Detecting profit shifting and measuring the CIT gap: Insights from South Africa, Kenya, and Uganda

On 20 May the 6th quarterly meeting of the ATI Tax Gap Community of Interest (CoI) took place virtually, gathering around 50 participants. The session was opened by Nikola Djuric of the ATI Secretariat and brought together practitioners from revenue authorities, ministries of finance, international organisations, civil society, and academia to discuss two complementary analytical approaches to measuring and addressing corporate income tax (CIT) non-compliance and profit shifting. The CoI provides a recurring space for ATI members to exchange methodological experiences and country findings on tax gap analysis.

A new CIT gap study for South Africa

Chloé van Biljon, Tax Justice Project Officer at the Alternative Information and Development Centre (AIDC), opened the substantive discussion with findings from a recent CIT gap study for the tax years 2016-2020. The study applies UNU-WIDERs Tax Gap Toolkit’s bottom-up approach to South African administrative data and is the first systematic, firm-level estimate of where CIT non-compliance is concentrated in the country. CIT is South Africa’s third-largest revenue source, contributing roughly 17% of total tax revenue and 4.3% of GDP.

The study draws on more than 1.3 million annualised CIT returns and 10,220 specialised CIT audits and uses a random forest machine learning algorithm structured in two stages: first predicting the probability that a firm is compliant or non-compliant and then estimating the adjustment amount for non-compliant firms. A total of 196 predictors are used, including firm characteristics, financial ratios, deviations from industry averages, and year-on-year changes.

The overall estimated CIT gap across the five-year period is 13.5% of potential corporate tax revenue, with substantial differences across firm types. Small firms (with revenue below ZAR 20 million) show an average tax gap of around 25%, while medium-large firms come in at about 2%. Yet in nominal terms, the average evasion amount per medium-large firm exceeds that of small firms, given the very different size distributions. The construction sector alone accounts for more than 20% of the aggregate estimated CIT gap, with accommodation, agriculture, mining, and broader services also showing persistently high relative gaps. Among multinationals, evasion rates were notably lower than for purely domestic firms in the same size class. As van Biljon noted, this likely reflects both heightened scrutiny of multinationals by tax authorities and the fact that much MNE tax minimisation operates through legal or legally grey channels that fall outside what a bottom-up CIT gap study can capture. A striking finding concerns Special Economic Zones (SEZs): firms qualifying for SEZ tax treatment show evasion rates significantly higher than those of non-qualifying firms in the same size category (52% vs 25% for small firms, 19% vs 2% for medium-large firms). 

On the policy side, van Biljon highlighted four possible approaches: differentiated audit strategies by firm type and sector; simplifying compliance for small firms alongside enforcement; addressing MNE tax advantages through legislative reform rather than enforcement (including measures such as reinstating withholding taxes on cross-border service fees, renegotiating problematic tax treaties, and excluding tax-haven-based entities from the participation exemption); and factoring the additional enforcement costs of preferential regimes into the cost-benefit assessment of SEZs and similar incentive programmes.

Detecting profit shifting in administrative data

Ronald Davies, Professor at University College Dublin and Director at Skatteforsk Centre for Tax Research, presented ongoing work using administrative data to detect profit shifting at the firm level, with applications in Kenya, Uganda, and South Africa. With over USD 1 trillion estimated to be shifted to tax havens by multinationals each year, roughly 7% of corporate taxes lost in Kenya, Uganda, and South Africa combined according to the Atlas of the Offshore World, the question is how revenue authorities can move from aggregate estimates to identifying concrete audit targets. 

The method is not designed to prove profit shifting but to flag data patterns unlikely to occur in its absence, helping authorities target audits without replacing them. The logic builds on a simple observation: multinationals are typically among the most profitable and innovative firms in any economy, so when a multinational reports near-zero or abnormally low profits relative to both its industry and other MNEs, this is a statistical anomaly worth investigating. The method identifies firms that show both abnormally low returns (average return on assets more than 1.25 standard deviations below the industry / ROA in the bottom 10%) and abnormally high conduit activity such as intra-firm debt, trade with havens, or payments for IP services.

Across all three countries, losses are highly concentrated: in South Africa, 44 firms account for 63% of the total estimated loss with most firms flagged on intra-firm debt rather than IP and licensing arrangements. Professor Davies concluded with what he called the “lesson of realism”: tax enforcement is hard and will never eliminate avoidance, but targeting efforts at the small subset of firms most likely to be shifting can deliver meaningful revenue gains.

Open discussion

A rich exchange followed both presentations. On the South African study, questions arose about how SEZ status is identified in the data, clarification on the two-stage methodology, the quality of the underlying audit data, and how the potential evasion amount is estimated for firms that have not been audited.

On the profit-shifting detection method, questions focused on practical application. Participants asked how the approach handles capital-intensive sectors such as mining, where long depreciation schedules can legitimately suppress profits for many years. Professor Davies explained that comparisons are made within industry and that firm age is also accounted for in the South African application. Another question concerned how the threshold values triggering red flags are chosen: Professor Davies acknowledged that the 1.25 standard deviation cut-off (and experimentation with 2 SD) is partly data-driven but ultimately involves some arbitrariness, similar to using the bottom 10% of ROA or a conduit-to-EBITA ratio of 0.3 as the other thresholds.

A broader theme that emerged across both presentations was the importance of proportionality in enforcement. Only a small share of multinationals (around 5%) appears to engage in profit shifting, but they account for a disproportionate share of revenue losses, with especially the top 1% driving cumulative losses. At the same time, overly broad enforcement can impose significant compliance costs on many compliant firms. Participants were invited to reflect on how targeted, data-driven audit strategies of the kind presented can help strike this balance, and on how such approaches could be adapted to their own country contexts.