rethinking social insurance (march...

Rethinking Social Insurance†

By MARTIN FELDSTEIN*

Social insurance is a subject I have beenstudying for nearly 40 years. The intellectualand policy revolution in social insurance that isoccurring around the world is among the mostsignificant and positive developments of currenteconomics.1

Social insurance programs have become themost important, the most expensive, and oftenthe most controversial aspect of governmentdomestic policy, not only in the United Statesbut also in many other countries, including de-veloping and industrialized nations. In theUnited States, these programs include SocialSecurity retirement, disability, and survivor in-surance, unemployment insurance, and Medi-care insurance for those age 65 and older.Together these programs accounted for 37percent of federal government spending andmore than 7 percent of GDP in 2003. Theseratios have increased rapidly in the past and areprojected to increase even faster in the futurebecause of the more rapid aging of thepopulation.

I will discuss how the major forms of socialinsurance could be improved by shifting to asystem that combines government insurance withindividual investment-based accounts: unem-ployment insurance savings accounts (UISAs)backed up by a government line of credit, per-

sonal retirement accounts (PRAs) that supple-ment ordinary pay-as-you-go Social Securitybenefits, and personal retirement health ac-counts (PRHAs) that finance a range of Medi-care choices. I think that such reforms wouldraise economic well-being and are also appeal-ing on broader philosophical grounds.

Several nations are now doing this for theirretirement programs, including such diversecountries as Australia and Mexico, England andChina, Chile and Sweden (Feldstein, 1998a;Feldstein and Horst Siebert, 2002). The focusby governments around the world on socialinsurance pension reform is driven in part by therealization that the aging of their populationsimplies that the tax rates required to fund socialinsurance pension benefits will rise rapidly ifthe programs are not changed.

The impetus for broader social insurance re-form comes from the recognition that existingprograms have substantial undesirable effects onincentives and therefore on economic perfor-mance. Unemployment insurance (UI) programsraise unemployment. Retirement pensions induceearlier retirement and depress saving. And healthinsurance programs increase medical costs. Gov-ernments are driven by a desire to reduce theeconomic waste and poor macroeconomic perfor-mance that these disincentives create and to avoidthe resulting tax consequences, as well as theincreased tax cost, of the aging population.

Economic research has helped policy offi-cials to recognize these undesirable effects andto redesign social insurance programs. The paceof reform and the nature of the program changesdiffer from country to country, reflecting initialconditions and local political realities. Reformsare inevitably only partial and part of an ongo-ing process. But the reforms generally make theprograms more economically efficient, provid-ing more protection relative to the financialcosts and the economic distortions. I will exam-ine some of the favorable changes that havealready occurred in U.S. unemployment, retire-ment, and health care programs.

Before looking at these specific types of so-cial insurance, I want to discuss three general

† Presidential Address delivered at the one hundred six-teenth meeting of the American Economic Association,January 8, 2005, Philadelphia, PA.

* Department of Economics, Harvard University, Cam-bridge, MA 02138 (e-mail: [email protected]) and Presi-dent of the National Bureau of Economic Research.Although there are some references in this paper to my ownwork and to that of others, there is no attempt to survey thelarge literature on social insurance. I have posted a full listof my own papers on social insurance at www.nber.org/feldstein.

1 Many of the ideas presented here were developed inteaching my graduate public economics course at Harvardsince 1969. I am grateful to the students in those classes forstimulating discussions and to Kathleen Feldstein, JohnGruber, Caroline Hoxby, Lawrence Kotlikoff, Jeffrey Lieb-man, James Poterba, Andrew Samwick, and Andrei Shleiferfor comments on an earlier draft.

1

questions. Just what is social insurance? Whydo, or should, we have such programs? Andwhat are the principles by which such programsshould be evaluated and redesigned?

I. Social Insurance and Welfare Programs

The word “insurance” is used to describethese transfer programs because they deal withrisks: the risk of job loss, of health care ex-penses, and of inadequate assets during retire-ment. But social insurance is very differentfrom private insurance. The key distinction isthat participation in social insurance programsis mandatory or is induced by substantial fiscalsubsidies.

Social insurance programs are also very dif-ferent from welfare programs. Welfare benefitsare means tested, i.e., they are paid only to thosewith incomes (and assets) below some level. Inthe United States, these means-tested programsinclude Medicaid, food stamps, subsidizedhousing, school lunches, and others.2 In con-trast, social insurance programs are “event con-ditioned.” Benefits are paid when some eventoccurs in an individual’s life regardless of theindividual’s income or assets. Unemploymentbenefits are paid to those who lose their jobs andMedicare benefits to those who are ill and over65. Social Security benefits are available tothose over age 62, disability benefits to thoseunable to work, and survivor benefits to thewidows and children of deceased workers.

Unlike welfare programs, social insuranceprograms are not designed to be vehicles forincome redistribution. Although some fractionof social insurance outlays is paid to those withlow incomes, most of the benefits go to middle-and higher-income households. This is particu-larly true in the United States, where cash ben-efits to retirees and the unemployed are positivelyrelated to previous earnings and where health careis provided by private hospitals and physicians,even when financed by social insurance.

Social insurance may appear to be redistrib-uting income to the poor because benefits arepaid to those who are temporarily poor due tothe event that triggered the payment of benefits.This ignores the permanent or lifetime income

of these recipients. It also ignores the effect ofthe social insurance on the incentive to accu-mulate funds for these rainy days. Social Secu-rity benefits that replace 50 percent or more ofafter-tax pre-retirement income reduce signifi-cantly the incentive to save for old age andtherefore depress income in retirement. Unem-ployment benefits with high replacement rateshave a similar effect on saving to finance spellsof unemployment.

The lack of redistribution is well illustratedby the Social Security retirement program. Lastyear, a new retiree who had annual earnings ator above the Social Security program maximumtaxable amount ($87,900 in 2004) for at least 35years received a benefit of $21,900. In contrast,someone with lifetime earnings in the middle ofthe earnings distribution received only abouttwo-thirds as much in retirement benefits. Andsomeone with low earnings (i.e., 45 percent ofthe average wage) received benefits of less than$9,000 a year. This lack of redistribution iscompounded by the rules governing benefits tospouses and widows. A retiree who previouslyhad maximum taxable income and who retiredwith a dependent spouse received more than$32,000 a year from Social Security, while thewidow of a low-income earner would receiveless than $9,000 a year.

The Social Security program appears to beredistributive because everyone pays the sametax rate, while the ratio of benefits to lifetimeearnings is designed to fall as those earningsrise. In practice, however, this apparent redis-tribution is offset by the longer expected life ofhigher-income individuals, their increased useof spousal benefits, and the later age at whichthey begin to work and to pay taxes. Researchby Jeffrey Liebman (2002), based on a largesample of actual individual earnings histories,showed that less than 10 percent of Social Se-curity benefits represented net redistributionacross income groups within the same birthcohort. Julia Lynn Coronado et al. (2000)showed that the combination of taxes and ben-efits for the Social Security program leaves thelifetime Gini coefficient of the population’s in-come essentially unchanged. In addition, thegeneral equilibrium effects of Social Securitytilt the pretax distribution of income towardhigher income individuals by reducing capitalaccumulation, which in turn lowers real wagesand raises the return to capital.

2 See Moffitt (2003) for detailed studies of a variety ofwelfare programs.

2 THE AMERICAN ECONOMIC REVIEW MARCH 2005

Unemployment insurance also does not redis-tribute to the poor. In Massachusetts, a state con-sidered to have a very generous UI program, theUI benefits were financed in 2003 by a payroll taxon only the first $10,800 of earnings (with a zeromarginal tax rate above that level), while basicbenefits were 50 percent of previous wages up tomore than $50,000 of wages per year.

An individual who earns $50,000 a year paysthe same tax as someone who earns $11,000 ayear but would receive benefits that are nearlyfive times as high.3 Taken by itself this wouldmean a substantial redistribution from low-wage earners to higher-wage earners. Moreover,since benefits are paid only to individuals whohave earned some minimum amount during thepast year, those with long spells of unemploy-ment may not be eligible for any benefits at all.

Although the same Medicare rules apply toeveryone over age 65, higher-income seniorsoften get substantially more benefits than thosewith lower incomes. Mark McClellan andJonathan Skinner (1997) concluded that Medi-care produced net transfers from the poor to thewealthy as a result of the higher annual expen-ditures and longer survival times of wealthierMedicare beneficiaries. In the same spirit, Skin-ner and Weiping Zhou (2004) found greater useof mammography screening, diabetic eye ex-ams, and other indicators of good care amonghigh-income Medicare groups than amongthose with lower incomes.

The very high level of spending on the mid-dle class social insurance programs hurts thelow-income population in another way: by put-ting a drain on the government budget in a waythat reduces the funds available for helping thepoor. Social insurance programs cost $800 bil-lion in 2003, while federal spending on allmeans-tested programs, except Medicaid, wasless than $150 billion.4 Over the past four de-cades, the spending on means-tested programs(except Medicaid) has remained relatively con-stant (rising from 1.0 percent of GDP to 1.3percent of GDP) while the social insurance pro-

grams that are not means tested rose from 2.7percent of GDP to 7.4 percent of GDP.

The negative effect of social insurance spend-ing on means-tested programs is not only anobserved fact but is also what optimal tax theoryimplies. The deadweight burden of an extradollar of taxes increases with the share of incometaken in taxes. The high level of taxes that isneeded to finance middle-class social insuranceprograms therefore increases the deadweight bur-den of any incremental taxes that would be used tofinance means-tested poverty programs. The largesocial insurance programs thus reduce the optimalsize of means-tested poverty programs.

II. Why Social Insurance?

Some writers see social insurance in broadphilosophical terms, reflecting their specificviews of the appropriate role of government insociety. One such view, more common in Eu-rope than in the United States, is that socialinsurance should be judged by its contributionto social solidarity, i.e., to the sense that all ofthe individuals in the nation are, in effect,viewed as a single family and treated equally.This leads to the principle of uniform health carefor all, although this is more often an assertedpolitical goal than a practical reality. Similarly,they may reject any role for company-basedprivate pensions in order that all workers par-ticipate in a common pay-as-you-go state plan.

The opposite philosophical view is that theprovision of health care or retirement income isnot a legitimate role for government because itforces individuals to participate in a commonprogram rather than taking personal responsi-bility and making decisions that reflect theirown preferences. Milton Friedman’s Capitalismand Freedom (1962) is the classic statement ofthis view that social insurance programs areinappropriate because they infringe upon indi-vidual liberty.

The social solidarity view is often combinedwith the statement that individuals are incapableof making the complex decisions required toplan for retirement income or to choose healthinsurance or health care. The opposite viewemphasizes that individuals differ in their tastesand are better able than governments to judgewhat is in their own best interest.

I believe in the diversity of individual pref-erences and the ability of most individuals to act

3 The unemployment tax is technically levied on theemployer but the incidence is likely to be primarily on theemployee.

4 Although Medicaid is a means-tested program, morethan half of its outlays are for nursing home care for thevery aged rather than care for those with low lifetimeincomes.

3VOL. 95 NO. 1 FELDSTEIN: RETHINKING SOCIAL INSURANCE

in their own self interest. But I also believe thatthere is a role for government that justifies theprovision of social insurance benefits. I come tothis conclusion on utilitarian grounds ratherthan from any philosophical commitment to so-cial solidarity.

There are two distinct reasons for providingsocial insurance. Both reflect the asymmetry ofinformation. The first is that asymmetric infor-mation weakens the functioning of private in-surance markets. The second is the inability ofthe government to distinguish between thosewho are poor in old age or when unemployedbecause of bad luck or an irrational lack offoresight from those who are intentionally“gaming” the system by not saving in order toreceive transfer payments. Both problems showthat the case for social insurance cannot be re-jected simply by arguing that such programs forcepeople to act against their own best interests.

But these problems of asymmetric informa-tion or any other market failures do not neces-sarily justify government action. While aperfect and benevolent government would bebetter than a private market burdened by marketimperfections, actual governments are neitherperfect nor necessarily benevolent. Political ac-tors do not maximize a social welfare function,but reflect political pressures and bureaucraticpreferences. Moreover, social insurance pro-grams impose costs that must be weighedagainst the benefits of overcoming market im-perfections. Both require empirical evaluation.

Consider first the asymmetry of informationin insurance markets. To be specific, considerthe case of private annuities. If individuals canbuy annuities on actuarially fair terms they mayincrease their expected utility by annuitizingtheir assets at retirement. But if individuals dif-fer in their life expectancy and know moreabout their mortality prospects than the insur-ance company can learn, those with shorter lifeexpectancy will want to annuitize a smallerportion of their wealth. Insurance companieswill recognize the resulting self-selection andoffer annuities with premiums that reflect themortality rates of the long-lived individualswho are their most likely customers. This pro-duces a downward spiral in the demand forannuities that is limited only when, at somepoint, the risk-reducing value of annuitizingoutweighs the less than actuarially fair pricingof individual annuities.

A mandatory social insurance program like tra-ditional Social Security circumvents this asymme-try of information by providing everyone with aretirement annuity rather than a lump sum at re-tirement age. But whether this is better than animperfect private annuity market, in which someannuitize little and others not at all, depends on theimplicit rates of return available on the socialinsurance annuity, on the private annuity, and onnon-annuitized saving. It also depends on the de-gree of diversity in preferred spending patterns inretirement and in attitudes about bequests, sincecomplete annuitization at retirement would notpermit the purchase of retirement homes or othermajor consumer outlays or the making of bequestsor inter vivos gifts.

The problem of information asymmetry inprivate annuities could be reduced if individualspurchased annuities at relatively young ages,before they could accumulate much informationabout their own likely mortality risks in old age.Alternatively, a mandatory annuity could bemore attractive if it were based on the higherreturn available in an investment-based pro-gram rather than in a mature pure tax-financedpay-as-you-go program.

Two conclusions follow from this. First, theexistence of asymmetric information may jus-tify a social insurance program (a governmentannuity in this case) but does not necessarily doso. The case for a mandatory annuity programdepends on calculations that could be done butthat have not yet been done. Second, the appro-priateness of a social insurance program and itsoptimal size can be increased if the cost of thesocial insurance option is reduced, somethingthat depends on how it is financed.

Consider now the second form of asymmetricinformation that might provide a rationale for asocial insurance program: the government’s in-ability to distinguish those who are poorthrough bad luck or inadvertence from thosewho deliberately choose to act in a way thatleads to eligibility for free benefits. A primaryreason for social insurance programs is thatsome individuals would not act in their owninterest, saving far too little for their retirement,for health care after they are no longer working,or to finance consumption when they are unem-ployed. Although some economists may rejectthe likelihood of such irrational behavior as abasis for policy analysis, as individuals we allrecognize that such irrationality exists in prac-


tice. Recent work on behavioral economics hashelped to make the possibility of such irratio-nality or myopic behavior a part of mainstreameconomics.

But such departures from rational saving by afraction of the population need not justify thegeneral provision of social insurance benefits.Why not simply provide means-tested benefitsinstead of the universal provision of social in-surance benefits? The primary reason for notdoing so is that some rational and farsightedindividuals would be induced by a means-testedsystem to act in a way that allows them toqualify for benefits. Doing so would impose taxcosts on the rest of the population that couldmake overall well-being lower than in a univer-sal (i.e., not means-tested) program.

Consider a simple example of a means-testedretirement program (Feldstein, 1987). Assumethat some fraction of the population is myopicand would not save anything for retirement. Ameans-tested program would provide a benefitfor all such myopic retirees. How would ratio-nal working-age individuals respond to such asystem? They would have the choice of eithersaving for their own retirement or consuming allof their income before they retire and receivingthe means-tested benefit. The potential means-tested benefit acts as a kind of tax on saving,reducing the incremental retirement consump-tion that saving would produce. A rational in-dividual would decide whether to act as if he orshe is myopic by comparing the lifetime utilitieswith optimal positive saving and with no sav-ing. Those with relatively high incomes wouldnot be tempted by the means-tested benefit. Butothers with lower incomes would have higherlifetime utility by increasing their consumptionduring working years even if the means-testedbenefits would only provide lower consumptionduring retirement than optimal saving wouldallow. Although they would achieve higher life-time utility through their action, their benefitswould be financed by tax-financed transfers,which would make others worse off.

There is no way for the government to dis-tinguish between the genuinely myopic andthose who are rational utility maximizers gam-ing the system. The government could, in prin-ciple, set the means-tested benefit so low thatvery few rational individuals would be tempted.My judgement is that our relatively affluentsociety would not accept that policy. The

means-tested benefits would be set at a higherlevel that would tempt many rational individu-als to save nothing.

A policy of forcing everyone to save for hisor her own retirement would eliminate the prob-lem of those who game the system. The onlyadverse effect of such a policy is that someindividuals might be required to shift more oftheir lifetime consumption to their retirementyears than they would prefer. For them, part ofthe mandatory saving would be a tax to theextent that they valued the saving less thancurrent consumption.

The choice between such a mandatory savingplan—essentially a kind of investment-basedSocial Security pension—and a means-testedbenefit should depend on numbers. How manypeople would receive means-tested benefits?How much deadweight loss would be involvedin financing those benefits? How many peoplein a mandatory saving plan would have to pro-vide more for their retirement than they want?

In the absence of such a mandatory invest-ment-based option, the policy choice is betweena means-tested program and a universal pay-as-you-go retirement benefit. Such a pay-as-you-go plan forces all individuals (after theinitial generation) to receive a lower rate ofreturn than they could obtain on private saving.As such, it also imposes a tax on labor incomesince each extra dollar of earnings would inducean additional pay-as-you-go tax liability. Thereduction in saving that is induced by the pay-as-you-go system also causes a fall in capitalincome and therefore in corporate and personaltax revenue that requires higher marginal taxrates to recover the lost revenue.

Both of these examples of asymmetric informa-tion show that a social insurance program may bean appropriate response to a market failure but thatit need not be. Even when there is a market failure,it may be better to do nothing or to have a means-tested welfare program. The choice depends inpart on the relative costs of the different options,and those in turn depend on the design of thepotential social insurance program. Whether sucha program is investment-based or purely tax fi-nanced on a pay-as-you-go basis is an importantfeature of that cost.

Economists can help to evaluate these choicesby estimating the relevant costs and benefits of thedifferent options. My own conclusion is thatinvestment-based social insurance programs for


retirement, unemployment, and health care of theretired population are more appropriate than pay-as-you-go programs, means-tested programs, or apolicy of doing nothing.

There is an important political economyreason for economists to work on improvingthe design of social insurance programs ratherthan advocating means-tested programs forunemployment and old age. Elected govern-ments will inevitably seek to create universalbenefits to capture political support from thelargest possible majority of voters. Otto vonBismark introduced social insurance in Prus-sia in 1881 in an attempt to win support forhis conservative government and to fend offthe appeal of the nascent social democrats.Even if it were economically desirable to doso, economists could not prevent the spread ofsocial insurance by arguing that means-testedprograms would be more efficient. If econo-mists don’t analyze the effects of social in-surance programs and recommend rules thatreflect good economics, the political processwill inevitably produce inferior programs.

III. Principles of Social Insurance

Accepting that there is a reason for mandatorysocial insurance programs does not imply the ap-propriateness of the programs that we have inher-ited from the past. Today’s Social Security andunemployment insurance were enacted nearly 70years ago. Economic conditions, administrativetechnologies, and assumptions about economicbehavior have all changed dramatically since then.And yet during these past 70 years, the key socialinsurance programs have expanded without fun-damental change.

Before I consider some of the specific waysin which our basic social insurance programscan be reformed and strengthened, I want todiscuss broader principles that can help us tothink about each of the specific programs. I’llbegin with three fundamental political princi-ples and then turn to four economic principles.

A. Three Political Principles

Political principles involve value judgmentsto a greater extent than the economic principlesto which I will turn later. I can explain thepolitical principles that shape my view aboutappropriate reforms but I cannot prove that they

should determine policy. You may or may notagree with them. Of course, you might agreewith me about specific reforms even if youreject some or all of these principles.

Permitting Individual Choice.—Individualsdiffer in their preferences. We do not all havethe same risk aversion, the same time prefer-ence, the same relative taste for goods and lei-sure. Letting individuals choose among optionsin a way that reflects their individual prefer-ences should be an important aspect of socialinsurance design. For Milton Friedman, suchfreedom to choose is paramount. For me, it isimportant but not decisive. In cases whereasymmetric information creates serious effi-ciency problems, I might restrict that choice.But I prefer to allow as much choice as possible.I think that allowing individuals to make theirown choices is morally correct and generallyimproves individual, and therefore social,well-being.

But allowing choice means that programsshould be designed so that choice enhanceseconomic efficiency rather than creating dead-weight losses. A good example of such a pro-gram redesign was the Social Security reformthat introduced the actuarial adjustment forearly and delayed retirement in a way that, inprinciple, will allow individuals to decide whenthey will start collecting benefits without chang-ing the actuarial present value of their benefits.5

Creating Program Transparency.—Socialinsurance programs involve complex rulesabout the benefits to be received, the taxes to bepaid, and the link, if any, between them. Whoamong you is confident about even the mostbasic Social Security rules that determine ben-efits at retirement? If you are a man, whatbenefit would your wife receive if she collectson her own rather than as your spouse? Howwould that change if she earned more or workedanother year? If she decides to retire at age 62rather than 65? I’m told that there are more than2500 separate rules in the Social Securityhandbook.

The complexity of the rules weakens the per-ceived link between the payroll taxes paid and

5 This adjustment will provide actuarially equivalentbenefits with a 3 percent real rate of return.


subsequent benefits. Many employees may sim-ply regard their Social Security payroll tax assimilar to the income tax, thereby increasing theperceived marginal tax rate and raising thedeadweight loss of the tax.

Lack of transparency also permits programsto have effects that might not be politicallyacceptable if they were more explicit. For ex-ample, some defenders of the current SocialSecurity system believe it permits a substantialamount of redistribution that Congress wouldnot be willing to build into an investment-basedsystem of individual accounts. Although theSocial Security rules do not actually achievethat redistribution, the important political prin-ciple is that it is inappropriate in a democracy touse a deliberately opaque system to achieve aredistribution of income that would be rejectedif proposed in a more transparent way.

The Social Security program lacks transpar-ency because it is a defined benefit systemrather than a defined contribution plan of thesort that now characterizes most private pen-sions. Converting Social Security to a definedcontribution plan—even an unfunded “no-tional” system such as Sweden and Italy nowhave—would allow individuals to see the linkbetween their taxes and the resulting benefits.An explicit decision by Congress to supple-ment the contributions of low-income earnersin such a defined contribution plan wouldachieve income redistribution without a lossof transparency.

Recognizing Political Dynamics.—When weeconomists talk about policy design, we gener-ally think about enacting permanent reforms.But experience shows that legislated rules dochange and that the initial conditions influencethe path of that change. When designing a par-ticular program or advocating a particular de-sign, economists should recognize that somedesigns are more stable than others and shouldanticipate how a program might evolve.

The Medicare drug legislation enacted in2003 is a good example. Medicare beneficiarieswill pay the first $250 a year in drug expenses,followed by a 25-percent coinsurance rate to amaximum benefit limit. Patients must then pay100 percent of the drug cost up to $3600 inout-of-pocket payments (in 2006). Above that,Medicare will pay 95 percent of any additionaldrug costs.

This rather strange design was accepted tolimit the total cost of the plan while deliveringbenefits to a very large number of senior citizenvoters. An economically more rational planwith the same budget cost would have insured atleast part of the range that is currently uninsuredand kept total costs down by a larger deductible.But that would have had the political disadvan-tage of giving benefits to fewer individuals. Itseems likely that future legislation will addressthe residual insurance gap in a way that willraise the total cost of the program.

There is another and potentially more signif-icant aspect of the future evolution of this Medi-care drug program. If all of the drugs consumedby seniors come to be covered by governmentinsurance, there will be strong pressure to reg-ulate the price of those drugs. Such price regu-lation is, in turn, likely to discourage thedevelopment of drugs for those diseases thatparticularly affect the elderly. It would be sadlyironic if an insurance plan initiated to improvedrug access for seniors led ultimately to a re-duced availability of new drugs for this group.

B. Four Economic Principles

Let me turn now from these three politicalprinciples—permitting individual choice, creat-ing program transparency, and recognizing po-litical dynamics—to four economic principles.

Recognizing the Economic Effects of SocialInsurance Programs and Their Taxes.—Non-economists who write about social insuranceprograms often implicitly assume that socialinsurance programs do not affect the behaviorof beneficiaries or the overall performance ofthe economy. Evidence shows that the oppositeis true. Social insurance programs have impor-tant and sometimes harmful effects on theeconomy that are not fully recognized by thepublic, Congress, or the politically responsibleofficials.

A substantial volume of work during the pastquarter century has shown the various ways inwhich social insurance programs do affect indi-vidual behavior and the overall economy. Theseeffects include reducing national saving, induc-ing early retirement, raising the unemploymentrate, pushing up the cost of health care, andcrowding out private health insurance. Any se-rious evaluation of social insurance programs,


and any attempt to improve their design, shouldtake these effects into account.

There is, of course, controversy about themagnitude of these effects, just as there is aboutmost other economic parameters. Decisionsabout program design have to use the evidencethat is available, even if parameter estimatescome with substantial uncertainty. But there isclearly room for economists to use new data,new statistical methods, and new natural policyexperiments to improve our knowledge and,therefore, to improve policy design.

Adverse effects result from specific programdesigns and are not inherent in the goals of theprograms. For example, John Gruber and DavidWise (1999) showed that the rules governingretirement benefits induced early retirement inseveral European countries but that differentrules at different times and in other countriesdid not induce early retirement. The U.S. benefitrules that now specify an almost actuarially fairrelationship between benefits and retirementage reduces substantially the perceived bias infavor of early retirement.

More generally, social insurance programsnot only distort economic behavior directly,thereby creating deadweight losses, but alsocreate further deadweight losses because of thetaxes that are levied to finance those programs.I believe that the deadweight losses of those taxesare much larger than is generally recognized.

I will illustrate this with the effect of the50-percent increase in the payroll tax rate thatcould occur if there is no change in benefitrules. Deadweight losses depend on marginaltax rates. Consider an individual who now facesa combined federal and state marginal rate ofincome tax of 30 percent without social insur-ance. The current 15.3 percent employer-employee payroll tax rate,6 when adjusted forthe interaction with the income tax7 and for the

present actuarial value of the additional retireeand survivor benefits that result from increasedtaxable earnings, now increases the overall mar-ginal tax rate from 30 percent to about 37.7percent.8 A 50-percent rise in the 15.3 percentmarginal tax rate, adjusted for the income taxinteraction, would increase this effective mar-ginal tax rate from 37.7 percent to 44.2percent.9

The increase in the deadweight loss thatwould result from this tax increase reflects boththe reduction in labor supply—broadly definedto include not just working hours but also theaccumulation of human capital, the choice ofoccupation, effort, etc., and the change in theform of compensation—away from taxable cashand to less valuable fringe benefits. Althoughneither behavioral change can be measured ex-plicitly, the resulting deadweight loss can becalculated empirically by estimating the extentto which the higher payroll tax would reducetaxable labor income. It is appropriate to focuson the decline in taxable labor income withoutevaluating the two separate effects because therelative price of the two components—the mar-ginal tax rate on the reward for increased laborsupply and the marginal tax rate that determinesthe net cost to the taxpayer of fringe benefits—remains the same when the tax rate changes.Taxable labor income is, therefore, a Hicksiancomposite good that can be used to assess thedeadweight loss (Feldstein, 1999a).10

6 The 15.3 percent rate includes Medicare as well as theSocial Security pension and disability taxes. Currently thepension and survivor insurance portion is 10.6 percent oftaxable payroll. The disability tax is 1.8 percent and theMedicare portion is 2.9 percent. The costs of these threecomponents will rise at different speeds but the combinedcost will eventually increase the required tax to 150 percentof the 15.3 percent rate.

7 Since the employer half of the 15.3 percent payroll taxis excluded from the personal income tax base, the effectivemarginal rate of tax in this example is 30 percent plus the7.65 percent paid by the employee plus 70 percent of the

7.65 percent payroll tax paid by the employer: 0.30 �0.0765 � 0.70(0.0765) � 0.43. The extent to which a laborsupply response causes the tax to be shifted does not alterthe appropriate calculation of the marginal tax rate.

8 Feldstein and Samwick (1992) show that the presentactuarial value of the incremental benefits varies substan-tially among different age and demographic groups, from novalue for young workers and some married women to morethan a 100 percent offset of the incremental tax for oldermen with dependent wives. If this offset is approximated by50 percent of the 10.6 percent of the old age and survivorsportion of the tax, the 43 percent marginal tax rate calcu-lated in footnote 8 is reduced to 37.7 percent.

9 The increase in the payroll tax rate is subject to theincome tax offset (reducing the additional 7.65 percent to6.5 percent). There is no incremental benefit associated withthe higher tax rate.

10 The situation is more complex when we deal withtaxes on capital income. These distort choices among finan-cial instruments by both issuers and purchasers, choicesabout the form of business (corporate vs. non-corporate,domestic vs. foreign) and choices about saving vs. spend-ing. The key elasticity that matters in the saving vs. spend-


The elasticity of taxable labor income withrespect to the net-of-tax share, i.e., to one minusthe marginal tax rate on labor income, is muchgreater than the traditional elasticity of laborsupply as measured by labor force participationand average hours worked. Estimating this elas-ticity is now a subject of very active researchamong public finance economists. Although awide range of estimates has been produced,some studies are more reliable than others. Ibelieve that a conservative estimate is that thecompensated elasticity of taxable income withrespect to the net-of-tax rate is one-half.

Using this elasticity and the 2004 taxablepayroll implies that a rise in the effective mar-ginal tax rate from 37.7 percent to 44.2 percentincreases the annual deadweight loss by $96billion, or nearly one percent of GDP.11 Sincethe 6.5-percent increase in the marginal tax rateapplies only to taxable labor income (about 40percent of GDP), the deadweight loss is equal toabout one-third of the incremental tax revenue.Even this understates the relative size of thedeadweight loss because it ignores the reductionin the tax base and therefore in the tax revenuethat results from the higher marginal tax. Whenthat reduction in taxable income is taken intoaccount, the incremental deadweight loss isnearly 50 percent of the incremental revenue.12

The true cost per additional dollar of payroll taxrevenue is therefore $1.50.

Note that this is just the deadweight loss orexcess burden—i.e., the pure waste—associated

with the incremental tax. It does not include thedeadweight loss of the existing tax or the directburden of the taxes themselves. And it does notinclude the deadweight loss caused by the pro-gram distortions.

Although scaling back the rise in future ben-efits would reduce the increase in the dead-weight loss, it would also reduce the protectionthat Social Security provides to future retirees.An alternative approach is therefore to redesignthe program so that the increased financing re-quired for the aging population has less of thecharacter of a tax.

One way to do that is to strengthen the tax-payers perception of the link between taxes paidand future benefits. That is one of the advan-tages of shifting from the existing complicateddefined benefit rules to a defined contributionsystem, even to a notional defined contributionsystem. Although a notional defined contribu-tion plan would remain a pay-as-you-go system,it would clearly link each worker’s social insur-ance tax payment to his or her resulting futurebenefits.

A defined contribution system would providea tax-benefit link for those groups in the popu-lation that now receive no extra benefits at all inexchange for their additional taxes. For them,the Social Security payroll tax is a pure tax justlike the income tax. These include both youngand older workers who are not in the top 35wage-indexed earning years of their life, thebasis on which Social Security benefits are cal-culated, as well as working women who willeventually claim benefits based on their hus-bands’ earnings.

Although an unfunded notional defined con-tribution system would provide some remedy,the very low implicit rate of return in an un-funded system implies that the payroll taxwould retain much of its distorting character. Apay-as-you-go plan that substitutes a 2-percentreal return for private saving that would other-wise earn a 5-percent real return is equivalentover a lifetime of saving and dissaving to a taxrate of about 75 percent.13

ing distortion is not the elasticity of saving with respect tothe net-of-tax return but the elasticity of future consumptionwith respect to that net of tax return; Feldstein (1978b).Saving is equivalent to expenditure on future consumption.The relevant elasticity is therefore much larger than the elas-ticity of saving. If saving does not respond to changes in the netinterest rate, the relevant compensated elasticity is equal to oneminus the marginal propensity to save.

11 The formula for the increase in the deadweight loss is0.5 E (t2

2 � t12) TLI/(1 � t1) where TLI is taxable labor

income, t2 � 0.442 and t1 � 0.377. With TLI equal to 40percent of GDP or $4.5 trillion and E � 0.5, the impliedincrease in the deadweight loss is $96 billion.

12 With a compensated elasticity of 0.5 and an incomeeffect of 0.15, raising the marginal tax rate from 37.7percent to 44.2 percent on an initial tax base of $4.5 trillionreduces taxable income by $199 billion and therefore re-duces tax revenue by $88 billion. The net tax increase isthus reduced from $293 billion to $205 billion and thedeadweight loss per dollar of incremental revenue increasesto 46 percent.

13 An individual who can get a real net of tax rate ofreturn of (say) 5 percent in an individual retirement account(IRA) or a 401(k) plan converts one dollar saved at age 45(in the middle of his working life) to $4.32 at age 75 (in themiddle of his retirement). But if the implicit return on apay-as-you-go Social Security tax is only 2 percent, the


A much more substantial reduction in theeffective tax rate would be achieved by financ-ing the increased cost of Social Security andMedicare by a funded system that would permitfuture benefits to be financed without a largeincrease in the tax rate. Moreover, to the extentthat the additional saving that individuals doearns a favorable rate of return, they might notconsider it a tax at all. I will return to this issuelater when I discuss Social Security reformmore fully.

Designing Programs by Balancing Protec-tion and Distortion, while Seeking Reforms thatImprove the Available Tradeoff.—Social insur-ance programs generally involve a tradeoff ofprotection and distortion. Social insurance pro-grams protect individuals against undesirablylow levels of consumption during old age, spellsof unemployment, or when hit by large medicalbills. They also protect individuals from theneed to work longer than health warrants, toaccept a job when additional searching wouldbe adequately productive, or to forego appropri-ate medical care because of an inability to pay.But the same social insurance programs alsodistort incentives in ways that cause inefficientuse of resources: early retirement, low saving,unproductively long job searches, and excessiveconsumption of medical care.

Social insurance program parameters shouldbe chosen to balance protection and distortion.The level of Social Security benefits shouldreflect the fact that high benefits relative toprevious income improve the protection againstreduced consumption in old age but also depresssaving and may induce early retirement. A highlevel of unemployment insurance benefits helpsthe unemployed to maintain consumption butalso encourages longer spells of unemploymentand the choice of jobs that have a greater like-lihood of leading to a layoff. Low co-paymentsin health insurance reduce the risk of foregoingneeded care or suffering a major drop in otherconsumption, but they also lead to an increaseddemand for care that is worth less than its costof production. More complete protection in

each program also raises the program cost and,therefore, creates greater distortions through thetax system.

As protection becomes more complete, themarginal value of protection declines and theincremental distortion rises. The primary goalof social insurance should, therefore, generallybe to prevent catastrophic losses: poverty in oldage, long-term loss of income when unem-ployed, very expensive out-of-pocket medicalcosts, and the consequences of permanentdisability. More generally, at the optimum,the marginal value of additional protectionshould just equal the marginal cost of thedistortion. Economists can help the policyprocess by evaluating the protection and dis-tortion created by different changes in pro-gram design.

Useful economic analysis can go beyond select-ing an optimal point on a protection-distortionfrontier. It is important to seek ways to shift thefrontier, permitting less distortion at each level ofprotection. Reforms based on individual accountsthat I describe later in the paper would achievethat improvement.

Redesigning Programs to Keep Pace withChanging Conditions.—Three important changesthat should influence the design of our social in-surance programs have occurred since thoseprograms began: a changed economy, new tech-nology, and a different understanding of the effectof government programs on individual behavior.

The Social Security and unemployment in-surance programs were created during the de-pression of the 1930s when individual savingshad been destroyed by widespread bank failuresand when many individuals had been unem-ployed for a year or more because of a lack ofaggregate demand. Keynesian economists in the1940s like Harvard’s Seymour Harris praisedthe unfunded character of the new Social Secu-rity program for its ability to depress nationalsaving and stimulate aggregate demand (Harris,1941). In contrast to those depression years,conditions in the past half-century have beenvery different, with relatively low unemploy-ment rates and a system of government depositinsurance that protects individual savings. Theunemployment insurance and Social Securitythat may have been appropriate in the 1930s isno longer appropriate for the economy of thetwenty-first century.

dollar paid into such a plan at age 45 only grows to $1.81.The mandatory saving in the form of the low-return SocialSecurity is therefore equivalent to a 75.6 percent tax (thatreduces the gain of $3.32 to $0.81).


A second relevant change has been in thetechnology of financial administration madepossible by the introduction of computers.When Social Security was created, PresidentRoosevelt wanted it to be a funded systemrather than a pay-as-you-go system.14 Therewas of course no way to have individually con-trolled personal retirement accounts and the Re-publicans in Congress did not want to trust thegovernment to manage a large pool of funds.And since the Congressional Democrats wereeager to start paying benefits, the result was a defacto shift to a pay-as-you-go structure. Thecreation of individual investment accounts forevery adult, which would have been technicallyimpossible in the 1930s, is no longer even dif-ficult. Today more than 90 million Americansown mutual funds, including IRAs and 401(k)plan accounts. In contrast to the formidable taskin the 1930s of keeping track of everyone’sSocial Security account without the help ofcomputers, the creation of a system of individ-ual investment-based accounts would now berelatively easy.

The third important change has been in theeconomic profession’s understanding of howfiscal incentives affect individual behavior. Inthe 1930s, economists assumed that individualswere so unresponsive to taxes and benefits thatany behavioral response could simply be ig-nored. Most economists continued to ignore theadverse incentive effects even when the topmarginal tax rate was over 90 percent, as it wasfrom 1944 to 1963. The adverse effects of highunemployment insurance benefits on job searchand on the choice of jobs were also ignored.Today economists recognize that high mar-ginal rates of income tax and the marginal taxrates implicit in various benefit rules reducetaxable income and create substantial dead-weight losses.

These three changes imply that if Social Se-curity and unemployment insurance were beingcreated now, the programs would likely be sig-nificantly different from those in current law.Economists today would regard the adverse ef-fect of Social Security on saving and capitalaccumulation as a deterrent to growth rather

than as a favorable source of Keynesian de-mand. The widespread ownership of mutualfunds, IRAs, and 401(k)s would be a naturalstarting point for any new social insurance pro-gram. And every aspect of behavior would beassumed to be more responsive to tax rates andprogram design.

More generally, the reforms that will be en-acted in the future will inevitably evolve aseconomists learn more and as the set of feasibleoptions changes. An interesting example of thischanging perception of what is feasible is thepossible transition to personal retirement ac-counts in an investment-based Social Securityprogram. About 20 years ago, when I served aschairman of the Council of Economic Advisersin the Reagan administration, the Social Secu-rity retirement program was on the verge of acrisis. The trust fund was about to reach zeroand the projected taxes to be collected over thenext few years were not large enough to pay thebenefits specified in law. President Reagan ap-pointed a bipartisan commission to find a solution.The resulting plan called for an acceleration ofscheduled future tax increases, the taxation ofSocial Security benefits, and a variety of othersmaller measures.

President Reagan was unhappy with theseproposals and asked a small group of us in theAdministration whether there wasn’t some-thing better to be done, perhaps along thelines of the Chilean reform that used in-vestment-based personal retirement accountsinstead of a pay-as-you-go system. None of uscould design a feasible transition to such aplan. It looked to me and to the others as ifaccumulating funds to finance such per-sonal retirement account annuities would in-volve a double burden on the transition gen-eration that was both unfair and politicallyinfeasible.

I now know that that was wrong. Researchthat Andrew Samwick and I have done in recentyears (Feldstein and Samwick, 1998a, 1998b,2002) shows that it would be possible to tran-sition gradually to a completely investment-basedplan without ever increasing the combination ofpay-as-you-go taxes and personal retirement ac-count (PRA) saving by more than 2 percent ofpayroll earnings, or about 1 percent of GDP, andwithout reducing the benefit that retirees receivefrom the combination of the traditional tax-financed program and the new investment-based

14 See Scheiber and Shoven (1999) for a valuable dis-cussion of the origins of the Social Security legislation.


annuities.15 The key, we learned, is to have atransition in which the personal retirement ac-count annuities gradually substitute for pay-as-you-go benefits, allowing the pay-as-you-go taxrate to decline and the PRA contribution rate toincrease until the transition is complete.

Of course, the demonstration that such a tran-sition is feasible doesn’t mean that it is desir-able. A pure investment-based PRA plan wouldinvolve more risk than many individuals mightwant, a subject to which I will return later. Buta shift to a mixed system that avoids an increasein the payroll tax rate or in private saving mightbe an economic improvement. I’m sorry that Icouldn’t offer that solution when President Re-agan asked for it.

My point in recalling this is that economicresearch has changed what we regard as feasi-ble. Similarly, future research can and will de-velop new ways to provide social insuranceprotection with greater economic efficiency andmore responsiveness to individual tastes. A ba-sic principle of designing social insurance pol-icies should be a willingness to accept suchideas when they become available.

Separating Social Insurance from IncomeRedistribution.—As I indicated earlier, socialinsurance programs are not means tested. Eligi-bility for benefits does not depend on the in-come or wealth of the recipient but on an eventlike reaching age 65, beginning a spell of un-employment, or incurring a medical problem.Not surprisingly, the evidence that I cited makesit clear that today’s social insurance programsdo not redistribute income to the poor. Indeed,the positive correlation of income and longevitytilts the net benefit of Social Security andMedicare to households with higher life-time incomes. The structure of unemploymentinsurance rules causes a similar shift in thatprogram.

There is, of course, a role for means-testedprograms that are more narrowly focused onindividuals who demonstrate that they have lowincome or assets. Although I doubt the desir-ability of the myriad of existing in-kind pro-grams like food stamps and housing subsidies

(Robert Moffitt, 2003), I have no doubt aboutthe appropriateness of transferring income tothe very poor.

There is, moreover, a clear case for beingmore generous to some demographic groupsthan to others. The existing Supplemental Se-curity Income program provides means-testedbenefits to those over age 65 whose Social Se-curity benefits plus private resources do nottogether reach some minimal level. A moregenerous means-tested program, targeted at in-dividuals over age 75, would not distort laborsupply to the same extent that it would foryounger ones. It is possible, therefore, to havemore protection with less distortion in such ameans-tested program. It is a shameful featureof our Social Security system that, even with theSupplemental Security Income program, 10 per-cent of those over age 65 are in poverty whileSocial Security provides nearly $500 billion ayear in benefits to individuals who are finan-cially more comfortable.

To the extent that distributional concerns mo-tivate the design of social insurance, the empha-sis should be on eliminating poverty and not onthe overall distribution of income or the generalextent of inequality. Like most economists, Iaccept the Pareto principle that an economy isbetter off if someone gains and no one loses.This is true even if the gainer has above-averageincome, causing a Gini coefficient measure ofincome distribution to shift to greater inequal-ity. Although there may be spiteful egalitarianswho reject this Pareto principle, I believe thatmost economists agree with me. To see if youdo, ask yourself whether you think it would bea good thing if everyone reading this articlereceived $50 by some magical process that didnot decrease the income or wealth of anyoneelse. Since we are an above-income group, na-tional inequality would rise. Nevertheless, Ithink there are few who would reject bestowingthis extra wealth on us all.

This brings me to the end of my four economicprinciples of social insurance. I turn now to dis-cuss how the three major forms of social insurancecould be improved by shifting to a system thatcombines government insurance with individualinvestment-based accounts: unemployment insur-ance savings accounts (UISAs) backed up by agovernment line of credit, personal retirement ac-counts (PRAs) that supplement ordinary pay-as-you-go Social Security benefits, and personal

15 Alternative plans could achieve the transition withoutany rise in taxes by allowing the Social Security Trust Fundto borrow for a temporary period.


retirement health accounts (PRHAs) that finance arange of Medicare choices.

IV. Unemployment Insurance

Although unemployment insurance is a rela-tively small program with total federal and stateoutlays in 2003 of $39 billion, it is particularlyimportant because of its impact on macroeco-nomic performance. It is also significant as anillustration of how reforms have been able toreduce distortion while retaining protection forthose who need it. Moreover, it is a form ofsocial insurance where further reforms throughinvestment-based accounts could achieve sub-stantial economic gains.

The unemployment insurance program in theUnited States was created in 1935 in the depthof the depression. The program is administeredby the individual states but under federal rulesthat substantially restrict the scope of state gov-ernments’ actions. Benefits of a typical recipientare 50 percent of previous earnings and can becollected for up to six months. The Europeanunemployment benefit programs are substan-tially more generous in both the relative leveland the duration of benefits, with clearly ad-verse effects on European unemployment rates.

Thirty years ago, when I began doing re-search on unemployment insurance (Feldstein,1973a, 1973b), there was a general perceptionthat unemployment benefits were relatively lowand that they had little or no effect on economicbehavior. People were assumed to be unem-ployed solely because there was inadequate ag-gregate demand. Reformers focused on seekingincreases in the level and duration of benefits tohelp those who were unemployed for what wereassumed to be reasons beyond their own control.

We now know that perception was wrong.Unemployment insurance benefits raise the un-employment rate in a variety of ways that econ-omists have now analyzed and measured. Butback in the 1960s and 1970s, the higher unem-ployment rates that were actually induced byunemployment insurance were instead incor-rectly perceived as due to inadequate demand.When the government tried to reduce this highstructural unemployment with expansionarymonetary and fiscal policies, the result was ris-ing inflation. Fortunately, this is now betterunderstood. Monetary policy no longer tries toreduce structural unemployment. But although

unemployment insurance is therefore no longera source of increased inflation, it continues toraise the rate of unemployment. This is a par-ticularly serious problem in Europe where un-employment rates remain close to 10 percent.

The old notion that unemployment benefitswere too low to affect the economy was theresult of a misleading comparison of the aver-age weekly unemployment benefit and the av-erage weekly wage. Although the averagebenefit was only about 30 percent of the averagewage of all workers, the unemployed had sub-stantially lower pre-unemployment wages thanthe labor force as a whole. Unemployment in-surance benefits actually averaged about 50 per-cent of the pre-unemployment income of thosewho received benefits, with even higher re-placement rates in states that supplemented thebasic benefit with payments for spouses andchildren. But even this substantially understatedthe relevant replacement rate because benefitswere not subject to the income and payroll taxesthat were levied on wages. Since the combinedmarginal rate of income and payroll tax for thespouse of a high-earning individual could theneasily exceed 50 percent, the ratio of untaxed UIbenefits to the individual’s net-of-tax potentialearnings could exceed 100 percent. For such aperson, it was possible to have a higher netincome by remaining unemployed than by re-turning to work.

Even significantly lower benefit replacementrates could have substantial adverse incentiveeffects, as a number of studies eventuallyshowed. Although macroeconomists came torecognize that much unemployment was not ofan involuntary Keynesian type but was a pro-ductive search for good job matches, the accu-mulating evidence showed that UI benefits wereinducing excessively long periods of searchingin which the gain from the marginal search wasless than the value of the foregone output. Forexample, Larry Katz and Bruce Meyer (1990)showed that the probability that an unemployedperson takes a job rises dramatically in the fewweeks just before their benefits would expire.Jim Poterba and I (1984) found that the medianvalue of the reported reservation wage of newUI recipients was actually higher than the wageon their previous job, that it was an increasingfunction of the UI replacement rate, and that itcame down only very slowly during their spellof unemployment.


Longer durations of unemployment are notthe only adverse effect of UI benefits. The prac-tice of temporary layoffs in which unemployedindividuals have a spell of unemployment butreturn to their original employer is substantiallyencouraged by high UI replacement rates (Feld-stein, 1976, 1978a). High benefits also encour-age individuals to accept work in firms withhigh seasonal or cyclical layoffs. That reducesthe wage that such firms have to pay and thussubsidizes the expansion of those high unem-ployment industries.

As all of this became clear, the most obviousfirst reform was to include unemployment ben-efits in taxable income. Although there wasinitially strong opposition to this idea, it washard to argue with the position that cash incomeis cash income and should be taxed. The notionthat taxing unemployment insurance would in-appropriately burden the poor was clearly con-trary to the fact that the income tax allows asubstantial exclusion of income before any taxis levied. A poor UI recipient would pay no tax.

The initial legislative compromise was to in-clude only half of UI benefits in taxable incomeand to do so only for relatively high-income tax-payers. This provided a natural experiment thatGary Solon (1985) used to show that the relativeduration of unemployment fell for those whosebenefits were taxed. Later, in the Tax Reform Actof 1986, the UI benefits were fully subject to theincome tax like all other forms of labor income.

Taxing UI benefits eliminated the possibilitythat an individual could have a higher net in-come from UI benefits than by working. It ishard to know what the aggregate effect on un-employment has been, but my personal estimateis that the unemployment rate probably fell byabout one-half percentage point after benefitswere taxed, an effect equal to more than500,000 jobs at any time.

The evidence that UI benefits cause substan-tial distortion led to analytic studies of the levelof benefits that optimally balances distortionand protection. Martin Bailey (1978) presentedan analytic model in which the optimal level ofbenefits depends on the individual’s coefficientof relative risk aversion and on the elasticity ofthe duration of unemployment with respect tothe UI benefit replacement ratio. John Gruber(1997) used this framework to derive an explicitoptimal UI benefit based on data on the effect ofunemployment on household food consump-

tion, concluding that the optimal replacementrate should be much less than the 50 percent incurrent law. More recently, however, RajChetty (2003) showed that the measure of riskaversion that is relevant to designing the opti-mal UI benefit may be substantially greater thanthe risk aversion that is relevant to financialinvestments, because many types of householdspending cannot be adjusted in the short-run,which is relevant to unemployment spells. Chet-ty’s analysis points to optimal UI replacementrates that are close to the levels that we observein the United States.

These calculations of optimal UI benefits as-sume that individuals have no financial assets.In contrast, if individuals save optimally, theoptimal value of UI benefits—especially forshort and moderate spells would be very muchless. Although there is evidence that individualswho face greater income uncertainty havesomewhat higher saving rates, it would bewrong to assume that in the absence of unem-ployment insurance everyone would saveenough to finance consumption optimally dur-ing spells of unemployment. Some individualswould be too short-sighted to save for potentialunemployment.

What is the optimal response to this problem?One possibility would be to continue the currentsystem of paying UI benefits but with the leveland time path of benefits selected to balance thegain from protection and the loss from distortion.Another possibility would be to shift to a means-tested program, although that would induce someindividuals to game the system, saving nothing sothat they could qualify for means-tested benefitswhen they became unemployed. The same prob-lem of asymmetric information would prevail, asin the case of Social Security retirement benefitsthat I discussed before: the government could notdistinguish individuals who were too short-sightedto save from those who were gaming the system.On efficiency grounds, the choice between thecurrent system and government means-tested ben-efits would depend on the response of unemploy-ment to the benefit level and on the relativenumber of those who would save optimally, thosetoo shortsighted to save, and those who wouldchoose not to save in order to qualify for themeans-tested benefits.

A third possibility is to require everyone tohave an unemployment insurance savings ac-count earmarked to pay benefits if unemploy-


ment occurs. Dan Altman and I (Feldstein andAltman, 1998) explored a variety of such pos-sible plans. In a typical plan, each individualwould be required to accumulate funds in anunemployment insurance savings account untilthe balance was enough to pay benefits for twospells of six months at 50 percent of the indi-vidual’s current wage. These funds would beinvested and would earn a market rate of return.After a transition period to accumulate accountbalances, anyone who would be eligible forunemployment benefits under today’s UI ruleswould instead be able to withdraw the sameamount from his unemployment insurance sav-ings account. If a balance remains in the ac-count when the individual reaches retirementage, the funds would be available for the indi-vidual to take and spend. An individual whodies before retirement bequeaths the accountbalance. In short, individuals would regard thefunds in the UISA as their own money. Forsomeone who expects to have a positive balancein his account until retirement, the UISA planwould provide the same income protection asthe current UI system, but without anydistortion.

What about individuals who experience somuch unemployment that they use up the fundsin their UISA? Such individuals would be ableto borrow from a government UI fund to receivethe same benefits that they would withdraw ifthey had a positive account balance. After theyreturn to work, they would again save to repaythe loan with interest and to rebuild their UISAbalance. If they expect their account to accumu-late a positive balance in the future, the dollarsthat they borrow would be a very real obligationand the incentives to return to work would notbe distorted by the government loan. Theywould have full protection and no distortionwhile unemployed and would accumulate per-sonal wealth after they returned to work.

It is only those who expect that they will havea negative balance in their account when theyretire for whom this plan would represent noimprovement over current law. For them, theprotection and distortion would be the same asit is with the current UI rules.

The extent of the gain from introducing unem-ployment insurance savings accounts therefore de-pends on the proportion of the unemployed whoexpect to retire with negative balances and on thesensitivity of unemployment to the change in in-

centives. Dan Altman and I did some preliminaryempirical analysis of this approach using a sampleof men in the National Longitudinal Survey. Wefound that, even with no favorable behavioral re-sponse of unemployment to the improved incen-tives, less than 10 percent of benefits would bepaid to those who eventually retire with negativebalances (or who had negative balances when ourdata sample ended).

Our analysis thus implied that the UI programcould be redesigned around individual unem-ployment insurance savings accounts in a waythat substantially reduces the current distortingeffect while not reducing either the availabilityof funds when unemployment occurs or theprotection against relatively large cumulativeamounts of lifetime unemployment. More re-search on this potential form of unemploymentinsurance would certainly be valuable.

V. Social Security16

Social Security is the largest social insuranceprogram in the United States, with expendituresin 2003 of $470 billion, or 22 percent of totalfederal government spending. It includes notonly annuities for retirees and survivors but alsoa separate program of disability insurance thataccounts for some 15 percent of the total SocialSecurity outlays. Since the disability insuranceprogram involves a range of very different is-sues, I will not be considering it in this article.

Social Security is now a defined benefit, pay-as-you-go program in contrast to the definedcontribution, investment-based structure of mostprivate pensions. In a defined benefit program,an individual’s benefit at retirement depends onhis earnings during his working years and noton the performance of asset prices during thattime. The program is a pay-as-you-go one be-cause most of the payroll taxes collected in eachyear are used to pay concurrent benefits. Thereis not the kind of asset accumulation and finan-cial investment that there would be in a private

16 Feldstein and Liebman (2002a) provides a survey forthe Handbook of Public Economics of the large theoreticaland empirical literature on Social Security. See also my ElyLecture to the American Economic Association (Feldstein,1996a) and a forthcoming article in the Journal of EconomicPerspectives (Feldstein, 2005b) that deals with the currentreform debate in more detail.


pension. The benefits are financed by a payrolltax on earnings currently up to about the eighty-fifth percentile of the distribution of wages($87,900 in 2004). The payroll tax rate devotedto the Social Security program other than dis-ability is now 10.6 percent, divided equallybetween employers and employees.

Benefits take the form of an annuity that isindexed to the CPI to retain its real value duringthe individual’s retirement years. The level ofbenefits at retirement depends on the averagewage-indexed earnings of the individual duringhis or her highest 35 earning years. The benefitformula provides higher annual benefits per dol-lar of previous earnings at lower earnings levelsthan at higher levels. Individuals who retirebefore their normal retirement age (now risinggradually from 65 to 67) receive an actuariallyreduced benefit, while those who retire aftertheir normal retirement age receive an actuari-ally increased benefit.

Couples may collect benefits either on the basisof their separate earnings records or as 150 percentof the benefits of the individual with the higherbenefit level. A surviving spouse can receive 100percent of the benefit of the higher earning spouse.These rules for spouse benefits have the effect ofcausing many women who pay Social Securitytaxes to get little or nothing back for their taxessince their benefits are based on their husbands’incomes. This is true not only for married womenbut also for younger women who expect to marryand for divorced women and widows who willalso expect to claim benefits based on their former(or future) husbands’ earnings.

I became interested in Social Security as agraduate student in the 1960s when I realizedthat the tests of consumption theory—Fried-man’s work on the permanent income hypoth-esis and Modigliani’s work on life cyclesaving—completely ignored the role of SocialSecurity even though it had become the majorsource of retiree income. I realized also that thetheory of Social Security’s effect on saving wasmore complex than a simple displacement offinancial wealth by Social Security. To the ex-tent that Social Security induces earlier retire-ment, it raises the desired level of financialwealth. The net effect of Social Security onsaving therefore depends on the balance be-tween the positive induced retirement effect andthe negative wealth displacement effect, an is-sue that could be settled only empirically.

My initial time series analysis (Feldstein, 1974)implied that Social Security “wealth,” the presentactuarial value of future Social Security benefits,significantly reduced personal saving. Reestimat-ing this equation 22 years later with a corres-ponding amount of additional data produced reas-suringly similar results (Feldstein,1996b) Theconclusion that Social Security depresses privatesaving was also supported by household data andcross-country analysis. Other researchers wholooked at this question generally supported theprimary conclusion, although with estimates ofvarying magnitudes (Congressional Budget Of-fice, 1998).

This adverse effect of Social Security on sav-ing is relevant to understanding the significanceof Paul Samuelson’s very important 1958 over-lapping generations paper (Samuelson, 1958).Samuelson showed that a pure pay-as-you-goSocial Security system in an economy withoutcapital and without technical progress wouldgenerate an implicit rate of return on each gen-eration’s taxes equal to the rate of growth of thepopulation. This occurs because the number oftaxpayers is larger by the rate of growth ofpopulation than the number of retirees. If tech-nical progress is added to this economy, theimplicit return in a pure pay-as-you-go systemis still the rate of growth of the tax base, nowthe sum of the growth rate of population and thegrowth rate of productivity.

Samuelson noted that this positive rate of returnmeant that an unfunded Social Security programcould raise welfare in an economy that has nocapital assets. But what happens when we recog-nize that all economies do have capital stocks andthat Social Security transfers act as a substitute forreal capital accumulation? With that more realisticdescription, the reduction in the rate of savingcaused by the provision of pay-as-you-go annu-ities can cause a reduction in the present value ofall current and future consumption.

To understand why, consider first an oversim-plified textbook economy in which there are nocapital income taxes. Each generation of workersreceives an implicit pay-as-you-go return equal tothe sum of population and productivity growth butforegoes the larger return equal to the marginalproduct of capital. For each such taxpayer gener-ation, there is, therefore, a cost of having a pay-as-you-go Social Security program rather thanproviding for retirement consumption by savingand investing in real assets (or the financial assets


that represent a claim on those real assets.) How-ever, the initial generation of retirees had the goodfortune to receive benefits without ever havingpaid taxes.

It can be shown that aggregating the consump-tion losses of all generations of taxpayers back tothe initial date using a discount rate equal to themarginal product of capital produces a presentvalue loss just equal to the windfall gain of theinitial retirees. (Feldstein and Liebman, 2002a;Feldstein, 2005b) In short, with these simplified“textbook” assumptions, the introduction of a pay-as-you-go Social Security program does not re-duce the present value of national consumptionbut rather redistributes it from later generations tothe first one. More generally, whenever the pro-gram is expanded, those who are about to retire orwho will soon retire receive a windfall gain at theexpense of all future generations but with nochange in the present value of consumption overall generations.

This neutrality result depends, however, on theimplicit assumption that there are no distortingcapital income taxes.17 With capital taxes, theappropriate intra-generational rate of discount ofconsumption is less than the marginal product ofcapital. Moreover, the appropriate rate for aggre-gating the consumption of different future gener-ations may not be a market rate at all but adiscount rate equal to the rate of decline in themarginal utility of consumption. With a realisticgrowth rate and any plausible value of the elastic-ity of marginal utility with respect to consump-tion, the appropriate rate of discount would be farless than the marginal product of capital.18 Withthat discount rate, the net present value of the lossof consumption due to introducing and then re-peatedly expanding a pay-as-you-go Social Secu-rity would be very large. Similarly, shifting froma pure pay-as-you-go program to a mixed programwith a substantial investment-based portion wouldcause a large rise in the present value ofconsumption.

The reduction in saving and in the presentvalue of consumption is not the only adverse

effect of a pay-as-you-go program. A secondimportant effect is the distortion of labor supplyand of the form in which compensation is paidbecause of the increase in the marginal tax rate.The relevant marginal tax rate is the statutoryrate net of the anticipated increase in the actu-arial value of benefits. The increase in benefitsis zero for many married women. It is also zerofor individuals who are not in one of their 35highest earning years, typically when they areeither young or old, so that the higher effectivemarginal tax rate comes when the individual’sattachment to work is relatively weak. Manyindividuals may also underestimate the effect ofadditional earnings on future benefits. In all ofthese cases, the payroll tax may create a sub-stantial deadweight burden.

This incremental deadweight loss from dis-torting the labor supply would be essentiallyeliminated if individuals earned a market returnon their Social Security savings, as they wouldin an investment-based system. That wouldmake the actuarial present value of their benefitsequal to their Social Security savings. The onlylabor supply distortion would result if someindividuals were forced to do more saving forretirement than they preferred or thought thatthe implicit actuarial terms of the annuity didnot reflect their own mortality risk. A mixedsystem that combines pay-as-you-go and invest-ment-based components would reduce but noteliminate the labor supply distortion.

A third distortion caused by a traditional pay-as-you-go system is the incentive to retire earlywhen an implicit tax results from the loss ofbenefits caused by delayed retirement. Althoughthe United States has now largely eliminatedthis by an appropriate actuarial adjustment, thisdistortion remains a major problem in Europeand elsewhere (Gruber and Wise, 1999). Earlyretirement increases the annual cost of SocialSecurity benefits and reduces the available laborincome tax base. This leads to a higher marginalrate of Social Security tax, further increasingthat source of deadweight loss. When combinedwith formal or informal restrictions that preventreducing wages to offset the high payroll taxrates, these taxes contribute to the high unem-ployment rates that we see in Europe. The U.S.experience shows that this problem can be elim-inated within the pay-as-you-go system. Itwould, of course, also be eliminated in aninvestment-based system in which retirement

17 For an explicit derivation of the neutrality result and anexamination of the implication of capital income taxes and ofother discount rates, see Feldstein and Liebman (2002a).

18 For example, with a per capita growth rate of twopercent and an elasticity of the marginal utility function oftwo percent per year, the rate of decline of the marginalutility of consumption would be 4 percent.


income is withdrawn from personal accounts,which can be bequeathed if the individual diesbefore exhausting the account or before annu-itizing the accumulated balance.

The analysis of these three types of distortionshould make it clear that the shift to a “notional”defined contribution system would help only alittle to reduce the adverse effects of the currentpay-as-you-go system. A notional defined con-tribution system is one in which each individualhas an account that is credited with his taxpayments and with a notional rate of return onhis accumulated balance, but in which there isno actual investment in financial assets. Thenotional rate of return that is feasible in the longterm is the modified Samuelson return, i.e., therate of growth of the tax base. Since there is noreal capital accumulation, the reduction in thepresent value of consumption is not changed.The distortion in labor supply and in the form ofcompensation is reduced (but not eliminated)because individuals can more clearly see thelink between their taxes and their future bene-fits. A notional defined contribution system alsoreduces the distortion in retirement decisionsbecause individuals reduce their future benefitsif they retire early and increase them by delayedretirement. But even with this improved trans-parency, the low implicit pay-as-you-go rate ofreturn leaves a substantial distortion in workand compensation incentives.

Although the scope for reducing the substan-tial deadweight losses of the pure pay-as-you-gosystem and for increasing the present value ofall future consumption should provide a strongincentive for a change in policy, they are not thereason that has driven the political process inmany countries to move from a pure pay-as-you-go to a mixed system or to consider such achange. The primary driving force is the recog-nition that the increasing age of the populationwill require a very large tax increase or benefitcut if nothing is done to change the existingsystem. This is not a temporary effect of thebaby boom generation reaching retirement agebut a permanent result of the trend to increasedlongevity. This demographic change is signifi-cant not only because it drives the politicalprocess but also because it increases the poten-tial gain of making such a change.

The desirability of shifting to an investment-based or mixed system depends on four issues: (a)the transition process and its cost; (b) the ongoing

administrative costs; (c) the riskiness of financialinvestments; and (d) the effect on the incomedistribution and especially on the poorest group. Ihave done work on these issues during the pastdecade, both alone and with colleagues JeffreyLiebman, Elena Ranguelova, and Andrew Sam-wick. I will now summarize what I have learned.

I commented earlier on the transition prob-lem when I discussed my experience with Pres-ident Reagan. The common view that thetransition from a pure pay-as-you-go system toa mixed system requires the transition genera-tion to “pay double”—once to save for theirown retirement and once to meet obligations toexisting retirees—is wrong. As examples ofwhat could be done, Andrew Samwick and I(1998a, 1998b, 2002) showed how the projectedrise in the pay-as-you-go tax rate to more than19 percent19 could be avoided if individualscontribute just 1.5 percent of wages out-of-pocket to personal retirement accounts. The keyto the transition is using personal retirementaccount annuities to supplement the pay-as-you-go benefits. The growth of the PRA annu-ities offsets the slowdown of the pay-as-go-benefits that results from not increasing the taxrate as the population ages.

There is no free lunch in this process. Thekey is that additional saving during the transi-tion years can reduce long-run costs. This extrasaving could be voluntary, induced by a match-ing PRA contribution out of existing payroll taxreceipts.20 Although the matching would reducethe “trust fund” balances, the transition need notinvolve borrowing by the Social Security sys-tem from general revenue (Feldstein and Sam-wick, 2002).21

19 The 19 percent is based on Social Security Administra-tion’s “intermediate” demographic and economic assumptionsbut ignores the effect of the higher tax rate of the size of the taxbase for the payroll and personal income taxes.

20 The matching rate could be higher for low incomeindividuals to assure very high levels of participation. In theextreme, the entire contribution for low-income individualscould come from payroll tax revenue, requiring no contri-bution of their own. High income individuals would gener-ally welcome a chance for more tax favored saving andcould therefore be induced to participate with little or nomatching.

21 Stated differently, the Social Security Trust Fundcould remain positive and be rising at the end of the 75 yearprojection period, showing that the system has long-runstability.


Even a transition in which the initial personalretirement account deposits are financed whollyby government borrowing could eventuallyraise national saving and the present value offuture consumption. Financing the initial de-posit to personal retirement accounts by gov-ernment borrowing would have no immediatedirect effect on national saving because the in-creased budget deficit would be offset by thedeposit of those funds into the personal retire-ment accounts. Over time, the availability of thePRA annuities could permit reducing the pay-as-you-go benefits (relative to those projected incurrent law) without lowering total retirementincome. This reduction in the pay-as-you-gobenefits would mean that the annual rise in thebudget deficit would be less than the amounttransferred to the PRAs. That difference wouldbe an increase in national saving.22

I turn next to the issue of the administrativecost of PRAs. Some critics of PRAs argue thatthe administrative costs of a PRA programcould offset all of their higher return relative tothe pay-as-you-go system. Although there havebeen bad cost experiences in some countries,this certainly need not be so. Sweden’s recentPRA program involves administrative costsequal to between 30 and 100 basis points of theassets, an amount that will come down furtheras the total assets grow, since the administrativecosts depend on the number of transactions andnot on the value of the assets. TIAA-CREFoperates an individual account system with avariable annuity for a charge of only 37 basispoints.23

The issue of risk is an important considerationin both the pay-as-you-go and investment-basedsystems. Although pay-as-you-go programs donot have asset price risk, they have the politicalrisk that future taxpayers may not be willing toraise taxes when demographic or economic

changes would make it necessary to do so in orderto finance promised benefits. The United Statesenacted benefit cuts in 1983 by increasing the agefor full benefits. Many Latin American countriescut cash benefits in the 1980s and 1990s. Morerecently, Germany, Italy, and Japan have an-nounced or enacted reductions in state pensionbenefits.24 Perhaps the most reliable way to avoidfuture legislation that causes an unexpected reduc-tion in retirement income is to develop a mixedsystem that does not require a future rise in thepayroll tax rate.

The issue of asset price risk is more complex.On the basis of a substantial amount of research,I believe that a suitable mixed system that com-bines tax-financed pay-as-you-go benefits withinvestment-based PRA annuities can satisfythree conditions: a substantially lower long-term cost of financing retirement income thanthe tax projected for the pay-as-you-go system;a higher expected level of benefits from thecombination of pay-as-you-go and the PRA an-nuities; and a very low probability that the ac-tual level of combined benefits will be less thanthe pay-as-you-go benefits projected in currentlaw.25 The low risk could be achieved by acombination of three things: the floor on retire-ment income provided by the pay-as-you-gobenefits in the mixed system, restrictions on theinvestments that can be made in the PRAs, andexplicit guarantees provided by either the gov-ernment or the private market. Because individ-uals differ in their risk preferences, the solutionthat can best reflect those different preferencesmay be the availability of a variety of alterna-tive guarantees from the private organizationsthat manage the PRAs and PRA annuities (Feld-stein, 2005a).

I turn, finally, to the issue of the distributionaleffect of the shift from our pay-as-you-go sys-tem to a mixed system. I have already discussedthe evidence that the current Social Securitysystem does very little redistribution and leavessome 10 percent of seniors with incomes belowthe poverty line. Women who were never mar-ried, widows, and particularly those who werewidowed or divorced at a relatively early agegenerally have quite low benefits under current

22 Feldstein and Liebman (2002a, section 7.1.3) used anoverlapping generation model to show how such a debtfinanced transition could raise the present value of con-sumption. The rise in saving does not happen in the firstperiod but begins after that. The debt service on the initialborrowing cannot be financed by additional borrowingalone; the growth of the debt must be less than the growthof the economy.

23 See Shoven (2000) for several discussions of admin-istrative issues. The chapter by Goldberg and Graetz (2000)shows how administrative costs can be reduced while main-taining the individual account structure.

24 See McHale (2001) for more evidence on this point.25 See Feldstein (2005a, 2005b) for a more extensive

discussion of the risk issue.


law and could do substantially better under amixed system (Feldstein and Leibman, 2002b).Divorced women would benefit if the total ofthe husbands and wives’ PRAs are pooled anddivided at the time of divorce.

While a PRA itself does not cause any in-come redistribution, the redistributive structureof the pay-as-you-go benefits could, in princi-ple, be changed to make the combined benefitachieve any degree of redistribution.

In summary, it seems clear from the researchthat has been done that the current pay-as-you-go system could be gradually replaced witha mixed system that includes investment-basedpersonal retirement accounts in a way thatmaintains or exceeds the benefits that are pro-jected in current law, while sharply reducing thelong-run cost of achieving those benefits. Thistransition could be financed with relativelysmall additional PRA saving by individuals orby using existing payroll tax revenue. Even ifthe tax revenue is used, the initial fiscal deficitwould not decrease national saving because ofthe concurrent increase in private saving. Na-tional saving would rise in the long run becausethe PRA savings would exceed the increasedfiscal deficits.

A mixed system would eliminate the need fora future increase in the Social Security payrolltax and would, therefore, avoid the political riskthat future taxpayers would be unwilling to raisetaxes to finance promised benefits. It could bedesigned so that, despite the asset price uncer-tainty, there would be little risk that the com-bined benefits would be less than the currentlyprojected pay-as-you-go benefits. The remain-ing asset-price risk could be substantially re-duced by guarantees that could be produced bythe private financial market.

VI. Medicare

Medicare, the federal health care program forthose over age 65, is more difficult to reformthan either unemployment insurance or SocialSecurity. The program is more complex and thereaction to proposed changes is often moreemotional. And yet without reform, Medicarecosts will rise even more dramatically than thecost of Social Security retirement benefits, re-flecting the increasing numbers of the very old(who consume relatively more medical care)and the changing medical technology that pro-

vides new opportunities to spend money to pro-long life and increase the quality of life.

Before looking at Medicare, it is useful toconsider the basic theory of health insuranceand the current way that the government pro-vides a kind of quasi-social insurance for thepopulation under age 65 by its favorable taxtreatment of employer payments for health in-surance. This is important in itself and suggestsan approach that may be useful for reformingMedicare.

The basic theory of insurance implies that arisk-averse individual will prefer an actuariallyfair insurance policy to an uncertain and exog-enous distribution of potential losses. But thedistribution of potential health spending is notexogenous and the gain from risk reductionmust be balanced against the distorting effect ofinsurance on the demand for care. As insurancebecomes more complete, the marginal gainfrom additional risk reduction declines and themarginal deadweight loss from distorting thedemand for health care rises. At the optimumlevel of health insurance (e.g., at the optimumcoinsurance rate) there is a deadweight losscaused by the distortion in the demand for carebecause individuals, advised by their doctors,make decisions about diagnosis and treatmentbased on a net price of care that is very muchless than the cost of producing that care. Thatlevel of insurance is, however, efficient becausethe deadweight loss from distorted demand isless than the gain from risk reduction.

In actual practice, the demand for health in-surance is greatly increased by the tax treatmentof excluding employer payments for health in-surance from the taxable income of employees.Allowing employees to buy health insurancewith pretax dollars in this way changes thenature of health insurance. For someone with amarginal tax rate of 40 percent, the ability tobuy health insurance at a cost of only 60 centsper dollar of premium substantially increasesthe demand for health insurance with low de-ductibles and low coinsurance rates. This in-creases the deadweight loss caused by thedistortion in demand for care.

Direct attempts to eliminate or even reducethe tax subsidy or to constrain it by requiringminimum deductibles or coinsurance rates havenot been politically successful. When the Rea-gan Administration proposed to limit the em-ployers’ deduction for health insurance premi-


ums, it was unable to get any member of thePresident’s own party to introduce the legisla-tion in Congress.

Recently, however, Congress enacted legisla-tion to shift the incentives away from excessivehealth insurance. The new health savings ac-count (HSA) rules, enacted as part of the 2003Medicare legislation, allow individuals or theiremployers to deposit up to $5,000 of pretaxincome into a health savings account if theyhave a health insurance policy with an equallylarge deductible and with protection against cat-astrophic expenses. The individual foregoes theadvantage of the tax-free income in the form ofthe employer-paid premium, but gets an evenlarger tax-free income in the form of the healthsavings account contribution. Assets in theseIRA-like accounts earn tax-free investment in-come. Funds not spent in one year automaticallycarry forward to the future and can be used tofinance any kind of health care without payingtax. Funds can also be withdrawn for any otherkind of spending by paying tax at that time.

The health savings accounts create a strongincentive to choose policies with high deduct-ibles instead of the current comprehensive low-deductible and low coinsurance policies. This,in turn, should change the nature of the demandfor care. Until individual health spendingreaches the deductible limit, money spent onhealth care is the individual’s own money andnot that of an insurance company. Spendingbelow the high deductible limit, therefore,would not have any of the distortion caused bycurrent policies with low deductibles and lowcoinsurance rates. And the requirement that thepolicies provide protection for catastrophic lev-els of spending means that the most importantform of protection is retained or increased at thesame time that the distortions are reduced.26

Of course, anyone who spends several daysin a hospital will exceed the deductible limit. Atthat point, the insurance company is paying forcare as it would today and the patient and hisdoctor no longer have the incentive to be costconscious. The favorable incentive effects ofthe HSA could be increased without reducingthe individual’s insurance protection by replac-ing the deductible with a 50-percent coinsur-

ance rate on spending up to twice the level ofthe HSA saving deposit. For example, the limitassociated with the $5,000 HSA deposit wouldshift from a $5,000 deductible to a 50 percentco-payment on the first $10,000 of care, causingsignificantly more individuals and health spend-ing to be in the cost-conscious range.

The shift from the current tax-induced com-prehensive insurance to large deductibles or co-insurance is not only a way to limit excessivehealth care spending, i.e., spending that individ-uals and their doctors recognize as less valuableto them than the cost of production. It is also away of making health spending reflect eachindividual’s preferences. While all of us wantgood health, the lifestyles choices that individ-uals make show that some of us value it morethan others. Although we all understand theadverse health effects of obesity, smoking, andthe lack of exercise, not everyone acts on thisinformation. Many people knowingly make thetradeoff to enjoy more eating, to smoke, and toavoid the rigors of exercise. Just as people makedifferent lifestyle choices, some are more will-ing than others to sacrifice more of other con-sumption to increase spending on health care.Health savings accounts will allow this expres-sion of taste in health care spending instead ofeffectively inducing almost everyone to pur-chase high-cost health care.

Health savings accounts may be a model forMedicare reform. If nothing is done, the cost ofMedicare to the federal government will risefrom 2.4 percent of GDP to about 6 percent by2030 and 8 percent by 2050. The rising cost ofMedicare is similar to, but even more dramaticthan, the rising cost of Social Security retire-ment benefits. The remedy for this problemshould have two components: changing thespending incentives to slow the growth ofMedicare outlays, and using a mixed financingsystem to raise the needed funds without thesharp tax increase that would otherwise beneeded (Feldstein and Samwick, 1997; Feldstein,1999b).

If health savings accounts are successful, thehigh deductibles and coinsurance that willevolve because of the HSAs for those under age65 may establish a precedent that will also af-fect future Medicare benefits and therefore thespending incentives of the Medicare population.

A mixed financing system for Medicarecould combine a tax-financed Medicare annuity

26 I discussed this type of health insurance reform inFeldstein (1971) and Feldstein and Gruber (1995).


for retirees geared to the then-current cost ofhealth care plus an opportunity for individualsduring their working years to accumulate fundsin retirement health savings accounts. Thesecombined funds could be used at retirement topay for the type of health plan that the retireeprefers: a comprehensive insurance plan of thetype that Medicare now provides; membershipin a health maintenance organization that pro-vides a managed care plan; or a lower cost planwith substantial deductibles and coinsurance.Alternatively, working age individuals coulduse the funds contributed annually to their re-tirement health savings account to purchasehealth insurance for their retirement years,thereby minimizing the problem of asymmetricinformation in policy choice at retirementthat would occur in buying insurance afterretirement.

VII. Conclusion

The reform of social insurance is clearly awork in progress in the United States and inother countries as well. Policymakers can domuch to improve the major social insuranceprograms that protect the unemployed, the aged,and the ill. Economists can contribute to thisprocess by improving our understanding of theeffect of social insurance rules and by derivingnew program designs. In this paper I have em-phasized the use of personal investment-basedaccounts created and regulated by the govern-ment and earmarked for unemployment bene-fits, for retirement income, and for health careduring retirement. Such accounts have the po-tential to provide a better tradeoff of increasedprotection and reduced distortion. They alsogive individuals greater discretion in tailoringbenefits to their own tastes.

I am an optimist about economic policy. Ihave examined what is wrong with our currentsocial insurance programs and what could bedone to improve them in the future. I believethat the policy process does evolve and thateconomists have contributed to that evolution.We see that in the important reforms of the pasttwo decades that I have described. But there isstill much for economists to do in designingbetter policies for the future and in educatingthe public and the political decision makersabout the desirability of making such changes.

REFERENCES

Baily, Martin Neil. “Some Aspects of OptimalUnemployment Insurance.” Journal of PublicEconomics, 1978, 10(3), pp. 379–402.

Chetty, Raj. “Optimal Unemployment InsuranceWhen Income Effects Are Large.” NationalBureau of Economic Research, Inc., NBERWorking Papers: No. 10500, 2004.

Congressional Budget Office. Social Security andprivate saving: A review of the empiricalevidence. Washington, DC: U.S. GovernmentPrinting Office, 1998.

Coronado, Julia Lynn; Fullerton, Don and Glass,Thomas. “The Progressivity of Social Secu-rity.” National Bureau of Economic Re-search, Inc., NBER Working Papers: No.7520, 2000.

Friedman, Milton. Capitalism and freedom. Chi-cago: University of Chicago Press, 1962.

Feldstein, Martin S. “A New Approach to Na-tional Health Insurance.” The Public Interest,1971, 23, pp. 93–105.

Feldstein, Martin S. “Economics of the NewUnemployment.” The Public Interest, 1973a,33, pp. 3–42.

Feldstein, Martin S. “Lowering the PermanentRate of Unemployment,” Washington, DC:Joint Economic Committee, U.S. Congress,1973b.

Feldstein, Martin S. “Social Security, InducedRetirement, and Aggregate Capital Accumu-lation.” Journal of Political Economy, 1974,82(5), pp. 905–26.

Feldstein, Martin S. “Temporary Layoffs in theTheory of Unemployment.” Journal of Polit-ical Economy, 1976, 84(5), pp. 937–57.

Feldstein, Martin S. “The Effect of Unemploy-ment Insurance on Temporary Layoff Unem-ployment.” American Economic Review,1978a, 68(5), pp. 834–46.

Feldstein, Martin S. “The Welfare Cost of Cap-ital Income Taxation.” Journal of PoliticalEconomy, 1978b, 86(2), pp. S29–51.

Feldstein, Martin S. “Should Social SecurityBenefits Be Means Tested?” Journal of Po-litical Economy, 1987, 95(3), pp. 468–84.

Feldstein, Martin S. “The Missing Piece in PolicyAnalysis: Social Security Reform.” AmericanEconomic Review, 1996a, 86(2), pp. 1–14.

Feldstein, Martin S. “Social Security and Sav-ing: New Time Series Evidence.” NationalTax Journal, 1996b, 49(2), pp. 151–64.


Feldstein, Martin S., ed. Privatizing Social Secu-rity. Chicago: University of Chicago Press,1998a.

Feldstein, Martin S. “Transition to a FullyFunded Pension System: Five Economic Is-sues,” in Horst Siebert, ed., Redesigning So-cial Security. Tubingen: Mohr (Siebeck),1998b, pp. 299–315.

Feldstein, Martin S. “Prefunding Medicare.”American Economic Review, 1999a, 89(2),pp. 222–27.

Feldstein, Martin S. “Tax Avoidance and theDeadweight Loss of the Income Tax.” Re-view of Economics and Statistics, 1999b,81(4), pp. 674–80.

Feldstein, Martin S. “Administrative Costs andEquilibrium Charges with Individual Ac-counts: Comment,” in John B. Shoven, ed.,Administrative aspects of investment-basedSocial Security reform. Chicago: Universityof Chicago Press, 2000, pp. 162–69.

Feldstein, Martin S. “Reducing the Risk of In-vestment-Based Social Security Reform.”National Bureau of Economic Research, Inc.,NBER Working Papers: No. 11084, 2005a.

Feldstein, Martin S. “Structural Reform of So-cial Security.” National Bureau of EconomicResearch, Inc., NBER Working Papers: No.11098, 2005b.

Feldstein, Martin S. and Altman, Daniel. “Unem-ployment Insurance Savings Accounts.” Na-tional Bureau of Economic Research, Inc.,NBER Working Papers: No. 6860, 1998.

Feldstein, Martin S. and Gruber, Jonathan. “AMajor Risk Approach to Health InsuranceReform,” in James M. Poterba, ed., Tax pol-icy and the economy. Vol. 9. Cambridge,MA: MIT, 1995, pp. 103–30.

Feldstein, Martin S. and Liebman, Jeffrey B.“The Distributional Effects of an Investment-Based Social Security System,” in MartinFeldstein and Jeffrey B. Liebman, eds., TheDistributional Aspects of Social Security andSocial Security Reform. Chicago: Universityof Chicago Press, 2002a, pp. 263–322.

Feldstein, Martin S. and Liebman, Jeffrey B. “So-cial Security,” in Alan J. Auerbach and Mar-tin Feldstein, eds., Handbook of publiceconomics. Vol. 4. Amsterdam: ElsevierScience, North-Holland, 2002b, pp. 2245–2324.

Feldstein, Martin S. and Poterba, James. “Unem-ployment Insurance and Reservation Wages.”

Journal of Public Economics, 1984, 23(1–2),pp. 141–67.

Feldstein, Martin S. and Samwick, Andrew.“Social Security Rules and Marginal TaxRates.” National Tax Journal, 1992, 45(1),pp. 1–22.

Feldstein, Martin S. and Samwick, Andrew. “TheEconomics of Prefunding Social Security andMedicare Benefits,” in Ben S. Bernanke andJulio J. Rotemberg, eds., NBER macroeco-nomics annual 1997. Cambridge, MA: MITPress, 1997, 115–48.

Feldstein, Martin S. and Samwick, Andrew. “Po-tential Effects of Two Percent Personal Re-tirement Accounts.” Tax Notes, 1998a, 79(5),pp. 615–20.

Feldstein, Martin S. and Samwick, Andrew. “TheTransition Path in Privatizing Social Secu-rity,“ in Martin Feldstein, ed., PrivatizingSocial Security. Chicago: University of Chi-cago Press, 1998b, pp. 215–60.

Feldstein, Martin S. and Samwick, Andrew. “Po-tential Paths of Social Security Reform,” inJames M. Poterba, ed., Tax policy and theeconomy 2001. Vol.16. Cambridge, MA:MIT Press, 2002, pp. 181–224.

Feldstein, Martin S. and Siebert, Horst., eds. So-cial Security pension reform in Europe. Chi-cago: University of Chicago Press, 2002.

Goldberg, Fred T., Jr. and Graetz, Michael J.“Reforming Social Security: A Practical andWorkable System of Personal RetirementAccounts,” in John B. Shoven, ed., Adminis-trative aspects of investment-based Social Se-curity reform. Chicago: University ofChicago Press, 2000, pp. 9–37.

Gruber, Jonathan. “The Consumption Smooth-ing Benefits of Unemployment Insurance.”American Economic Review, 1997, 87(1), pp.192–205.

Gruber, Jonathan and Wise, David A. “Social Se-curity and Retirement around the World: Intro-duction and Summary,” in Jonathan Gruber andDavid A. Wise, eds., Social Security and retire-ment around the world. Chicago: University ofChicago Press, 1999, pp. 1–36.

Harris, Seymour. Economics of Social Security:The relation of the American program toconsumption, savings, output, and finance.New York: McGraw Hill, 1941.

Katz, Lawrence F. and Meyer, Bruce D. “TheImpact of the Potential Duration of Un-employment Benefits on the Duration of


Unemployment.” Journal of Public Econom-ics, 1990, 41(1), pp. 45–72.

Liebman, Jeffrey B. “Redistribution in the Cur-rent U.S. Social Security System,” in MartinFeldstein and Jeffrey B. Liebman, eds., Dis-tributional aspects of Social Security and So-cial Security reform. Chicago: University ofChicago Press, 2002, pp. 11–48.

McClellan, Mark and Skinner, Jonathan. “TheIncidence of Medicare.” National Bureau ofEconomic Research, Inc., NBER WorkingPapers: No. 6013, 1997.

McHale, John. “The Risk of Social SecurityBenefit Rule Changes: Some InternationalEvidence,” in John Campbell and MartinFeldstein, eds., Risk aspects of investment-based Social Security reform. Chicago: Uni-versity of Chicago Press, 2001, pp. 247–82.

Moffitt, Robert A., ed. Means-Tested TransferPrograms in the United States. Chicago: Uni-versity of Chicago Press, 2003.

Page, Ben. Social Security and Private Saving: AReview of the Empirical Evidence. Washing-ton, DC: Congressional Budget Office, 1998.

Samuelson, Paul A. “An Exact Consumption-Loan Model of Interest with or without theSocial Contrivance of Money.” Journal ofPolitical Economy, 1995, 66(6), pp. 467–82.

Schieber, Sylvester J. and Shoven, John B. Thereal deal: The history and future of SocialSecurity. New Haven: Yale University Press,1999.

Shoven, John B., ed. Administrative aspectsof investment-based Social Security re-form. Chicago: University of Chicago Press,2000.

Skinner, Jonathan and Zhou, Weiping. “TheMeasurement and Evolution of Health In-equality: Evidence from the U.S. MedicarePopulation.” National Bureau of EconomicResearch, Inc., NBER Working Papers: No.10842, 2004.

Solon, Gary R. “Work Incentive Effects of Tax-ing Unemployment Benefits.” Econometrica,1985, 53(2), pp. 295–306. Presidential Ad-dress delivered at the one hundred sixteenthmeeting of the American Economic Associ-ation, January 8, 2005, Philadelphia, PA.


rethinking social insurance (march...

Documents