What an economic model of South Africa tells us about trade, unpaid work, and gender equality in Africa
By Alfred Latigo, Founder and Chief Innovation Architect, ScaleX Labs, and Omar Abdourahamani, UNDP
In 2003, at the United Nations Economic Commission for Africa (UNECA), we built one of Africa’s first economic models to count the unpaid work women and men do at home, such as cooking, caring, and fetching water and fuel, alongside paid work in the market economy. Working with Statistics South Africa and Université Laval, we used South Africa’s first national time-use survey to estimate this unpaid work. We then brought it into the country’s economic accounts through a gender-aware computable general equilibrium (CGE) model. Our question was practical: when South Africa removes its import tariffs, who gains, who loses, and who pays in time?
Twenty-three years later, we have returned to that question. The original data file had been lost, so we rebuilt it entirely from the figures published in the 2003 report. We then re-coded the CGE model, tested it against the original results, and extended it with Responsible AI. The rebuilt model confirms our earlier findings. It also shows what policies need to do if they are to change outcomes for women rather than simply describe them.
Why Count Unpaid Work?
Most economic models treat the economy as if it stopped at the market. Yet our earlier estimates suggested that household production, the unpaid work that keeps families fed, clean, and cared for, is equivalent to roughly 44% of total economic activity in South Africa. Women do most of it. In the data underlying our model, women carry about 1.7 times as much unpaid domestic work as men.
When a policy changes wages, prices, and jobs, it also changes how women and men divide their time between paid work, unpaid work, and rest. A model that ignores the household cannot see these effects. That was the gap our UNECA work set out to fill, and it is why the model was endorsed by an Expert Group Meeting of African policymakers in 2003.
Rebuilding the CGE Model 23 Years On
The 2003 report published every share needed to reconstruct the original data, sector by sector and household by household. These include:
- Trade and tax rates;
- How value added is split between workers and capital;
- Men’s and women’s wage shares;
- Household income and spending;
- How men and women in each population group spend their time.
From these we rebuilt a complete picture of South Africa’s 1998 economy: 27 sectors, four household groups, women’s and men’s labour, and R 355 billion of unpaid domestic work valued at market wages.
The rebuilt model reproduces the original results closely. Sector by sector, the match with the published figures ranges from 94% to 99.9%. The original gender findings all reappear:
- Removing tariffs grows the economy slightly, with output rising by 0.16% and consumer prices falling by about 1.7%.
- Men’s wages and paid work rise more than women’s. Men are concentrated in the expanding, export-oriented mining sectors, while women work mainly in textiles and services, which contract.
- Women continue to do about 1.7 times as much unpaid work as men, because both men and women adjust mainly by giving up leisure rather than housework.
- African women, the poorest group, reduce their paid work by about 0.8%, almost exactly as the original study found (0.75%).
Reproducing these results from the published record alone, more than two decades later and with an entirely new code base, shows that the original UNECA work was sound.
From Gender-Aware to Gender Transformative
A gender-aware model describes how outcomes differ for women and men. Gender Transformative Approaches go further. They ask which policies change the structures behind those differences: the care burden women carry, the segregation of jobs into “men’s work” and “women’s work,” and the gap in access to skills, credit, and technology.
We tested four such policies alongside full tariff removal.
1. How Lost Tariff Revenue Is Replaced
Removing tariffs costs the government revenue. Replacing it through a progressive income tax protects the poorest households. Using VAT instead shifts the burden onto them and cuts the poverty gain by almost 90%. It also reduces women’s paid work, because VAT erodes the real value of wages.
2. Supporting Sectors That Employ Many Women
A 5% subsidy to health and social work, textiles, hospitality, and other services raises women’s paid work and reduces poverty. But because men also work in these sectors, women’s share of income barely changes.
3. Investing in Water, Energy, and Care Services
This brings the largest welfare gain of any policy, at 3.7% of household income. It is also the only measure that shifts unpaid work. Women’s domestic time falls by 3.4%, men’s rises by 3.4%, and the ratio of women’s to men’s unpaid work falls from 1.73 to 1.62. Women gain both paid work and rest.
4. Closing the Gender Productivity Gap
Better access to skills, credit, and technology for women raises GDP more than any other single policy and reduces poverty the most. But when men’s and women’s jobs remain separate, much of the gain still flows to men.
The Key Finding: Freed Time Must Meet Open Doors
Every policy improved welfare. But when the four were combined into a single package, women’s share of labour income still fell by about one percentage point. Their time was freed, but the doors to better-paid jobs stayed closed. More women seeking work crowded into the same occupations, which pushed down women’s wages.
The decisive factor was occupational segregation. When we allowed men’s and women’s jobs to become less segregated, the same package produced very different results:
- Women’s paid work rose by 4.2%.
- Women’s share of labour income rose by 0.54 points instead of falling.
- The poverty headcount fell more than under any other scenario.
In short, easing women’s time burden must go hand in hand with opening male-dominated occupations to women. That means vocational training in mining, manufacturing, and technical trades; enforcement of equal-pay and anti-discrimination laws; and changes in workplace norms and public procurement.
How We Used Responsible AI
We used artificial intelligence carefully and transparently, in three steps:
- Testing the conclusions. AI ran the full CGE model about 6,000 times, each time with a different combination of policies and assumptions about how the economy works, to see which conclusions hold.
- Learning the model. A neural network learned to reproduce the model’s answers with over 99.5% accuracy on runs it had never seen. Policymakers can now explore scenarios in seconds rather than hours.
- Finding what matters most. A statistical sensitivity analysis showed which assumptions drive which results.
The analysis confirmed that the segregation finding is robust. Occupational segregation explains 99% of the variation in women’s income share across all 6,000 runs. Women’s income share rose in 90% of runs where jobs were less segregated, and in fewer than 0.1% of runs where segregation stayed high.
Throughout, the CGE model, not the AI, produces every reported result. Every assumption is documented, and every number can be regenerated from the code. Interpretation and policy judgement remain with people. This reflects the founding principle of ScaleX Labs: AI must strengthen development professionals and never replace their expertise, context, judgement, oversight, and accountability. People lead. AI strengthens.
From UNECA to ScaleX Labs and GIDE
This CGE model is the macroeconomic root of the work we continue today. The Global Institute for Development Evidence (GIDE) is an independent non-profit institute founded in 1990. There, this line of work produced the Gender Transformative Tool and the Scalability and Replicability Analysis Tool, later applied in climate-finance portfolios across more than 25 African countries. ScaleX Labs, an integrated Responsible AI platform built for the development professionals of the Global South, now carries this work forward.
The rebuilt model will become the analytical engine of GenTraQ™, our platform for both designing and measuring gender transformative change. GenTraQ™ places a policy or programme on a five-level continuum, from gender-blind to gender-transformative. Its forecasting component, GenTraQ Forecast, will let governments and programme designers test a policy’s likely gender transformative impact before adopting it, along with the conditions that impact depends on. GenTraQ™ and our AI-Powered MEL System will then measure what actually changed after implementation. Forecasts and outcomes can then be compared and the models improved.
Working with Partners in the Global South
Our aim is for each country to own its data, its models, and its forecasts. We will work with national statistics offices, ministries of finance, gender, and planning, universities, and research institutes. Together we will build national time-use accounts, country-specific CGE models, and the skills to use them. Through the ScaleX Labs Academy and GIDE’s research and training, we will help strengthen the national capacity that the 2003 UNECA Expert Group Meeting called for.
Our next steps are to update the South African model with current data, add unemployment and the informal economy, and extend the approach to other African countries that have time-use surveys.
A Note on the Evidence
The model uses data from 1998 to 2000, so it shows the method and the direction of effects rather than today’s magnitudes. Some policy effects, such as the productivity gains from care investment, are illustrative assumptions, and the AI analysis tests how much they matter. The full working paper sets out every assumption.
We gratefully acknowledge UNECA’s African Centre for Gender and Development, Statistics South Africa, and the authors of the original 2003 report, including Ismaël Fofana, Bernard Decaluwé, and John Cockburn of Université Laval.
The full working paper, “AI-Powered Gender Transformative Macroeconomic Computable General Equilibrium (CGE) Model for Evaluating Impacts of Policies on Poverty Reduction in Africa: The Case for South Africa,” is available on request and is being prepared for journal submission.
Learn More or Partner With Us
To learn more or to partner with us, visit scalexlabs.io or write to info@scalexlabs.io.
