A forty-point spread inside one authority
This is the twelfth paper in the African Revenue Intelligence Research Series, instantiating ARIM (Paper 1) for Namibia. It tackles the widest sector-rate spread documented in the series (12% to 55%), establishing sector-reclassification detection models that later inform threshold and rate-cliff models across subsequent country studies.
The Namibia Revenue Agency (NamRA) applies a 30% general corporate rate (falling to 28% from FY2026/27), 37.5% for non-diamond mining, 35% for petroleum, and 55% for diamond mining. Long-term insurers sit at the opposite end: 40% of gross investment income is taxed at 30%, an effective 12%. It is the widest sector-rate differentiation documented in this research series.
Around it, the base is being tightened. A 10% dividend tax took effect on 1 January 2026, the 3:1 thin-capitalisation ratio gives way to a 30% limit on interest deductions, assessed-loss carry-forwards are capped, and a new provision targets substance loans disguised as preference shares — an estimated N$41 million of leakage. Reported compliance rates have previously sat below 60%.
Data graded by what NamRA actually holds
ARIM grades every candidate source. Essential: VAT and corporate returns tagged with sector codes, historical audit adjustments including existing arrears, and mining and petroleum licence records. Desirable: dividend and preference-share issuance records, plus company-registry data. Potentially unavailable: transaction-level VAT e-invoicing data, anticipated around April 2026 but not assumed to be live.
The core models are therefore built to work without it, with an explicit upgrade path the day invoice-level data arrives.
Matching licences to sector codes
Deterministic matching starts from the NamRA Taxpayer Identification Number. Fellegi–Sunter probabilistic matching then links TIN, registry, director, shareholder and preference-share-holder evidence into resolved entities.
For Namibia the resolution layer does one more job: it cross-references mining and petroleum licence-holder records against the sector code declared on the corporate return. Unlike a two-way ownership test, this is a five-category classification problem — and it is the prerequisite for measuring any rate gap at all.
Three risks, three Namibia-specific models
Model NA1 scores classification integrity. It compares the rate implied by the declared sector code with the rate implied by the licensed activity, and adds a flag where a code change lowered the applicable rate without any documented change in activity. Genuinely diversified companies receive a proportion-weighted expected rate instead of a forced single label.
Model NA2 looks through legal form. Preference shares with a fixed redemption date, a fixed non-discretionary return and no real equity or voting participation score as debt-like, whatever they are called — substance over form, expressed as a detectable pattern.
Model NA3 blends recovery with detection. Existing arrears are divided by a collectability discount that grows with age, then combined with the standard new-risk score, so one queue serves both known liabilities and fresh signals.
Six domains, one weighted toward sector
ARIM scores each taxpayer as a domain-weighted sum. For Namibia the weights tilt hard towards sector behaviour — the highest illustrative weight of any country in the series — while the transaction weight stays near zero pending e-invoicing.
Tax behaviour
VAT-to-sales deviation and corporate effective-rate gaps against the sector-appropriate rate.
Transaction behaviour
Minimal for now; expected to rise sharply once the VAT e-invoicing rollout goes live.
Network behaviour
Director and shareholder centrality, plus preference-share holder structure feeding Model NA2.
Sector behaviour
The classification-integrity score: the paper's central domain, given the width of the rate spread.
Historical compliance
Arrears magnitude and age, plus the count of prior audit adjustments.
External data
Licence cross-references for mining and petroleum, and dividend and preference-share filings.
A flag is a signal, never a finding
On a fully synthetic set of 30 taxpayers across eight sectors, three anomalies were seeded. The substance-loan profile and the sector-misclassification profile ranked first and second, each driven by a clearly different domain. The aged-arrears case ranked lower despite a N$420,000 balance, because the collectability discount stops raw arrears from dominating the score.
Every case file carries a domain-tagged reason code. A diversified company with a verified activity split reads as no classification risk, which is the safeguard against over-flagging legitimate multi-sector businesses. The reason-code standard should also be shaped alongside the planned Tax Court. And the framework claims no real-world detection figure — the N$41 million estimate is government-reported and unverified here — until NamRA validates it on its own data.