Short-term revenue effects of overall limits on exceptionally large retirement accounts

Introduction

The U.S. has long provided tax-preferred treatment for both employer-sponsored retirement plans and Individual Retirement Accounts (IRAs). The purpose of these tax preferences is mainly to help people achieve an adequate living standard in retirement. The preferences are not, however, intended to promote unlimited accumulation of wealth.¹

Nevertheless, in recent years, it has become apparent that the private retirement saving system permits the accumulation of vast sums of tax-favored benefits. For example, the congressional Joint Committee on Taxation (JCT) reports that over ten thousand individuals have balances exceeding ten million dollars each, with over a thousand exceeding $25 million (JCT 2024, p. 232). Other reports indicate that at least one of these “mega-IRA” balances is on the order of $5 billion (Elliott, Callahan, and Bandler 2021).  Many policymakers and others view these sums as far exceeding what could reasonably be described as an adequate retirement nest egg.

In response to these issues, legislative proposals have been advanced to limit additional tax-favored contributions to such unusually large retirement accounts and/or require disgorgement of balances that exceed certain high thresholds. These proposals would not limit the overall amount of saving that individuals can do; they only would restrict the extent to which tax preferences would partially subsidize balances of this magnitude. However, such proposals raise many issues, including the impact on overall saving, employers’ willingness to offer plans, federal revenues, the distribution of income, and administrative considerations.²

The revenue effects alone have several dimensions. Limiting exceptionally large retirement accounts by prohibiting further contributions and/or requiring disgorgement of “excess” balances would generate immediate revenue for the government. At the same time, such proposals would also change future revenues, both upward and downward. For example, earnings on the disgorged balances and prohibited contributions would be subject to annual taxation rather than being left in tax-preferred accounts, leading to higher future revenues. But the policies would reduce future taxes owed on withdrawals that would otherwise have occurred in future years by instead requiring taxable withdrawals sooner. A full revenue estimate would consider all these effects.

In this report, we provide estimates of one aspect of the overall calculation: the short-term revenue that would be generated from placing limits on exceptionally large retirement accounts. To estimate these effects, we use data from the 2022 Survey of Consumer Finances (SCF)—the most recent available high-quality data—coupled with analysis using the NBER’s TAXSIM model. The SCF is a cross-sectional wealth survey conducted by the Federal Reserve Board that contains high-quality data on household assets and income as well as an oversample of affluent households. TAXSIM allows calculations of households’ tax liabilities under a variety of assumptions.

We obtain several results. First, eliminating tax-deductible contributions for retirement accounts that already contain at least $5 or $10 million would raise very little revenue on an immediate basis (less than $0.15 billion per year or $0.05 billion per year, respectively). Second, and in contrast, requiring immediate disgorgement of excess balances would raise larger amounts of revenue on an immediate basis—$84 billion for accounts in excess of $5 million, or $11 billion if the limit were set at $10 million. In all the examples, however, future revenue would change depending on rates of return, future tax rates, and the timing of distributions that would have been taken in the absence of the policy. Thus, we emphasize that these estimates should be understood not as stand-alone figures but as just one of the several inputs that would be needed to undertake a full revenue estimate. Understanding the full revenue impact and the broader implications of proposals to limit large retirement accounts represents an important area for future research.

Tax preferences for retirement saving

How do tax preferences for retirement saving plans work?

When individuals save in conventional, fully taxable vehicles (like a bank saving account), (a) their deposits or contributions are not tax-deductible, (b) earnings generally are taxable as they accrue, and (c) withdrawals are exempt from taxation³ (see Table 1). Retirement plans—including IRAs and defined contribution (DC) plans like 401(k)s—receive more favorable treatment. In traditional (pretax) IRAs and DC plans, employee contributions are made with pretax dollars and account balances accrue earnings tax-free, but withdrawals are fully taxed as ordinary income. Roth accounts—both IRAs and 401(k)s—also receive preferential treatment but in a different form than traditional/pretax retirement accounts. For Roth accounts, the contribution is not deductible, but returns accrue tax-free and withdrawals are generally tax-free. Employer contributions are deductible when made to the plan—even though they are not taxable to employees until distributed from the plan—and are excluded from employees’ income for payroll tax purposes.  These contributions, along with any growth and earnings, are taxed as ordinary income when distributed from the plan.

There are a variety of constraints on retirement contributions and withdrawals. In most cases, contributions can be made only by people with earned income and are subject to annual limits. In 2026, annual contribution limits for traditional and Roth accounts are $7,500 for IRAs and $24,500 for employee contributions to 401(k) plans. DC plan participants and IRA owners who have reached specified ages can make limited additional “catch up” contributions. Employers can also contribute to DC plans on behalf of their employees. The sum of employer and employee contributions to a DC plan is subject to a combined 2026 annual limit of $72,000 plus catch-up contributions allowed to older employees (Internal Revenue Service 2025).

Roth IRAs allow direct contributions only by those with income below certain thresholds, but those with higher income have indirect means of establishing and adding to Roth balances.⁴ Traditional/pretax IRAs are available to those who do not participate in an employer-sponsored retirement plan and also to plan participants whose income falls below certain thresholds.

In traditional retirement plans and IRAs, withdrawals prior to age 59 ½ for non-qualified purposes are subject to a 10% penalty in the form of an additional tax on the withdrawal. In Roth plans, penalties apply to funds held for only short periods of time.

How large are the tax expenditures for retirement saving?

As noted, the tax treatment of retirement saving plans is more generous than the treatment of conventional saving instruments, meaning that less tax revenue is collected than if the tax preference for retirement did not exist. The reduction in tax revenue attributable to these types of special provisions in the tax code is known as a tax expenditure. The Office of Tax Analysis (OTA) at the U.S. Department of the Treasury and the JCT estimate that the 2025 tax expenditure was $293.4 billion and $384.3 billion, respectively, for the entire private pension and retirement saving system (OTA 2024; JCT 2025).⁵ Our focus in this paper is on IRAs and tax-qualified DC plans (such as 401(k)s, section 403(b) plans, and section 457(b) plans) rather than defined benefit plans. For those items, the OTA and JCT estimates are $224.5 billion and $249.6 billion, respectively.

These figures show the overall magnitude of subsidies for retirement saving plans, but they are not directly related to the estimates that we present below for several reasons. We consider the immediate revenue effects of limiting contributions for those with balances above a given threshold and, in some scenarios, disgorging excess balances from those accounts. In contrast, the tax expenditure estimates noted above for a given year consider the differences in the tax treatment of contributions, accruals, and withdrawals and include the tax-related effects of accruals (tax-free or tax-deferred inside buildup of earnings or gains) and of withdrawals resulting from contributions made in previous years.

How do exceptionally large balances in retirement accounts exist?

The annual contribution limit was $1,500 when IRAs were created in 1974 and has been increased gradually to $7,500 in 2026 (Myers 2020). As an illustrative calculation, suppose someone born in 1960 saved the maximum IRA contribution he or she was eligible for—including catch up contributions—every year between 1980 and 2025 and invested in the S&P 500 index using a “buy and hold” strategy. The $181,000 in total contributions made over the saver’s working years would have accumulated to approximately $3 million at the end of 2025. 

Yet data show that many savers have accumulated much larger balances than this simple calculation would suggest. JCT reported that 3,625 people had accumulated traditional or Roth IRA balances above $10 million as of 2019; that number increased to 10,196 by 2021 (JCT 2021, 2024). Among the 1,049 taxpayers with balances above $25 million, the average balance is approximately $90 million, with the amounts in Roth accounts permanently exempt from future taxation. In 2021, ProPublica reported that one Roth IRA had accumulated over $5 billion from an initial $2,000 investment (Elliott, Callahan, and Bandler 2021). How is this possible?

While the path to exceptionally large balances likely differs from account to account, in the case described by ProPublica, the IRA owner reportedly used his Roth IRA to purchase shares of early-stage startups at what turned out to be extremely low valuations. Because those shares exploded in value, all the gains accrued inside the Roth account—without triggering capital gains taxes or income tax on earnings.⁶

Proposals for reform

Historically, there have been various caps on the combined aggregate value of, or benefits from, retirement accounts, but several pieces of legislation removed them all (see Appendix A for a summary). In light of reports of exceptionally large retirement accounts, proposals to reinstate some form of overall combined limit on retirement accounts have received attention from recent administrations and legislators. A brief review of several prominent proposals indicates the variety of issues that arise when designing such a limit (see Appendix B for more on this).

The Obama administration proposed to prohibit further contributions or benefit accruals for people whose aggregate retirement plan balances (across all plans in which they participated) were large enough to provide a life annuity at least as large as the maximum life annuity permitted to be paid under a defined benefit plan (U.S. Department of the Treasury 2013, pp. 165-67).⁷ Contributions that caused the plan to exceed the limits would be taxable, and accruals within the plan that caused the plan to exceed the limits would be required to be withdrawn and taxed.

Building on the Obama-era proposal, the Biden administration proposed to prevent what it called “excessive accumulations” in tax-favored retirement accounts by requiring distributions of aggregate vested balances from DC plans and IRAs when the total exceeded $10 million. Under the proposal, at least 50% of the excess was required to be paid out each year. If the aggregate vested account balances exceeded $20 million, the required distribution would be the excess over $20 million or, if smaller, the portion of the saver’s balance that is held in Roth IRAs or Roth DC plan accounts. However, distributions from Roth IRAs or accounts would still be tax-free (JCT 2024). 

In 2016 Senator Ron Wyden (D-OR) circulated a draft proposal that would have imposed a $5 million limit on Roth IRAs only, grandfathering existing balances that exceeded that amount (Senate Finance Committee 2016). In 2021, House Ways and Means Committee Chair Richard Neal (D-MA) put forward—and the Committee approved—a somewhat similar proposal with a $10 million limit but no grandfathering of existing balances (Congressional Research Service 2021, p. 53).

One way to frame limits on exceptionally large balances is through the lens of Required Minimum Distributions (RMDs), the legally mandated withdrawals that people must take from most tax‑deferred retirement accounts once they reach a certain age. Congress created these rules to ensure that tax revenue is eventually collected on tax preferred retirement contributions and their earnings and to promote the use of these accounts for retirement security rather than for building wealth for future generations. Current law requires that most individuals begin taking RMDs from non-Roth accounts at age 73, and the starting age is scheduled to rise to 75 in 2033 (Myers 2024). Withdrawing less than the required minimum results in a tax penalty. 

Imposing limits on exceptionally large balances can be thought of as an alternative to an age-based RMD policy: Instead of requiring distributions when an account owner exceeds a certain age, such a policy would require distributions when an individual’s accounts together exceed a given dollar threshold. In other words, a policy that imposes RMDs for exceptionally large balances could potentially raise enough revenue to pay for a broad simplification by fully exempting the millions of seniors whose aggregate retirement savings are below a specified balance from the current, age-based RMD rules.⁸ This change could be designed to be budget-neutral from the perspective of tax revenue and could simplify and increase the efficacy of the retirement tax subsidy in improving retirement security.

Estimated effects of imposing overall limits on short-term revenue

We aim to estimate the short-term revenue impacts of two forms of limits: 1) a limit on further contributions to retirement accounts that takes effect when aggregate balances exceed a “contribution threshold,” and 2) a requirement to distribute aggregate balances exceeding a “disgorgement threshold.” To establish a baseline, we first determine the short-term revenue impacts of the status quo—i.e., contribution limits that would apply regardless of account size—and then see how short-term revenue would change with the imposition of the two forms of limits we consider.

Data and methods

We begin with data from the 2022 SCF, a nationally representative, cross-sectional sample that contains comprehensive information on households’ retirement account balances, including both IRAs and DC plans. The SCF oversamples affluent households, providing good representation of wealth measures in the upper tail of the wealth distribution. The survey also provides high-quality data on income and other information necessary to determine a household’s tax liability.

To calculate the revenue effects of alternative policies, we use the SCF data in conjunction with NBER’s TAXSIM model, which allows us to generate household-level income tax liability under each policy considered. To estimate the short-term revenue impacts of limiting contributions or requiring disgorgement of funds, we simply add the amounts in question to taxable income.

We emphasize that our estimates measure only the immediate revenue effects of changes in policy. They do not account for the taxes that would be owed at some future date in the absence of the policy, nor do they incorporate the effects of tax-favored treatment of inside build-up (investment returns within accounts) over time or the effects of contributions made in prior years. In addition, we assume that any required disgorgement of funds occurs immediately. In practice, such a policy would likely allow taxpayers to spread the tax burden over a few (e.g., three or five) years.

Appendix C describes our methods in more detail, including how we allocate account balances among household members and between Roth and non-Roth accounts, and how tax liabilities were estimated using the TAXSIM model. We also report estimates under alternative assumptions. Appendix D describes the results of a variety of benchmarking exercises to assess the SCF’s representation of large retirement accounts and compares our estimates of forgone tax revenues to tax expenditure estimates reported by OTA and JCT.

Short-term revenue effects of limiting contributions

To establish a baseline, we first report the short-term forgone revenue from a policy that would limit all contributions regardless of the size of one’s existing balances. We estimate that the immediate forgone tax revenue from retirement contributions to DC plans and IRAs totals $183.1 billion.⁹

Figure 1 displays the distribution of this baseline forgone tax revenue across account size categories at the individual level. The immediate tax benefits from allowing tax-deductible contributions to retirement accounts as currently legislated are heavily concentrated among individuals with the largest retirement account balances. In particular, while only about 2% of individuals have account balances above $1 million, these individuals account for approximately 19.5% of the total forgone revenues from retirement contributions.

A policy that prohibits contributions to tax-favored retirement accounts with balances above $5 million would raise only a very small amount of revenue: $147.8 million, or less than one tenth of one percent of the total forgone revenue from all contributions to tax-favored retirement accounts. Even less immediate revenue would be raised if the contribution threshold were higher, as shown in Figure 2. Specifically, raising the contribution threshold to $7.5 million or $10 million would raise $101.4 million or $47.1 million, respectively.

Mandating withdrawals of the excess from accounts with balances exceeding a given threshold has a much larger impact on immediate revenue collections. For example, our estimate suggests that requiring immediate disgorgement of balances in excess of $5 million would raise $84.3 billion in immediate revenue as of 2022. We also illustrate the estimated revenue impacts of a contribution threshold of $5 million combined with disgorgement thresholds of $7.5 million and $10 million in Figure 3. As shown in the figure, a higher disgorgement threshold raises a smaller amount of immediate revenue, $37 billion and $11 billion, respectively.  Note that these amounts are in addition to those raised from limiting contributions.

Conclusion

The rationale for tax preferences for retirement savings is to provide incentives for people to save for retirement, ultimately leading to improved financial security. While annual contributions to retirement accounts are already limited by law, current law imposes no overall dollar limit on the amount of tax deferral or exemption of inside buildup from outsized investment gains, and the tax preferences disproportionately benefit those who have already accumulated large amounts of retirement wealth. Thus, it has been argued that limiting tax preferences for exceptionally large retirement accounts would reduce inequity and raise revenue with virtually no impact on retirement security.

We analyze the immediate revenue impacts of freezing contributions and required withdrawals from accounts that exceed certain thresholds. Our estimates suggest requiring disgorgement of funds in retirement accounts that exceed certain limits can raise substantial revenue in the short run. In contrast, limiting contributions to large accounts would raise almost no revenue. In both cases, however, future revenue would change, though the amount is uncertain and depends on how the funds are invested, when withdrawals would otherwise be taken, the affected savers’ marginal tax rates, and the appropriate discount rate. These considerations are some of the important inputs that would go into an overall assessment of the social value of limiting accumulation of funds in tax-preferred retirement accounts.

Appendix A: Historical limits on tax preferences for retirement

Combined plan limit

Enacted as part of the Employee Retirement Income Security Act (ERISA) in 1974, a “combined plan limit” was imposed on an individual who participated in both a defined benefit (DB) and a defined contribution (DC) tax-qualified plan of the same employer. Plan administrators had to look back and track each plan participant’s cumulative employee and employer contributions to that employer’s DC plan(s), project the participant’s future DB benefits under the employer’s DB plan(s), and annually calculate each year’s permissible increment to the participant’s aggregate, cumulative lifetime benefits.

15% excise tax on excess distributions and 15% estate tax on excess accumulations

In the Tax Reform Act of 1986, Congress tried to improve significantly on the combined plan limit by adding a new provision designed to take a more comprehensive approach (JCT 1987, p. 754-60). It applied cumulatively to the aggregate of all plans covering an individual, including those sponsored by all of the individual’s employers over the course of the worker’s career, as well as any of the individual’s IRAs and tax-sheltered annuities. The provision imposed a 15% excise tax on the aggregate amount of “excess distributions” in a given year from an individual’s retirement plans or IRAs and a 15% estate tax on the aggregate amount of “excess accumulations” in an individual’s retirement plans or IRAs. Excess distributions were defined as the aggregate amount of an individual’s retirement plan payouts in a given year that exceeded a specified, inflation-indexed dollar amount (or lump-sum payouts that exceeded five times that amount). Originally, as enacted in 1986, the tax applied to the extent that the individuals’ retirement plan payments for the year exceeded $150,000 or to the extent that an aggregate lump-sum distribution exceeded $750,000 (both amounts indexed for inflation).

Pension simplification initiative and repeal of combined plan limit and excise tax

In 1995, as part of a broad pension simplification initiative in the executive branch and in Congress, the U.S. Treasury Department developed a set of legislative proposals that included repeal of the combined plan limit.¹⁰ The rationale for repeal was threefold. First, the combined plan limit was inadequate in a number of respects. Although it aggregated DB and DC plan benefits, the limit applied only on an employer-by-employer basis, so an individual who changed employers or worked for multiple employers got the benefit of a fresh start with the new employer. The limit also failed to include IRA contributions or balances and did not apply to investment earnings in DC plans.

Second, the combined limit suffered from administrative complexity and therefore presented an ideal target for repeal as part of a pension simplification initiative. Universally known to pension professionals by its tax code section—“415(e)”—the combined limit was sufficiently difficult to understand and administer that it was widely regarded as perhaps the single most complex provision of the famously complex body of pension tax law.  

Finally, the combined plan limit was largely redundant with the excise tax and estate tax imposed by the Tax Reform Act of 1986 described above, which was more cumulative, aggregative, and comprehensive. The new provision corrected the single employer focus of the combined plan limit by measuring career-long pension accumulations and distributions from plans sponsored by all of an individual’s employers over time as well as from the individual’s IRA(s).¹¹

The White House thus announced pension simplification reforms in 1995 that proposed to repeal the combined plan limit but retained the more comprehensive 15% excise tax provision in order to potentially broaden it to become a more aggregative, cumulative, and effective restriction on “excessive” retirement benefits. In 1996, Congress enacted legislation that adopted various executive branch pension simplification proposals including repeal of the combined plan limit (Congress.gov 1996, section 1452(a)). To avoid the redundancy of limits, Congress suspended the 15% excess distributions excise tax temporarily, until 2000, when the repeal of the combined plan limit was scheduled to take effect (Congress.gov 1996, section 1452(b)).

But before a new proposed expansion of the 15% excise tax provision could be developed and put forward, the political winds shifted decisively with the 1996 congressional elections, which led to more deregulatory and anti-tax policies under the leadership of House Speaker Newt Gingrich. In 1997, just a year after voting to repeal the combined plan limit, largely because of its considerable redundancy with the 15% excess distributions tax, Congress repealed the 15% excess distributions tax and the 15% excess accumulations estate tax (JCT 1997, pp. 262-63). Congress cited three reasons for repealing this remaining aggregate limit. First, acknowledging the need to limit tax-deferred saving by individuals, it deemed the existing annual limits on contributions and benefits to be sufficient. Second, Congress stated that “[a]dditional penalties are unnecessary, and may also deter individuals from saving.” Finally, it concluded that “the excess accumulation and distribution taxes also inappropriately penalize” successful investing (JCT 1997, p. 263)

Appendix B: Selected considerations in designing a comprehensive limit on exceptionally large retirement balances

As illustrated by the various proposals that have been put forward to date, structuring a comprehensive cap involves a range of design options, choices, and policy issues. The following is a non-exhaustive summary of some of the key decisions in designing a proposal: 

• Plans covered by limit: Should the new dollar cap apply to all tax-qualified plans (DB and DC plans) and all IRAs, or a subset? If DB benefits are included, how will they be compared to DC plan balances?
• Individuals covered: Should the rules apply only to certain more highly compensated employees?
• Size of the limit: What should the dollar threshold be? 
• Consequences of exceeding the limit: Should the consequence be merely a freeze on new contributions or also mandatory distribution/disgorgement of excess amounts? Should the dollar threshold that triggers a contribution freeze also serve as the threshold triggering disgorgement, or should the latter be set at a higher level? 
• Grandfathering: Should existing accounts be exempt from required disgorgement of excess balances if they already exceed the threshold as of a specified effective date?
• Enforcement mechanisms: Should compliance with required disgorgement be enforced through the RMD excise tax or by some other method? 
• Compliance safeguards: Should tax-free rollovers to certain other qualified plans that might not be subject to the overall limit (such as DB plans under some proposed limit designs) be prohibited once the threshold has been crossed?
• Information reporting: How will the IRS obtain the data needed to administer the rule? Should plan administrators, IRA trustees (beyond their existing reporting requirements), or individual savers be responsible for reporting account balances and related information?  How costly or difficult would such required reporting be?
• Interaction with Roth transactions: Should the proposal address conversions or rollovers into Roth IRAs or Roth accounts in DC plans? If so, how? 

Appendix C. Estimating forgone tax revenues from the SCF

We use data from the 2022 wave of the Survey of Consumer Finances (SCF) to run the National Bureau of Economic Research’s TAXSIM microsimulation model, a tax calculator that allows us to estimate taxes owed under both present and alternative tax rules. Running TAXSIM with SCF data first requires separating households surveyed in the SCF into tax units, a process described by Gale et al. (2022) and Gale and Sabelhaus (2024). We remove non-filing tax units from our analysis.

To simulate taxes owed if tax preferences for retirement plans did not exist, we assume all retirement contributions (employer and employee contributions to DC plans and deductible IRA contributions) are instead earned as income. To make this adjustment in TAXSIM, we add these contributions as “other non-property income.” The difference in federal income taxes owed under this scenario and taxes owed under present law (where contributions are not included in one’s income) represents the forgone revenue from allowing taxpayers to deduct these contributions. When simulating policies that mandate disgorgement, we similarly treat disgorged balances as taxable income and compare tax liabilities under this alternative rule to present law. We add disgorged balances to the “pension” field when running TAXSIM.¹²

Our measure of retirement account balances is based on the SCF Bulletin household level summary variables that combine DCs and IRAs for the respondent and spouse and do not distinguish between Roth and non-Roth IRA balances and contributions. The alternative policies we consider, however, would apply any restrictions to accounts at the individual level and would treat disgorged balances differently depending on account type (i.e., Roth distributions would not be subject to taxation). We must therefore take a stance on how balances and contributions are divided between account types and among individuals in married tax units.

We address uncertainty in how total account balances—the sum of DC, Roth IRA, and non-Roth IRA balances—are allocated across individuals in married tax units by simulating two scenarios for each policy change we consider: one in which balances are held entirely by one individual in the tax unit, and another where balances are split equally between members. For example, consider a policy that limits contributions by individuals with over $10 million in total account balances. Allocating balances equally to both members of a tax unit with $12 million in balances means neither individual’s balances ($6 million each) fall above the $10 million threshold. However, if we assume the entire $12 million belongs to one person, then that individual could no longer contribute to his or her accounts on a tax-preferred basis under this policy. The allocation method we select particularly matters when estimating the revenue effects of policies that require account disgorgement. In the first scenario, with balances divided evenly between spouses, no disgorgement would occur. Under the second approach, however, the individual holding the balances would need to withdraw and pay taxes on $2 million in balances.

We similarly approach the question of how IRA balances and contributions are split between Roth and non-Roth accounts by considering two scenarios. In the first, we assume all individuals hold a fixed share of total IRA balances in non-Roth accounts. The fixed share in non-Roth accounts is taken from the Joint Committee on Taxation’s “Description of the Revenue Provisions Contained in the President’s Fiscal Year 2025 Budget Proposal” (p. 232), which estimates that 63.9% of aggregate IRA balances owned by taxpayers with total IRA balances of $10 million or higher were held in non-Roth IRAs (JCT 2024). We likewise assume a share of IRA contributions are directed toward Roth accounts. We estimate the share based on data reported by the IRS Statistics of Income that provides information on aggregate contributions into Roth and non-Roth IRAs in 2022 by age (Myers 2025).¹³

In our second approach, we assume all IRA balances are held in non-Roth accounts and all contributions are made on a pretax basis. Under policy alternatives that require disgorgement of account balances, we assume individuals disgorge balances proportionally to balances in each account type. If an individual holds half of their balances in DC plans, we assume half of the balances he or she disgorges will come from DC accounts and the remainder from IRAs, with the share of IRA balances that are taxed depending on how we allocate IRA balances between Roth and non-Roth accounts. Thus, the tax revenue from disgorged balances is maximized under the second approach where we assume all IRA balances are held in non-Roth accounts and all contributions were made on a pretax basis.

The main text presents a revenue estimate of the first approach outlined above, which allocates balances equally between spouses in married tax units and assumes a fixed share of total IRA balances and contributions are attributable to non-Roth accounts. This approach represents our best estimate of the revenues that would be raised from alternative policies. Estimates from the second approach, where we assume all balances are held by one individual and all IRA balances and contributions belong to non-Roth accounts, represent an upper bound revenue estimate using data from the SCF and are provided below.

Appendix D. Benchmarking exercises

We benchmark our data and methods in two ways. We first compare SCF data on large IRA balances to similar data from JCT. Next, we compare our tax expenditure estimates to figures published by JCT and the U.S. Treasury Office of Tax Analysis (OTA).

Comparison of SCF and IRS data on large balances

JCT’s “Description of the Revenue Provisions Contained in the President’s Fiscal Year 2025 Budget Proposal” (p. 232) documents the number of taxpayers in 2021 with $10 million or more in total IRA balances and the total wealth held in these accounts, distinguishing between traditional and Roth IRA balances and classifying taxpayers into one of three balance ranges (at least $10 but less than $15 million, at least $15 but less than $25 million, and at least $25 million) (JCT 2024). We replicate this analysis using SCF data in Appendix Table D.1 to assess how data from the survey compares to data reported by JCT. We focus here on traditional (non-Roth) IRAs. As described in Appendix C, we calculate the share of total IRA balances held in traditional accounts in the SCF by assuming all individuals hold a fixed share of total IRA balances in traditional IRAs , equal to the rate implied by the JCT reported aggregates for accounts above $10 million.

We begin by gauging how well the SCF captures the number of taxpayers with account balances of at least $10 million. In the SCF, we calculate the implied number of taxpayers by summing the weights assigned to those observations that fall into that balance range. In comparing the first and second columns, the SCF overcounts the number of account owners in the first two balance ranges ($10 to $15 million and $15 to $25 million) and undercounts those with at least $25 million. Whereas JCT finds 617 taxpayers with at least $25 million in traditional IRA balances, the SCF suggests that 372 individuals exceed this threshold.

We next sum the traditional IRA balances held by these taxpayers and compare the aggregate balances to those from JCT’s report, as shown in columns three and four. Among those in the first two balance ranges, aggregate balances in the SCF are comparable to JCT’s figures. For those with $10 to $15 million in traditional balances, aggregate balances are roughly $7 billion larger in the SCF than JCT’s estimate, while our estimate for the next range ($15 to $25 million) is around $7 billion lower. However, aggregate balances held by individuals with over $25 million in total IRA balances fall well below JCT’s estimate ($17.68 billion vs. $51.42 billion). This disparity holds even if we assume all reported IRA balances in the SCF are held in traditional accounts by one member of the tax unit.

We conclude this comparison by using JCT’s reported balances to estimate the revenue effects of requiring disgorgement by those with over $10 million in total IRA balances. We use average effective tax rates from TAXSIM to perform this calculation, as reported in the table notes.¹⁴ Given the lack of data on DC balances in the JCT report, we consider only revenue from IRA disgorgement when estimating revenue in the SCF to make the calculation comparable. Since disgorged revenue is drawn heavily from those with large balances, the revenue estimates calculated using JCT’s aggregate balances are predictably larger than those estimates in the SCF data. Overall, this comparison suggests that our revenue estimates of disgorgement based on the SCF data are likely to be understated.

Comparison with external tax expenditure estimates

We compare our estimate of the forgone tax revenue under the baseline policy to tax expenditure estimates published by JCT and OTA. In both cases, we compare estimates calculated using the 2022 wave of the SCF—which reflects information from the previous year—to external estimates from the same year. In comparing our results to other tax expenditure estimates, we discuss more generally how our methodology contrasts with other approaches to calculating tax expenditures.

Appendix Table D.2 summarizes how our tax expenditure estimates differ from those provided by JCT and OTA. We estimate forgone revenue from the tax subsidy for retirement contributions to be $183.1 billion in 2021. For the same year, JCT estimated a total tax expenditure of $204.2 billion from the net exclusion of contributions to and earnings on DC plans, traditional IRAs, and self-employed plans, and OTA estimated a total tax expenditure of $153.05 billion (JCT 2020). While we cannot isolate contributions to self-employed plans in the SCF, our total tax expenditure estimate nonetheless aligns closely with JCT’s total estimate. OTA’s total tax expenditure estimate, however, falls somewhat below both our estimate using the SCF and JCT’s figure, a difference primarily explained by OTA’s estimate for DC plans (OTA 2021).

Our estimate and the values reported by JCT and OTA represent annual cash-flow estimates of tax expenditures, or the “difference between tax liability under present law and the tax liability that would result if the tax expenditure provision were repealed,” leaving the rest of the tax code unchanged (JCT 2025). However, important differences exist between the three estimates, with JCT cautioning that their and OTA’s estimates of the same provision are not necessarily comparable. We briefly consider how the three methodologies differ.

JCT and OTA employ a more comprehensive approach to estimating tax expenditures for retirement plan contributions. Namely, they calculate tax expenditures as “the income taxes forgone on current tax-excluded pension contributions and earnings less the income taxes paid on current pension distributions,” (JCT 2025). Our approach does not consider lost revenue from taxes on current distributions nor lost revenue from taxes on currently tax-excluded earnings.

Like JCT and OTA, we do not consider behavioral changes by taxpayers, such as changes to saving behavior. However, JCT and OTA differ in whether taxpayers can take advantage of other parts of the tax code following the initial rule change. Under the JCT’s methodology, taxpayers can take advantage of any remaining provisions. The OTA, on the other hand, only models “tax form behavior” changes, such as switching from itemizing to instead claiming the standard deduction.¹⁵ The OTA clarifies, for example, that their approach would not allow taxpayers to direct contributions to IRAs if the tax-preferred treatment of DC plan contributions were repealed. Our approach thus aligns more closely with OTA’s methodology in this respect.

In addition to reporting annual cash-flow estimates of tax expenditures, OTA also constructs present-value estimates for tax provisions that involve long-term deferrals of tax payments. In the context of retirement contributions, cash-flow estimates can overstate tax expenditures, as taxes will eventually be paid once contributions made today are withdrawn. Present-value estimates address this concern by netting out the discounted value of future revenue that results from decisions made today.

Estimating the present value of tax expenditures, however, presents several challenges: Performing a present-value calculation requires one to take a stance on the discount rate, the growth rate on contributions, and the tax rate that will apply to future distributions. For policies that require disgorgement, one must also consider how taxpayers will use disgorged balances, such as whether they would place them in investments in which growth is taxed annually. Determining when and by whom retirement accounts will be inherited also matters if heirs face different tax rates than the individual making contributions today. For these reasons, we leave calculating present-value tax expenditure estimates to future work.