What is Beamm?

Beamm stands for BElgian Arithmetic Microsimulation Model, and is a tax-benefit microsimulation model. A tax-benefit microsimulation model simulates the tax-benefit system on micro-data, i.e., it simulates the rules of the tax system and social security for each individual or household in a representative sample of the whole population. This allows you to study how the impact of a policy reform varies with the socio-economic characteristics of all individual and households in the sample.

To build Beamm, and to allow you to analyze the impact of a policy reform, we proceeded in 5 steps:

1. Creation of a synthetic micro-dataset
2. Reweighting and uprating the synthetic micro-data
3. Simulating the tax-benefit system
4. Analysis and visualization
5. Behavioral modelling.
6. More about microsimulation modelling and Beamm

We summarize each step on this page. Click on the buttons for more information.

The objective

The first step consists of building a suitable micro-dataset. The Beamm model needs detailed micro-data (i.e., data on an individual level) allowing it to compute the individual taxes, social security contributions, transfers as well as the socio-economic background information needed to present a detailed impact analysis.

Statistical matching

Because no existing representative dataset covers all the required pieces of information for Belgium, we need to merge different existing datasets, each containing parts of the required information. However, all individuals and households are anonymous in each dataset that we have at our disposal, such that exactly connecting the information from different datasets to the correct individual is impossible. Fortunately, this is also not necessary, because we need our dataset to be correct at an aggregate level, and are not interested in being correct at the level of single individuals. To merge the different datasets into one single synthetic dataset with all the necessary information, we employ advanced statistical matching techniques (machine learing algorithms).

The data

The foundation of this synthetic dataset is an administrative dataset covering the entire Belgian population, and containing some key tax data as well as important socio-economic data. To complete the information about socio-economic background, we statistically merge these data with a series of additional datasets: the Survey on Income and Living Conditions (EU-SILC), the Household Finance and Consumption Survey (HFCS), the Household Budget Survey (HBS), the Labour Force Survey (LFS), BELDAM, MONITOR, the time use survey (HETUS), etc. Because all individuals in all our datasets are anonymous, the information that we impute from these additionnal surveys into the main dataset will almost always be incorrect at the individual level, but we aim to make this imputation as statistically correct as possible at the level of the population, such that the dataset represents the real world as closely as possible.

Simulation of a fictitious dataset

Once all micro-datasets are integrated, we have a highly realistic and very rich dataset, that may still contain some blocks of real data. For online use in the present interface, we use machine learning algorithms to create an entirely fictitious dataset that is very realistic and gives very similar simulation results as the real dataset, in technical terms: that has the same joint distributions as the ‘real’ synthetic dataset. This synthetic 100% fictitious but highly realistic dataset is used in the present online simulation platform.

What?

The synthetic micro-dataset created in the previous step necessarily represents a moment in the past: the moment at which the data were collected. However, the Beamm tax-benefit micro-simulation model is most often used to evaluate policy reforms in the present or future. Hence, we need to bring the micro-data ‘closer to the present or future’.

How

We use the Federal Planning Bureau’s projections of economic growth and the demographic evolution of the population to adapt the population weights in function of the demographic projections and we uprate the prices and wages in function of the economic projections. This process brings us a synthetic micro-dataset that reflects the present or future as well as possible.

How?

Beamm then applies all the rules of the Belgian tax-benefit system to each individual in the micro-dataset constructed in step 1. Hence, Beamm computes for each individual or household all the due taxes, the transfers that the individual is entitled to, the disposable income etc. This simulation of the tax-benefit system is performed twice: once for the current fiscal system prior to runtime, and once for a reform scenario, where a reform is chosen by you, the user, in the Beamm graphical user interface.

What?

Beamm covers at present the personal income tax (Fantasi), social security sontributions, VAT & excise duties, inheritance taxes, car taxes, property taxes, investment income taxes, gift taxes, birth allowances, income support, child benefits, pensions and maternity leave. Besides this standard set of policies, the researchers at CAPE can also use Beamm as a polyvalent platform for policy support and scientific research, by adding custom public policies. More details of how Beamm model handles these different taxes and transfers can be found in our documentation on the tax-benefit system.

What?

Once the Beamm microsimulation model has twice calculated for each individual or household all the covered taxes and benefits - once for the tax-benefit system before the chosen policy reform and after the reform - it can proceed with a detailed analysis and visualization of the reform’s impact from a variety of viewpoints.

How?

By aggregating the changes in paid taxes, Beamm can precisely compute and decompose the budgetary impact of a reform. Likewise, Beamm can compute for each individual or household the change in disposable (i.e., net) income due to the reform. Because we have for all individuals a lot of information about their socio-economic characteristics (age, household composition, education, work, housing characteristics, income, mobility, risk of poverty, disability etc.), Beamm can present a detailed picture of how the reform affects different groups of citizens differently, who wins or loses etc. Beamm can also compute many other policy impact dimensions, such as changes in labor costs for employers etc.

Combined reforms

Note that from an equity perspective, Beamm can not only depict a very detailed picture of how different socio-economic groups are impacted differently by a policy reform, or how vulnerable citizens are impacted. Beamm also allows policy makers, citizens and researchers to combine a targeted policy reform with detailed compensating policy measures to mitigate possible inequitable effects of a reform. Indeed, given that Beamm is a detailed model of (a big part of) the Belgian tax-benefit system and institutional context, Beamm users can combine a policy reform with detailed compensation measures at the right institutional level and compute the combined impact of such combination of measures in a single simulation.

Why behavior?

The tax-benefit micro-simulation model itself does not account for any behavioral reactions of citizens w.r.t. changes in the tax-benefit system. It merely applies the rules of the tax-benefit system, and computes as such the ‘day after effect’ of a reform, i.e., the impact before citizens can adapt their behavior to the reform. This is not only unrealistic, but changing the behavior of citizens is often an important motivation of policy reforms. In order to evaluate a policy reform whilst taking behavioral reactions of citizens into account, Beamm must be connected with behavioral models.

How does Beamm integrate behavior?

The integration of behavioral models with Beamm is conceptually relatively straightforward. The integration occurs principally at the level of the micro-data, and proceeds in the following steps:

1. The policy reform is translated (possibly with the help of Beamm) into changes in disposable incomes, generalized costs and other key parameters of the behavioral models.
2. The behavioral model then operates separately from the tax-benefit microsimulation model to produce predictions in terms of behavioral changes in mobility for different profiles of citizens.
3. These behavioral predictions are then brought to the synthetic micro-data: the model reconsiders for each individual in our synthetic micro-dataset the behavior in light of these predictions, and changes the information about mobility accordingly if necessary.
4. After changing the synthetic micro-data to fit with our behavioral predictions, we apply all the rules of the reformed tax-benefit system on the altered micro-data, and compute for each individual all the transfers and taxes. Bringing all this together, the model then presents an analysis and visualization of the impact of the reform from a variety of viewpoints, this time taking the behavioral reactions of people to the reform into account.

What behavioral models?

Tax-benefit microsimulation models are very flexible in terms of their integration with behavioral models. At present, CAPE has integrated 3 sorts of behavioral models in the Beamm platform.

• Labor market model: an econometric model of how individuals adapt their labor market participation in function of changes in the tax-benefit system.
• Consumption model: an econometric model of how individuals and households adapt their consumption patterns in function of changes in prices and disposable (i.e., net) income.
• Transport models (transport-beamm platforms only): models built by transport engineers to predict how changes in prices (e.g., due to taxes or subsidies) changes how people move by car, public transport, bicycle etc.

Microsimulation

Microsimulation modelling is a methodology that uses micro units (e.g., firms, individuals, households) to run ex-ante simulation analysis of public policies and economic or social changes.
Two main characteristics of microsimulation models:

1. They use surveys or administrative data.
2. They intensively use computers
   Microsimulation modelling has existed since the late 1950s and it has its formal roots in a proposal by Orcutt (1957, 1960). While its progress was relatively limited until the 1990s, micro-based simulation analysis is now used extensively around the world for policy analysis and design.

Beamm

Beamm is a tax-benefit microsimulation platform developed and maintained by the Center for Applied Public Economics (CAPE) at the UCLouvain Saint-Louis - Bruxelles since 2020. Beamm is almost entirely built in R, and makes intensive use of the R Tidyverse set of R packages for data science. The online interfaces are built with the R Shiny, R Blogdown, R Bookdown packages as well as the Hugo static site generator.