A reform year on a fragmented ledger
This is the fourth paper in the African Revenue Intelligence Research Series, instantiating ARIM (Paper 1) for Nigeria. Its data-maturity profile sits between Kenya (Paper 2) and South Africa (Paper 3), introducing a two-tier transaction model for transitional e-invoicing and adapting extractives reconciliation from South Africa's mining model (Paper 3) to production-sharing contracts in oil and gas.
From 1 January 2026, small companies with turnover under NGN 100 million are exempt from company income tax, capital gains tax for companies is harmonised with CIT at 30%, a 4% development levy applies to assessable profits, and large or multinational groups face a 15% minimum effective tax rate with new controlled-foreign-company rules. VAT stays at 7.5%.
Administration is split. FIRS runs VAT and CIT; 36 State Internal Revenue Services run PAYE separately. That fragmentation, together with petroleum receipts that still account for most federal revenue, is what makes Nigeria a distinct case within the series.
Data that depends on where a taxpayer stands
ARIM grades Nigerian data as essential, desirable, optional or potentially unavailable. FIRS VAT and CIT returns, the TIN and CAC registration numbers, and past audit outcomes are essential. State PAYE data, MBS e-invoices and NUPRC lifting volumes are desirable. Customs declarations are optional. A cross-state taxpayer-matching bridge does not yet exist.
Transaction behaviour is also two-tier. Large taxpayers onboarded to the Merchant Buyer Solution (MBS) are checked against verified invoice chains; the rest receive a lower-confidence VAT-ratio proxy, and every score states which tier it came from.
Two identifiers, one owner
The FIRS Tax Identification Number and the Corporate Affairs Commission (CAC) registration number are issued by different bodies. The series treats that as a first-class entity-resolution challenge, not a data-quality defect, and the state PAYE registration adds a third identifier.
Deterministic matching on TIN and CAC is followed by Fellegi–Sunter probabilistic matching on names, directors and, where available, state registrations. Ambiguous pairs go to a joint FIRS and state review queue rather than being auto-linked.
The hypothesis: cross-referencing all three identifiers will surface materially more common-control relationships than any single identifier alone.
When missing links are themselves a signal
The taxpayer graph connects companies, directors, shareholders, petroleum licence holders, MBS buyer–seller transactions and, where cross-referenced, TIN-to-state-PAYE links. Louvain community detection and betweenness centrality then score clusters that share directors across inconsistent identifiers.
For that reason Nigeria's model adds identifier cross-reference completeness as a network feature in its own right, a step neither Kenya's nor South Africa's instantiation needed.
Three models, one explainable score
ARIM still scores each taxpayer as a weighted sum across its behavioural domains. Nigeria adds three models: NG1 reconciles NUPRC lifting volumes and commodity prices, less cost-oil, against the declared profit-oil share; NG2 applies invoice-chain checks to MBS taxpayers and a labelled proxy to the rest; NG3 detects federal-state entity networks.
Eight typologies guide the models. Each may equally reflect legitimate commercial behaviour, administrative error or lawful planning under the new rules, so six of them are shown here as risk patterns, not accusations.
Oil and gas leakage
Under-declared lifting volumes, cost-oil and profit-oil misallocation, related-party crude pricing.
VAT and MBS missing traders
Schemes that exploit the gap between onboarded and not-yet-onboarded taxpayers.
CIT threshold manipulation
Turnover held just under the NGN 100 million small-company boundary.
Minimum-tax avoidance
Profit-shifting built around the 15% minimum effective tax rate.
Identity fragmentation abuse
Inconsistent TIN, CAC and state registrations used to obscure common control.
Shell-company networks
CAC-registered clusters sharing directors and addresses.
A flag is a signal, never a finding
On a fully synthetic set of 30 taxpayers across eight sectors, all three seeded anomalies ranked in the top five. The top-ranked missing-trader profile was driven jointly by transaction and network behaviour; the oil-and-gas lifting-gap profile at rank five was driven mainly by sector behaviour. No single generic score would have surfaced both.
Every reason code is domain-tagged and states whether it rests on verified MBS data or the proxy tier, so auditors can weight it accordingly. Federal-state sharing should run under a formal FIRS and state agreement consistent with the Nigeria Data Protection Act 2023. The study claims no detection or recovery figures; only FIRS's own validated deployment can establish those.