The analysis of the impact of tax and benefit transfers is an important field in the evaluation of public policies. While many studies assess the effect of a particular transfer, it is difficult to obtain an overall view of the tax and benefits system and its impact on standards of living. In fact, the legislation on taxes and benefits in France is complex (based on numerous legal codes: the General Tax Code, the Social Action and Family Code or the Social Security Code) and there are many interactions between tax and benefit transfers. Microsimulation models make it possible to identify the effects of these different measures on the incomes of individuals and their families, taking into account their specific characteristics. These models provide an overall analysis of transfers according to family configuration, household living standards or the employment status of individuals. They also make it possible to take into account all the interactions that exist between the measures. Microsimulation models have thus acquired “a central, though slightly misunderstood, position in the field of analysis of tax and benefit policies” (Legendre, 2019).

This article provides a presentation of the INES microsimulation model. Part of the family of static microsimulation models, INES simulates the main taxes and benefits in France. It makes it possible to carry out in-depth studies to shed light on areas of debate in the fields of monetary redistribution, taxation or social security protection.

“Static” microsimulation models simulate, for each individual, the taxes they pay and the benefits they receive based on the legislation’s schedules, allowing their disposable income to be determined. To do so, they rely on real data (age, family composition, income, employment status of individuals, etc.). It is possible to vary the schedules and study the effect of those variations on household disposable income or on public finances, and thus to shed useful light on the effects of these public policies.

These models thus make it possible to assess the effect of a variation of the schedules for certain social security taxes or benefits on the living standards of the people concerned, to identify the winners and losers of the reforms, or even to measure the effect of the reforms on the poverty rate, living standards inequality and public finances. To do so, two situations are compared: one in which the schedules have been modified, compared to a “counter-factual” one, generally with the legislation currently in force.

This was the goal behind the creation of the INES model in 1995. Having initially been developed by INSEE (David et alii, 1999), DREES came on board in the early 2000s and CNAF joined in 2016 (Box 1). From the beginning of its story, the INES model has been used as a decision support tool to calibrate major tax and redistribution reforms. The model has thus been used to calculate figures for the introduction of the prime pour l’emploi (an earned-income tax credit) in 2001, the creation and implementation of the revenu de solidarité active (RSA – a statutory minimum income coupled with an in-work benefit) in 2009, the prestation d’accueil du jeune enfant in 2014 (PAJE – an early childhood benefit) and the prime d’activité (an in-work benefit) in 2016. In 2019, it was used to calibrate the response measures to the social emergency following the gilets jaunes (yellow vest) movement and to fuel work on the revenu universel d’activité (RUA – a merger of means-tested benefits).

Unlike the models of the case studies for representative agents, which simulate legislation for a fictitious individual, the microsimulation models are based on the observation of real situations and are, therefore, closely linked to the data used. Since 2006, the INES model has mainly used data from the Tax and Social Incomes Survey (ERFS) (Box 2). This source contains hundreds of detailed information entries on income, employment or family status for a sample of more than 50,000 households and 130,000 individuals. The ERFS is INSEE’s preferred source for the analysis of living standards. This richness of information makes it possible to finely simulate the social security benefits, taxes and contributions that depend on many variables, which are not always present in fiscal sources alone: family profile, labour market history, number of hours worked, type of job and business, rents, place of residence, disability status, etc.

The sample from the ERFS is representative of the French population living in ordinary housing and in metropolitan France (Box 2). This makes it possible to produce analyses by dividing the population into ten groups of equal size according to their living standards (from the poorest 10% to the wealthiest 10%), or according to employment status or family configuration. Some results can be given at finer levels, by dividing the population into 20 groups, for example, in order to estimate the effect of certain specific measures that mainly affect the wealthiest, such as the replacement of the impôt de solidarité sur la fortune (ISF – solidarity tax on wealth) with the impôt sur la fortune immobilière (IFI – real estate wealth tax) (Paquier et alii, 2019). However, the estimated effect at the extremes is less robust than those estimated for each tenth and the sample size does not allow for more detail, such as per hundredth for example, unlike exhaustive sources.

Finally, it should be noted that data on income for year N are only available in year N+2: for the purposes of the model (simulation of legislation for the year N+2), this requires “ageing” the ERFS data (see below).

For each person, home or household, depending on the level of analysis required, the INES model uses the ERFS data and applies all the rules of the legislation to them.

It is thus possible, for example, to identify the households eligible for the prime d’activité (in work benefit), using the family composition (number of dependent children, whether the parents are together or it is a single-parent family) and people’s monthly employment situation reconstituted based on an “employment calendar” and the household income from ERFS database. By using all this information for each potential household, the model makes it possible to simulate an entitlement to the benefit which is then integrated into that household’s disposable income, by taking into account the possibility of non take-up of the benefit, i.e. that people do not claim the benefit even though they are entitled to it.

Income is crucial for simulating the various transfers. The incomes used as input data for the model, and which remain constant in the event of a change to the amount of the various transfers received or paid, are the posted net incomes in the tax return (this is referred to as the fixed point of the model). Based on these posted incomes, the INES model simulates most tax and benefit transfers, as illustrated in Figure 1. The social security contributions and taxes are simulated, whether they are payable by the employer, the employee or the self-employed person. Similarly, all other taxes and subsidies based on the total payroll are simulated, such as the crédit d’impôt compétitivité-emploi (CICE – tax credit for employment and competitiveness) or the tax on employees or transport. For direct taxation, the model simulates income tax, all tax credits and flat taxes3. Recently, the development of a module on household wealth (Paquier et alii, 2019) has made it possible to simulate tax on wealth (impôt de solidarité sur la fortune). Finally, the indirect taxation module (Ouvrir dans un nouvel ongletAndré et alii, 2016) makes it possible to add indirect taxes such as value added tax (VAT), the taxe intérieure de consommation sur les produits énergétiques (TICPE – domestic tax on consumption of energy products) or the taxes on tobacco products.

With respect to the benefits paid to households, the model makes it possible to simulate means-tested and non-means-tested family benefits, personal housing benefits, statutory minimum incomes, the prime d’activité and the garantie jeune (a benefit for NEETs aged 16 to 25). For some of these benefits, such as the prime d’activité, non take-up is taken into account. Finally, certain in-kind transfers can also be simulated, to allow an analysis of “expanded” disposable income (Figure 1).

Most often, these transfers are simulated based on the legislation of the previous year, but they may also be simulated based on other legislation (see below).

Figure 1. The Different Concepts of Income and the Tax and Benefit Measures Simulated Using INES

The INES model simulates these tax and benefit transfers thanks to a sequence of more than 100 computer programs. To do so, it uses as its input a version of the ERFS which is first reworked in the “upstream” part of the model, then used in the “downstream” part to simulate the various transfers of the tax and benefit system for a given legislation year, based on the legislative schedules. The “upstream” part makes it possible to ensure that the format of the data will be usable by the rest of the model. The “downstream” part contains all of the programs that simulate the legislation. In this part, the parameters of the legislation (and potential non take-up) are applied to the data output of the upstream part. This is the operation to which we are referring when we use the term “simulation”. It differs from the “imputation” present in the upstream part, which are intended to add to or rely on external data.

As an output, the model makes it possible to determine the number of recipients or persons subject to the various transfers and the financial sum distributed, the disposable income and the living standards of households, as well as the various usual inequality and poverty indicators (Figure 2). The fact that the transfers are simulated makes it possible to compare all the results of the model in different situations, in accordance with legislative changes, and to analyse the redistributive effects.

In order to function, the model needs two important parameters: the reference year (the year of the input ERFS) and the legislation year to be simulated. As it is currently used, this pair is generally composed of the latest version of the ERFS available (year N) and the year of legislation corresponding to two years later (year N+2).

Regardless of the legislation year selected, the ERFS data are “aged” by two years in the upstream part of the model, so as to best represent the population for years N+1 and N+2. To that end, the model takes into account a multitude of socio-demographic (age pyramid, socio-professional categories, etc.) or economic (inflations, salary evolution, etc.) developments between the reference year and two years later.

Specifically, this ageing process is carried out in two stages, one calibration weighting stage in which the weightings of each household in the ERFS are changed and one individual income evolution stage:

Figure 2. Data and Parameters, the Inputs of the INES Model

The model is updated each year between February and October to simulate the previous year’s legislation (Figure 3). In 2020, the model thus evolves to incorporate the 2019 legislation. This update consists, on the one hand, of updating all the parameters of the various legislative schedules. For example, in 2020, the flat amount of the allocation adulte handicapé is updated, which rose from €860 to €900 in November 2019. It also involves coding work on the new tax and benefit measures, which can be significant when major reforms are introduced.

In addition to these annual updates, the model is continuously amended to be improved or corrected. In addition, supplementary modules are created on a regular basis. These supplementary models are optional in the execution of the model and make it possible to extend the scope of the measures covered and to flesh out the outputs with, in particular, an analysis of indirect taxes (mainly VAT and excise duties on tobacco and alcohol) and taxes on wealth.

Figure 3. Calendar for the ERFS and Updating INES

As it is most commonly used, the model does not take into account any behavioural effects in the evaluation of tax and benefit reforms, it is purely static.

However, behavioural responses may result from the reforms implemented on simulated transfers: for example, a fall in tobacco consumption following the tax rise, an increase in dividends paid following the introduction of the prélèvement forfaitaire unique (flat tax on capital income – PFU) in 2018 or even the change in the labour supply following the reforms to the prime d’activité.

The effects of macroeconomic closures are also not addressed, such as the positive impact of higher incomes on consumption and GDP and, in turn, on employment and incomes. Introducing these impacts requires making many assumptions, relying on work that is often divergent and questionable, however serious it may be. Static analysis is less contestable and thus the choice to not take behavioural effects into account is “virtually systematic by most governments and independent institutions responsible for costing economic measures” (Ouvrir dans un nouvel ongletBach et alii, 2020).

Thus, it must be borne in mind that the effects measured are generally short-term effects, not taking into account dynamic effects. However, some studies have estimated behavioural elasticities based on the INES model (Ouvrir dans un nouvel ongletFugazza et alii, 2003; Ouvrir dans un nouvel ongletLehmann et alii, 2013; Sicsic, 2019). The evaluation of the introduction of the prime pour l’emploi (earned income tax credit) in 2001 was thus carried out taking into account the effects on the labour supply (Ouvrir dans un nouvel ongletFugazza et alii, 2003). Current work is measuring the impact of the PFU and the reform to the ISF in 2018 by integrating the effects of these measures on amounts of wealth and dividends.

INES is the first microsimulation model based on representative data to have entered the open source era: since 2016, the model’s source code and documentation are publicly available online on the ADULLACT (Association des développeurs et utilisateurs de logiciels libres pour les administrations et les collectivités territoriales) website (see Ouvrir dans un nouvel onglethttps://adullact.net/projects/ines-libre).

Access to data from the ERFS remains strictly controlled due to the fiscal data included. However, anyone with this right of access can now use the model to carry out studies or evaluate reforms. This openness gives the model “a central position that is demonstrated, for example, by its use by the OFCE” (Legendre, 2019). Indeed, the OFCE is one of the very first users of the model and heralded this availability in 2017 as a “silent revolution in official statistics” (Ouvrir dans un nouvel ongletMadec and Timbeau, 2017).

The richness of the documentation is key: both the development team and the open source users must be able to understand the model and be aware of the quality of its results in regards with the measures studied.

At the end of the update period, the development team produces a validation note for the stabilised version of the model. This version is also referred to as the “fixed” model for the legislation year that has just been incorporated. Thanks to this “fixed” version, or tag in the vocabulary of version management tools, the constant evolution of the model does not hinder the reproducibility of the results This is particularly useful, as the figures must be reproducible with a certain degree of consistency from week to week or even year to year.

The annual validation note provides an overview of the model’s strengths and weaknesses: in particular, it compares the model’s outputs with the number of benefit recipients and the amounts provided or paid, obtained from external sources and called “external targets”; the latter are corrected to a field equivalent to that of the ERFS.

The results of the simulations are not “aligned” with the external sources, which makes it possible to judge their quality. It is thus shown that simulations for income tax, family benefits, housing benefits or the prime d’activité are very close to the targets, while simulations for the allocation aux adultes handicapés (AAH) or benefits for young children (CMG, PrePare, AEEH) are of lower quality. In most cases, measures are simulated with less precision because they are aimed at a very specific population that the data does not identify perfectly. This is particularly true in the case of the allocation aux adultes handicapés.

More general documentation for the model can be found by registered users on the ADULLACT website in a “wiki” specific to the INES model. In addition, each development of new modules or new transfers simulated in INES gives rise to working papers documenting the method and results of those additions.

Since the mid-nineties, the expansion of the INES team and the progressive enrichment of the model through the addition of new modules able to expand the simulations have led to a diversification of uses and publications related to the model. Five important uses of the INES model are detailed below. The first three focus on an analysis of the tax and benefit system as it exists or of the effect of reforms that have taken place (ex-post analysis), while the next two are used to study the effects of avenues of reform or fictitious reforms (ex-ante analysis). Two examples of more selective studies are also presented in Box 3.

One of the main objectives of the INES model is to describe the redistribution performed by the tax and benefit system. Since the start of the twenty-first century, an annual overview of the redistribution has been published in the INSEE publication “France, Social Portrait”. The amounts of the various transfers per decile of living standards are indicated therein, as well as disposable income before and after redistribution, which makes it possible to quantify the reduction in inequalities due to the tax and benefit system. Thus, in 2018, the poorest 10% of the population have average living standards before redistribution of €3,290 per year, compared to €73,130 for the wealthiest 10%, i.e. 22 times more (INSEE, 2019). After redistribution, this ratio is reduced to 5.6 times more.

This redistribution assessment also makes it possible to calculate the contribution of each transfer to the reduction of inequalities. It is thus shown that, although social benefits represent monetary transfers corresponding to half the amount of the taxes, in 2018 they account for 63% of the reduction in inequality due to their targeting of the poorest, compared to 37% for taxes.

The advantage of a microsimulation model for the analysis of redistribution is threefold, it makes it possible to:

     (i) carry out redistribution assessments over a long period, as has been the case in the past
     with the INES model: see (Murat et alii, 2001) for the period 1990-1998, (Amar et alii, 2007)
     for the period 1996-2006, and (Eidelman et alii, 2013) for the period 1990-2010;

     (ii) take into account transfers not present in the ERFS data:
     for example, the non-contributory social security contributions in the usual analysis, the impôt sur la fortune, and even a broader scope from redistribution to health cover or education in some publications (Amar et alii, 2008);

     (iii) provide more recent results than the ERFS data based on previous years.

The previous application provides an overview focused on the field of tax and benefit transfers relevant to the redistribution analysis. One may also wish to analyse the effect of all legislative and regulatory changes on household living standards and on various inequality indicators.

To that end, every year since 2014, INSEE and DREES have published a redistribution and budget assessment for the new tax and benefit measures that came into force the previous year in the review France, Social Portrait. The assessment includes the same analysis of the measures implemented in the year preceding publication (for example, for 2018 in the 2019 edition of the publication; (Biotteau et alii, 2019)) stemming from the initial and amending laws on finance and the funding of the social security system, as well as decrees and national inter-professional agreements..

The principle of this evaluation is to compare the disposable income obtained with the new legislation against a so-called “counter-factual” legislation, if no legislative change had occurred during the year. The difference between the actual situation and the counter-factual situation corresponds solely to the effect of the reforms, regardless of any changes in the economic outlook that may have taken place at the same time and the effects of reforms decided on previously. The comparison makes it possible to identify households whose living standards rise or fall as a result of the reforms and to describe them according to their position in the scale of living standards or employment status. It also makes it possible to measure the effect of each measure on public finances.

The measures taken into account are the direct changes to the way transfers are calculated, allowing for a shift from “gross” income to disposable income. Indirect taxes are taken into account in some analyses.

Based on the ERFS, INSEE publishes the poverty rate and the main indicators for inequalities in living standards for year N in September of year N+2. In order to more rapidly assess the effectiveness of public policies to combat poverty and inequality, INSEE is using an INES-based method, known as “nowcasting”, to produce the advanced indicators for year N in autumn of N+1 (Figure 3).

Thus, the 2018 poverty rate was estimated in October 2019 (Cornuet and Sicsic, 2019). The exercise is more global than the previous one, it takes into account not only the reforms implemented but also the evolution of the economic situation, the usual revaluations or the effect of earlier reforms. This method is also used by Eurostat for advance publications of inequality indicators or by academics (Ouvrir dans un nouvel ongletGasior and Rastrigina, 2017).

To carry out this exercise, the ERFS of year N is used first to simulate year N, then to simulate year N+1 (legislation and evolution of the different types of income between N and N+1) (Note). The evolution of the indicators between N and N+1 which is published is, therefore, based on the use of a single version of the ERFS. The method thereby gains in robustness, as it makes it possible to overcome sampling errors from one version of the ERFS to the next. This is the fifth year that INSEE has published this advance estimation exercise and the results have always been very close to the final results published one year later (Fontaine et Sicsic, 2015).

As indicated by (Legendre, 2019), one of the important outputs of microsimulation models is the ability to simulate fictitious tax and benefit reforms.

Since 2018, INSEE has been making available, for the first time in France, simulations of fictitious tax and benefit reforms based on the INES model. It simulates the effect of fictitious rises and falls of 1%, 3% and 5% in the main transfers on household living standards per decile, inequality, public finances and the beneficiaries of the transfers (see Fontaine and Sicsic, 2018 for the methodology). This shows, for example, that a 5% increase in the flat rate of the RSA would reduce the poverty rate by 0.1 percentage points, reduce poverty intensity by 0.7 percentage points and cost €850 million in public finances. These simulations are released in a publication and a file presenting all of them (Cornuet and Sicsic, 2019).

The nature of the exercise is different from the annual assessment of the effects of tax and benefit reforms: the effects are calculated on the basis of fictitious measures, defined in a theoretical way, and not on measures that have actually come into force; however, the results of both exercises are consistent.

The INES model is regularly used at the request of government departments or organisations responsible for the evaluation of public policies (Parliament, the High Council for Family, Childhood and Old-Age, the Court of Auditors, the evaluation committee on taxation of capital, the expert group on the minimum wage, etc.), i.e. before deciding on its implementation. This use meets one of the main objectives the INES model has had since its creation, and it continues each year to represent an important part of the work produced by the development teams, particularly at DREES.

Recently, the model has been used a great deal to respond to the social emergency of the gilets jaunes movement. In December 2018, DREES has simulated several avenues of reform that would make it possible to focus the aid announced by the President of the Republic (€100 per month) on the targeted population (employees on the minimum wage). This work contributed to the decision in favour of an increase in the individual bonus of the prime d’activité. The teams were again called upon when the National Assembly evaluated this reform in the summer of 2019.

INES has also been used for almost three years to provide technical insight into the consequences of merging certain benefits and statutory minimum incomes. This project for the merger of means-tested benefits (RUA) is still under investigation at the time of writing.

This work requires a very precise simulation of the transfers that are the subject of the reform and, therefore, contributes to improving the model. They are usually published in the form of notes appended to the public reports.

A great strength of the INES model is the experience accumulated since its creation more than 20 years ago. The model has been used to develop or change many tax and benefit transfers affecting household living standards and has therefore been constantly improved to simulate those transfers as precisely as possible. The uses of the model have diversified over the years and the model is used today, beyond the traditional analysis of redistribution, to provide advance estimates of inequality indicators and to evaluate previous reforms, or shed light on planned reforms and reforms that could take place. These applications are also an opportunity to expand and improve the model.

Bolstered by this experience, the INES model is based on extensive documentation, which is a guarantee of its reliability. One of the assets of INES, which ensures the accuracy of the simulations, is linked to the data used, in particular the very rich information from the Labour Force survey. However, this is also one of its limitations, as the survey sample sizes on which the model is based are small and do not make it possible to provide results for small sub-populations. The new comprehensive government databases that are being made available to official statistics should be a source of improvement in this respect in the long term. These data, particularly the monthly DSN data, could also improve understanding of infra-annual trajectories and incomes. This could make it possible to improve the accuracy of the simulation of certain transfers for which an understanding of the income trajectories is essential, in a context in which the tax and benefit system is evolving to take into account resources in a more “contemporary” (Note) manner. An amendment to the ageing method, which is not very well suited to large-scale economic crises such as the one caused by COVID-19, could also be an area for development of the model. This is enough to predict a long future for the INES model!