Two currencies, one tax ledger
This is the tenth paper in the African Revenue Intelligence Research Series, instantiating ARIM (Paper 1) for Zimbabwe. It models multicurrency tax compliance across USD and ZiG environments and addresses non-resident digital services, linking digital payment withholding checks to the multi-tier cross-referencing principles established in Nigeria (Paper 4).
The Zimbabwe Revenue Authority (ZIMRA) administers VAT at 15.5%, raised from 15% on 1 January 2026 under the Finance Act, 2025, and corporate income tax at an effective 24.72–25.75% — a 24% base rate plus a 3% AIDS levy applied to tax payable — under a source-based system.
The structural feature is the currency. The US Dollar remains the primary transactional currency for much of the formal economy, a legacy of the 2008–2009 hyperinflation and subsequent dollarisation, while the ZiG is the domestic currency of record. USD amounts are converted at the Reserve Bank of Zimbabwe’s official rate for tax computation, so declaring in whichever currency, or at whichever conversion timing, minimises apparent liability becomes a new way to under-report.
Two further instruments complete the picture: fiscalisation, ZIMRA’s electronic tax-invoicing requirement with six-year record retention, and the Digital Services Withholding Tax (DSWT), under which local financial intermediaries withhold on payments to non-resident digital-service providers whose Zimbabwe-sourced revenue exceeds US$25,000 a year.
Data that anchors the anomaly
ARIM grades every candidate data source as essential, desirable or optional. For Zimbabwe, the essential tier holds ZIMRA VAT and CIT returns (ZiG-denominated, with USD conversion detail), fiscalised electronic tax invoices carrying currency, rate and value fields, the Reserve Bank’s official exchange-rate history, and audit-adjustment outcomes restricted to the period after April 2024.
DSWT intermediary-withheld records and company-registry data are desirable, feeding reconciliation and entity resolution. Pre-2024 compliance data is optional, retained but weighted down or excluded.
That regime break is a design constraint, not a footnote. Because pre-dollarisation and pre-ZiG history may say little about present behaviour, HistoricalCompliance is calibrated on post-April-2024 data by default — a deliberately short, but relevant, historical base.
Six typologies, none of them proof
Zimbabwe’s fraud typologies mix the familiar with one that is entirely new. Each shows up in a different part of the data, and each has an innocent explanation that the model must leave room for.
Currency-conversion manipulation
Strategic choice of transaction currency, conversion timing or exchange-rate reference to shrink ZiG-denominated liability.
Fiscalisation non-compliance
Non-issuance or manipulation of electronic invoices, including currency and rate fields that do not add up.
DSWT under-withholding
Payment intermediaries failing to withhold correctly on payments to non-resident digital-service providers.
CIT under-declaration
Suppressed margins inconsistent with sector benchmarks, in either currency.
Shell-company networks
Company clusters sharing directors or addresses.
Customs under-invoicing
Under-declared import values, potentially compounded by currency-conversion choices at the border.
Any of these patterns may equally reflect legitimate commercial behaviour, administrative error or a lawful response to a genuinely volatile currency. In Zimbabwe that caution carries extra weight, because there are real, non-strategic reasons to hold or transact in either currency.
Three models and a graph for a dual-currency ledger
ARIM still scores each taxpayer as a domain-weighted sum across six behavioural domains. Zimbabwe adds a feature class no earlier country paper needed — currency-mix stability and rate-consistency — inside TransactionBehaviour, while ExternalData carries the Reserve Bank volatility index and the DSWT reconciliation.
Model ZW1: currency anomaly
Scores how much the USD/ZiG mix shifts and how closely those shifts track rate movements, then subtracts how well the customer base explains the mix.
Model ZW2: rate consistency
Counts fiscalised invoices whose recorded conversion rate strays from the Reserve Bank rate on the invoice date by more than a tolerance.
Model ZW3: DSWT reconciliation
Compares what intermediaries report withholding with the non-resident provider’s own declared Zimbabwe-sourced revenue.
Taxpayer graph
Adds correlated-currency-switching edges between entities whose mixes move together, resolved through ZIMRA Business Partner numbers and probabilistic matching.
Five hypotheses, thirty synthetic taxpayers
The central question is which combination of ARIM domains best detects strategic choice of currency, conversion timing or rate reference. Five hypotheses frame the testing: currency-mix shifts that track exchange-rate movements (ZWH1); risk that rises with ZiG/USD volatility (ZWH2); invoices whose rate fields disagree with the official rate drawing more later adjustment (ZWH3); DSWT withholding that diverges from declared revenue (ZWH4); and post-ZiG history out-predicting pooled history (ZWH5).
The demonstration uses a fully synthetic, explicitly hypothetical dataset of 30 taxpayers across eight sectors, with three seeded anomalies. The currency-timing profile ranks first with a score of 1.141, driven almost entirely by TransactionBehaviour. The fiscalisation rate-inconsistency profile ranks third at 0.810, driven instead by the rate-gap term.
Five hypothetical cases illustrate the workflow: a retail business timing its currency mix around rate moves; a tourism business with a stable, USD-dominant mix that correctly scores low; a hospitality business with divergent invoice rates; a DSWT reconciliation gap; and a cluster of related companies switching currency in step. No real taxpayer is identified, and no detection or recovery figure is claimed.
A flag is a signal, never a finding
Every Zimbabwe case file carries domain-tagged reason codes. A currency flag might read that the mix shifted materially in four of the last six periods, each time alongside a Reserve Bank rate movement, with no change in the declared customer base. The tourism case would instead record a high USD share consistent with foreign-currency customers, scored low despite the volatility.
The Data Protection Act, 2021 (Chapter 11:12) governs ZIMRA’s use of personal data. Given the risk of penalising ordinary, lawful multicurrency commerce, the paper recommends that no enforcement action proceed on a currency-anomaly flag alone, without independently verified evidence of strategic intent.
Two cautions remain. The Reserve Bank rate feed drives the whole currency score, so it should arrive through a formally governed channel with integrity verification. And the results show mechanical coherence on synthetic data, not real ZIMRA detection performance; intent is hard to establish even with real data, and the post-ZiG base is still short.