Statistics Canada, Canada’s statistical agency, has a long history of using the software Modgen—and, more recently, its open source successor OpenM++—to develop and maintain dynamic microsimulation models in areas such as demography, health, and retirement income. Beyond Canada, OpenM++ is used in several European countries to model population aging and interactions with the social welfare system, demography and pensions. Building on the strong foundation of the OpenM++ open-source microsimulation technology, stcopenmpp is a dedicated fork of OpenM++ that will be released and maintained by Statistics Canada to support sustainable microsimulation model development, as well as experimentation and community contribution.
OpenM++ is a powerful and versatile tool that provides modern, portable, scalable, and cross platform capabilities for dynamic microsimulation. It allows users to model complex life-course events, including aging, employment trajectories, and changes in health status. The stcopenmpp repository offers a space for those interested in experimenting with this software, while also creating an avenue for users to contribute feedback and suggestions to its maintainers.
This presentation will offer a high level overview of the evolution of microsimulation programming tools at Statistics Canada, tracing the progression from Modgen to the development of stcopenmpp. It will also highlight the importance of sustaining and expanding open source tools and infrastructures to support ongoing microsimulation research and development. Finally, we will provide an expanded introduction to the stcopenmpp repository and the open source tools available within it.
Alignment is a critical calibration technique in microsimulation, ensuring individual-level transitions aggregate to known macro-targets. While indispensable for updating populations to match demographic projections or macroeconomic forecasts, the statistical properties of various alignment algorithms remain under-researched. This paper provides a systematic evaluation of alignment methods for discrete-time models to guide researchers in method selection.
We categorize existing methods based on a robust taxonomy: variable type (continuous vs. discrete), the nature of the outcome (exact match vs. stochastic), and time-step logic (continuous vs discrete). Building on the work of Stephenson (2018), we demonstrate that most contemporary alignment methods can be unified under a single constrained optimization framework, differing only in their choice of objective functions. This unification allows us to establish previously unrecognized relationships between seemingly disparate techniques.
Furthermore, we derive closed-form expressions and Taylor series approximations for computationally intensive iterative methods. These mathematical shortcuts allow modellers to reduce simulation run-times significantly while converting exact alignment processes into stochastic ones without losing desirable statistical properties. We then study the impact of these methods on the predicted distributions and covariance of outcomes between simulation runs.
Our analysis reveals that common techniques, such as Sorting by Predicted Probability (SBP) and Sorted By Difference between predicted Probability and a random number (SBD), are undesirable for microsimulation modelling as they can introduce biases and path dependence into the simulation outcomes. In contrast, parameter variation methods (i.e. recalibration of model coefficients such as intercept shifting) offer modellers control over the impact of alignment and better interpretability. Furthermore, methods like logit scaling, variance-weighted additive scaling for probabilities, and additive scaling for continuous variables offer greater robustness due to their foundations in probability theory. We conclude by providing a decision matrix for modellers, weighing the trade-offs between computational efficiency, distribution preservation, and variance reduction.
This paper analyzes the reduction in total labor costs induced by an increase in the minimum wage in France. Using the Ines microsimulation model developed by the French National Statistical Institute (Insee), I simulate a 2% increase in wages for all workers paid at the minimum wage. Due to the complex of exemptions of the French socio-fiscal system, an increase in the minimum wage leads to a reduction in employers’ social security contributions (SSCs) for workers earning between 1.01 and 3.5 times the minimum wage. I confirm existing results from L’horty (2000): overall, a 1% increase in the minimum wage reduces employers’ SSCs by approximately €1.67 billion.
The BELMOD microsimulation model of the Belgian Federal Public Service Social Security is based on an administrative input dataset. While the model’s policy simulations are updated twice a year, the most recent input dataset currently available refers to 2019. As a result, the discrepancy between the input data and the present situation has grown to several years. This gap can be reduced through nowcasting methods, by updating the outdated input data with more recent information to bring it more in line with the present situation, thereby making the data more suitable for simulations relating to the most recent years. We applied nowcasting to the BELMOD input dataset by incorporating both demographic changes and changes in individuals’ labour market status. To assess which nowcasting approach is most suitable for BELMOD, we developed and compared three different methods. In the first two methods, labour market status transitions are modelled using, respectively, a parametric and a non-parametric approach. For individuals experiencing a labour market status transition, the relevant variables in the dataset are subsequently adjusted. In addition, demographic changes are incorporated in these two methods through reweighting. In the third method, both labour market status and demographic changes are implemented through reweighting. We validated these three methods using external statistics on the number of income and benefit recipients, as well as aggregate income and benefit amounts.
We introduce the use of microsimulation modelling in the context of survey experiments, coined “simu-survey experiments.” They are a specific type of information experiment (i.e. a research approach for studying how specific pieces of information affect attitudes or intentions) in which the informational treatments are based on empirically grounded microsimulation estimates.
We argue that simu-survey experiments are particularly well suited to research contexts that involve some form of change, either a prospective policy reform (e.g. pension reform) or a life course event (e.g. taking up work). In both cases, the core principle is the same: respondents are asked in a survey to evaluate a situation that departs from the status quo, and the experiment identifies how their expressed attitudes or behavioural intentions shift when they are provided with microsimulation based information about the likely consequences of that change. Depending on the research question, this information can be situated at the household level, providing respondents with personalized estimates tailored to their socio demographic profile (e.g. the predicted change in disposable income), and/or at the societal level, offering general information that applies uniformly to all respondents (e.g. the expected change in poverty).
Simu-survey experiments therefore rest on a dual promise of causal inference and empirical realism. Respondents are not reacting to fabricated, unrealistic or vague information, but to concrete numbers that could actually apply to them or to their society, meaning that their responses will more closely approximate to how people think and behave in the real world. As such, simu-survey experiments represent a versatile methodological innovation for studying how individuals evaluate certain hypothetical changes in either policy context or personal circumstances (and their consequences) across a wide range of policy domains. They also allow researchers to more accurately assess how informed individuals form opinions, revise their beliefs or update their behavioural intentions.
To illustrate how simu-survey experiments can be applied in practice and to demonstrate their effectiveness, we present the second wave of the Basic Income in Belgium (BABEL) survey, our own simu-survey experiment aimed at uncovering how support for a universal basic income (UBI) is causally impacted by microsimulation-based information regarding its specific household-level and societal-level policy outcomes.
Synthetic populations are a key component of many transportation and urban analysis frameworks, as they provide disaggregated representations of individuals or households used to feed traffic simulators and exposure models. Beyond mobility studies, they are increasingly mobilized to assess territorial sensitivity to external factors such as environmental nuisances or construction noise. Traditionally, synthetic populations are generated using calibration-based methods such as Iterative Proportional Fitting (IPF), which adjust a micro-sample to match aggregated census constraints. While robust and interpretable, these approaches are limited in high-dimensional settings and can only reproduce individuals that are already present in the initial sample.
Recent advances in machine learning offer new perspectives for synthetic population generation. In particular, Variational Autoencoders (VAEs) have demonstrated strong capabilities in learning complex joint distributions and generating realistic synthetic data in domains such as image and text generation. Applied to population synthesis, VAEs allow the modeling of rich, multidimensional dependency structures between socio-demographic attributes and enable the generation of more diverse populations. However, VAEs alone do not naturally enforce consistency with known marginal distributions derived from official statistics, which remains a key requirement in applied territorial studies.
This contribution presents a hybrid methodology that combines the generative power of VAEs with the statistical guarantees of IPF. First, a VAE is trained on a microdata sample to learn a low-dimensional latent representation of individuals and to generate synthetic agents capturing complex correlations between attributes. In a second step, IPF is used as a post-processing procedure to adjust the generated population so that selected marginal distributions strictly match external constraints. This decoupled strategy leverages the strengths of both approaches: the flexibility and scalability of VAEs in high-dimensional spaces, and the ability of IPF to enforce consistency with official statistics.
The presentation will detail the architecture of the VAE and the integration of IPF as a calibration layer. Results will illustrate how this approach improves diversity, realism, and statistical coherence of synthetic populations compared to traditional methods.
Spain records one of the highest rates of in-work poverty in the European Union (Eurostat, 2024). Despite this, the development of policies specifically targeted at supporting low-income workers has been limited, especially when compared to other European countries (Laun, 2019). This policy gap, combined with the expansion of minimum income guarantees schemes, may weaken work incentives by narrowing the income gap between employment and nonemployment, thereby increasing the risk of poverty traps among low-income households (Domínguez-Olabide & Zalakain, 2023).
This paper proposes the introduction of an in-work benefit in Spain, inspired by the American Earned Income Tax Credit (EITC), with the aimed of improving the living conditions of the low-wage workers while strengthening labour market incentives. As in the EITC, the benefit is implemented through the Spanish personal income tax system as a refundable tax credit, varying according to household type. In particular, we propose a more generous credit for families with dependent children, as they face higher poverty rates. The reform is simulated using the EUROMOD microsimulation model, based on 2022 EU-SILC data. Labor supply responses are estimated using a structural labour supply model.
The main results suggests that the proposed in-work benefit would not generate negative labour supply effects. On the contrary, the reform is associated with an overall increase in labour supply, particularly in full-time employment. Positive effects are found for both single and couples, with stronger impacts among households with dependent children.
We introduce MIDAS DE, a LIAM2 based microsimulation model for analysing German pension incomes under current law and counterfactual policy scenarios. The model reproduces the statutory formula for earnings point accrual, access and type factors, and the current pension value, and is designed to evaluate distributional, gender, and adequacy effects of reforms such as pension splitting and survivor benefit adjustments within a unified framework. MIDAS DE is implemented in LIAM2 using discrete time processes over entities (individuals, households), typed fields (e.g., insured status, pension points), and explicit links (spouse/partner, parent–child) necessary for survivor pensions and splitting eligibility. The model combines SOEP RV administrative insurance records with SOEP survey microdata. Linkage relies on rv_id (SOEP RV↔SOEP) and pid to reconstruct households and partnerships from ppath/ppathl, household matrices from pbrutto/pl, and family histories from biofam/biomars. This enables (i) identification of spouses; (ii) retrieval of pension relevant histories for groups under represented in DRV (e.g., civil servants, self employed) via biowork/biojob; and (iii) construction of household attributes needed for survivor benefit means tests. To harmonise labour income for accrual, we estimate gender and occupation specific Heckman selection models for three groups—salaried employees, self employed, and civil servants—ensuring segment specific participation mechanisms and wage processes. Predictions are selection corrected (inverse Mills ratio) and back transformed with lognormal adjustments; observed wages replace predictions when available. This captures institutional heterogeneity (e.g., civil service pay scales, self employment volatility) and mitigates bias from missing or misreported earnings, feeding consistent contributory bases into earnings point calculations. Robustness checks consider exclusion restrictions (household composition and partner status), outlier trimming, and alternative retransformation (smearing). Technically, MIDAS DE shows how LIAM2 can host a law consistent German pension engine calibrated on linked RV–SOEP microdata with explicit household links, enabling faithful simulation of survivor pensions, pension splitting, and income offsets. Substantively, the model structures policy scenarios along current law splitting, VersAusglG style variants (with and without 25 year conditions and cross pillar coverage), and a universal splitting regime, providing outcomes on the gender pension gap, poverty at retirement, and fiscal effects.
Unemployment benefits and contributions are often uniform across partnered and single individuals. In this paper, we study whether this constitutes an optimal policy, given that partners within couples benefit from risk sharing. Our contribution is threefold. First, we document the channels through which unemployment benefits have a first-order impact on aggregate welfare. Second, we express the welfare impact in terms of a few sufficient statistics, which can be derived from observational data. Third, we quantify the optimal policy in an empirical application using US data.
Background and Motivation Regional microsimulation models face critical data challenges: survey data lack local representativeness, administrative data access is often restricted, and commercial datasets are costly. Increasingly, researchers turn to synthetic data generated through statistical models, but questions remain about their validity for policy analysis compared to established commercial alternatives like Experian.
This study evaluates synthetic population data against Experian commercial data for Essex microsimulation applications. Both datasets represent different forms of modelled data: Experian combines administrative records, commercial sources, and modelled estimates, while synthetic data is generated through statistical algorithms. Although both are typically validated against official statistics (ONS Census, household size, income benchmarks, age structure), validation against marginal distributions does not guarantee equivalence for policy analysis, where joint distributions and correlation structures matter critically. Our existing work developed UKMOD-aligned weights for Essex by reweighting FRS survey data to match Experians joint distribution of household characteristics. This provides a validated benchmark against which to assess synthetic data alternatives.
Methodology The comparison evaluates synthetic data against our established reweighted UKMOD variant by applying identical tax-benefit policy scenarios through UKMOD using each dataset. Policy simulation outputs - including income distribution changes, gains and losses by household type, and demographic impacts, are compared to assess whether synthetic data replicates the distributional patterns produced by our machine learning-based reweighted data. This direct comparison provides practical evidence on whether synthetic data offers comparable analytical utility to established reweighting methodologies for regional microsimulation applications.
Expected Contribution This research provides practical evidence on whether synthetic data produces comparable policy analysis results to established reweighting methodologies for regional microsimulation. The study addresses a fundamental question: can synthetic data adequately substitute for more resource-intensive data alignment approaches while maintaining analytical reliability? Findings will inform data strategy decisions for researchers and local authorities conducting sub-national distributional analysis, particularly where resource constraints, data access limitations, or timeliness considerations favor synthetic data solutions.
