At the last meeting of the Community of Practice on Gender and Tax (CoPGT), held on 10th September 2024, the conversation focused on microsimulation models and their role in helping us understand the gendered effects of tax policies in the context of the wider fiscal system. As a community, we’re committed to tackling some of the toughest questions and the often-overlooked areas in the field of gender and tax research, and this session presented a good opportunity to engage in that shared commitment.

This was a collaborative discussion to explore what these models have taught us in our work in gender and tax. We shared concrete examples to illustrate how we (could) apply such models in our work, what their limitations are, and what can be learned from real-world applications of these models. Three CoPGT members, Laura Abramovsky (ODI, IFS and TaxDev), Pia Rattenhuber (UNU-WIDER) and Edoardo Magalini (OECD), provided an overview of such models and examples of applications to organise our discussion, which was moderated by Caren Grown (Brookings) and Giulia Mascagni (ICTD).

Caren Grown: “If microsimulation models are to be used in policy discussions, it is important to understand the underlying assumptions, data, and methodological decisions.” They may provide insights on magnitudes and effects and should be thought of as one tool in a broader toolkit.”

Microsimulation models allow simulating the distributional impact of tax policies and public spending on individuals and households with different characteristics, including income, gender and age. They can, for example, be used to synthetise how tax and benefits affect (formal) work incentives through their impact on take-home pay.  As such, they can be valuable tools when trying to figure out how these policies affect men and women differently.

Data often reveals starkly different outcomes for men and women in terms of paid and unpaid hours worked, earnings, income, consumption and other dimensions. Do these differences arise due to tax and benefit policies? Or do underlying differences drive gendered impacts of tax and benefits? What is the role of other factors, such as gendered social norms? It is difficult to tease these mechanisms apart. Most tax-benefit microsimulation models can provide one important piece of the puzzle, often capturing the interaction between fiscal policies and existing gendered differences in economic activity. More sophisticated and data-demanding models can incorporate how fiscal policies affect specific margins of behaviour such as decisions on whether to work and the amount of work hours. In any case, it is important to understand the main drivers of observed gender inequalities and identify the best policy instrument to address specific policy aims.

Laura Abramovsky kicked things off by diving into the different types of models typically used by researchers and policy analysts:

Tax-benefit microsimulation models are partial models. General equilibrium effects are thus muted. Microsimulation models can be either micro-level data driven or hypothetical models. Static models based on individual or household level data give us a snapshot of tax policy incidence or impacts from policy reforms.  Basically, they show how people’s incomes would change if a tax policy were changed overnight. They are thus a good starting point for understanding first-round “day-after” effects of policy reform, and for modelling relatively small policy changes.

But as Laura pointed out, they do not capture the whole story, which is relevant also when it comes to gender. Static models do not consider how people might change their behaviour in response to policy reform. For example, evidence clearly shows that women and men often respond differently to personal income tax reform in terms of whether to work or not and how many hours. A static model will thus not pick up if a woman decides to work less hours if her personal income tax burden increases due to a reform.

Behavioural models go one step further by considering how people might adjust their behaviour in response to tax changes, resulting in so-called “second-round effects”. Using a behavioural model, a study for Uganda, for example, suggests that the 2012 personal income tax reform led to lower reported incomes for some taxpayers affected by the reform but not all. Such models thus give us a better sense of the medium or long-run effect of a policy reform. Similarly, some workers might decide to leave the formal sector if tax burdens increase. This is a margin of behavioural response particularly important in the developing world. Analysis of a hypothetical in-work benefit for mothers in the formal sector in Ecuador would have a positive effect on female formal employment. Taking these behavioural reactions into account, the policy would reduce income poverty more than if one would just consider the static impact of the policy. The response is also not uniform between single women and women in couples, adding further context to the potential impact. These are valuable insights for policy makers when considering how a tax and/or benefit policy reform might affect men and women, and other groups of population differently, and ultimately also affecting tax revenues.

Dynamic models are long-term models, tracking how individuals and households are affected by policies over time. They are important tools for understanding future distributional consequences of potential policy reforms. They are also complex and data-intensive, which poses a challenge, especially in countries where data is limited.

Laura Abramovsky: “Microsimulation models help to answer questions such as who are the winners and losers of policy reform, including by income, age and gender; and whether a policy had its intended effect, for example in terms of poverty alleviation or redistribution. When thinking about tax and gender, it is key to consider the tax and spending system as a whole. Not every tax and spending policy has to address gender gaps. Across countries, women are often over-represented among the poor. Most of the impact of tax and benefits arises because of gendered differences in income, employment, entrepreneurship and consumption patterns, in turn mediated by gender gaps in a wide array of legal rights, unpaid work and protection and social norms more generally. We need to understand key barriers and best solutions to address these, and often this means to look beyond tax policy.”   

One major topic in the discussion was data gaps and data quality. As set out by Pia Rattenhuber microsimulation models can only go as far as they are powered by sufficient, suitable and high-quality data. Both administrative and survey data are essential for building accurate and meaningful microsimulation models. But both data types have their drawbacks. Administrative tax data in developing countries often do not include information on gender. By nature, they do not include information on informal workers; as women, particularly, tend to engage in informal work, administrative data cannot help answer questions in that area. While survey data capture informal economic activities to a certain extent, the quality of income data is lower and high-income individuals tend to be underrepresented. Some survey data do offer interesting information on informal markets, allowing to proxy in informal vs formal consumption, and thus incidence of VAT on households.

Pia Rattenhuber: Microsimulation models are only as good as the data they run on. Frequent, high-quality survey and administrative data are key for producing meaningful insights for policy. While administrative data provide the highest quality of income data reported to revenue administrations, they often lack information on gender, income outside of the tax bet, and household context. Survey data in turn provide rich socioeconomic fabric and can provide information on both income and consumption tax bases but are not collected often enough in most developing countries. Combining microsimulation analyses based on both types of data can overcome some of these drawbacks and is an important area of innovation.

Another challenge that came up in the discussion was the measurement of income or expenditure at the individual level and given the lack of data at the individual level, how to model intra-household resource allocation. How do we account for the fact that resources are not always equally shared within households? Women may have less access to household income, but many models estimate distributional impacts at the household level and still assume equal distribution, which means we are likely underestimating the true gender disparities in our analysis. This data issue needs to be solved collectively.

The World Bank’s Engendered-Commitment to Equity (CEQ) models produce static fiscal incidence analysis looking at the combined distributional impact of tax and spending policies for different households according to gender characteristics, though excluding how women and men may change their behaviour as a result of policy changes. These models were highlighted in our discussion as another tool for understanding the gendered incidence of tax policies given observed patterns of income and consumption, in the context of the wider fiscal system.

Edoardo Magalini (OECD) led the conversation on secondary earners and how household-level personal income taxation rules often disproportionately burden women. Using the OECD’s Taxing Wages model, he showed how most OECD countries adopt fiscal rules resulting in higher effective tax burdens for secondary earners, which discourages women from entering or remaining in the workforce. He explained that joint filing and household-level personal income taxation rules in countries like Germany place a heavier tax burden on secondary earners, creating a disincentive for women to work. On the other hand, he also highlighted that the analysis carried out through the Taxing Wages model focuses on just one dimension of a very complex policy challenge and additional perspectives on tax fairness and gender inequalities should be taken into consideration before elaborating clear policy recommendations.

Edoardo Magalini: While the strict methodology of Taxing Wages ensures the internal validity of the results, external validity should always be assessed carefully. Static simulation models can provide interesting insights only when they are contextualised, and their limitations are acknowledged. Across OECD countries, in 2023, second earners generally face a higher effective tax burden than a comparable single worker at the same earnings level. Second earners do not face disincentives to work from tax rules only in those countries where the individual is the tax unit and without household-level tax reliefs.

The CoP’s discussions underscored the value of microsimulation models in analysing how tax systems may impact women and men differently.  It is important to keep in mind the caveats such as that most models are static models, based on imperfect data and that they are traditionally oblivious to important underlying drivers of inequality between men and women, such as unpaid work and social norms. These drivers may interact with tax systems but are likely best changed using other policy levers. Evidence using other, suitable methodology is best placed to inform policy in those matters as a complement to microsimulation analyses.

Overall, however, microsimulation models can provide critical insights into how tax policy and their reforms may impact women and men differently, providing policymakers with key evidence when designing tax policies. As microsimulation models continue to evolve, they can add to the   research base that enables a fruitful discussion on what both inclusive and equitable tax reforms for men and women should be.

Daisy Attu

Daisy Attu has extensive experience coordinating stakeholders and facilitating engagement, including amongst women leaders on taxation in the African continent. She is currently the coordinator of the Community of Practice on Gender and Taxation hosted by the International Centre for Tax and Development.