient reason to increase assumed longevity. But they sensibly decided to wait for the revolution to show up in the numbers before making even larger shifts.

As I noted, expert judgment may change a lot—and quickly. If the Trustees promptly and completely incorporated the latest, best wisdom of economic or demographic experts, the 75-year projections would oscillate crazily from year to year. The result would be alternate bouts of euphoria and panic and a breakdown of trust in the projections. Instead of conveying the important message that Social Security faces a projected long term deficit that we should close promptly, people would look at current surpluses and shrug off the long-term projections as undependable.

I have focused on projected mortality, but the same cautionary note applies to other key assumptions. Productivity growth is dramatically above what the actuaries assume. Rather than incorporating these higher levels into long-term projections, the Trustees raised assumed productivity growth a modest 0.1 percentage point. The "new economy" is great fun. Let's enjoy it as long as we can. But only if it lasts a decade or more, are the Trustees likely to incorporate it fully into long-term Social Security projections. Given the history of trend reversals, the Trustees' practice of cautious and highly damped adjustments to new events is the only prudent course.

Rates of Return Oceans of financial data wash over us. Every mutual fund routinely reports its annualized rates of return for various past periods. Business school professors have computed rates of return on common stocks, 30-year Treasury bonds, Treasury bills, and just about everything else. Shouldn't Social Security provide such data to each worker and as part of the annual Trustees Reports' The answer, I believe, is a clear and unambiguous "No!" There are three reasons:

  • the enduring effects of Social Security's past history on the continuing operation of the system;
  • the character of the Social Security benefit package; and
  • the fallacies that arise from choosing incorrect perspectives for measuring rates of return.

Social Security is a combination of annuities, insurance, and income redistribution. Furthermore, the program's history shapes its present and future. For workers who live to retirement, Social Security is an annuity. For people who become disabled or die, it is insurance that provides payments to workers or their dependents. For low-earners and large families, the program provides social assistance, financed by high earners and small families. Furthermore, the program today must deal with the consequences of the decision to pay early beneficiaries larger benefits than their contributions merited. Each of these factors should be taken into account in computing rates of return. Yet no currently available analysis has done so.

Past policy. The extremely generous benefits that Social Security paid to early beneficiaries were financed by payroll taxes collected from active workers. As a result, reserves accumulated for these younger workers were tiny. Benefits subsequently paid to those younger workers have had to be financed by later workers. The reserves that were not accumulated on behalf of today's workers are the unfunded liability of the current system. Unless Congress reneges on these benefits—which is inconceivable—it will be necessary to collect taxes to pay those benefits. This obligation is inescapable, whether Social Security continues in its current form or is replaced. No mutual fund or asset group carries such an unfunded liability. Should some system of individual accounts replace Social Security, taxpayers would not escape this unfunded liability.

For this reason, it would be misleading, at best—meaningless, at worst—to compare rates of return on Social Security with returns on individual accounts that omit the cost of paying this unfunded liability. No such individual accounts plan could exist, unless Congress was prepared to renege on promises to current workers. Yet, such comparisons would be natural and would even be encouraged by the ignorant.

The benefit package. Several analysts have presented estimates of the rate of return on Old-Age Insurance—that is, of retirement benefits. Unfortunately, the estimates of returns on Old-Age Insurance are incomplete and misleading. In contrast, no one has ever presented estimates of the rate of return on Social Security as a whole. The reason is that reliable data and defensible methods for estimating the rate of return on Social Security as a whole are currently unavailable.

Let me make clear that I am not throwing stones simply at others. More than a quarter century ago, I prepared the first empirical estimates of the rate of return on Old-Age Insurance. They were the best I could do at the time, but they weren't good enough, as the discussant of my paper pointed out. The problem was that, like many later analysts, I was forced to make a crude and highly distorting assumption. Because Social Security consists not only of Old-Age Insurance but also of Survivors Insurance and Disability Insurance, one cannot attribute the entire payroll tax to support of Old-Age Insurance. On the average about two-thirds of the payroll tax goes to support Old-Age Insurance.

I and, subsequently, others have assumed that each worker bears the burden of about two-thirds of the payroll tax. We have compared the present discounted value of that tax with the present discounted value of the expected retirement benefits that workers receive. The interest rate that equilibrates these present expected values was the rate of return.

The purpose of my study was to test an hypothesis originally presented by Milton Friedman, that blacks do less well than whites under Social Security. The problem, Friedman noted, was that blacks have briefer life expectancies and collect retirement benefits for a briefer period than do whites. This condition, he suggested, more than offset the disproportionately generous benefits paid to low earners, among whom blacks were—and are—over-represented. I found that blacks' shorter life expectancies about offset the effects of the benefit formula and that rates of return for blacks and whites were rather similar.

My discussant pointed out that I had not found what I had thought I had found. The problem, he pointed out, is that Social Security is an integrated package of retirement, survivors, and disability insurance benefits. The shorter life-expectancies of blacks mean that they receive proportionately more survivors benefits than do whites. In addition, the shorter black life-expectancies are correlated with higher disability rates, so that blacks receive proportionately more disability benefits than do whites. For this reason, he pointed out, I was wrong to assume that the same fraction of the payroll tax supports retirement pensions for blacks and whites. To compute the rate of return for blacks as a group or for whites as a group, one would have to take into account the differential value of disability and survivors benefits. I hadn't done that. So, my computations may have been interesting, but they did not mean what I thought they meant.

Data to solve this problem were lacking a quarter century ago, and they are lacking still. This fact has not prevented analysts from repeating the same mistake I made and, unfortunately, adding new ones of their own. The Heritage Foundation, for example, compared rates of return to blacks and whites using life-expectancies at birth, rather than examining when labor force entrants died and how long those who reached retirement age actually received benefits.

The result was a gross distortion in the relative lengths of benefit payments. For example, Heritage estimated that the average duration of benefit receipt for black men aged 20 in 1997 would be 2.2 years. The true expectation was 8.1 years. This distortion severed any connection between the calculations and reality. Both former chief actuary Robert Myers and current deputy chief actuary, Stephan Goss have written devastating critiques of this egregious study.

If one should not present bad estimates of returns on Old-Age Insurance as estimates of returns on Old-Age Insurance and certainly not as returns on Social Security, the question remains as to whether it is possible to prepare good estimates of returns on Social Security. I imagine that one day such studies will be done. But they haven't been done yet, and a number of very difficult problems will have to be solved before such studies merit inclusion either in the Trustees Reports or in annual statements to workers. Among these questions are the following:

  • What is full protection against inflation risk worth? No private asset provides this protection. Hence, it is necessary to value this protection if fair comparisons are to be possible between Social Security and other assets.
  • What is full protection against financial market risk worth? All private assets analogous to individual accounts carry financial market risk, which the account holder must bear. Social Security spreads such risks over all workers and through time. How much is such risk diversification worth?
  • How should one value political risk? Social Security is subject to political risks that benefits or tax rates may be changed. Individual account holders are subject to the risk that income or estate tax rules may be changed in ways that affect the value of their accumulations. How should one value each kind of risk?
  • How much is the insurance protection in Disability and Survivors Insurance worth? People regularly pay premiums that exceed the expected pay-out for property and casualty, disability, life, and health insurance. They do so because such insurance spares them risks they feel ill-equipped to bear. This fact means that the value of such insurance exceeds its cost although the payments are less than the cost. How should one estimate this extra value in the case of the insurance protection provided by Social Security against loss of income from disability or premature death of a breadwinner.
  • How much is the "earnings" insurance provided by Social Security worth? When workers enter the labor force, it is usually unclear whether they will be high or low earners. Social Security's benefit formula provides higher replacement rates if earnings turn out to have been low, rather than if they turn out to have been high. This is a form of earnings insurance. How much is this insurance worth?

None of these questions has been answered well with respect to Social Security. Many more equally difficult questions could be listed. Of course, any good analyst can come up with answers to almost any question. Unfortunately, the answers will differ enormously. This fact means that responsible people will make radically different estimates. To mandate reporting of rates of return would therefore be to mandate highly uncertain, even arbitrary, estimates.

Perspective for Measuring Rates of Return. When workers enter the labor force, they typically do not know anything more than their race and sex. They do not know what they will earn. They do not know when or whether they will marry. Or get divorced. Or remarry. Or have children, and, if so, how many. They do not know how often or how long they will be unemployed. They do not know if they will become disabled or how long they will live and, if married, how long their spouse will live. They do not know future asset yields or rates of inflation. They do not know how averse they will be to risk and how much they will pay to avoid these risks.

Not only do workers not know these things, neither does any actuary or economist. Yet without knowing these things it is impossible—not just difficult, but completely impossible—to construct meaningful estimates of the rates of return on a complex set of contingent payments such as Social Security. Let me be clear. Some analyst can make assumptions about each of these variables, plug them into a computer and come up with a numerical answer. I am simply asserting that this numerical answer would not be worth the powder to blow it to hell.

Of course, as workers age, they will learn answers to many of these questions. They will still not know how long they will live or whether they will become disabled or what rate of inflation will prevail when they become beneficiaries. But they will know if they are rich or poor; married, divorced, or single; parents or childless; well or ill; and so on. Once they know these things, however, they will—by definition—no longer be benefitting from the protections that are an essential part of Social Security—the insurance against these risks.

For that reason, it makes no sense to measure the value of Social Security to, say, a fifty- five year old by the taxes that worker has paid and will pay and the benefits that worker will receive. To see why, consider a fifty-five year old homeowner who has had fire insurance since buying the home at, say, age twenty-five. This homeowner has paid premiums for thirty years and will continue to pay premiums. It would surely be complete nonsense to value that insurance by comparing the present value of all premiums the homeowner has paid and will pay against the expected pay-outs in the event of a fire. Most of the value of the insurance has already been achieved, in the form of peace of mind over three decades. The best outcome would be one in which the homeowner looks back at a lifetime of premium payments and no cash benefits. The financial rate of return in that case would be negative infinity. The economic return is presumed to be positive; otherwise people would not have bought the insurance.

In the same sense, most of us would like to look back at a lifetime of payroll tax payments and to find that we never collected disability insurance, our children never collected survivors benefits, and we did not benefit from the high replacement rates paid to low earners.

What this means is that the only potentially meaningful calculations of rates of return would have to be made for new labor force entrants. At that point, workers know their race and sex, but not much else. If one thought one could answer the questions I listed in the preceding section—as well as many others that I could have included—then one might prepare estimates of rates of return for race/sex groups. But I don't think that these questions can be answered with satisfactory confidence. And so I come to my conclusion—mandating estimates of rates of return would be ill considered. Analysts can make such estimates, but they would not be meaningful.

Conclusion

I have focused on two issues: legislative mandates regarding particular assumptions and legislative mandates regarding estimates of rates of return as part of annual reports to workers or of the Trustees Reports. Each issue illustrates a larger class of issues. My major purpose is to warn about the dangers of legislative micro-management of the contents of statistical reports, such as that of the Social Security Trustees, or annual reports to workers. The Trustees Reports are subject to annual review by outside experts. The methods used in making these reports have changed. Statistical capabilities change over time as new analytical methods arise, as data sources expand, or, as has been increasingly the case in some areas of late, as data sources vanish because of penny-wise cuts in budgets of statistical agencies.

In addition, I hope that the Trustees will continue the work now under way to construct representative life-time earnings profiles. I would urge them to reconsider the linkage of various assumptions now grouped in the high- or low-cost projections. But these and other problems are best addressed by the Trustees with the aid of outside professional consultants. Congress has mandated, and will mandate, one-time studies of particular questions. Where these questions are sensitive—and virtually everything seems to be sensitive to someone—it often mandates a responsible organization whose objectivity is not in question, such as the National Academy of Sciences or the Institute of Medicine to carry out such a study. If Congress wishes to explore in greater depth the issues raised in these hearings, it might consider a similar approach, mandating a study by the National Academy of Sciences or the National Academy of Social Insurance.