# 1. Introduction

;; 1.01
Consider a household in the top 0.1 percent of the income distribution in 2021. In the Survey of Consumer Finances, such a household more likely than not reports tax-exempt interest income: {{N:top01_share_with_munis_pct}} percent of tax units in that group do, against {{N:units_with_munis_pct:.1f}} percent of all tax units.{{FN: The top-0.1 percent cell rests on a few hundred oversampled observations, so the participation rate is a point estimate with nontrivial sampling error; the SCF also excludes the Forbes 400 by design, so the very top of the group is absent. The concentration pattern across groups, which is what the argument uses, is robust. Section 4 describes the data.}} The household's tax return reports that interest on line 2a of Form 1040, where it is counted as income and taxed at zero. Every maintained distributional framework in the United States records the same thing: municipal bond interest enters the income base at the yield the household received, and no tax is recorded against it. The measured conclusion is that this income escapes tax.

;; 1.02
The household bears a real burden anyway. When it bought the bond, it accepted a yield below what a comparable taxable bond paid. That forgone yield is a tax in everything but remittance: the household gives up income because of the tax system, and the issuing government collects the difference as a lower borrowing cost. Galper and Toder (1984) named it four decades ago: savers bid down the yield of tax-exempt assets, and "this decline in the before-tax yield constitutes an implicit tax." What the holder remits to the Treasury is zero; what she pays is not.

;; 1.03
No agency corrects for this, and the most direct confirmation comes from authors who defend the practice. Saez and Zucman (2024), in a methodological paper that treats municipal bonds as a litmus test, write that a tax-incidence approach "should logically assign muni investors a higher counterfactual pre-tax income... and then fictitious taxes they do not pay," and that "to the best of our knowledge... in practice US agencies and think-tanks follow our methodology and do not gross up muni interest income by fictitious taxes." They regard the convention as the right default for tax-reform analysis; Section 2 takes up their argument directly. Here it serves as certification of the gap: the correction this paper performs is recognized as the logical incidence treatment, and nobody does it.

;; 1.04
This paper does it, for the asset where it can be done from market prices. We combine household portfolios and incomes from the 2022 Survey of Consumer Finances (income year 2021), statutory marginal tax rates computed for each tax unit with NBER's TAXSIM calculator, and a market-implied municipal tax rate identified from the term structure of municipal yields. With these three inputs we recompute average effective federal income tax rates by income percentile, with and without the implicit tax on municipal bonds. Throughout, the statutory rate decomposes as τi = τ* + (τi − τ*): the market-implied rate τ* is the burden every holder pays through the depressed yield, and the excess of the holder's own rate over the market rate is an inframarginal rent, a windfall the tables also fail to record. We report both parts separately, because they answer different questions: the first corrects measured burdens, the second measures what the exemption gives away.

;; 1.05
The paper reports three findings. First, the correction is concentrated exactly where measured top rates bind, and it is material there. Across the paper's grid of market-implied rates — the term structure plus the working-paper version of the anchor — the correction to the 99th-to-99.9th percentile group, the paper's headline group (the top tenth is reported separately), runs from {{N:grid_min_top1_delta_pp:.2f}} to {{N:wp_row_top1_delta_pp:.2f}} percentage points; at the anchor — the short-end identified rate of 38.0 percent (Longstaff 2011, the published estimate; the working-paper version reported 41.6) — it rises by {{N:top1_delta_pp_headline:.2f}} points, and the largest correction belongs to the 95th to 99th group, at {{N:p95_99_delta_pp:.2f}}. Below the 90th percentile no group moves by more than a tenth of a point. The anchor is near the top of the paper's own grid, chosen for identification cleanliness rather than size; its federal component gives {{N:fed_component_top1_delta_pp:.2f}} points.

;; 1.06
Second, the aggregate unrecorded burden is material under every construction the paper prefers, and never less than ${{N:grid_min_agg_2021_bn}} billion on the survey income base. Grossing up reported tax-exempt interest at the anchor gives ${{N:agg_implicit_tax_2021_bn:.0f}} billion for tax year 2021 on the survey income base, ${{N:agg_implicit_tax_2021_soi_scaled_bn:.0f}} billion scaled to the administrative record, and ${{N:fed_component_agg_2021_bn}} billion at the federal-component rate; each construction is stated with its base in Section 5. At today's stocks and yields the range is ${{N:burden_current_spread_bn}} billion (the observed ten-year spread) to ${{N:burden_current_anchored_bn}} billion (the anchor crossed with the ten-year yield, an upper-bound pairing). For scale, Treasury scores the entire tax expenditure for public-purpose municipal interest at ${{N:te_muni_bn:.0f}} billion for fiscal 2024, and calendar-2021 federal personal current taxes were ${{N:cy2021_personal_tax_receipts_bn:,}} billion, so the unrecorded burden is roughly {{N:burden_pct_of_cy2021_receipts}} percent of federal personal current taxes that year.

;; 1.07
Third, whether the exemption is a windfall to the rich turns out to be a question about maturity — one no distribution table has recorded in either direction. Measured against the short-end anchor, with holder and market rates paired on the same tax basis and holders' rates computed as average rates on the full counterfactual taxable block, the aggregate inframarginal rent in 2021 was approximately zero: plus ${{N:rent_blockavg_combined_bn:.1f}} billion on the expected-state pairing, minus ${{N:rent_blockavg_fedcap_bn_abs:.1f}} billion on the federal-only pairing, against ${{N:agg_implicit_tax_2021_bn:.0f}} billion of burden. Pairings that compute holders' rates at the margin rather than on the block widen the range to minus ${{N:rent_fedcap_pairing_bn_abs:.1f}} through plus ${{N:rent_blockavg_hightax_bn:.1f}} billion, which is still small against the burden and still not the large windfall the exemption is usually said to deliver. The short-end market prices the average municipal bond dollar closely: no aggregate windfall. At the thirty-year implied rate, by contrast, most of the statutory rate is rent.

;; 1.08
The measurement stakes are specific. After-tax income is recorded correctly under either convention, because the household's cash income and its remitted tax are both observed. The error is confined to the ratio of tax to income, the effective rate, and effective rates by income group are precisely the object on which progressivity claims are made. On the municipal margin, correcting the ratio makes the tax system look more progressive than the tables say. We state that plainly, and we also state its converse: correcting only municipal bonds, the asset held at the top, while leaving other capitalized preferences uncorrected, biases the overall picture toward more progressivity. Section 6 sizes the two large neighbors, owner-occupied housing and corporate equity, and states which way each pushes, rather than pretending the correction is complete.

;; 1.09
The contribution is a corrected effective-rate schedule built from market prices, with the burden and the rent reported separately by percentile. Implicit taxes have met progressivity measurement before. Iglarsh and Allan (2003) imputed a municipal implicit tax from tax-return data and recomputed aggregate progressivity indices; we know that paper only from its published abstract, and the differences we assert — a market-identified rate rather than an assumed spread, a percentile-level rate schedule rather than an index, and the burden-rent decomposition — are conditional on a full read, which is the first item for the coauthor round.{{FN: Advances in Taxation 15 (2003); paywalled at the time of writing. The engagement here is deliberately limited to what the abstract supports.}} Galper, Rosenberg, Rueben, and Toder (2013) distributed the benefit of the exemption under incidence assumptions, the other side of the ledger from measured burdens. Neither produces what the distribution tables would need: corrected rates by income group.

;; 1.10
The data marriage is not new either, which is a point in its favor. Poterba and Ramírez Verdugo (2011) combined SCF portfolios with TAXSIM marginal rates to estimate the revenue cost of the exemption under portfolio-substitution counterfactuals. We use the same marriage for a different object: measured burden rather than counterfactual revenue. The founding empirical fact goes back further. Feenberg and Poterba (1991), using the first tax returns on which tax-exempt interest had to be reported, found the holdings concentrated at high marginal rates. Municipal bonds sit at the top of the distribution; that is why recording their tax as zero moves measured progressivity where it is most watched.

;; 1.11
Everything turns on the implicit tax rate, and the naive measure of it fails. The ratio of municipal to Treasury yields embeds credit risk, liquidity, and call provisions along with tax; read off the Bond Buyer index, it implies a marginal investor whose tax bracket turned negative after 2008. Longstaff (2011) provides the identification the paper leans on: at the very short end of the municipal swap market, where credit and call effects are negligible and liquidity is stripped through the swap term structure, a naive reading in the mid-teens becomes an identified marginal tax rate of 38.0 percent. We anchor the headline on that rate and treat the maturity term structure — down to roughly 14 percent at thirty years today — as the robustness dimension, because the long end of the municipal market is a genuinely open puzzle, not a nuisance to be assumed away. Section 3 takes the puzzle up directly.

;; 1.12
Three limits are stated here and maintained throughout. The exercise is partial-equilibrium: it reallocates measured burden at observed prices and makes no claim about what yields would be in a different tax system; Section 7 takes up the general-equilibrium question, including the transfer to tax-exempt institutions that a Miller-style gross-up would imply and that no one has quantified — the open-economy stance implies no such economy-wide transfer, and the paper declines to choose between them. The income year is 2021, when pandemic-era credits pushed measured rates at the bottom of the distribution to unusual lows; the municipal correction barely interacts with that, but the baseline schedule wears it. And the two available measures of aggregate tax-exempt interest disagree: the survey base is ${{N:agg_taxexempt_interest_bn}} billion, the administrative base ${{N:soi_2021_taxexempt_interest_bn}} billion. We carry both rather than adjudicate, and because the correction is close to linear in tax-exempt interest, the choice moves the rate correction roughly in proportion — {{N:top1_delta_pp_headline:.2f}} points on the survey base against {{N:top1_delta_pp_soi_scaled:.2f}} on the administrative one, with the shape of the schedule unchanged.

;; 1.13
Section 2 sets out the accounting identity and documents the convention it corrects. Section 3 identifies the implicit tax rate. Section 4 describes the data and the construction. Section 5 reports the corrected schedules, the aggregates, and the rent decomposition. Section 6 sizes the housing and equity corrections and states the direction of each. Section 7 states what the correction assumes about incidence and what remains unknown. Section 8 concludes.
