Tax fraud schemes and detection

Tax fraud schemes and detection covers intentional violations of tax law to obtain unlawful tax benefits and the tools used to identify and them. Schemes include false or misleading returns, fictitious invoices and records, identity-theft refund filings, deliberate non-registration or non-filing, and organised VAT fraud.^[1] Detection relies on third-party reporting and data matching, targeted audits and investigations, statistical tests (e.g., Benford's Law), regression and supervised models, anomaly detection, graph/network analytics on invoices and entities, and document forensics, alongside preventive controls like e-invoicing and stronger taxpayer identification.^[2][3][4]

Underreporting and non-filing

Taxpayers may conceal cash sales, skim receipts, or fail to register or file returns.^[5][6] Businesses may use electronic sales-suppression tools ("zappers," phantomware^[7]) that can delete or alter point-of-sale data to understate turnover.^[8]

For example, a small café exceeds the registration threshold but stays off-register, with the owner removing part of each day's cash before it reaches the till (skimming) and failing to file income-tax returns.^[9][10]

Overstated deductions and credits

This fraud involves the inflation of expenses or use of false invoices to fabricate deductible costs or claim unwarranted credits.^[11]

Gambling

Taxpayers claim large "gambling losses" without records, or misclassify personal betting losses as business expenses to erase taxable income. Courts and IRS require contemporaneous logs and cap losses at gambling winnings.^[12] Gambling is also use to launder untaxed cash. Cash is converted to chips, minimal play occurs, chips are redeemed, and funds are deposited as ostensible "winnings," obscuring the original source and enabling evasion and money laundering.^[13][14]

An example: A bookmaker-paid influencer runs wagers through a nominee on an offshore site that lacks KYC. He deposits undeclared business cash as "bankroll", cashes out intermittently as "winnings", and files a return claiming large gambling losses to zero out income. Bank deposits and platform records don't match any W-2G reporting.^[15][16]

Art market schemes

Art is used to evade tax by inflating charitable-donation appraisals, over- or under-invoicing related-party sales to shift value cross-border, and storing works in freeports to defer VAT/duties and obscure ownership.^{[17][18][19][20][21]}

An example: A collector donates a painting to a charity at an appraised value far above market, claims an inflated deduction, and files Form 8283 without a substantiated qualified appraisal. Parallel pieces are warehoused in a freeport to avoid import VAT and hide beneficial ownership.^[19][22]

Refund fraud and identity theft

Stolen or synthetic identities are used to file returns and secure refunds. Since 2017 the IRS, state tax agencies, and industry have operated the Identity Theft Tax Refund Fraud Information Sharing and Analysis Center (ISAC)^[23] to share indicators, issue rapid alerts, and coordinate responses.^[24][25]

Invoicing rings and VAT/GST carousel fraud

"Missing trader" and carousel schemes exploit cross-border VAT rules: a trader buys VAT-free, charges VAT on resale, then disappears without remitting, often within circular supply chains.^[26][27][28]

For example, company A in member state 1 sells phones VAT-free to company B in member state 2. B sells to company C in the same state, charges VAT, pockets it, and disappears. C exports the goods back to member state 1, zero-rates the sale, and reclaims input VAT. The same consignment then cycles through new buffers, repeating the refund claims.^{[29][30][31][32]}

Payroll and employment tax fraud

The scheme include off-the-books wages, ghost employees, and misclassifying employees as independent contractors to avoid withholding and social contributions.^[33][34]

Transfer pricing abuse and profit shifting

Related-party transactions may be mispriced to shift profits to low-tax entities. The OECD's BEPS project and Action 13 reporting seek transparency via master/local files and country-by-country reports.^[35]

For example, a U.S. parent licenses a high-value algorithm to its Irish subsidiary for a below-market royalty. The subsidiary books most global sales and pays the low royalty, so residual profit accumulates in Ireland at a low rate while the parent shows thin margins.^[36][37][38]

Shell entities and opaque ownership

Anonymous or poorly documented ownership structures can hide beneficial owners and facilitate evasion and refund fraud.^[39][40]

For example, a promoter forms "Alpha Exports Ltd." with nominee directors and a mailbox address, held through a Belize company and a trust to obscure the real owner. Alpha submits large VAT refund claims supported by fabricated invoices, then dissolves after payment. Investigators pierce the structure by tracing beneficial ownership and bank beneficiaries, consistent with FATF and OECD guidance on transparency and IMF descriptions of fictitious-entity refund fraud.^[41][42]

Residence, nexus, and permanent establishment avoidance

Businesses may fragment operations or use commissionaire arrangements to avoid creating a taxable presence.^[43]

For example, a multinational sells into country X through a local "commissionaire" that signs contracts in its own name but on behalf of the foreign principal. Warehousing and marketing are split among closely related local entities so each activity looks "preparatory or auxiliary" when viewed alone. The group claims no permanent establishment in country X and books profit offshore.^[44][45][46]

Cash-intensive sectors and excise/customs fraud

Smuggling and diversion of highly taxed goods (tobacco, alcohol, fuels) erode revenue.^[47][48]

Typical patterns include moving duty-unpaid goods into domestic markets ("diversion"), re-routing consignments while in bond, counterfeit or undeclared production, and exploiting weak controls at warehouses and borders. Fuel fraud includes "laundering" rebated diesel to remove chemical markers and resell at full price.^[49][50][51]

For example, a logistics firm loads duty-suspended beer for export from state A to state B. En-route, the load is diverted to domestic warehouses and retailed without paying excise. Paperwork is recycled to mask the missing consignment. In a parallel scheme, a gang launders marker-dyed gas oil with acids, then blends and sells it as road diesel.^[52][53]

Platform and gig-economy non-compliance

Platform sellers may underreport income or fail to register for VAT/GST.^[54][55]

For example, a parcel reseller on a marketplace ships low-value goods to EU consumers but does not register for VAT. Under the 2021 EU e-commerce package the marketplace is treated as the deemed supplier and must collect and remit VAT on qualifying sales. It must also keep transaction records. Separately, under DAC7^[56] the marketplace reports the seller's identifying details, consideration, and taxes to its member-state authority for exchange across the EU.^[57][58]

Real-estate and capital-gains manipulation

Real-estate and capital-gains manipulation often involves recording a lower deed price than the true consideration to reduce transfer and gains taxes, using nominees or shell companies to obscure beneficial owners, and holding property through opaque cross-border structures that hinder detection.^[59][60][61] Policy analyses note limited tax-authority visibility over foreign-owned property and call for digital ownership registers and cross-border information exchange.^[62][63]

In a typical scheme, a condominium's deed records a price of $1,000,000 while an additional $300,000 is paid off-record, lowering assessed transfer duties and the seller's capital gain. Title sits with "Delta Properties Ltd.", itself owned through an offshore company and a trust that obscure the beneficial owner. During audit, authorities reconcile bank deposits, escrow and loan payoffs with conveyance records and then compel beneficial-ownership disclosures. Discrepancies and nominee links support reassessment and penalties. International standards and policy work describe these vulnerabilities and prescribe remedies: FATF requires timely, accurate beneficial-ownership information for companies and trusts and risk-based controls in the real-estate sector, while the OECD urges digital ownership registers and cross-border exchange to improve tax transparency over foreign-owned property. The United States has moved to mandatory reporting of all-cash residential transfers to entities or trusts with beneficial-owner details.^{[64][65][66][67][68]}

Charities and non-profit abuse

Abuses include fabricated or inflated donations, ineligible relief claims, and "fake charity" solicitations that exploit disaster events and donor confusion.^[69][70][71] Regulators warn that some promoters market abusive contribution schemes, including "charitable LLC" arrangements and donation tax shelters later denied on reassessment. Charity regulators also cite Gift Aid and other relief claims made through sham or misused structures, and have issued sector risk assessments and fraud strategies.^[72][73][74]

For example, a promoter sells a leveraged "donation program" promising credits exceeding cash outlay. Participants receive receipts from an affiliated charity, claim inflated deductions, and the organizer retains fees.^[75][76][77]

Timing and deferral schemes

Timing and deferral schemes manipulate when tax liabilities, income, credits, or refunds are recognized to delay payment or extract cash before detection. Typical enablers include short-lived entities, fabricated documentation, and gaps between filing and third-party data arrival.^[78][79][80]

An example: a newly formed exporter records large input VAT in March, then files an early refund claim before suppliers' e-invoices and customs data are available. Supporting invoices are back-dated to straddle the period end, and an amended return in April increases the refundable balance. The firm's directors are nominees, and it dissolves after the refund is paid. Audit later matches bank flows, e-invoices, and export manifests, identifies fictitious documentation, and reverses the refund with penalties.^[81][82]

Measurement

Administrations measure the tax gap as the difference between theoretical liability and amounts paid voluntarily and on time. They also report voluntary compliance rates and net gaps after enforced/late payments.^[83]

Approaches

Bottom-up

Uses audit results, operational data, and microsimulation (randomized auditing) to estimate underreporting, non-filing, and underpayment across taxes.^[84][85]

Top-down

Compares potential VAT (from macroeconomic bases) with actual receipts to derive a compliance gap at national or bloc level.^[86][87]

IMF RA-GAP

The Revenue Administration–Gap Analysis Program provides standardized methods to estimate tax gaps - the differences between potential revenue and actual collections. It separates the policy gap from the compliance gap. It implements a sectoral VAT gap using value-added by industry.^[88]

Shadow-economy proxies

Where administrative data are limited, researchers infer non-observed activity using currency-demand and MIMIC models^[89], satellite night-lights, or hybrids. These proxies are imprecise and conceptually broader than tax non-compliance.^[90][91][92]

Recent estimates

United States

US IRS projects a gross tax gap of $696 billion for Tax Year 2022 and a voluntary compliance rate of 85.0%.^[83]

European Union (VAT)

EU VAT compliance gap estimated at 7.0% of VAT Total Tax Liability in 2022 (≈ 89.3 billion Euro).^[86][93]

United Kingdom

UK's HMRC's 2025 edition reports a total tax gap of 5.3% for 2023–24, with detailed tables by tax and behaviour.^[84][94]

The use of AI for detection

Tax administrations apply machine learning, graph analytics^[95], and natural-language processing to flag risky returns, networks, and documents at scale. OECD surveys report AI is now used for evasion/fraud detection, decision support, and service delivery across many jurisdictions.^[96][97]

Supervised machine learning

Classification models^[98] score returns and refund claims using labelled audit outcomes and third-party data. Surveys report widespread use in fraud detection, and U.S. programs cite AI-based targeting for complex partnerships and high-income cases.^{[99][100][101]}

For example, a gradient-boosting model flags abnormal loss carrybacks given industry, seasonality, and filer history, and high-risk cases are queued for review.^[102]

Regression and risk scoring

Generalized linear models estimate expected ratios (e.g., refund-to-sales, input-VAT-to-turnover) and rank deviations for audit selection.^[4] Foe example, a Poisson regression /negative-binomial model predicts invoice counts by sector and month and outliers trigger seller verification.^[4]

Benford's law

According to Benford's law, first-digit and second-digit distributions screen for fabricated amounts in invoices, expenses, and payouts.^[103] For example, repeated 9,500–9,999 entries inflate 9-leading amounts, focusing auditors on a subset of claims.^[103]

Graph analytics

Network models on invoices, directors, bank accounts, and IPs identify missing-trader/carousel motifs and short-lived nodes. EU and academic work describe graph-informed classifiers for VAT fraud.^[104][105] For example, a community-detection score surfaces a ring with rapid circular trades and zero-rated exports. The refunds are held pending verification.^[104][105]

LLM

Tax administrations explore LLMs to parse unstructured descriptions, classify goods/services on e-invoices, and extract entities from filings. For example, an LLM maps free-text "item description" to a controlled taxonomy, improving VAT-rate validation and reducing false positives.^[3][106]

Tax authorities using AI and reported results

United States (IRS)

The IRS uses AI to select complex partnerships and high-income non-filers. The initiatives opened exams on 76 of the largest partnerships and recovered hundreds of millions in early phases.^{[107][108][109]}

United Kingdom (HMRC)

The HMRC connect big-data/AI system supports risk selection. The HMRC reported £4.6 billion additional revenue in 2024–25, up ~35% on prior averages.^[110][111]

See also

• Tax evasion
• Tax
• Tax haven
• Tax noncompliance
• Tax inversion
• Taxation as theft