Nine billion euro. Last place in the Union. The figure has been in circulation for thirteen years and was a payment condition for one billion euro of the National Recovery and Resilience Plan (PNRR) funds. It is worth asking what it actually measures, and what is done with it. An essay in three parts: the first two factual, the third an opinion, marked as such and closing with the arguments against it.
The term has a precise birth date: in September 2013, the European Commission published the first study quantifying the “VAT Gap” in the 27 member states, for the year 2011, and Romania already appeared among the states with the largest gap relative to GDP. Thirteen years have passed.
The study was not carried out by Eurostat, nor by any statistical institution: it is a contractual deliverable, commissioned by DG TAXUD from a private consortium, led at the time by CPB Netherlands Bureau for Economic Policy Analysis, a Dutch institute. CASE, a Warsaw think tank, was a member, not the lead; the role of project leader passed to it from around 2015 and it kept the role for nearly a decade, and the 11 December 2025 edition states that the report was prepared “by Syntesia and CASE”. The contractor's nationality explains nothing, the selection is made through a European tender. What matters is the status: the figure a minister reads out in front of the press does not come from the Union's statistical authority, but from a private contractor.
The method is called top-down and it is a subtraction. From the national accounts, the VTTL is estimated: the VAT that would exist if everyone complied perfectly. From it, the VAT actually collected is subtracted, a real figure from the treasury. The difference is the compliance gap.
“This method calculates the VAT Total Tax Liability (VTTL) [...] often requiring the calculation of around 10,000 parameters per year.”
This is the crux, and it is not an objection, it is a description. Of the two terms in the subtraction, only one is measured; the other is the product of ten thousand parameters estimated every year. The gap is not the difference between what should have been collected and what was collected, but between a model and a collection figure. The authors do not hide this.
The report, pp. 8 to 9: “[...] such revisions may affect the resulting VAT gap estimates. For this reason, each annual edition systematically reassesses all data inputs and recalculates past estimates where needed.” And on the pandemic years, p. 35: “the lowered quality of national statistics impacted the reliability of the estimates for that period”.
This is statistical hygiene, not failure: an institute that did not revise its estimates when the input data are revised would be a poor institute. The problem is not in the report, it is in what is done with it.
If you have an inspection under way or have just received a notification, the related analyses are grouped under tax inspection and the review of individuals. For how such matters are handled, see tax assistance.
In December 2024, the Commission published the estimate for 2022: Romania, 30.6%. In December 2025, the same institution published, for the same year, 26.7%. Between the two publications, nothing had changed in the 2022 economy, it had long since ended.
Sources: 30.6% and EUR 89.3 billion, the report and, respectively, the Commission's press release of 18 December 2024; 26.7% and EUR 100.8 billion, the 2025 Report, Table 7. The panels use different scales.
The difference is 3.9 percentage points, around one billion euro. At Union level, the movement was the reverse and larger: 89.3 billion became 100.8 billion, retroactively, without any European taxpayer having done anything. The report explains at p. 35: the estimates for 2021 and 2022 “were revised upwards by 0.6pp and 0.9pp”. The national accounts had been revised.
How far the model can drift is seen precisely where drift becomes impossible, Box 2, p. 33:
“Negative VAT compliance gaps, implying that collected VAT exceeds theoretical liability, are not feasible in practice.”
A negative gap would mean that the state collected more VAT than its entire economy theoretically owed. It is physically impossible. The model produced it anyway: in the 2025 edition, Luxembourg has a negative or near-zero gap for several consecutive years, after its revenue data was revised upwards by up to 8.5%, without the liabilities in the national accounts being revised to match. The results “cannot be considered statistically plausible”, the report states, and Luxembourg was removed from the Union aggregates.
And now, the absence. The figure is published to the point, 30.0%, to one decimal place, but with no confidence interval. There is no “±” anywhere. A number given with the precision of one decimal place, with no stated margin of error, yet later revised by nearly four percentage points, is a strange object, not for whoever builds it, but for whoever reads it.
The most widespread error about the gap does not concern the method, but the definition. On page 7, the report states what goes into it: from “the legal exploitation of loopholes” to organised fraud, passing through “administrative errors, omissions, non-fraudulent bankruptcies”. The European Parliamentary Research Service (EPRS) groups them, in briefing PE 767.221 of January 2025, into four sources.
Including carousel fraud. The only component that concerns the criminal law.
Companies that can no longer remit VAT because they have gone bankrupt. Not tax evaders: dead companies.
The lawful use of differences in rates and exemptions, purchases made in a state with a lower rate.
Errors, omissions, miscalculations. No intent, no criminal act, but the same figures, in the total.
So: the figure read as “the money stolen by tax evaders” includes an honest bankruptcy, an accountant's error, a lawful cross-border purchase and a statistical adjustment, and the Commission acknowledges that the gap can never be eliminated, precisely because of insolvencies. How much of it is fraud? About a quarter, says EPRS, adding that “precise figures for the fraud component are challenging to determine”: three quarters is not criminal fraud. And footnote 2 of the same document contains the sentence that deserves printing on the front page of every press report.
“The VAT gap remains a proxy estimate, based on various economic parameters, and does not indicate the ‘actual’ amount of revenue lost.”
The Commission also has a dedicated study on carousel fraud (MTIC), from December 2024: between EUR 12.5 and 32.8 billion a year in 2010 to 2023, even the Commission does not know, within a factor of nearly three, how much of it is fraud. But the most important result is not a figure:
“The study found no correlation between Member States' VAT gaps and MTIC gaps.”
The consequence is logical, not rhetorical: if there is no statistical correlation between a state's gap and the carousel fraud in that state, then reading the gap as an indicator of fraud is not an exaggeration of speech, it is a methodological error.
The same report also contains a second indicator, rarely discussed: the policy gap, the VAT not collected because the law does not require it, through reduced rates and exemptions voted in parliaments. For the Union, in 2023: 50.5%, against a 9.5% compliance gap; for Romania, 24.7%. No tax evader answers for it, it is the legislature, and no one points a finger at it.
The conclusion of this part is not an opinion of mine: the gap is an aggregate trend indicator, built to compare 27 tax systems with the same ruler. It is not a measure of fraud, and the institutions that produce it have said so in their own documents.
A comparative indicator does not, on its own, produce consequences: it produces them when it is quoted. What follows are dated facts.
The most concrete consequence is not a speech, but a milestone. Romania committed, through the PNRR, to reducing the VAT gap. It did not meet the target. In the autumn of 2025, the Ministry of Finance renegotiated with Brussels the replacement of the milestone with structural measures, avoiding the loss of up to one billion euro in 2026 grants (Profit.ro, 23 October 2025). This fact demolishes a frequent objection, including within my own profession: the gap does bear on Romania's national accounts, and it was a payment condition for one billion euro. The problem is not that the figure produces no effects, but that it produces massive effects without the precision that would justify them.
The second episode is from 17 July 2024. The then Minister of Finance, Marcel Boloș, met with representatives of Romanian Business Leaders and the Association of Romanian Businesspeople, in the presence of the president of ANAF. Asked to present the breakdown of the gap, the minister, according to the account given by Gabriel Biriș, a former State Secretary at the Ministry of Finance (cursdeguvernare.ro, 21 July 2024), could not present it immediately, and after a few minutes of searching through his notes, indicated: about 35% carousel fraud, 30% insolvencies, 15% evasion. The total comes to 80%. It is the account of an identified participant, not an official transcript, and I quote it as such, but around it stands a fact verifiable by absence: Romania's official breakdown of the gap has never been published.
It is not an academic question. Directive (EU) 2018/2057 allows a member state to request a derogation for generalised reverse charge between companies, the mechanism that eliminates carousel fraud by construction. Two of the cumulative conditions: a gap at least 5 points above the Union median, and at least 25% of the gap from carousel fraud. Romania comfortably meets the first (30.0% against a median of 8.2%); the second depends precisely on the breakdown that is not published.
This is a factual observation, not an accusation. I am not saying anyone is hiding anything; I have no evidence of that. I am saying only this: Romania's right to request a derogation that would eliminate carousel fraud depends on a breakdown that the state has not made public, even though it invokes the total of that figure, whole, for thirteen years.
The third episode concerns the way the figure circulates. On 11 December 2025, the day the report was published that put Romania in last place in the Union, ANAF issued press release No. 1299, focused on revenue stability and digitalisation: the same report, on the same day, read as failure and as success. The press went further, RFI, picked up by Rador on 16 December 2025: “The truth about the VAT gap: billions of euro remain in the accounts and pockets of certain companies and citizens”. And in politics: “Romania loses more than 9 billion euro every year in uncollected taxes.”
This is where the slippage happens. “Remain in pockets” does not describe a compliance gap, it describes an appropriation. Between an estimate built from ten thousand parameters and the claim that someone is keeping the money in their pocket there is no methodological bridge. There is only a sentence.
It would be dishonest to close the factual part with only what goes wrong.
In Romania, the gap has justified e-Factura, e-Transport, SAF-T (Declarația 406), e-VAT and e-Case de marcat; at Union level, VIDA and CESOP. And the results exist: Italy introduced mandatory electronic invoicing through the Sistema di Interscambio in 2019, and its gap fell from around 35 billion euro in 2018 to around 16 billion in 2022, EPRS notes, cautiously, that the timing “suggests it likely played a considerable role”. An imprecise indicator has thus produced real reforms, and that cannot be denied in good faith. There remains, however, a question of proportionality, also factual: the European Court of Auditors estimates that 2% of organised crime could be behind 80% of carousel fraud cases, while generalised compliance measures apply to 100% of taxpayers.
From here on, the text changes in nature. What follows is my professional opinion, a minority position within the profession, which contains no claim about the intentions of any institution or person.
My starting point is not the gap, but the institutional position of the one doing the evaluating. A state that finances part of its current spending with borrowing, with a deficit target of 6.2% of GDP for 2026 and rising defence commitments, is not a neutral observer when it assesses an entrepreneur's tax compliance. I am not saying it acts in bad faith, I do not believe that. I am saying that it is not neutral, in the same way that a judge with a financial interest in a case is not neutral, however honest he may be as a person. Neutrality is not a personal virtue, it is a position.
A point that works against me: in 2026, budget execution is much better than in the previous year, at four months, a deficit of 1.17% of GDP against 2.92% in the same period of 2025 (Ministry of Finance). My argument is not that the state is in a liquidity crisis and is looking for money through prosecutions; that would be conjunctural, and the conjuncture contradicts it. It is structural: pressure on revenue has been a constant of this decade, and the gap is the figure that translates it into a target.
Now, the mechanism. Legea nr. 126/2024 (Official Gazette No. 437 of 13 May 2024) introduced, for the offences under Article 6¹, 8 and 9 of Legea nr. 241/2005, a ground for non-punishment: if, within 30 days of the completion of an inspection identifying a loss of up to EUR 1,000,000, the loss increased by 15%, plus late-payment charges, is made good in full through actual payment, the act is not punished, and the competent bodies no longer refer the matter to the criminal investigation bodies. Public reaction went in a single direction: “tax evaders get off”. There is, however, also a reverse reading, flagged in the business press, StartupCafe called it the “poisoned apple”:
The 30 days run from the completion of the inspection, not from a court judgment.
The loss unilaterally set by the tax authority, increased by 15%, is what is paid. In other words, the figure is accepted.
Whoever does not pay goes down the usual route: referral, criminal investigation, seizure, years of trial.
It is between paying and being tried, and whoever is in the right can choose, rationally, just as much as whoever is not.
And now the question. Not the accusation, the question: how many entrepreneurs would prefer to pay rather than face a trial with an uncertain outcome? And if the answer is “many”, what institutional safeguards exist to ensure that this conversion of criminal pressure into revenue stays within the bounds of justice and does not, imperceptibly, become a revenue policy?
I put this forward as a hypothesis to be checked, not as a claim about anyone's intentions. The hypothesis does not even need bad faith to work, and that is precisely what is unsettling about it: a system in which paying is cheaper than the trial will produce payments, regardless of the intentions of those who run it. It is a loop: an imprecise estimate justifies pressure; the pressure produces payments; the payments are reported as loss recovered from evasion; and that confirms the narrative that justified reading the estimate in the first place. I am not claiming that anyone designed the loop, only that, once closed, it sustains itself without a designer.
The hypothesis is verifiable, at least in principle: it would be enough to know how many files have been closed by payment under Article 10 since 16 May 2024, and how much has been recovered that way. Two years after it came into force, these statistics are not public, I have not identified any official dataset. For now, the hypothesis remains a hypothesis.
On safeguards and public memory. Assessing the safeguards a state offers cannot be done in the abstract: it takes account of its history, just as assessing a debtor takes account of past conduct. And the Romanian state has, in its recent past, documented episodes in which a need for hard currency shaped policy towards its own citizens. The best documented is the emigration of ethnic Germans: between 1968 and 1989, through channels kept secret by both sides, “Geheimsache Kanal” in Bonn, Operation “Rückgewinnung” at the Securitate, the Federal Republic of Germany paid the Romanian state sums per person for the right to emigrate, reaching 8,950 marks in the last settlement, in August 1989; 226,654 people left in this way, and the total, never established exactly, is conservatively estimated at over one billion marks. I am not equating today's ANAF with that regime, that would be a crude and false comparison. The argument is about safeguards: that arrangement worked for two decades precisely because it was, formally, a procedure, with categories, rates and settlements. Its memory does not justify suspicion, but it does justify a requirement: that the mechanisms by which state pressure converts into money be transparent and published. The missing statistics are exactly the missing safeguard.
An essay that does not engage with the arguments against it is a pamphlet. I take them at their strongest.
“The gap is the only comparable indicator available.” True, and it is the best of all the arguments. The top-down method was not chosen for precision, but for comparability: it needs little data and can be standardised across 27 states. Judged against the purpose for which it was designed, the indicator works. The alternative, bottom-up, is expensive and requires microdata and sustained administrative capacity; contrary to a widely held idea even within my own profession, not even the United Kingdom uses it for VAT, and when HMRC ran a random audit in 2020 to 2021 the result was a substantial upward revision of its own tax gap. Even the “precise” method produces large revisions.
“The methodology is transparent, and the criticism misses the target.” This too is true: the 2025 report has 376 pages, a methodology chapter, an annex devoted exclusively to the limits of its own approach, two independent external evaluations and downloadable raw data under every figure. It is the opposite of a figure with obscure origins. A Commission economist could rightly reply: “I never said it was a measure of fraud; page 7 states what it contains. If Romanian politicians read it wrongly, criticise the politicians, not the report.” The reply is correct, and it is exactly what Part II does.
“VAT fraud is real.” It is. Carousel fraud has a known mechanism, files, indictments and final convictions, and the Commission estimates it at around 23 billion euro for 2023. Nothing I have written suggests that fraud does not exist or is negligible, only that its size cannot be read from the gap.
“Non-punishment on payment is not extortion, it is recovery.” This would be a prosecutor's argument, and it is a serious one. Extinguishing criminal liability by making good the loss exists in many legal systems and serves a legitimate purpose, repairing the harm. Moreover, it benefits the taxpayer: someone who really did get it wrong is given an exit they did not have before, at 15% instead of years of trial and a criminal record. A mechanism does not become abusive for being efficient, and the fact that some of those who pay might have won at trial is the price that any transaction over liability, including the plea agreement, pays for speed.
“The state has a duty to collect.” It does, and it is not a duty I would ever dispute. A state that does not collect VAT is not being lenient, it is shifting the burden onto those who do pay: if the gap really is 30%, someone bears the difference, and that someone is the compliant taxpayer, not the tax evader.
All of these arguments carry weight, and I believe each of them, in part. What I cannot believe is the leap that links them: from a difference between a macroeconomic model and a collection figure, to a claim about the criminal conduct of a category of people. I am not disputing Romania's national accounts, I am disputing that step, and only that step. I ask readers not for agreement, only that, the next time they hear “Romania loses nine billion euro to tax evaders”, they know that the sentence contains three claims, and that none of them is fully verifiable.
This text is a professional opinion, given in a personal capacity; it does not constitute legal or tax advice. Updated on 18 July 2026.
If you have received a figure, a 30-day deadline and a choice between paying and a criminal file, the useful discussion is not about the VAT gap. It is about how the loss in your file was calculated, and what can be dismantled from it.