Wealth Tax Enforcement in Sweden: Filing Requirements and Pre-Populated Returns

Why it matters, in detail. With counterfactual growth g the shortfall is 1-(1+g)N1996/N1997: g in [0.053, 0.094] gives 15.1-18.3% (vs 22.4%) and an increase of 17.8-22.3% (vs 28.8%); the implied effect on all taxpayers is 11.9-15.0% (vs 19.3% = window gain / 1996 total); netting 1997->98 growth from official counts gives +15.8% for families and +19.3% for individuals. Four independent indicators (third-party control, paper's above-2x band, official upper brackets, post-reform growth) all point to g of 5-10%. Because the same adjustment raises the 1994 effect to about 28%, the two reforms are less symmetric than claimed: pre-population regained about 58-70% of the 1994 window loss in numbers.

Evidence. Paper Sec. 3.2 p.6 ('these natural annual shifts are very small ... making our estimate slightly higher than the true causal effect'); Fig. 2 notes ('minimal change above twice the threshold'). Recovered counts (scripts/reproduce_core.py): N1996=166,477, N1997=214,451 in [900K,1800K); [1.8M,2.2M) 27,628 -> 30,240 (+9.5%); third-party control [900K,1.8M) +5.3%; window 1997->98 +7.2%, 1998->99 +8.9%. Official (RSV Skattestatistisk arsbok 1998 Tab. 4.16 and 1999 Tab. 4.18, source SCB): families with taxable wealth above SEK 2M 73,270 -> 80,145 (+9.4%); all families 275,822 -> 342,301 (+24.1%) vs 342,301 -> ~367,000 (+7.2%) in 1998 (arsbok 2000); individuals 457k -> 578k (+26.5%) vs +6.1% in 1998. Drift-adjusted window shortfall 1-(1+g)N96/N97: 18.3% (g=5.3%), 16.7% (7.2%), 15.1% (9.4%); total effect relative to counterfactual: 15.0%, 13.5%, 11.9% vs 19.3% raw; official-count-based: +15.8% (households), +19.3% (individuals) (scripts/derived_checks.py). Countervailing: the listed-share taxable fraction fell from 100% to 80% in 1997, leaving a fixed listed portfolio's tax value flat (-0.04%) despite +25% prices, and upper-band growth may contain intensive-margin reform effects.

Why this severity. Under drift benchmarks drawn from the paper's own data and from official statistics, the 1997 magnitudes, including the abstract's 'almost 20%', fall by roughly a fifth to a third (central about a quarter); the existence of a large jump and the near-threshold doubling survive.

The authors' response. Partly anticipated in the paper (Sec. 3.2 p.6): 'we do not adjust our estimate for such annual drifts, making our estimate slightly higher than the true causal effect.'

Why it matters, in detail. The printed 32.9% is N1993/N1994-1 (high/low compliance ratio minus one), mislabelled as a reduction. In the conventional 1-N1994/N1993 metric the 1994 effect is 24.8% raw and about 27.6-28.6% after a 3.9-5.3% drift adjustment, i.e. roughly a quarter, not a third. In a common metric the two reforms are also closer than printed (24.8% vs 22.4%, or 32.9% vs 28.8%).

Evidence. Dofiles/Fig_1a2a2b3a.do l.82-93: mtem_1 = sum of 1993 counts, mtem_2 = sum of 1994 counts, dd=(mtem_1-mtem_2)/mtem_2. Adjudicator's own extraction of the exact rarea polygons in Output/Figure_2a.eps: N1993=194,725, N1994=146,504, loss 48,221 -> 32.9% of 1994 vs 24.8% of 1993 (scripts/reproduce_core.py; work/reproduce_core.json). Text: Introduction p.2, Sec. 3.1 p.5, Fig. 2 notes p.6. NBER rev0/rev1 printed '33%'. Official RSV counts of individuals debited wealth tax: 474k (1993) -> 379k (1994), -20.0% (Skattestatistisk arsbok 1998 Tab. 4.14). With positive counterfactual drift (third-party control +3.9%; post-reform +5.3%) the conventional decline is 27.6-28.6%.

Why this severity. Overstates one headline window statistic by about a quarter (8.1 pp), but the abstract's 1994 claims (about half just above the threshold; about 20% of all taxpayers) are unaffected, the number is correct under its unstated definition, and the unadjusted drift biases the conventional estimate the other way.

Why it matters, in detail. From 1995 most filers could sign and return a pre-printed form; a taxpayer with taxable wealth had to supply wealth as an addition. The 1993->1994 contrast therefore changes the legal duty and, simultaneously, the default and marginal hassle of reporting wealth, so a default/hassle-cost model can also generate concentrated non-reporting just above the threshold. The new interest-expense reports increased the agency's information on debts and therefore cannot explain the drop under a deterrence model; if anything they reinforce the paper's point that information returns alone did not secure compliance. The estimate's external validity is to a threshold change within a pre-printed-return system.

Evidence. Prop. 1992/93:86 (5 Nov 1992), Sec. 6.2: wealth to be reported 'endast om nettoformogenheten uppgar till skattepliktigt belopp' in order to limit the need for those using the simplified procedure to supply supplementary wealth information (the wealth tax was then scheduled for abolition from the 1995/1996 assessment); Sec. 6.3: pre-printed simplified return to all concerned taxpayers, RSV estimate 5.5-6 million forms, about a third needing additions; Sec. 5.1 and 3 kap. 25 a LSK: lender control statements on interest paid; SFS 1992:1659 applies 'i ovrigt vid 1995 ars taxering'. Pre-reform wording (bill's comparison table): general returns reported assets and debts 'even if the wealth does not reach a taxable amount'. OECD (2008) para. 14: 'in 1995 the first version of the pre-filled tax return was made available for some personal taxpayers'. Paper p.2: 'no direct evidence showing effects of a pure change in the filing threshold like in our study'; 'This large change in compliance happens despite no changes in information returns.' Table 1 note (3): cadastral values pre-printed for the property tax from 1995.

Why this severity. The measured 1994 response is unchanged, but it is the response to a bundle (threshold rule plus pre-printed returns that omitted wealth plus new interest reports), which qualifies its attribution to filing requirements and weakens the argument that filing/hassle costs cannot explain the universal-versus-threshold contrast.

Why it matters, in detail. In 1996->97 the taxable fraction of listed shares fell from 100% to 80% under a new law; relative to constant valuation this lowers 1997 taxable wealth (conservative for the jump), while relative to the no-drift assumption it removes stock-driven drift, so the net effect on the 22.4% is ambiguous. In the 1986 design the treatment pair (1985-86) is at 100% valuation and the control pair (1987-88) at 75%, so the control understates stock-driven growth and the 8.9% is biased upward, reinforcing 'comparatively small'. The p.8 inference from the flat 1986->87 density fails because the tax value of a fixed listed portfolio fell about 31% that year and life insurance entered the base.

Evidence. Paper Sec. 2 p.3 ('stocks were valued at 80 % of market value'); Table 1 notes p.5. Prop. 1996/97:117 Sec. 4.2.4 (75% 'fr.o.m. 1988 ars taxering'; full value at the 1997 assessment) and Sec. 4.2.10 (life insurance from the 1988 assessment); RSV Skattestatistisk arsbok 1998 p.70, 1999 p.79, 2000 p.85 ('t.o.m. 1995 ... 75 procent ... Ar 1996 ... fulla vardet. Fr.o.m. 1997 ... 80 procent'); SFS 1997:954 transitional provision 2; Lag 1997:323 first applied at the 1998 assessment; the paper's own source Du Rietz & Henrekson (2014) pp.2-3 give 75% for 1978-1996. RSV 1998 p.64 and Tab. 4.5 (rolling recalculation from 1996; freeze 1998-99; rental/industrial reassessment 1994). Waldenstrom (2014) price index x taxable fraction: fixed listed portfolio tax value 1987 -30.9%, 1996 +84.0%, 1997 -0.04%; recovered Fig. 4(a) density in [400K,800K) +3.1% in 1986->87 (scripts/derived_checks.py, reproduce_core.py).

Why this severity. An important undisclosed caveat at two of the three identifying comparisons; the resulting biases mostly favour the paper's conclusions, but one of the two 'reassuring' arguments for the 1986 design is invalid.

Why it matters, in detail. Identifying a psychological cost of commission requires holding perceived audit probability and filing costs fixed across regimes; neither is measured, and both plausibly changed (1994: pre-printed return; 1997: the agency could compute net wealth and signalled it). The A.2 placebo (95-97% non-zero reporting under universal filing even at low wealth) is suggestive of non-pecuniary motives but is equally consistent with compliance with an explicit legal duty. The claim is best read as a hypothesis consistent with, not required by, the reduced-form evidence.

Evidence. Introduction p.2; Sec. 2 p.3 ('This was equally true during universal filing requirement 1984-1993, filing above threshold 1994-1996, or pre-populated returns 1997-1999'); Sec. 4 pp.8-9 ('we think ... can best explain'). NBER rev0 (30 Jan 2025) fn.15: 'According to discussions with tax administration officials from that period, Sweden did not systematically audit taxpayers ...' (removed in later versions). 1997 bundle: Lag 1997:329 (debt balances in lender reports), Lag 1997:956 (pre-printed wealth), Lag 1997:323 and SFS 1997:954 (new wealth-tax law; listed shares at 80%), all first applied at the 1998 assessment. RSV Skattestatistisk arsbok 2000 ch.10 p.194: surtax 'i regel' 40%, lower rates 'om myndigheterna har visst kontrollmaterial', exemptions. Partial rebuttal of one auditor argument: interest-expense reports existed from wealth year 1994 (SFS 1992:1659), so the agency had indirect debt information in 1994-96, though no debt balances or computed stock values.

Why this severity. Affects the paper's main interpretive claim, which the Introduction states strongly, but not the measured effects; alternative rational explanations are not excluded by the design.

The authors' response. Section 4 (p.8) is more tentative than the Introduction: 'This is why we think that it is the combination of filing costs and psychological costs of providing explicitly false information that can best explain all our results.'

Why it matters, in detail. Because selection into the observed gross-wealth density depends on net wealth and on pre-population status, equal gross and net responses cannot decompose the extensive-margin response of the 1997 bundle into a pre-population effect and a debt-information (deterrence) effect; the conclusion should be narrowed to 'reduced debt over-reporting by existing filers does not explain the jump'.

Evidence. Footnote 3 p.3; App. Fig. A.3 notes. Reproduced: gross-wealth counts G1996=164,813, G1997=213,833 in [900K,1800K) -> 22.9% shortfall (29.7% increase) vs net 22.4% (28.8%) (scripts/reproduce_core.py). Pre-population was defined on third-party net wealth, which required the new debt reports (paper p.3; SFS 1997:329/956).

Why this severity. The test is valid for the alternative the authors target; the remaining ambiguity (pre-population per se versus the agency's new knowledge of net wealth) concerns the mechanism and is covered by F4, not the size of the reform-level effect.

Why it matters, in detail. In a common 1-low/high metric the raw effects are 24.8% (1994), 22.4% (1997) and 8.2% (1986); in a high/low-1 metric they are 32.9%, 28.8% and 8.9%. The mislabels do not change any estimate but invite misreading of the 1997 and 1986 magnitudes.

Evidence. Fig_1a2a2b3a.do l.91 vs l.151; Fig_4.do l.55-62 and l.80; AFig_3.do l.56; paper p.5, p.7, Fig. 4 caption p.7, App. Fig. A.3 notes. Window gain 47,974 / 1996 total (about 248,300 at paper scale) = 19.3% (scripts/derived_checks.py).

Why this severity. Presentation only: each number is arithmetically correct under some convention, but the mixture obscures comparisons (the printed 32.9% vs 22.4% makes the reforms look less symmetric than the raw data).

Why it matters, in detail. Because the control's definition changes at the 1993/94 boundary and its content may include securities, its stability cannot be read as a like-for-like test of 'no change in actual wealth' across the 1994 reform; across 1996->97 it is like-for-like but moved +5.3%, which is itself evidence of drift (F2).

Evidence. Dofiles/A1_build.do l.56-95 (asset_3 includes fkuvp with comment 'Stocks / funds, third-party reported'; formfors/ffors only in 1991-93; formfast/ffstpr in 1991-93 vs asma* from 1994; asma075 absent in 1996-97). Paper p.4 fn.8 and Fig. 1/3 notes (securities excluded). Recovered third-party window changes: +3.9% [800K,1.6M) 1993->94; +5.3%, -2.2%, -6.7% [900K,1.8M) 1996->97->98->99 (scripts/reproduce_core.py).

Why this severity. The control is supporting evidence; the core identification rests on the sharp discontinuities in reported densities and the longitudinal A.2 evidence, and the observed control changes are small relative to the reported-density changes.

The authors' response. The paper discloses that debts and security values are excluded from the depicted control densities (Fig. 1 and Fig. 3 notes); the code-text discrepancy on fkuvp is not addressed.

Why it matters, in detail. The total-taxpayer effects can only be approximated by combining recovered window counts with official totals (about 19% relative to 1993 and 23% relative to 1994 for 1994; 19.3% relative to 1996 for 1997); the 1986 '7%' cannot be checked at all.

Evidence. README.txt (Stata 16, gtools; data confidential, access via SCB); master.do l.8-9 vs l.14-32; AFig_1.do l.22, l.29; graphexportpdf in all figure do-files.

Why this severity. Prevents exact verification of the aggregate numbers but no error in them is implied; the confidential-data restriction is disclosed.

The authors' response. README: the data are confidential and may be obtained from Statistics Sweden after a confidentiality assessment.

Why it matters, in detail. The early-period revision changes levels by about 15-18% with little change in the 1986 estimate, suggesting a change in sample or unit construction for 1984-89 that is not documented in the final paper or README.

Evidence. Adjudicator's SVG extraction of rev0/rev1 Fig. 2: red/grey polygon ratio 0.0890 in both (scripts/rev_svg_check.py); rev0 Fig. 2(b) polygon spans 91 bins vs 90 in the final; Fig. 4(b) 1985 window mass in rev0/rev1 218,591 vs 184,894 in the final (39 matched bins; scripts/rev_fig4_check.py), consistent with auditor A's 228,675 -> 186,498. Text: rev0 lines with '33%', '9.2%', '22.6%', '8.6%', fn.15.

Why this severity. Headline percentages moved by at most 0.3 pp and the revisions corrected errors; the issue is transparency about a sizeable early-period data revision.

Why it matters, in detail. Defensible alternative counterfactuals give 6-13% versus 8.9%; the choice of 1987->88 is motivated by similar stock returns but is not validated, and the 100% vs 75% valuation regimes bias the chosen design upward.

Evidence. Adjudicator recomputation from Fig. 4(a) bins (scripts/reproduce_core.py): 8.92% (paper's 87->88), 6.17%, 13.43%, 20.53%; raw 1985->86 +24.2%. Stock price changes (Waldenstrom 2014): 1985 +25.7%, 1986 +50.6%, 1987 -7.9%, 1988 +51.9%, 1989 +24.5%. Paper p.8 calls this 'a strong identification assumption'.

Why this severity. The qualitative conclusion that the 1986 effect is small relative to the 1994/1997 effects survives every defensible alternative; only the point estimate is fragile.

The authors' response. Paper p.8: 'This finding relies on a strong identification assumption.'

Why it matters, in detail. The omitted finding (pre-filling reduces self-reported, non-pre-filled items) suggests that pre-population can shift reporting toward pre-filled components, which would make taxpayer counts an incomplete measure of compliance with respect to non-third-party wealth.

Evidence. Paper p.2. Kotakorpi & Laamanen, 'Prefilled Income Tax Returns and Tax Compliance: Evidence from a Natural Experiment', Tampere Economic Working Papers (IDEAS RePEc:tam:wpaper:1604, 2016 version) abstract; the paper cites a 2017 working-paper version that was not retrieved.

Why this severity. Literature framing only; 'almost no effect' is accurate for total taxable income and taxes, the outcome most comparable to the paper's.

Why it matters, in detail. Low-stakes, third-party-dominated margins maximize the relative role of defaults and hassle costs; the main enforcement problems for high-threshold wealth taxes (valuation of private business and illiquid assets, offshore holdings, avoidance planning) are outside the variation studied, so the results do not speak to avoidance elasticities at the top.

Evidence. Paper p.2 (opening paragraph; 'Wealth taxation for the rich ... whether such taxes can be successfully enforced'); Sec. 2 p.3 and Table 1 notes (private business wealth only until 1990; large shareholders' exemptions); RSV Skattestatistisk arsbok 1998 p.70 (unlisted and OTC/O-list shares untaxed); Sec. 4 p.8 ('pre-population is a net societal gain if taxpayer filing costs are substantial'). Auditor A noted Kopczuk's valuation point in its corpus section without logging an issue.

Why this severity. A caveat on generalization rather than an error in the estimates; the paper does not claim quantitative transfer to taxes on the very rich.

Why it matters, in detail. The 1-2 pp change in coverage between 1996 and 1997 implies the official increase in [0.9M,2.0M) (+29.4%) slightly exceeds the paper's (+27.2%); the cause (identifier drops, unit definitions, SCB sampling for the published tables) cannot be determined without the microdata.

Evidence. RSV Skattestatistisk arsbok 1998 Tab. 4.16 and 1999 Tab. 4.18 (source SCB, Inkomster, skatter och bidrag 1996/1997) vs recovered Fig. 3(a) counts (scripts/derived_checks.py). Paper p.3 ('administrative tax data for the full population'), Fig. 1 and Fig. 3 notes. A1_build.do l.28-50, l.97-105.

Why this severity. Ratios and shapes are essentially unaffected (official bracket growth in 1996->97 is, if anything, slightly larger than in the paper's data); only levels and the population description are.

Why it matters, in detail. If a share of pre-populated households with true taxable wealth below SEK 900K passively accepted the pre-filled values, part of the measured jump is over-reporting, biasing the evasion interpretation upward; A.1 refers to 1999 and to a third-party measure whose valuation basis cannot be verified, so this remains a hypothesis.

Evidence. Recovered App. Fig. A.1: third-party/taxable = 1.110 in [900K,1.8M), 1.058 in [1.8M,2.2M) (scripts/reproduce_core.py); Van Dijk et al. (2020) abstract: returns that overestimated liabilities yielded lower compliance than correct or blank returns; paper Fig. 3 notes ('due to reduced tax evasion').

Why this severity. A possible alternative channel with no direct evidence of its size; the symmetry of the 1994 and 1997 responses limits how large it can be.