Challenges of Monitoring Tax Compliance by Multinational Firms: Evidence from Chile

Why it matters, in detail. Panel C is an IPW-weighted mean of firm-level T/W and Panel A of EBIT/W, so their ratio is an aggregate tax/EBIT with firm weights w_i/W_i: 10.0% (local) vs 10.8% (MNE). The -33.5% tax/payroll gap is therefore accounted for by the -38.4% EBIT/payroll gap, and no ETR differential is identified. The caveats cut both ways. The diagnostic uses separately p99-winsorised means and includes loss-makers in the denominator, which are more common among MNEs (P(pay) 0.587 vs 0.760), so it cannot rule out a lower ETR among profitable MNEs. In the authors' 2010 sample of medium+ internationally active firms, the ratio of winsorised means is 7.4% (MNE) vs 10.5% (domestic; w30114 Table 1). Reading 'effective tax rate' as tax relative to true activity (payroll) would require assuming equal true profit/payroll across groups, i.e. assuming the profit-shifting conclusion.

Evidence. Abstract on the AEA article page (also Crossref and DataCite metadata), verbatim: 'Novel data from Chile show that multinationals make up a large share of GDP but report lower profit and effective tax rates than local firms.' Figure 1 (p. 503 left col.) and text (p. 503 right col.). Adjudicator's re-extraction from the PDF vector stream (scripts/fig1_values_adj.py -> outputs/fig1_values_adj.json; axis-fit residual < 5e-5): EBIT/payroll 1.9016 vs 1.1710 (-38.4%); tax/payroll 0.19038 vs 0.12657 (-33.5%); implied tax/EBIT 10.01% vs 10.81% (+0.80 pp); delta-method 95% CIs assuming zero covariance [9.3, 10.7]% and [8.2, 13.4]%. Both auditors obtained the same values. The paper uses 'effective corporate tax rate' in the tax/profit sense on p. 500 (Wright & Zucman). The authors' follow-up paper (NBER w30114, 2022 and 2026 versions) never calls tax/payroll an effective tax rate; it reports 'corporate income tax/payroll'.

Why this severity. Within this paper's evidence it removes one of the abstract's two descriptive claims about MNE taxation: the implied 'lower ETR' becomes a statistically indistinguishable +0.8 pp. It is not critical because the other claim (lower reported profit relative to payroll) and the body's correctly labelled results survive, and the authors' later data suggest MNE ETRs may be lower in other samples.

Why it matters, in detail. 15x exceeds every reconstruction from the later per-firm counts: 11.9x for four firms and 10.0-14.2x for three-firm subsets. It also exceeds the Revelio series: 7.4x, or 7.2x with the 2008-2010 average baseline. One unverified reconciliation is that the 2019 number came from three firms (the {24, 26, 35} subset gives 14.2x) with different endpoint years; the later note that post-reform counts are from 'either 2014 or 2015' shows the endpoints are not fixed. Headcounts are in any case a proxy for compliance and advisory costs, not a measure of them.

Evidence. P&P p. 503 right col. NBER w30114 (June 2022): p. 4 ('grew from 8 to 95 between 2010 and 2014'); Sec. 4.1 and Sec. 6, pp. 16 and 27 ('In 2014, we conducted in-person interviews with senior transfer pricing consultants who work in the Chilean branches of three of the Big 4 consulting firms. In 2021, ... transfer pricing experts in each of the Big 4'); p. 28 ('12-fold increase from 8 to 95'); fn. 10 (overlap with Bustos et al. 2019). w30114 (Jan. 2026): Sec. 3.3 ('20 confidential interviews (3 in 2014 and 17 in 2021-2022)'); Sec. 6 and Figure 4 (p. 34: 2 per firm pre-reform; 24, 10, 26, 35 post-reform, 'either 2014 or 2015'; Revelio 9 to 67). Arithmetic (outputs/consistency_adj.json): 95/8 = 11.9x; three-firm subsets 10.0x, 11.5x, 11.8x, 14.2x; 15/11.9 = +26%.

Why this severity. Affects a secondary quantitative claim. The qualitative conclusion, a large boom specific to transfer-pricing advice, is corroborated by the later interview and Revelio evidence. But the multiplier is about 26% above the authors' own later interview count and about twice the Revelio figure, and the stated evidence base ('the four largest') appears inaccurate at the time of writing.

The authors' response. Not addressed explicitly. The later paper reports 12-fold (interviews) and 7-fold (Revelio) without reconciling the 2019 figure or the four-firm description.

Why it matters, in detail. The reported gaps (-38% EBIT/payroll, -34% tax/payroll) are conditional, firm-weighted IPW contrasts. In the authors' own 2010 data the unadjusted tax/payroll gap is +0.6% (mean of ratios), +26% (ratio of means) and -44% (medians). This is consistent with MNEs sitting in high tax/payroll size cells with fat right tails, and with the P&P gap being driven by composition adjustment and the mass of small and medium MNEs. The small CI-implied effective MNE sample (~1,070 against ~21,000 MNE firm-years) fits unequal weights, sample restrictions (positive payroll and costs exclude non-operating entities) or firm clustering; the code would be needed to tell which. For revenue questions activity-weighted comparisons matter, and none are reported.

Evidence. Figure 1 notes (p. 503); p. 503 right col.; abstract. NBER w30114 Table 1 (2022 p. 45; identical in 2026 p. 37), 2010, 11,333 domestic vs 2,755 MNE medium+ internationally active firms: mean [median] tax/payroll 0.162 [0.064] vs 0.163 [0.036]. Ratios of winsorised means: EBIT/payroll 0.69 vs 1.24; tax/EBIT 10.5% vs 7.4%. MNEs are ~6.4x larger by mean sales (outputs/consistency_adj.json). Panel B CI-implied effective n = p(1-p)/SE^2: 18,655 local and 1,073 MNE (outputs/fig1_values_adj.json), against ~5,300 MNEs x 4 years. SII PUB_GEN_TRAM 2010: 2,515 'Grande 3-4' legal-person firms (0.83%) make ~75.6% of legal-person sales (outputs/sii_tram_2010.txt). WP 2022 p. 8: MNEs 'pay over 60% of all corporate income taxes'.

Why this severity. A robustness and representativeness concern with limited demonstrated impact. The alternative evidence is not like-for-like (no size or industry adjustment; different year and sample), so it does not refute Figure 1. It does show that the sign of the MNE-local tax comparison depends on the statistic and the population, which the abstract's unconditional statement ignores.

The authors' response. No direct response. The follow-up paper (NBER w30114) documents its own sample and reports means and medians, which partly fills the documentation gap but shows no mean tax/payroll gap in 2010.

Why it matters, in detail. Without the estimand and N, the bars cannot be mapped to a population: ATE weights target the overall firm distribution, which is dominated by small firms, while ATT targets the MNE covariate distribution. If pre-reform MNE status is inferred from withholding affidavits on payments to related non-residents, the MNE group is partly selected on having cross-border related-party payments. Such payments (royalties, services, interest) are deductible and directly lower EBIT. Conversely, MNEs that never revealed their status would sit among 'local' firms and attenuate the gap. The net direction cannot be signed (speculative).

Evidence. Figure 1 notes (p. 503); p. 502 right col. ('Firms indicate in their tax forms whether the relationship with a given foreign affiliate is one in which ...'). openICPSR folder listing archived 2022-01-10 (Wayback): Bustos-Pomeranz-VilaBelda-Zucman.do 12.2 KB and README.txt 2.5 KB, deposited 2019-12-07. The live pages returned HTTP 403 (Cloudflare) to the adjudicator and both auditors. NBER w30114 (2022) Sec. 4.1 p. 15 (affidavits 1850, 1912 and 1907; Orbis/D&B merge to identify MNEs 'that did not reveal their status'); notes to the heterogeneity figure (MNEs classified as revealing status pre-treatment, post-treatment or never). SII form F1912: 'Codigo Tipo de Relacion'. DJ 1850 instructions 2008-2012: withholding of the Additional Tax on payments to non-residents.

Why this severity. A documentation problem that changes no result by itself. It does prevent verification of the paper's central figure and leaves unclear which firms the 'MNE' bars describe (see F2). The confidentiality of the data and the P&P page limit partly excuse it.

Why it matters, in detail. A sales share does not identify a value-added share without relative VA/sales ratios. A 40% sales share maps to roughly 25-57% of corporate value added if MNE VA/sales is 0.5-2 times that of local firms (codex sensitivity). Moving from corporate VA to GDP requires further coverage assumptions.

Evidence. P&P p. 502 right col.; p. 503 right col.; abstract. SII PUB_GEN 2010: legal-person sales 15.07 bn UF x average 2010 UF of CLP 21,171.80 (SII daily values) = CLP 319.1 tn, or 2.88x GDP (CLP 110.78 tn; World Bank NY.GDP.MKTP.CN). 40% of that is 1.15x GDP.

Why this severity. A labelling error in the abstract. The qualitative point that MNEs are economically important is unaffected, and the body uses the correct concept.

Why it matters, in detail. The reform is staggered: SII enforcement capacity from 2011, the statutory burden-of-proof and documentation regime from 27 Sep 2012, and the first DJ 1907 filings in 2013. Attributing the reporting requirements to 2011 compresses this sequence. Nothing in the P&P's numbers depends on it.

Evidence. SII Circular 29 (14 June 2013), Sec. III.2: the new Art. 41 E rules 'rigen a contar del 27 de septiembre de 2012'. NBER w30114 (2022) p. 8: 'Starting in 2011, the tax authority hired specialized auditors and created a new specialized unit ... This was followed in 2012 by the passage of a transfer pricing law ... The new reports were filed for the first time in 2013, covering transactions conducted in 2012.' SII DJ 1850 instructions for 2008, 2010, 2011 and 2012 (shared cache) have essentially the same scope.

Why this severity. A dating imprecision of one to two years for the legal and reporting components. Figure 1's 2007-2010 pre-period is pre-reform under either dating, and the authors consistently treat 2011 (the new enforcement unit) as the start of the reform.

The authors' response. The later paper (NBER w30114) gives the precise staggered timeline without flagging the difference.

Why it matters, in detail. The citation presents the high (macro) end of a contested magnitude as settled, while citing a survey that concludes the better-identified micro literature finds much smaller shifting. It affects how the reader calibrates priors for interpreting Figure 1, not the figure itself.

Evidence. Crossref abstract of Dharmapala (2014), Fiscal Studies 35(4): 421-448, doi:10.1111/j.1475-5890.2014.12037.x: 'A major theme that emerges from this survey is that in the more recent empirical literature, which uses new and richer sources of data, the estimated magnitude of BEPS is typically much smaller than that found in earlier studies.' Dharmapala (2019), AEA P&P 109: 488-492 (corpus offshore-havens-10): micro-implied <20% vs macro ~40% of foreign profits shifted.

Why this severity. Background framing only; it does not touch the Chilean evidence.

Why it matters, in detail. Debt shifting is well documented elsewhere (DFH 2004; Mintz & Weichenrieder 2010, also cited), and 'more than 6' versus '6' is rounding-level. The TWZ point matters only in that the paper's 'in line with TWZ' remark has no Chile-specific support in TWZ's data, which are partly imputed for Chile and so weak evidence either way.

Evidence. Crossref: DFH 2004 J. Finance abstract ('internal borrowing being particularly sensitive to taxes'); DFH 'Dividend Policy Inside the Multinational Firm' (SSRN 317040) abstract on dividend remittances. Wright & Zucman (2018) NBER w24983, Sec. 4.1, pp. 9-10: 'the rise of profit shifting to tax havens (6 points)'. TWZ (2018) NBER w24701 abstract and Sec. 4 ('Local firms are all firms that are not foreign'); TWZ appendix Table A.7, Chile: pi_f 0.9907 (identical to the US-affiliate value 0.9908, suggesting imputation), pi_l 1.0106, shifted profits -0.2 bn. Crossref: Clausing 2016b, NTJ 69(2): 353-385.

Why this severity. Citation and transcription slips that do not affect any finding.

Why it matters, in detail. On a non-double-counted basis the haven share of US MNE foreign profits may be nearer one third than one half. B&R's 'where earned' attribution may also strip out genuine foreign-to-foreign shifting (Clausing 2020), so the corrected figure is plausibly a lower bound. The P&P quotes its source accurately but without caveat.

Evidence. P&P p. 500 left col. Clausing (2016a), working-paper version (shared cache): 'these seven countries account for 50.1% of all foreign profits' (gross income, 2012) and 52.3% of direct-investment earnings. Blouin & Robinson (SSRN 3491451, 2020; J. Public Econ. 252 (2025) 105512): six-haven share falls from ~55% to ~35% after removing equity income (2020 version; ~38% for the 2025 version, per corpus, not verified). Corpus offshore-havens-08, status partially_accepted.

Why this severity. A background statistic that the Chilean analysis does not use.

The authors' response. Tørsløv, Wier & Zucman (REStud 2023, Sec. 4.2.1) accept that BEA net income double counts but say their own estimates use BEA value-added (profit-type return) tables. Clausing (2020, SSRN 3503091) disputes B&R's subtraction method (per corpus entry).

Why it matters, in detail. With profit/payroll as the outcome, any conditional MNE payroll-per-worker premium maps directly into a lower ratio: a 20-30% premium accounts for roughly 40-60% of the gap. Productivity premia work the other way, so the net bias is unsigned. For most MNE links the statutory-rate differential ran from Chile's 17% toward higher-rate partners; the exception is the 35% Additional Tax on distributions to non-residents. A split by haven links (630 firms) or by foreign vs Chilean ownership would be the natural test, and none is reported.

Evidence. Figure 1 notes; p. 503 right col. Wage-premium arithmetic (outputs/consistency_adj.json), holding EBIT per worker fixed: a 10/20/30% payroll-per-worker premium alone yields 24/43/60% of the -38.4% EBIT/payroll gap, and a 62% premium yields all of it. OECD Corporate Tax Statistics (scripts/cit_rank_check.py): Chile 17% in 2007-2010; median 25-27%; 72-77% of 146 jurisdictions higher. 2010 partner rates: US 39.2, AR 35, BR 34, CO 33, PE 30, MX 30, ES 30, NL 25.5, CH 21.2, IE 12.5. The 35% Additional Tax with First Category Tax credit is described in J. Espinosa (Espinosa & Asociados), 'Managing Corporate Taxation in Latin American Countries - Chile', Mondaq (c. 2010; https://www.mondaq.com/tax-authorities/109414/): 'There is a 35 % remittance tax on any distribution, dividends, profit income and branch profits remitted abroad ... The 17% First Category Tax may be offset to 35 % Additional Tax'. The same guide gives a 35% general withholding on foreign-debt interest (4% for bank loans). NBER w30114 Table 2: semi-elasticity of payments to affiliates with respect to the destination tax rate -0.055 (s.e. 0.012). w30114 (2026) p. 7: 'Multinationals have incentives to shift profits out of Chile, even though it is not a high-tax country'. w30114 Table 1: payroll/sales 12.9% (MNE) vs 16.0% (domestic).

Why this severity. Concerns the interpretation, which the paper itself hedges ('suggestive evidence consistent with profit shifting'), not the descriptive gap. The authors' later within-firm evidence independently supports tax-motivated payments, so the interpretation is weakly evidenced here but not contradicted.

The authors' response. Partly addressed in later work: NBER w30114 shows payments to foreign affiliates respond to destination-country tax rates and notes that Chile 'is not a high-tax country'.