IRAS AND NATIONAL SAVINGS*** - National Tax Association · Equivalence of Different T=-Favored 30.' In the example, the taxpayer is in the Accounts 28 percent tax bracket throughout

IRAS AND NATIONAL SAVINGS***

LEONARD BURMAN,* JOSEPH CORDES** AND LARRY OZANNE*

Introduction tended effect. This paper explores the evi-dence on how tax-favored saving accounts

HE. low rate of saving in the U.S. is a affect national saving. Looked at in termsTsenous problem. While the infusion of of the tradition@l life-c le modgLgf,,iav-fot@ilif c@a@itaT-Iki@cu-SWIFn-ed@ii@v!M' t in-gs -s@ue -@cc;i@@ -UTd 'ti@Cto;@expe@t@deii , fi@ ti7wouspending from the effectr. of a lower sav- to increase national savings; because ofing rate, there are disadvantages to in- the limitation on aimual contributions, thecreased dependence on foreign investors. tax incentive might well reduce personalForeign investment represents a future savings. Moreover, the tax preferenceclaim on U.S. production of goods and ser- would tend to increase the federal deficit,vices. As a result, investments financed reducing public savings as well. We findby inflows of foreign capital will contrib- substantial evidence that is consistent withute less to raising the future standard of the life-cycle interpretation of these taxliving in the U.S. than capital that is fi- incentives.nanced out of domestic U.S. saving. Recently, an alternative school of

Since national saving is the sum of thought has suggested that HUB or otherpublic and private saving, there are two tax-favored accounts might affect savingsbroad approaches to raising the national through psychological rather than eco-saving rate. First, the current level of nomic mechanisms. Such a premise is dif-public dissaving could be reduced by cut- ficult to reject in economic terms, becauseting the federal budget deficit. Second, it questions some of the basic tenets ofpolicies could attempt to increase the sup- modem economic analysis. h is also dif-ply of private saving. ficult to prove. The evidence on IRAs that

Creating additional tax incentives for has been posited in support of these al-savers continues to enjoy considerable ternative models does not refute the life-support as a means of encouraging pn- cycle model. Thus, the argument in favorvate saving. In particular, there contin- of tax incentives for saving is tenuous atues to be interest in expanding access to best. Direct deficit reduction would seemtax-favored saving accounts. Some Pro- to be a surer path to higher national sav-posais would do so by allowing all tax- ings.payers to make contributions to Individ-ual Retirement Accounts (IRAs) that areat least partially deductible. Others would Current Law and Proposals fordo so by creating so-called backloaded ChangeIRAs. Contributions to a backloaded IRA,like those to the nondeductible IRA gen- Under current law, deductible IRA con-erally available under present law, would tributions of up to $2,000 are allowed tonot be deductible, However, unlike the workers who are not covered by an em-nondeductible IRA, earnings accumu- ployer pension plan or have adjusted grosslated in a backloaded IRA account would incomes (AGIS) of less than $25,000 if sin-be exempt from tax on withdrawal in re- gle or $40,000 if married and filing jointly.tirement (or, for other qualifying pre-re- For AGIs that exceed these limits by uptirement purposes). to $10,000, ceilings on contributions are

There is a lively debate about whether phased down to zero. Contributions to IRAssuch tax incentives will achieve their in- for an eligible worker and a spouse not in

the labor force can total $2,250.*U.S. Congrmi?nal Budget Office, Washington, DC Individuals not entitled to the maxi-

20515 mum deductible contribution may make**U.S. Congressional Budget Office, WashingtonDC, 20515 and George Washington UniN,ersity: nondeductible contributions. Nondeduc-Washington, DC 20052. tible contributions are limited to the dif-

259

260 NATIONAL TAX JOURNAL [Vol. XLIII

ference between allowable deductible con- Eligibility to contribute to FSAs wouldtributions and $2,000. For example, a be limited to taxpayers with adjusted grossworker with pension coverage who files a income below certain levels-$60,000 forsingle return with an AGI of $30,000 could single taxpayers, $100,000 for heads ofdeduct an IRA contribution of $1,000 and households, and $120,000 for couples fil-make an additional nondeductible IRA ing joint returns. As with an IRA, contri-contribution of up to $1,000. If the work- butions to an FSA could not exceed a per-ees AGI were $36,000 or more, he or she son's compensation. Couples filingjointly,could make a contribution of up to $2,000, however, could each contribute up to thebut could not deduct any of the contri- $2,500 limit provided their total contri-bution. butions did not exceed their combined

IRA earnings are not taxed until funds compensation. For example, a couple inare withdrawn. At that time, earnings and which one spouse worked outside the homeall deductible contributions are included earning $30,000 could contribute up toin income. Withdrawals before age 59 and $2,500 to each of two accounts. Depen-one-half are generally subject to an ad- dents would be ineligible.ditional 10 percent income tax. FSAs would have the same investment

options as IRAs, and contributions could

Proposals for Expanding IRAs be made to them in addition to any othertax-advantaged saving for which a person

Several proposals have been made to qualified, such as an IRA, Keogh plan,expand the availabiliity of tax-free or tax- employer pension, or 401(k) plan.preferred savings in the form of MAs. Oneproposal would allow a 50 percent deduc- Revenue Effectstion for IRA contributions that would notbe deductible under current law. The The proposals described above wouldamounts of fully deductible contributions permanently lose revenue and raise theallowed under current law would remain deficit unless offset by increases in otherunchanged. taxes or reductions in spending. For ex-

Another option is the so-called "back- ample, the Joint Committee on Taxationloaded IRA." A backloaded IRA would be has estimated that a proposal to allow allsimilar to the nondeductible IRA avail- taxpayers to deduct at least half of theirable under present law in that contribu- IRA contributions would lose over $12tions would not be deductible. But unlike billion in tax revenue over the periodthe nondeductible IRA, earnings accu- 1990-1994. The JCT has also estimatedmulated in a backloaded IRA account that the Treasury would lose $300 millionwould be fully exempt from tax on with- in revenues from FSAs during 1991, anddrawal in retirement (or, for other qual- that the annual loss would rise to $1.8ifying pre-retirement purposes). billion in 1995. The revenue loss from

A variation on the backloaded IRA is FSAs would continue to grow rapidly af-the Family Savings Account (FSA) pro- ter 1995 as people shifted otherwise tax-posed in the President's 1991 budget. Un- able savings into these accounts. If peopleder the proposal, people could make non- ultimately smfted 10 percent of their tax-deductible contributions of up to $2,500 able savings into FSAS, the annual rev-annually to individual accounts, and the enue loss in 1991 dollars would reachinvestment income of amounts left on de- about $8 billion. If they shifted as muchposit for at least seven years could be as one quarter of their taxable savings intowithdrawn tax free. The investment in- FSAS, the annual revenue loss would-l@@achcorij_p.^Amau_at**Ioft -on.@@n A-gu-cWa re@u-It$9040ionln--1991 dollthree and seven years would be subject to would not be extreme: contributions toincome tax but no penalty. A 10 percent deductible IRAs accounted for about onepenalty on investment income would be quarter of taxable private savings in theadded for amounts left on deposit less than early 1980s. In fact, FSAs provide taxthree years. benefits over a lifetime that are similar

No. 31 LEONARD BURMAN, JOSEPH CORDES AND LARRY OZANNE 261

to deductible IRAs, but with much higher ing this. The first is to show why a de-effective contribution limits and fewer re- ductible and backloaded HU are finan-strictions on withdrawals. cially equivalent under certain

assumptions. The second is to show why,

How Tax-Favored Accounts Affect in the life-cycle model, financial equiva-

National Savinglence implies that each type of IRA willhave the same effect on national saving.

Un-losa,these revenue losses are fi- Pi rtce. Overvwfa-ff Mnanced by offsetting increases in other of an IRA-which may be 30 years ortaxes, they add to the federal deficit, which more-a backloaded and a fully deduct-increases public dissaving. Thus, such tax ible IRA have identical revenue effects inincentives for saving could be counter- terms of present value as long as the tax-productive. If the incentives did not prompt payer's marginal tax rate remains con-private saving to rise by as much as the stant and the taxpayer is allowed to de-increase in the deficit, national saving posit an equivalent amount of savings intowould decline rather than increase. each type of account. A 50 percent de-

One important question, therefore, is ductible IRA would cause a smaller rev-whether tax-favored accounts can gener- enue loss because the initial tax subsidyally be expected to provide an effective in- is only one half that of the fully deduct-centive for households to increase private ible IRA. The nondeductible IRA causessaving. Another question is whether some the smallest revenue loss because bothvariants of tax-favored savings accounts contributions and withdrawals are tax-may be more effective than others at rais- able; the only tax benefit is the deferraling national saving. of tax on accrued earnings.

How one answers these questions de- These points are illustrated in Figurepends very much on one's "prior" model 1, which shows how cumulative revenueof why and how households save. The losses (measured in terms of present value)model commonly used by economists is the change over time for a fully deductiblelife-cycle model of saving. This model has IRA, a nondeductible IRA, a backloadeda number of specific implications about IRA, and a 50 percent deductible IRA. Inhow tax-favored savings accounts can be this example, contributions are assumedexpected to affect national saving, to be made in year 0 and funds are with-

drawn as an annuity from year 21 through

Equivalence of Different T=-Favored 30.' In the example, the taxpayer is in the

Accounts 28 percent tax bracket throughout the en-tire period, and all forms of savings pay

Deductible IRAs, and in many cases 8 percent, which is also assumed to be theeven partially deductible IRAs lose more government's discount rate.tax revenue in the near-term than back- The example assumes that the tax-loaded IRAs. The apparent difference in payer would contribute $1,000 to a fullythe near-term revenue effects of front- taxable savings account if IRAs were notloaded and backloaded IRAs has made the available. A contribution of $1,389 to alatter type of IRA politically more attrac- deductible IRA has the same out-of-pockettive because it seems to offer a way of in- cost as the $1,000 contribution to a fullytroducing a saving incentive without at taxable account because the IRA contri-the same time raising the federal deficit. bution reduces taxes by $389 (28 percentThe first important implication of the life- of $1,389). Since the backloaded IRA andcycle model is that, under certain condi- the nondeductible IRA do not receive antions, seemingly different types of tax-fa- immediate tax deduction, a $1,000 con-vored accounts will yield equivalent re- tribution to these accounts has the samesults even though they can have very out-of-pocket cost as a $1,000 contribu-different effects on government revenue tion to the fully taxable savings account.and the federal deficit in the short term. Under the 50 percent deductible IRA, the

Two steps are involved in demonstrat- taxpayer could contribute $1,163, because

262 NATIONAL TAX JOURNAL [Vol. XLHI

this contribution reduces taxes by $163 "Young" taxpayers save out of labor in(28 percent of 50 percent of $1,163). come in the first period to provide for their

The fully deductible IRA and the 50 consumption when they become "old" tax-percent deductible IRA initially lose rev- payers in the next period. All taxpayersenue because of the tax deduction, as are assumed to have the same tastes forshown in Figure 1. Over time, the cu- present and future consumption, and themulative discounted revenue costs of all number of young taxpayers is assumed tofour types of IRAs increase because tax grow at the rate of n percent per year.liability on the accruing interest is de- Thus in each time period there are (1 +ferred.' However, because withdrawals n) times as many young taxpayers as therefrom the fully deductible IRA, the 50 per- are old taxpayers. In the initial period, thecent deductible IRA and the nondeducti- govenunent finances $G spending by a taxble IRA are taxable, the cumulative rev- on labor and capital income. For simplic-enue losses for these IRAs fall when the ity, government spending is then as-taxpayer withdraws these funds in years sumed to grow at a rate of n percent in21 through 30. Upon withdrawal of the each period.entire IRA balance, the cumulative rev- The first panel in Table 1 shows whatenue loss from the deductible IRA is equal would be observed in two "representativein present value to the revenue loss from time periods" in such an economy. For ex-the backloaded IRA. In other words, the ample, Table 1 shows that in time t thefully deductible and the backloaded IRAs old would dissave an amount St, which isare equivalent: they have the same rev- the amount saved by the old when theyenue loss in terms of present value, and were young.' The young, in turn, save (1the taxpayer ends up with the same + n) St. The govemment surplus or defi-amount of income in retirement. In each cit at time t equals taxes collected on la-case, the person earns the pre-tax return bor income received by the young duringto saving on money in IRA accounts and time t (TYt) plus taxes collected on capi-lifetime after-tax income is higher by an tal income received by the old (7@-So,whereamount equal to the present value of taxes r is the return to saving, minus govern-that would have been paid on interest that ment spending (G). National saving inwould otherwise have been taxed. period t equals the sum of the private

The revenue loss from the 50 percent saving of the young and the old plus gov-deductible IRA is lower because the im- ernment saving (+surplus or -deficit).mediate tax deduction was smaller. Not The pattem is repeated in period t + 1.surprisingly, the lowest revenue loss oc- Because there are (I + n) times as manycurs in the case of the nondeductible IRA. young workers as there were in period t,

Behavioral Equivalence. Financial national saving is n percent greater thanequivalence of backloaded and deductible in period t. The capital stock would growHW is in4portant because in the life-cycle at a rate equal to the growth rate in theframework individuals are presumed to labor force.save in order to provide for future con- The second panel in Table 1 illustratessumption for themselves and, possibly, the effects of allowing savings to be de-their heirs. If individuals save for these posited into backloaded IRAs. Under thepurposes, savings incentives that are fi- assumption that the saving incentive isnancially equivalent over a lifetime should introduced at time t, only the young in thatprompt people to make the same lifetime period are affected. As a result they wouldconsumption choices. If this is the case, change their desired saving from (1 + n)the two "different" types of IRA will also St to a(l + n) St. The coefficient a cap-hu-@w,.thesame eff6d"n 4oW national tuPeOffie-4ndividuals@-r@6spftse@tothe7-9-ftv-saving. ing incentive, where a > I represents the

This point may be illustrated using a case in which private saving increases insimple life-cycle model with overlapping response to the saving incentive, a = 1generations. In this model, all individuals represents the case in which private sav-are assumed to live for two periods. ing does not change, and a < 1 represents

No. 31 LEONARDBURMAN,JOSEPH CORDESAND LARRYOZANNE 263

Figure 1. Discounted Cumulative Revenue LossesFrom a $1,000 After-Tax IRAContribution

(FourTypesof IRAs)

Ful@Deductible

50 peroontdeducuble

400 BackloadedE

2Wt4ondeductible

00 5 10 15 20 25 w

Number of Years

the case in which private saving declines. The row in Table 1 labeled "Change inNote that in period t, the old would not Saving" shows how the introduction of thebenefit from the saving incentive, and thus backloaded IRA affects national saving.would not change their behavior. The old At time t, saving would increase, stay thewould still pay taxes on their capital in- same, or decline relative to saving undercome so that introduction of a backloaded no IRA depending on whether a wasIRA would not affect the government greater than, equal to, or less than 1. Inbudget deficit/surplus in time t. the longer run represented by period t +

In time period t + 1, however, the be- 1 national saving would be higher, thehavior of those who are old would reflect same, or lower, depending on the sign oftheir response to the backloaded IRA in (1 + n) [(a - 1)nSt - TrSt]. National sav-time period t. Moreover, in time period t ing could go up, remain unchanged, or go+ 1 withdrawals from backloaded IRA ac- down, depending on the response of pri-counts would no longer be taxed. As a re- vate saving. If, for example, a = 1, na-sult the government deficit would be larger tional saving would decline by an amountby TrSt, the amount of tax revenue that equal to the revenue loss resulting from

L

would have been collected from the old in the backloaded IRA (e.g., by (1 + n)l'rS).time t + 1 in the absence of the back- If a < 1, national saving'would decline byloaded IRA. Period t + I can thus be an amount equal to the revenue loss plusthought of as representing the longer-run the decline in private saving prompted bycase in which the behavior of both the the backloaded IRA. If a > 1, nationalyoung and the old reflect the creation of saving could decline, go up, or remain un-backloaded IRAs. changed depending whether the in-

264 NATIONAL TAX JOURNAL [Vol. XLIII

TABLE 1EFFECTS OF DIFFERENT TAX-FAVORED ACCOUNTS ON NATIONAL SAVINGS

Period t Period t + 1

. . ..... . ...................... . ..............................................................................................................................

No IRAs

Young (1+n)St (1+n) (1+n) S,

Old -st - (1+n) S,

Govt. TY,+TrS,-G (I+n)(Y,+TrS,-C)

Natl.Saving ((I+n)S,-S,)+(TYt+TrS,-G) (1+n) [ ((l+n)St-S,)+ITYt+Trst-C) I

..............................................................................................................................................................

Back-Uaded IRAs

Young a(l+n)s, a(l+n)(l+n)st

Old -st -a(l+n)St

GovL TYt+TrSt-G (1+n) (TY, -G)

Nati, Saving (a(l+n)St-St)+(TYt+TrSt-G) a(l+n) f((I+n)S,-S,)+(TY,-G)

Change inSaving S, (1+n) (a - 1) (1+n) [ (a-l)nS,-TrSti

.............................................................................................................................................................

Deductible IRAs

Young a(l+n)St/(l-T) a(l+n) (I+n)St/(I-T)

Old -st -a(l+n)St/(l-T)

Govt. TY,+TrSt-G-[Ta(l+n)S,/(l-T) (I+n) (Y,-G)-(1+n)nTaS,/(l-T)

Natl. Saving t [a(l+n)St/(l-T) ]-St) a(l+n) ([ (1+n)St-St)/(l-T))

+(TYt+TrS,-G) +(I+n) (TYt-G)- (Ta(l+n)St/(l-T) - (I+n) [nTaSt/(I-T]

Change inSaving S,(I+n) (a- 1) S,(I+n) [(a-1)n-Tr]

. ........................................................................................................................

crmmd sttving by the young exceeds the be--d@Ooslti@d into a deductible IRA. Therevenue lost from not taxing interest in- analysis above has already shown that thiscome. would be financially equivalent to a back-

The third panel of Table 1 shows what loaded IRA. This, in turn implies thatwould be observed if funds could instead taxpayers should change their saving be-

No. 31 LEONARD BURMAN, JOSEPH CORDES AND LARRY OZANNE 265

havior so as to be able to consume an in time period t the difference betweenamount in their old age that is the same private saving would offset the differenceas under a backloaded IRA. in the govenunent surplus/deficit. Na-

This would require young taxpayers in tional saving would be the same as undertime t to deposit an amount into their ac- a backloaded IRA.counts that was 1/(1 - T) times as large Thus, like the backloaded IRA, the de-as that deposited into a backloaded IRA.' ductible IRA would change national sav-Thug yaupg t4Q&ya,&@ I iag-ia thej@r rua 4*4moun"quW+ n)S,/(l - T)I. In time t the old would to (1 + n) [(a - I)nSt - trst]. The twonot change their behavior, and as in the types of IRA would differ in how each af-case of the backloaded IRA, the govern- fected the components but not the level ofment would collect taxes on the capital national saving. To the extent that publicincome of the old in time t. Unlike the case dissaving (e.g., the deficit) was higher un-of backloaded MAs, however, at time t the der the deductible IRA than under thegovernment would lose revenue equal to backloaded IRA, private saving would bethe deductions claimed by the young for higher by an offsetting amount. The nettheir IRA. effect is that the total amount of saving

The upshot is that in time period t the would be the same under either type ofgovernment deficit would be higher un- IRA, and would ultimately depend on theder the deductible than under the back- sensitivity of private saving to the tax in-loaded IRA. Private saving, however, centive.would be higher than under the back- Differences Between Tax-Favored Ac-loaded IRA. The difference between the counts. The assumptions that make theamounts of private saving observed under two different types of IRAs financiallythe deductible and the backloaded IRA is equivalent may, however, not be satis-equal in magnitude to the difference be- fied. The tax rate on withdrawals maytween the government surplus/deficit ob- differ from the tax rate on contributions.served under each type of IRA, so that na- If the tax rate on withdrawals is lower,tional saving would be the same. the present value of the revenue loss from

A similar though more complex pattern a backloaded IRA will be lower than thewould be observed in the longer-run state comparable revenue loss from a fully de-represented by period t + 1. In this pe- ductible IRA. The opposite is true if ariod, the old would make taxable with- higher tax rate applies to withdrawalsdrawals from deductible IRAs, while the than to contributions. Whether differentyoung would make tax deductible contri- types of tax-favored account are equiva-butions. The government would also make lent, therefore, depends partly on as-interest payments on the amount it needed sumptions about the future tax rates ofto borrow in period t to finance the tax participants compared with their currentrebates on initial deductible contribu- rates.tions. Another source of difference between

The net effect would depend on the rel- deductible and backloaded IRAs is thatative numbers of young and old. If, for ex- similar contribution limits affect differ-ample, there were more young than old, ent types of IRAs in different ways. Anthe taxes collected on HU withdrawals equal dollar limit on the contribution towould be less than the sum of the tax re- each type of IRA would result in an un-bates on new IRA contributions plus in- equal amount of revenue forgone fromterest payments on the added govern- taxpayers who contribute the dollar limit.ment debt. In that case, even in the longer The reason is that each dollar contributedrun, one would observe that the govern- to a backloaded IRA is out of after-tax in-ment deficit was larger under the de- come, while a dollar contributed to a fullyductible IRA than the backloaded IRA. deductible IRA is essentially pre-tax in-Private saving, however, would again also come. As a result, a backloaded IRA al-be greater under the deductible IRA. As lows a greater amount of retirement sav-

266 NATIONAL TAX JOURNAL [Vol. XLIII

ings to be sheltered from tax, thereby question. The first, traditional approach,conferring greater tax benefits over the is to look at IRAs in the context of thelife of the IRA than does a fully deduct- life-cycle model of savings. This approachible IRA with the same dollar limit on leads to the conclusion that IRAs cannotcontributions. be counted on to increase savings. The

These differences mean that private second, more novel, approach looks at RWsaving may respond differently to a de- as a special kind of commodity-savings,ductible IRA than to backloaded IRA, but but with some features that make con-do not change the basic conclusion. The F;iimers value these accounts in their owneffect on national saving of introducing right. Under this view of behavior, IRAstax-favored saving accounts does not de- may increase national savings.pend on the form of the account. It de-pends on whether and by how much pri- IRAs in the Life-Cycle Modelvate saving can be expected to change inresponse to the general type of financial If saving is motivated by life-cycle con-incentive provided by such accounts. For sumption choices, two conditions must beexample, suppose the expectation of higher satisfied if IRAs are to stimulate privatefuture tax rates and higher effective con- saving. IRAs must change the after-taxtribution thresholds make backloaded return to the additional dollar saved forIRAs a better financial incentive than de- a significant number of savers and pri-ductible IRAs. In essence this means that vate saving must respond to such changes.individuals will enjoy a higher after-tax The task is then to determine whether bothlifetime income and earn a higher effec- of these conditions are likely to be met.tive after-tax return per dollar deposited In the life-cycle model, savings is sim-into a backloaded IRA than into a de- ply a way of attaining future consump-ductible IRA. This means that private tion. People dowt save because they likesaving will respond more to the back- to save, per se, but because they are rel-loaded than to the deductible IRA; it also atively more interested in future con-means that over time, the backloaded IRA sumption than otherwise similar peoplewill be a larger revenue loser. If house- who save less.holds are likely to respond to the saving An IRA without a contribution limitincentive by maintaining or reducing pri- would provide an incentive to save byvate saving, so that a < 1, the implication raising the after-tax rate of return onis that a backloaded IRA would produce savings for retirement. The effect of such• larger decline in national saving than a savings incentive is illustrated in Fig-• deductible IRA. However, even if house- ure 2. Line AOB illustrates the budgetholds respond to the savings incentive by constraint facing an individual withoutincreased private saving, national sav- IRAs. The line represents the trade-offings may still decline because of the effect between current and future consumption.on the deficit. A positive effect on na- At point 0, the individual consumes all oftional savings requires a high degree of his or her wealth plus current income andresponse to the savings incentive. saves nothing for retirement. To the left

of point 0, the individual will have some

IRAs and Private Saving: Evidenceretirement savings (from current or past

from the 1980ssavings or both), which will earn the pre-tax rate of return, r, and the after-tax rate

Can IRAs be counted on to increase pri- of return, r(l - t). To the right of pointvate * by gno t dd to a I 0,"jndividual is a-uet-d$*,@ isiajlqL_g-g_ _W..o I i@i- - -@-@,-- -r@ ----_O@s i

-@LW wsaving? conomists isagree i@i * t b6rr6wi against retirement income atcan and cannot be inferred from the re- rate, rb. The borrowing rate is assumed tosponse of taxpayers to the universal be greater than the pre-tax rate of return,availability of IRAs from 1982 through r; otherwise individuals could always in-1986. There are, broadly speaking, two crease current and future consumption bydifferent ways of attempting to answer this borrowing to buy IRAs.

No. 31 LEONARD BURMAN, JOSEPH CORDES AND LARRY OZANNE 267

The introduction of IRAs is represented upon their preferences.by line AOI. Several features of this un- Figure 2 depicts a more generous sav-restricted IRA are salient. First, unre- ing incentive than has ever been adopted.stricted IRAs will have no effect on net In fact, IRAs have been subject to limi-debtors, i.e., those on the interior of line tations on contributions that would be ex-segment OA, since this part of the budget pected to lessen their effectiveness asconstraint is unchanged. In practical compared to an across-the-board increasetems@*Armeatw-&at ymuWrVftle are in the-retwm to -gaving@ Figuiti@3 sy6wsless likely to buy IRAs. Second, IRAs may the IRA budget constraint when IRAs areinduce some of those who neither save nor subject to limits on contributions, lineborrow to start saving. Persons who nei- AOLI. The limit causes the second kinkther save nor borrow are at point 0 in the in the line at point L. At this point, theFigure. Some of these persons may be on maximum amount of fimds has been in-the verge of saving at the after-tax inter- vested in an IRA and any additional sav-est rate and introduction of a tax-free rate ings (on segment LI) earn the after-taxmay induce them to save. Others at point rate of return. As in the unlimited case,0 may be on the verge of borrowing. Since there is little or no effect on the savingsthe borrowing rate, rb, is assumed to be behavior of individuals on segment AO.greater than the tax-free rate of return, Individuals who contribute less than thesuch individuals would not be tempted to limit (maximize utility on the interior ofsave by the higher rate of return to sav- segment OL) have the maximum incen-ings.' Third, IRAs would have an ambig- tive and are unaffected by the limit. How-uous effect on individuals who are al- ever, individuals who maxinlize utility onready saving for retirement, those on the interior of segment Ll have no incen-segment OB. They might increase or de- tive to save more. To them, the tax in-crease present consumption depending centive in IMs is equivalent to a gift fi-om

FIGURE 2IRA Budget Constraint Without Contribution Limit

80

.L) 60 -

E IRA Budget Constraint

c:0 40

Tmble savings Budget Constraint

E0

B

zu

A

020 25 30 35 40 45

Current Consumption

268 NATIONAL TAX JOURNAL [Vol. XL11I

the government. It has no direct effect on pensity to save are not evident in a sim-their savings decisions because they would ple tabular analysis, the first two char-save more than the limit amount in any acteristics can be examined. In fact, thecase. However, the indirect effect of this prediction of high income and wealth seemgift is that savings will decline. This is to fit well with the observed data. The databecause the taxpayer can aclueve the same are from a panel of tax returns filed fromor higher retirement savings with higher 1979 to 1984 (see Appendix for details).present consumption thanks to the gift Based on the panel data, Table 2 showsfrom the government. Since the individ- IRA contributions, average positive in-ual values present consumption (it in- come, and interest and dividend income ofcreases with income), the individual will IRA contributors and noncontributors forconsume more with IRAs than he or she 1983. 1983 is the year for which the panelwould have without them. has the largest sample; results are simi-

So the incentive effect of IRAs depends lar in the other years. IRA contributorscrucially on what segment of the budget are divided into those who contribute theconstraint in Figure 3 most taxpayers find limit amount and those who doet. The fintthemselves. The evidence from a sample striking fact is that almost all contribu-

6of tax returns filed in the three years fol- tions are made by limit contributors. Inlowing widespread availability of IRAs, 1983, 75 percent of contributions were1982-1984, is that the overwhelming ma- made by households at the limit ("all re-jority of IRA contributions were made by turns/at limit" in Table 2). An additionaltaxpayers on segment 01, i.e., limit con- 11 percent were made by households fil-tributors. These are taxpayers with rela- ing joint returns where the contributiontively high incomes, higher than average equaled the limit for one spouse ("all re-wealth, or a higher propensity to save than turns/one spouse at limit"). While someaverage. While the differences in the pro- authors have argued that these house-

FIGURE 3IRA Budget Constraint With Contribution Limit

80

c 4B

.L) 60 - /R4BudgetConsftint4.1

CL

E

c0 L

C) 40 Tamble Savings Budget Cons&aint

c

E(DL-'t 20

cc

A

0 -1

20 25 30 35 40 45

IRA Bud

Current Consumption

TABLE 2CHARACTERISTICS OF IRA CONTRIBUTORS AND NONCONTRIBUTO

IRA CONTRIBUTIONS POSITIVE INCOME IIN 1983 1983

7 OF OFCONTRI- POSITIV

TOTAL BUTIONS MEAN INCOME

RETURN TYPE

NoN-jorNT RETURN NO CONTRIBUTION 0 0.0% 10761 21.

LESS THAN LIMIT 136553 2.5% iazss 0.

AT LIMIT 986275 18.3z 34121 4.

TOTAL 1122828 20.97 12204 ZS.

I EARNER JOINT RETURN No CONTRIBUTION 0 0.0% 25935 21.

LESS THAN LIMIT 108298 2.OZ 37902 0.

AT LIMIT 945000 17.67 121904 12.

ONE SPOUSE AT LIMIT 346000 6.4% 58096 1 2.

TOTAL 1399298 26. 0*. 37272 36.

2 EARNER JOINT RETURN NO CONTRIBWION 0 0.0% 30040 23.

LESS THAN LIMIT 365452 6.SX 38077 2.

AT LIMIT 2109000 39.3% 73722 19.

ONE SPOUSE AT Limrr 368000 6.8% 42339 1.

TOTAL 28424S2 52.97 36s27 37.

ALL RETURNS NO CONTRIBUTION 0 0.07 18491 65.

LESS THAN LIMIT 610303 11.37 '32044 4.

AT LIMIT 4040275 75.3% 74189 25.

ONE SPOUSE AT LIMIT 714000 13.3Z 4997S 4.

SUMMARY WITHOUT IRA Gj 0. 0;' 1 18491 65.

WITH IRA 536457a 100.OZ 60SS6 34.

TOTAL S3645781

100.0%1

242241 100.

270 NATIONAL TAX JOURNAL [Vol. XLHI

holds are "falsely constrained,"' suggest- contributors, the average was essentiallying that the taxpayers may have been the same as for noncontributors, $1,242.fooled by advertising to misunderstand the The concentration of contributionsavailability of IRAs for both spouses, we among limit contributors is troublesome.doubt that this is the primary explana- As pointed out above, taxpayers who wouldtion. It was easy to determine from in- save at least the after-tax equivalent ofcome tax forms (and probably from banks their IRA contribution receive the equiv-and financial institutions as well) that alent of a gift from the IRA tax prefer-IRAs were available for both spouses. It ence. These taxpayers have an unambig-seems more likely that households where uous incentive to save less. Taxpayers whoonly one spouse contributed represented save just the limit amount (at point L onsome independence in financial decision- Figure 3), may have some incentive to savemaking within the household. Moreover, due to IRAs. However, the data suggestbecause the incremental contribution al- that only a small fraction of limit con-lowable to a nonworking spouse was rel- tributors are likely to be in this situation.atively small, $250, it is likely that some First, limit contributors have muchspouses decided that the second IRA was higher wealth and income than other tax-not worth the transaction costs of setting payers. This means, in terms of Figure 3,up the account. The percentage of returns that they are likely to be consuming toat the limit for one spouse is much higher the left of point L. Table 3 shows the dis-in the case of a nonworking spouse-25 tribution of interest and dividends andpercent-than for two-earner house- Table 4 shows the distribution of positiveholds-18 percent.' Thus, the spouse at income. Both distributions are for singlethe limit for a single contributor is simi- taxpayers in 1983, which avoids the issuelar to the single taxpayer facing a con- of one or two limits discussed above. Non-straint. Including returns with a single limit contributors dominate at the low in-spouse at the limit, the percentage of con- come levels; limit contributors dominatetributions made at the limit is 89 percent at higher levels. For example, 64 percent(75.3 plus 13.3 percent). of limit contributions are made by single

Limit contributors had significantly taxpayers with over $1,000 in interest andhigher incomes from all sources than non- dividends compared to 23 percent of non-limit contributors and non-contributors, limit contributions. In all, 52 percent ofas would be suggested by Figure 3. For single returns with IRA contributions hadexample, Table 2 shows that, in 1983, the over $1,000 in interest and dividends,average income of non-contributors in the compared to 9 percent for non-IRA con-sample was $18,491 compared-to $60,586 tributors (not shown).for IRA contributors. Households that Second, if taxpayers are induced to in-contributed at the limit had average in- crease their savings to reach the IRA limit,come of $74,189 ($49,974 when only one then their contributions might be ex-spouse of a joint return was at the limit). pected to fall off in subsequent years.' InThe average for contributors below the fact, the majority of contributions madelimit was $32,045. in 1982 were by taxpayers who were at

Limit contributors also have much the limit in every year, between 1982 andhigher taxable interest and dividend in- 1984. Tables 5 and 6 show that 61 percentcome, suggesting higher wealth, than of contributions on single returns and 66noncontributors and nonlimit contribu- percent of contributions on joint returnstors. IRA contributors in 1983 had an av- were made by taxpayers at the limit in4a@@@gf_of $4 634 of interest and dividends every year.10

Tesg @@payers would al-

compare&-i@@$1,232 f6i- nori@@,itiibulors. most c@A7ainly@t'l@ave coniWuTe-d- -m@or-e-'mAmong IRA contributors, the average 1982 if the limit were higher. Thus, theseamount of interest and dividends for limit taxpayers are unlikely to have receivedcontributors was $6,070 ($4,104 where one any marginal incentive under the IRAspouse was at the limit). For nonlimit program.

TABLE 3IRA CONTRIBUTIONS BY CAPITAL INCOME IN 1

SINGLE@ HEAD OF HOUSEHOLD, OR MARRIED FILING(NON-AGED RETURNS WITH IRA CONTRIBUTION

IRA CONTRIBUTION IN 1983

LIMIT STATUS

NOT AT LIMIT AT LIMIT

TOTAL X OF TOTAL X OFCONTRIBU- CONTRI- %-OF X OF CONTRIBU- CONTRI- Z OF

TIONS BUTIONS ROW RETURNS TioNs sLiTroNs Row

INTEREST AND DIVIDENDSIN 1983

$0 23215 17.0% 40.5%1 20.8% 34000. 3.4Z 59.

$1 - $999 82028 60.OZ 20.3% 62.5% 322000 32.6% 79.6

$1,000 - $1,999 7100 5.17 4.6% 4.lZ 146275 14.8% 95.3

$2,000 - $4t999 18441 13.57 7.77 10.1% 220000 22.3% 92.2

$5*000 - $9,999 4269 3.1% 2.6Z l.7Z 156000 15.8X 97.3

010@000 AND OVER 1500 1.07 1.3Z 0.57 1080001 10.9% 98.6

TOTAL 136SS3 100.OZ 12. 1;1 100.0 9862751 100.0% 87.8

TABLE 4IRA CONTRIBUTIONS BY POSITIVE INCOME

SINGLE, HEAD OF HOUSEHOLD, OR MARRIED FrLt(NON-AGED RETURNS WITH IRA CONTRIBUT

IRA CONTRIBUTION IN 1983

LIMIT STATUS

NOT AT LIMIT AT LIMIT

TOTAL 7- OF TOTAL X OFCONTRIBU- CONTRI- 7 OF X OF CONTRIBU- CONTRZ-

TIONS UMIONS RON RETURNS TIONS BuTrONS R

POSITIVE IN60ME IN 1983

UNDER $0 4000 0.4Z 10

$0 - $9,999 22251 16.27 23.OY 15.4Y 7427.51 7.5% 7

$10,000 - $19,999 53441 39.17 18.4% 4 7. 0;1 236000 23.97 8

$20,000 - segt999 46735 34.2Z 13.77 28.5% 294000 29.8% 8

$309000 - $49,999 12626 9.27 4.9X 8.3% 242000 24.57 9

$50,000 - $4)91999 1500 1.07 1.37 0.5% 106000 10.7Y 9

$100,000 - t'Agg,999 16000 1.6Z 10

$200,000 ANd, OVER 14000 1.4% 10

TOTAL 1365531 100.07 12.IX 100.07 986275 100.OY 8

TABLE 5PERSISTENCE OF IRA CONTRIBUTIONS AND Limr

-- JOINT RETURNS IN 1982-1984 --

ANNUAL IRA CONTRIBUTIONS BY IRA CONTRIBUTION IN IRA CONTRIBUTION IN IRAYEARS AT LIMIT 1984 1983

X OF Z OFCOt4TR1- CONTRI-

MEAN BUTIONS MEAN BUTIOM

CONTRIBUTED TO YEARS AT BROADIRA IN ANY YEAR IRA LIMIT

WITHOUT IRA NEVER 0.00 0.0% 0.00 O.OX

WITH IRA NEVER 676.30 7.97 486.91 6.37

1984 ONLY 2624.00 17.2Z 327.83 2.3x

1983 ONLY 543.22 1.4% 2S30.61 7.2%

1983 AND 1994 2718.14 14.6% 2703.43 16.IX

1982 ONLY 393.62 1.3Z 492.66 1.87

1982 AND 1984 2670.45 3.07 783.77 1.OX

1982 AND 1983 434.68 0.9x 2445.12 S.SY

ALWAYS 3050.85 S3.3x 3049.70 59.1%

ALL 562.09 ioo.ox 506.91 100.0%1

TABLE 6PERSISTENCE OF IRA CONTRIBUTIONS AND L

SINGLEs HEAD OF HOUSEHOLDP OR MARRIED FI(CONSTANT FILING STATUS IN 1982-

ANNUAL IRA CONTRIBUTIONS BY IRA CONTRIBUTION IN IRA CONTRIBUTION INYEARS AT LIMIT 1984 1983

7 OF X OFCONTRI- CONTRI-

MEAN BUTIONS MEAN BUTIONS

CONTRIBUTED TO YEARS AT BROADIRA IN ANY ',YEAR IRA LIMIT

WITHOUT IRA NEVER 0.00 0.0% 0.00 0. 0>.,

WITH IRA NEVER S77.20 11.0% 4S2.01 10.5%

1984 ONLY 1966.41 24.4Y 164.80 2.4%

1983 ONLY 162.50 0.4% 2000.00 6.8%

1983 AND 1984 2000.00 20.1% 2000.00 24.0%1

1982 ONLY 223.33 0.8% 235.29 1.0%

1982 AND 1984 2000.00 2.8% 366.50 0.6%

1982 AND 1983 187.13 0.5% 2000.001

6.8%

ALWAYS 1979.47 39.5% 2000.001

47.7.7

ALL 194,76rlOO-0-7 163.271 100. 0;1

No. 31 LEONARD BURNUN, JOSEPH CORDES AND LARRY OZANNE 275

Does It Matter Whether IRAs Provide a between current and future consumption:Marginal Incentive? the percentage change in the ratio of fu-

ture to current consumption observed inEven if IRAs raise the return to the ad- response to a percentage point change in

ditional dollar saved for some limit con- the ratio of the price of current consump-tributors and nonlimit contributors, pri- tion to the price of future consumption. Avate saving will rise only if saving recent study by Hall estimated an elas-responds to changes in the real -aftar--tax ti-city gf,.substitution of c_Wqnt for future

U16@ n- 'C'tvWreturn. Such a response cannot be counted cons ptio of e en-@6 @n-di-on. An increased rate of return has two eating little or no substitutability be-offsetting incentives-one that raises tween current and future consumption.saving and another that reduces saving. These elasticities of substitution wouldOn the one hand, a higher rate of return suggest that the savings elasticity is neg-raises the amount of future spending that ative.can be achieved per dollar saved now. This Thus, even for the minority of contrib-encourages people to save more, just as a utors for whom IRAs raise the marginallower price for a product encourages peo- rate of return on savings, the effect of theple to buy more of it. On the other hand, incentive on savings is far from certain.the higher return also means that pre- The available empirical evidence wouldvious levels of saving will allow an in- suggest that a higher return is as likelycrease in future spending without further to decrease as increase personal savings.sacrifice. The availability of that higherfuture spending encourages people to di- Implications for Saving of IRAvert some of the increase to current E rience from 1982 to 1986spending by saving less. The net effect of lp,these offsetting incentives could be either Based on the data reported above, IRAsto raise or lower saving. probably reduced national savings from

12The evidence on how people respond is 1982 to 1986. For example, on the panelinconclusive. Numerous studies have been tax return sample in 1982, $1.998 millionundertaken to determine how a higher of IRA contributions were made by tax-return to saving changes the personal payers at the limit compared to $0.286saving rate. The results of this research million by taxpayers facing a marginalare frequently summarized in terms of the incentive." As explained above, the bestestimated elasticity of private saving to we could hope for among the taxpayers atthe rate of return. Many studies have the limit is that their savings did not de-found elasticities near zero, which means crease. Suppose, to continue making thesavings is unresponsive to rates of return. strongest case for the effectiveness of IRAs,However, a few have found sizable sav- that all of the contributions made by non-ings elasticities, such as 0.4." limit taxpayers represented new savings.

A more recent set of studies has taken The tax reductions granted on all of thea different approach. These studies have IRA contributions were about $.762 mil-estimated how the rate at which individ- lion. This is more than twice the amountuals are willing to trade off consumption of the "induced savings@' of $.286 million.today for consumption tomorrow changes Thus, every dollar of tax reduction cor-in response to changes in the relative price responded to $0.62 of net dissaving. Thisof current and future consumption. This ratio remained essentially constant in 1983approach is appealing to economists be- ($0.62) and 1984 ($0.60). 14

cause it is more directly related to the In fact, since the response to the tax in-choices that individuals are presumed to centive is likely to be only a fraction ofmake in the life-cycle model. the $.286 million and the limit contribu-

The willingness of individuals to sub- tors may well have reduced their savings,stitute future for current consumption is the dissaving per dollar of tax reductionmeasured by the elasticity of substitution may be closer to one dollar.

276 NATIONAL TAX JOURNAL [Vol. XLHI

Empirical Evidence of the Life-Cycle results they report in their paper, they findModel that IRA contributions substitute for other

savings at an average rate of about oneThe tables and accompanying analysis for one. GS simulations suggest that an

presented above cannot fully identify the increase in the IRA contribution limit (insavings effect of IRAs because everyone 1985) would have reduced other savingswas eligible for IRAs during the years ex- by about 93 cents for every dollar in-amined. Full identification of the savings vested in IRAs. Since the resultant taxresponse through such a tabular analysis reductions would be worth an additionalwould require two comparable groups of 37 cents, net consumption would actuallypeople, one that was eligible for IRAs and rise by 30 cents for every additional dol-another that was not. lar contributed in IRAs. Put another way,

When everyone is eligible for MAs, their every additional dollar of tax reductionsavings effect must be inferred using a would have reduced total savings by $0.81.model of behavior that attempts to ex- These results are consistent with the in-plain the differences in IRA utilization ferences based on the tax return data. Theybased on observed differences among the imply that IRAs were counterproductive.population. Gale and Scholz (GS) devel-oped such a model that incorporates dif-

Non-Life-Cycle Models of IRA Behaviorferences in tastes and age in a life-cyclemodel of IRAs and savings. This model Other studies have attempted to assesstakes the conventional view that IRA whether IRAs stimulate private saving bysavings is similar to other savings except comparing the private saving of IRA con-for the penalty that is imposed on early tributors with that of "otherwise identi-withdrawals. GS show that the limit on cal" noncontributors. These models relyIRA contributions may be used to econo- on an alternative view of savings behav-metrically identify the relationship be- ior. Under the alternative view, IRAs aretween IRAs and savings as long as all a special kind of commodity that hascontributors to IRAs have similar pref- characteristics that transcend its value aserences for savings. Put simply, we know a savings instrument.that many limit contributors have no Statistical Evidence on the Commoditymarginal incentive to save under IRAs Model of IRAs. The papers by Venti andbecause they would have saved at least Wise (1987a, 1987b, and 1989) are thethe limit amount anyway (more gener- most technically sophisticated represen-ally, there is no marginal incentive if liq- tatives of the commodity view of MAs. Alluid assets are greater than the contribu- of the Venti and Wise (VW) papers usetion limit). However, taxpayers who invest - data from the period between 1982 andless than the limit in IRAs have the full 1986 when IRAs were universally avail-incentive effect of IRAs on savings. If they able. Like GS, VW have to deal with theinvested another dollar in an IRA, they lack of a control group that didwt havewould earn the tax free rate of return. This access to IRAs.difference allows the identification of the VW attempt to identify the savings re-parameters in an empirical model of IRAs sponse to IRAs by assuming that all peo-in which there are two operative differ- ple have the same preferences for sav-ences: IRA contributors may be different ings, and that they view MAs and otherfrom noncontributors (for example, be- savings as different commodities. Accord-cause they value future consumption rel- ing to VW, people may view IRAs andatively more) and limit contributors are other saving as separate commodities forconstrained whereas nonlimit contribu- seyMI_A#W.,-Jt -,mVy xaay, v ioN.M_Aatofs afb un,@6n-gfiii@@d. bein@ f6i -r6fiiement and their other sav-

GS find strong evidence that contribu- ing as being for shorter-term needs, theytors to IRAs view their contributions as a may view IRAs as a separate kind of sav-substitute for other savings, as would be ing because of the extensive advertisingpredicted by the life-cycle model. In the focused on IRAs, they may value the pen-

No. 31 LEONARD BURMAN, JOSEPH CORDES AND LARRY OZANNE 277

alty on withdrawals as a kind of control in IRAs-25 to 30 percent in the most re-mechanism, or they value the up-front tax cent study-is outside the range of his-deduction for IRAs disproportionately. torical experience. Since World War II,

When IRAs and other savings are savings rates have averaged 4.4 percent.viewed as separate commodities, they en- Also, VW seem to underestimate theter people's utility functions separately. marginal propensity to save outside ofPeople chose how much to "spen&' on each IRAs for higher income people.in'

ein @h,

tond An !4eq@nal problem with the VW,a_

- - -a0 s, s

-"@vho fifiibiite toon Ch-?rp-@'@sy@t4rgila4 iicht a mdet arises iftif6sesports car and a sedan. Just as a person U@Ashave different attitudes toward sav-might buy a sports car even if the person ings from those who do not. To illustratemight not otherwise buy any car, so a this problem, the IRA budget constraintperson might save in an IRA even though in Figure 3 is reproduced in Figure 4. Thethe person might not otherwise save. Just indifference curves of three individuals areas a person might spend more to buy both superimposed at the point where eacha sports car and a sedan than the person maximizes utility subject to the IRA bud-would otherwise spend on just a sedan, a get constraint. The individuals are iden-person might save more in both an IRA tical in terms of current income andand other accounts than the person would wealth; thus, they face the same budgetsave if IRAs were not available. constraint. They might well also be of the

The view of IRAs and other savings as same age and family size (factors VW useseparate goods is a fundamental depar- to control for preferences). There is onlyture from the classical life-cycle model of one apparent difference. Individual A is asavings. In the classical model, only cur- saver. He values retirement consumptionrent and future consumption enter the relatively highly. He would save for re-utility function; IRAs and other savings tireraent in the absence of a tax incen-are viewed as alternative means of pro- tive. Individual B values current con-viding for future consumption. Since HUB sumption relatively more. He is a netprovide more future consumption per dol- debtor and would not save for retirementlar of current consumption forgone (be- with or without a tax incentive. Individ-cause of their tax advantage) people save ual C values future consumption morein IRAs first, and then, if they plan to save than B but less than A. When IRAs aremore than the IRA limit, save in other ac- available, C finances all of his or her re-counts. tirement savings with IRAs.

Applying their commodity model of In a cross-section of such seeminglysaving, VW get consistent results with "identica I" people with MAs (A and C) andthree different data sets. Based on data without IRAs (B), the people with IRAsfrom Survey of Income Program Partici- will have more non-IRA savings than thepation in 1984 and 1985 (VW, 1989), for people without IRAs and the people con-example, VW estimate that 66 percent tributing below the limit (C) will haveof IRA contributions were financed more savings than the non-IRA people, butby reduced consumption, 31 percent less than the people at the limit. Aby reduced taxes, and only 3 percent by regression of IRAs against savings wouldreduced saving elsewhere. These results find a positive correlation, even thoughsuggest that IRAs are an astonishingly the IRA contributors are saving less out-effective savings incentive. Essentially, side of IRAs than they would in the ab-every dollar of tax reduction results in two sence of IRAs. IRAs would seem to bedollars of net additional savings. complements to, rather than substitutes

Others have questioned VW's charac- for, other savings. This might explain theterization of the individuals's decision VW result.problem (for example, Gravelle) and the As Manegold and Joines (1990) pointedplausibility of the results (for example, out, if the life-cycle model holds, thereDeaton). In particular, Deaton argues that should be a positive correlation betweenthe implied marginal propensity to save savings and IRA contribution levels. One

278 NATIONAL TAX JOURNAL [Vol. XLIII

FIGURE 4Savings, IRAs and Preferences

80

.0 60 -

E A

0040

c

E20 -

B

020 25 30 35 40 45

Current Consumption

cannot use this correlation to demon- to contribute to IRAs than otherwise sim-strate that the availability of IRAs in- ilar individuals who arewt. FS interpretduces savings. La the life-cycle model, just this as evidence that taxpayers wouldthe opposite occurs. Savers will use IRAs rather contribute to an IRA than contrib-and nonsavers won't. A regression of the ute to the IRS. This begs the question ofchange in interest and dividends against where the taxpayers found the money toIRA contributions may find that IRA contribute to the IRA. In fact, taxpayersholders seem to increase their non-IRA are typically underwithheld due to non-wealth relative to non-IRA holders, as wage income such as a capital gain or aFeenberg and Skinner flid. This could oc- commission or bonus. In the first case,cur because IRA people save more than selling a capital asset and reinvesting thenon-IRA people. proceeds in an IRA represents pure sub-

Other Anomalies. While the positive stitution of one form of savings for an-correlation between IRAs and other sav- other. The second case may represent newings is the centerpiece of the evidence in savings, but it may well be savings thatfavor of IRAs, there are other anomalies would have occurred in the absence ofthat may suggest that IRAs are somehow IRAs. A well-known implication of em-different from regular savings. Feenberg pirical life-cycle models of consumption isand Skinner (FS) produced the most com- that individuals have a much higher pro-pelling catalog of such evidence. How- pensity to save out of transitory (unex-eve T. virWWly,,All Xay pe-ptodl aQWtWm-.o".pormanent in-be explaiii@d-in the context of me life-cycle come. P8?s observation is consistent withmodel as traditionally viewed by econo- that explanation.mists. A related observation is that IRA con-

For example, FS find that taxpayers who tributions exhibit a high degree of persis-are underwithheld are much more likely tence except when individuals suffer a drop

No. 31 LEONARD BURMAN, JOSEPH CORDES AND LARRY OZANNE 279

in income. This is also consistent with the responsive to changes in the rate of re-life-cycle model. Individuals offset tran- turn. Second, the limit on contributions tository drops in income by drawing on sav- IRAs removed any incentive to increaseings. The drop-off in IRA participation savings for most contributions. Indeed,simply represents part of the predicted taxpayers who contributed the limitdecline in savings. amount may bave viewed the tax reduc.

Another cited anomaly is the signifl- tion as pure income and responded by in-cant prbo@rtidi@-bf j6int'retuims -*ith-6nly efegMWthEir con-gffiiiprii5n. -$2,000 in IRA contributions. Since these If universal availability of IRAs werehouseholds could have contributed at least reinstated, more taxpayers would face a$2,250 and possibly as much as.$4,000 if marginal incentive to save, largely be-both spouses earned income, FS call these cause of the growth in availability ofpeople "falsely constrained." FS view this 401(k)-type accounts since 1982. Further-as evidence that individuals were igno- more, recent proposals for backloaded IRAsrant of the actual benefits available un- and FSAs have had higher effective con-der IRAs and were affected by advertising tribution limits than deductible IRAs.that suggested that an IRA was $2,000. Nonetheless, most empirical evidenceThis model of household decision-making suggests that saving is unresponsive toseems simplistic. First, the choice be- higher rates of return, so a new IRA, oftween a $2,000 IRA and a $2,250 IRA for whatever type, could not be counted on toone-earner households may not have increase national saving.seemed worth the trouble of setting up a While the traditional economic modelsecond account. As mentioned earlier, one- of savings decisions has come under crit-earner households who contributed to IRAs icism in recent years, it seems to do re-were about 35 percent more likely to be markably well at describing who would"falsely constrained" than two-earner respond to IRAs and how. The data sug-households with IRAs suggesting that gest that IRAs were most attractive totransaction costs may be part of the story. taxpayers with higher incomes who likedA more fimdamental problem is that cou- to save and most contributions were madeples are comprised of two decision-makers at the limit amount year after year. Andwhose objectives are interrelated but not the only direct- empirical test of the life-necessarily identical." One spouse may cycle model with IRAs, Gale and Scholz,choose to contribute or not contribute for found strong evidence that taxpayers treatsimilar reasons that a single person makes IRAs as simply another form of savings.that choice. For example, one spouse might This model suggests that universal avail-have a pension and not feel the need for ability of IRAs reduced net national sav-additional retirement consumption, ings because private savings increased (ifwhereas the other spouse may not feel that at all) by only a fraction of the amount ofher expected retirement income is ade- individuals' tax reductions.quate without an IRA. The risk of divorce There is an alternative view of how IRAsand uncertainty about the division of affect saving that cannot be rejected outpension benefits might argue for separate of hand, but much of the evidence that hasaccounts. been offered to support this view is also

consistent with the life-cycle model. In-

Conclusiondeed, the response of savers to IRAs pro-vides a reasonably good test of some of the

A conventional economic analysis of implications of the life-cycle model. Re-IRAs suggests that they did not increase gardless of individuals' perceptions, IRAsnational savings and may well have re- were a substitute for other savings thatduced savings through the effect on the individuals were free to make. The wide-deficit when they were widely available spread advertising that is often cited inbetween 1982 and 1984. These conclu- favor of the alternative model made it rel-sions are based on two findings. First, in- atively easy for individuals to make in-dividual savings does not seem to be very formed choices. And individuals' choices

280 NATIONAL TAX JOURNAL (Vol. YLUI

were largely as would be predicted by the tions on joint returns were by taxpayers at the limit

life-cycle model. in each year."See Bosworth for a survey.

If private saving is primarily motivated 121f UW were financed by tax memms, they mightby factors outside the hfe-cycle model, then raise national saving because the tax could offset theeconomists have to reexamine much Of income effect.

what we think we know about savings and 13Contributions on joint returns where one spousewas at the limit are assumed to be constrained as ex-

much of what we think we know about la- plained above. The qualitative conclusion would notbor and taxes. However, at least in the case be affected if a stricter definition of limit contributorsof IRAs, the life-cycle model seems to be were -d-

a useful framework for analyzing saving "It should be noted that the value of the up-fronttax deduction may over- or under-state the long-term

behavior. Within that framework, the data revenue depending on how long the account issuggest that MAs are not a promising held and interest rates. In the example in Appendixpolicy option to raise the low rate of na- Table A.1, the long-term revenue loss is slightly

tional savings.greater than the immediate tax reduction. Nonethe-less, since the long run effect of any increase in per-sonal savings also depends positively on how longsavings are held, the comparison is likely to be valid

ENDNOTES over a wide range of assumptions."FS mentioned this as a possible caveat to their

***We gratefully acknowledge helpful comments conclusions.

from Rosemary Marcuss and Jonathan Skinner. The "Christian and Frischmann discuss the panel indetail and focus on the problem of attrition bias. Ma-

views expressed are the authors' alone and should not negold and Joines, who used the same panel in theirbe attributed to the Congressional Budget Office.study of IRAs, concluded that attrition bias did not'The calculations underlying Figure 1 are pre- seem to be a serious problem for the analysis of MAs.sented in Appendix Table A-1 and are explained more Since our results are very similar fi-om year to year,fully in Congressional Budget Office (1989). attrition is unlikely to atli.@etour conclusions.'Me cumulative discounted revenue cost at year T

is the silm of all revenue losses from time period 0 totime period T, discounted back to time period 0 at adiscount rate of 8 percent.

'Saving is the difference between after-tax income REFERENCESand consumption. The old spend all of their wealthby assumption (there are no bequests). Thus, in this Bosworth, Barry, Tax Incentives and Economic Growthsimple model, spending exceeds income (interest on Washington, D.C.: The Brookings Institution, 1984.savings less tax) by the amount of the prior period's Christian, Charles and Frischmann, Peter, "Attritionsavings. Savings in period t is thus the negative of Bias in the Statistics of Income Individual Panelsavings in period t - 1. Tax Data," Arizona State University, manuscript,

'Ihe reason is that withdrawals are taxed under a October 28, 1988.deductible IRA, so every dollar invested is subject to Congressional Budget Office, "The IRA Proposal Con-tax of T in the future. Thus, as compared to a back- tained in S. 1682: Effects on Long-Term Revenuesloaded IRA, $1/(I-T) invested in a deductible IRA and on Incentives for Saving," Staff Memorandum,Yields IlAl - T)I(l - T), which equals one dollar. November 2, 1989.

'Ihese individuals (at point 0) have a marginal rate Deaton, Angus, "Comment" in M. Feldstein (ed.) Theof substitution (MRS) of current for future consump- Effects of Taxation on Capital Accumulation, Chi-tion between 1 + rb and 1 + r(l - t), In other words, cago: University of Chicago Press, 1987, pp. 48-51.the MRS equals I + r', where rb @r' a: r(l - t). They Feenberg, Daniel and Skinner, Jonathan, "SourceA ofwill be induced to save after the introduction of ERM IRA Saving," NBER Conference Volume, Novem-if r > r'. ber 1988.

'This observation was first reported by Galper and Feldstein, Martin, and Feenberg, Daniel, "Alterna-Byce. See Galper and Steuerle. tive Tax Rules and Personal Savmg Incentives: Afi-

'See Feenberg and Skinner, for example. This issue croeconomic Data and Behavioral Simulations," inis discussed in more detail in the section on "non-life- M. Feldstein (ed.) Behavioral Simulation Methodscycle model of IRA behavior," below. in Tax Policy Analysis, Chicago: University of Chi-

sthe percentage of returns with one spouse at the cago Press and NBER (1985).limit divided by the percentage of returns with con- Gale, William and Scholz, John, "IRAs and Saving"tributions for each return type (from the last coluum UCLA Department of Economics, manuscript, Aprilof-4'abb@4). For onc4y*bor-rothms, this is 1 -p"nt 26,@19M.-divided by 4 percent; for two earners, it is I percent Galper, Harvey and Byce, Charles, "Individual Re-divided by 5.7 percent. Numbers don't correspond ex- tirement Accounts: Facts and Issues," Tax Notes,actly to numbers in the table because of rounding. Jim 2, 1986, pp. 917-921.

'See Feldstein and Feenberg. Galper, Harvey and Eugene Steuerle, "Tax Incen-'00ne or two spouses at the limit on joint returns. tives for Saving," The Brookings Review, Vol. 2, no.

Even adopting a strict definition of limit contribu- 2, Winter, 1983.tions (both spouses at limit), 53 percent of contribu- Gravelle, Jane, "Do Individual Retirement Accounts

No. 31 LEONARD BURMAN, JOSEPH CORDES AND LARRY OZANNE 281

increase Savings?" Journal of Economic Perspec- use panel of individual income tax returns as-tives, forthcoming. sembled by the Office of Tax Policy Research

Hall, Robert E., "Intertemporal Substitutions in Con- at the University of Michigan. The panel cur-smption;'Journal of PolUical Economy, April 1988, rently includes data for the years 1979 through96, pp. 339-357.

Manegold, James and Joines, Douglas, "IRAs and 1984. The returns were randomly selected basedSaving. An Empirical Analysis with Panel Data," on the social security numbers of the primaryUniversity of Southern California School of ic- taxpayer. The number of returns varies fromcoahiftr, manuscript, January 1990-. ofiiWWptc-,ut-

Venti, Steven and Wise, David A., (1987a) "IRAs and backs at the IRS, but also because of attri-Saving," in M. Feldstein (ed), The Effects of T-- tion.' Table A2 shows the sample sizes for 1982tion on Cqpitat Accumulation, Chicago: University through 1984 (the years studied for this paper)of Chicago Press, pp. 7-48. before and after the data selection steps de-Venti, Steven and Wise, David A., (1987b) "Have MAsincreased U.S. Savine.: Evidence from the Con- scribed below.sumer Expenditure Survey," National Bureau of Observations were retained if they had validEconomic Research Working Paper No. 2217. wage information for both primary and second

Venti, Steven and Wise, David A., (1989) "The Sav- taxpayer (if joint return), if the reported wnountmg Effect of Tax Deferred Retirement Accounts: of MA contribution was less than or equal toEvidence from SIPP," Dartmouth College Depart- the appropriate contribution limit, and if nei-ment of Economics, manuscript, March 1989. ther the primary nor the secondary taxpayer

had taken an age exemption, Wages were es-

Appendix timated as the sum of reported wages plus es-timated self-employment income, defined as the

This appendix presents the assumptions and self-employment tax divided by the self-em-numbers underlying the hypothetical revenue ployment tax rate. The imputation of 8elf-em-losses illustrated in Figure 1 and describes the ployment income would underestimate that in-data used for the Tables in the text. come for some high-income taxpayers, but that

is not a problem because wages were only used

Assumptions for Hypothetical Revenue @Ocompute IRA limits (so the only real issueis whether each spouse's wage was greater than

Loss or equal to the maximum IRA contribution ofFigure 1 illustrates the pattern of revenue $2,000). Wages for the secondary eamer on a

losses attributable to a single contribution to joint return were imputed by dividing the cmffitan IRA or savings account with an after-tax for two-earner joint returns by the credit rate.cost of $1,000. For the example, it is assumed Wages for the primary earner were estimatedthat the account is withdrawn in equal por- as total wages less wages for the secondarytions (after tax) at the end of years 21 through eamer. In nine cases, the estimated wage for30. Thus, the IRA or savings account is with- the primary earner was less than the wage fordrawn as a fixed annuity. the secondary earner and also less than $2,000.

The interest rate and the discount rate for Since this inconsistency would affect the esti-computing present values are both assumed to mate of the IRA limit, these returns werebe 8 percent. The taxpayer's marginal tax rate dropped.is 28 percent and is assumed to remain con- The IRA limit was computed using the fol-stant over the taxpayer's life. lowing formula:

Table Al shows the data underlying Figure1. The bottom portion of the Table illustrates IRA limit = min 12000, Primary Wage)the equivalence of backloaded and fully de-ductible IRAs; both provide the same after-tax + max (250, min (Secondary Wage, 2000)1annuity for equal after-tax contributions. The x JOINT,50 percent deductible and nondeductible IRAsprovide successively smaller annuities, whichcorrespond exactly to the smaller revenue losses where JOINT is a zero-one dummy set to oneto the government. for joint returns. For most single, head-of-

Detailed explanation of the methodology un- household, and married-filing-separate re-derlying Table Al is provided in the Appendix turns, the computed limit was $2,000; for mostto Congressional Budget Office (1989). joint returns, it was $2,250 if the spouse didiit

have earned income or $4,000 if both spouseshad income. About one percent of returns had

Data Underlying Tables reported IRA contributions greater than theThe data used in our analysis and summa- computed limit. These amounts probably were

rized in Tables 2 through 6 are from a public- rollovers of pensions into IRAs. Since there was

Table Al. Data Underlying Figure 1: IRABalance and Re

Type of AccountBadd=dW FulyDeductble 50% Deducbble

IRA Revenue IRA Revenue IRA Revenue IRYear3alance Loss Balance Loss Balance Loss Balan

0 $1,000 $0 il,389 $389 $1,163 $163 $11 $1,080 $22 $1,500 $22 $1,256 $22 $12 $1,166 $24 $1.620 $24 $1,356 $24 $13 $1,260 $25 $1,750 $25 $1.465 $25 $14 $1,360 $26 $1,89D $26 $1,582 $26 $15 $1,469 $28 $2,041 $28 $1.709 $28 $16 $1,587 $30 $2,204 $30 $1,845 $30 $17 $1,714 $31 $2.380 $31 $1.993 $31 $18 $1,851 $33 $2,571 $33 $2,152 $33 $19 $1,999 $35 $2,7"76 $35 $2,324 $35 $1

10 $2,159 $37 $2,999 $37 $2,510 $37 $211 $2.332 $39 $3,238 $39 $2,711 $39 $212 $2,518 $41 $3.497 $41 $2,928 $41 $213 $2,720 $44 $3,777 $44 $3,162 $44 $214 $2,937 $46 $4.079 $46 $3.415 $46 $215 $3,172 $49 $4,406 $49 U689 $49 $316 $3,426 $52 $4,758 $52 $3,984 $52 $317 $3,700 $55 $5,139 $55 $4,302 $55 $318 $3,996 $58 $5,550 $58 $4,647 $58 $319 $4,316 $61 $5,994 $61 $5,018 $61 $420 $4,661 $65 $6,474 $65 $5,420 $65 $421 $4,339 $69 $6,027 ($201) $5,046 ($141) $422 $3,992 $63 $5,544 ($207) $4.642 ($146) $323 $3,616 $58 $5,023 ($212) $4,205 ($152) $324 $3,211 $52 $4,460 ($218) $3,734 ($158) $325 $2.773 $46 $3,852 ($224) $3,225 ($164) $226 $2,301 $39 $3,195 ($231) $2.675 ($171) $227 $1,790 $32 $2,486 ($238) $2,082 ($178) $128 $1.239 $25 $1,720 ($245) $1,440 ($185) $129 $643 $17 $893 ($253) $748 ($193)30 $0 $9 $0 ($261) $0 ($201)

Pm-seM Vakie $407 $407 $268AnnuityAniount

Pre-Tax $695 $965 $808Exclusion % 100.00% 0.00% 7.20%Tax $0 $270 $210After-Tax $695 $695 $598

No. 31 LEONARD BURMAN, JOSEPH CORDES AND LARRY OZANNE 283

Table A2. Annual Sample Sizes for Michigan Panel, 1982-1984(Before and After Selection Criteria)

Observations After Selection CriteriaOriginal

Year Observations Delete Bad Delete Age Returns WithWages ExeTZ[ons Valid IRAs

1982 9,235 9,229 8,352 8,262

1983 19,186 19,176 17,489 17,292

1984 9,783 9,777 8,911 8,798

no way to know in these cases what portion of We also computed positive income-a mea-the reported IRA was a new contribution and sure intended to better represent actual in-what portion was a rollover, the returns were come than adjusted gross income (AGI). It isdropped. similar to the measure of income used by the

Returns where an age exemption was claimed Congressional Budget Office for its distribu-were dropped because these taxpayers faced a tional analyses of tax provisions. Positive in-different set of IRA constraints. While the ideal come is the sum of AGI, the excluded portionwould have been to drop returns where one of capital gains, losses from partnership andtaxpayer was over 59 and one-half (and with- closely-held businesses, and net losses reporteddrawals could be made without penalty), age from rental housing, farining, and sole propri-information was not available on the panel. etorships.