Welcome to TÁRKI homepage!

Microsimulation | Households, Work and Flexibility | Generational accounting | Gender Data Archive | TÁRKI—Andorka Library | TÁRKI Homepage

A Tax-benefit Microsimulation Model for Hungary

Péter Szivós, Tamás Rudas, István György Tóth


Paper prepared for the Workshop on Microsimulation in the New Millennium: Challenges and Innovations (22-23 August 1998, Cambridge)


This study was prepared with support of the PHARE-ACE Programme grant No. P96-6014-R “Household Welfare and Behaviour during Transition in Bulgaria, Hungary and Poland”

Correspondence may be sent to the authors at

Tamás Rudas
Péter Szivós
István György Tóth
H-1118 Budapest, Budaörsi út 45.


Understanding the likely effects of economic and social policy changes in a transition society is difficult. Possible causes for this can be of two different kinds. On the one hand, the social and economic environment is changing very rapidly – and, therefore, designing an appropriate decision making and analytical instrument is essential. On the other hand, lack of appropriate and comprehensive data sets and deficiencies of the government information systems also pose difficulties. Yet the need for ex ante simulations of the effects of welfare reforms is there.

TÁRKI (Social Research Informatics Center – Budapest, Hungary, http://www.tarki.hu) has developed a tax-benefit microsimulation model as an appropriate tool for such analyses. Since no single data set with a sufficiently detailed variable structure was available, the first step in model building had to be the creation of the data set. Three different surveys with individual records were used. The income and demographic variables from the Hungarian Household Panel Survey, consumption variables from the Household Budget Survey of the Hungarian Central Statistical Office and tax variables from actual administrative tax records. A dynamic multiple matching and multiple imputation procedure was developed to virtually combine these sets of data. Based on this, the microsimulation models and analyses were implemented for three different areas such as income tax, indirect taxes (VAT and excise taxes) and centrally regulated cash social benefits. The static model allows the analyst to redesign the rules of eligibility or liability, and produce outputs showing the gains or losses due to policy changes. The main feature of microsimulation models in relation to social and economic policy analysis is that they produce results that can be analysed at the individual level. Thus, the distribution impact of a policy measure on different types of families or income levels etc., can be assessed. At the same time, estimates of the aggregate outcomes can be easily derived by combining the individual results.

The paper explains the database and model building. Three applications are also presented. The first models the effects of three different income tax schedules for 1997. The second presents computations about a possible shift from personal income tax to a splitting system. Finally, a third application models likely effects of the introduction of a new cash benefit.

Introduction: the history of microsimulation in Hungary

Microsimulation procedures are designed to examine social and economic changes by assessing the effects of various policy changes on small units – in our case, households – and the description of macro-level effects is achieved through aggregation of micro-level changes. The relevance of the results for the entire society depends largely on the nature and quality of the survey data which are used as the basis of the microsimulation study. survey: Naturally, the possible scope of various social and economic developments to be modelled this way depends on the range of information available on the units of the database. A rich enough dataset should, on top of the generally used social and demographic variables, contain detailed household and individual level additional information on five areas: income, consumption, labour market status, taxation and take up of various social transfers.

Developing microsimulation databases in Hungary has been attempted by two different institutions, the Central Statistical Office (CSO) on the one hand and TÁRKI (Social Research Informatics Center) on the other.

The CSO started to set up a microsimulation system in 1986. The need for this type of technique was mainly driven by the public sector reform attempts of the time. An increasing importance of financial transactions could then be experienced, and a growth of differences within the society was recognised. The transformation of government institutions and the tendency towards more professionalization within and outside government, the increasing importance of professional approaches brought the need for impact analyses into the forefront.

At first, a dynamic model was planned. However, these plans soon proved to be overly ambitious and static, rather than dynamic, microsimulation applications were developed at the end of the 1980's.  The simulation system had a modular structure. The first applications used mainly the income and tax modules of the system. (Szivós,1993)

Shortly after the introduction of the personal income tax system in Hungary in 1988, an extended debate started on the type of income tax system (should it be family-based, personal, or should it contain splitting elements). The first use of the microsimulation technique was a test of these design issues. CSO prepared a paper for the Ministry of Finance analysing the effect of 13 different tax- and income distribution variants.

There were also two other analyses prepared at that time. Attempts to simulate developments in income inequalities and living standards based on Household Budget Survey and income survey data were made jointly by the CSO and the Economic Research Institute. The simulations were carried out using a system of hypotheses based on macroeconomic forecasts to analyse the overall and detailed effects of the simulated measures. These investigations were carried out for the year of 1991 and 1992 separately. (Szabó-Szivós, 1992)

The microsimulation database for the years 1987-1992 served as a source of information for income distribution statistics. This was the basic information for re-weighting the Household Budget Survey income data, for the Incidence Study for 1990 to investigate the effects of social transfers in kind and consumer subsidies, for the estimation of household consumption in national accounts and for the weighting system of Consumer Price Index.

In summary, it can be said that the potential uses of this instrument between 1989 and 1992 were as follows:


  • improving quality of statistical surveys (by re-weighting)
  • updating statistical surveys
  • making forecasts
  • investigating effects of measures.

All of these tasks used the microsimulation technique as a tool. However, contrary to experiences in other countries, mostly descriptive, rather than evaluative (impact analysis type) uses were dominant in that period.

The first generation of the TÁRKI microsimulation model was developed in 1995 jointly with the Ministry of Finance. The aim was to build a dataset containing income, tax and consumption information and producing mostly tax simulations. The second generation of the model, built in 1997 and containing data for 1996 is a tax/benefit model and is applicable for indirect taxes and cash transfers analyses, as well. It was an explicit aim to build a user friendly model and, with that, make the modelling activity accessible for non-professional clients.

Data sources

The range of possible social and economic changes that can be modelled is defined by the information available on the microsimulation units in the database. In addition to the generally used social and demographic variables, the microsimulation database contains detailed information on households related to five areas: income, consumption, taxation, participation in various social transfers and job market position.

When starting the model building, certain immediate trade-offs emerge. One is between detail and reliability of data. No survey is available that examines the above five areas in the required detail from all aspects. The execution of a survey like that is not only very costly and time consuming, but would not be reasonable due to reliability aspects, either. For these reasons we have created the microsimulation database by merging various data sets. The following datasets were used are: the Hungarian Household Panel of TÁRKI, the Household Budget Survey of CSO and actual (anonymous) tax returns from the APEH (Inland Revenue Service) database. These are datasets from different samples. Accordingly, the most difficult technical task when creating the database was merging the different data sets.

The virtual merging of the datasets was based on a procedure, where attributes were assigned to the actual households of one set of data, that were not available in the given data set, but were available on many other households of another dataset. The matching can be supposed to work well if the attributes assigned to the given household come from a household of the other data set that is similar to the one we are assigning to in the most significant aspects. Since the assigned data are not actually known on the given household, these have to be values that are not constant and show some distribution. We have selected the multiple imputation method, based on dynamic matching, for this task from among the many available statistical procedures.

Multiple imputation is a procedure generally used for managing the problem of missing data, while the missing data are substituted by not a single value, but by several values (Little, Rubin, 1986). The distribution of these should be as close as possible to the distribution from which the missing data would come. In practice, multiple copies of the dataset (hence the term “multiple”) to be complemented, that is the one containing missing data are created, and the missing data are replaced by different values in each set (hence the term “imputation”). The analysis to be performed is executed on all datasets created in this way, and the average of the results will be the result of the analysis based on multiple imputation.

The basic idea behind multiple imputation is used in the creation of the microsimulation database so that the data of households existing in one data set (TÁRKI Hungarian Household Panel) that already exist in other sets (such as consumption data in the CSO Household Budget Survey) are considered as missing data and are replaced from these sets. For households providing supplementary data to be as much like the original households as possible, we defined categories based on social and demographic data existing in both surveys, based on which we could impute the missing attributes of the original household from the data of the households suitable from the most aspects. This procedure is called matching.

The application of multiple imputation is optimal from the point of view of the errors created during matching, but results in a complex data structure that requires special programs to manage it, and these were also developed.

Below,a detailed description of the data sources applied and the matching and imputation procedures is given, the theoretical background, however, is not discussed.

The 1997 set of the Hungarian Household Panel data formed the basis of creating the microsimulation database. This contains the data of 1392 households. The dataset was collected using individual questionnaires for all members of the households in the sample. The sample for the first wave of the Hungarian Household Panel in 1992 consisted of adult inhabitants of 2600 housing units and the households they formed, selected from the housing unit data of the 1990 census using multi-stage random selection procedure. The sample became distorted every year due to the panel member dropouts, and therefore the 1997 Household Panel can not be considered as a random sample all Hungarian households. In order to obtain unbiased estimates, each household was to be weighted each year starting with the second wave. When creating the microsimulation database we have adapted the weights that are already in use in the Hungarian Household Panel. Note, that the proper selection of weights half time between two censuses is not evident, since the deviation from census data may be due to either the sample being distorted (in this case weighting is justified) or changes in the demographic situation (in this case weighting is not justified).

The CSO Household Budget Survey dataset contains very detailed consumption data on approximately 10,582 households. The surveys were created using diaries. The data collection lasted for a year (1995), using 1/12 of the sample each month. Therefore, the effective sample size is essentially 10,500/12, that is approximately 880 households. One of the advantages of self-filled diaries as a data collection method compared to other possible methods of data collection is that more detailed information may be gathered, but the drawback is that the data may be less reliable.

The original survey contained approximately two hundred consumption items, that in the most cases provided information on sums spent by the household for different goods and services in the month in question, while some of them were annual data. According to the imaginable goals of microsimulation modelling such detail of data is not required (e. g. handling different fruit types individually is not justified), so we used variables that were appropriately amalgamated.

APEH (Inland Revenue Service) provided two samples of the personal tax returns received. The first one is a .45 per cent random sample of tax returns prepared by employers for their employees containing 10,425 cases. The second one is a 1 per cent random sample of the full range of those submitting personal income tax returns, that is tax payers who had had incomes other than wages or salaries, containing the data of 19,398 individuals.

The microsimulation database consists of two separate data sets. One is the household dataset that contains core data and imputed data on 1,392 families of the 1997 TÁRKI Hungarian Household Panel (HHP) research. The other is the personal dataset that contains core and imputed data on 3,084 individuals questioned by the HHP.

The personal data file also contains data on the household (from HHP and CSO data) from the household database so that these variables are identical for those belonging to the same household. For example, consider a household of four. The personal dataset contains information regarding the ownership of consumer electronics equipment of the household. This will be identical for all four members of the household. The personal dataset contains most variables of the household set in this structure. It also contains personal data originating from the personal questionnaire of the HHP, and the (imputed) data of tax returns.

Each piece of the (imputed) income and expense data originating from the CSO survey refers to a household. These are also contained in the personal and household sets with identical names and identical values. The (imputed) data originating from the tax returns are only contained in the personal set. A part of the HHP income data originate from questioning the individuals; these variables can be found in the personal set. The sum of the same variables can be found in the household set. The household set also contains certain personal data on the head of the household.

The multiple imputation procedure of data not recorded in the panel and the basic principles for the matching required for imputation are identical in the case of imputing data from the HBS or from the tax returns. The only difference is in the selection of variables used for matching.

In the course of multiple imputation, ten households from the CSO sample are assigned to each household in the panel, and it is assumed that the consumption data of the actual household that are not included in the panel, but are included in the CSO records, are of a distribution that is like the distribution of the consumption data of the ten selected households. Whether this assumption is correct depends on if there are any characteristics being identical in the case of the actual household from the panel and the ten selected households from the HBS that result in their consumption being identical or similar; and if such variables exist, can there households be found in the CSO sample for each panel household that are sufficiently similar according to these variables. The reliability of this matching procedure therefore requires for as many variables to be considered as possible; the execution, however, requires that the matching is performed based only on variables that exist in both data sets, and that using these variables each household is only specified in the panel to a level so that several households of the same specification exist in the CSO sample.

We tried to comply with these contradictory requirements by applying a dynamic matching algorithm (that matches households belonging to a larger group based on several variables and households belonging to a smaller group using fewer variables). We used the following variables:

A: number of people in the household

B: number of supported children (highest category = “5 or more children”)

C: what decile the household belongs to based on its income per capita

D: education of the head of household (three categories)

E: settlement type (three categories)

If the number of CSO households corresponding to a panel household according to the five variables above is at least ten, then ten of these were randomly selected and their data imputed to the panel household. This could be performed for 903 households. If the above condition was not satisfied, then only the households identical according to the first four variables were considered, and the data of ten of these, selected randomly, were imputed (for 243 households). In the case of households for which even this was not possible, only the first three variables were used for matching (88 households); where even this was not possible, only the first two variables were used (19 households). There were 21 among the total of 1,392 panel households that we did not find enough households for in the CSO sample. In the case of these families, imputing was carried out based only on the household size.

As the result of the above procedure, ten data files with identical structure but different data content were obtained. The 1,392 households from the Panel are the observational units. Each dataset contains the variables of both the Panel and the HBS of CSO. The values of the panel variables are identical in the ten copies of the dataset, but the values of the CSO survey differ (since they are based on ten different actual families). We imputed the HBS of CSO household data for 3,804 individuals of the Panel in a similar manner to the above.

The matching of APEH data to Panel data was performed similarly. The individuals in the Panel records were divided into three groups: those not paying income tax (no APEH data were assigned to them), independent taxpayers (having income other than wages or salaries) and employed taxpayers (having only salary or wage income). In the case of independent taxpayers the variables forming the basis of selection were the following:

A: Are there any pensions, child care allowance, or scholarships (1: No pension, child care allowance, or scholarship, 2 = on pension, 3 = on child care allowance, 4 = on scholarship, 5 = on child care allowance and belongs to deciles 8, 9 or 10 according to net income)

B: Net income decile ( 0: No income)

C: Net income decile from enterprises ( 0: No income from enterprises)

D: Income from royalties, quartile (0: No income from royalties)

E: Small scale agricultural production income, quartile ( 0: No agricultural income)

In total, we imputed self-filled tax forms to 606 individuals. Matching was implemented according to 5 variables in 484 cases; based on four, three, two and one variables in the case of 50, 49, 13 and 10 individuals respectively.

In the case of workplace tax returns the variables forming the basis of selection are the following:

A: Net income decile (0: No income)

B: Other incomes, decile ( 0: No other income)

C: Pension, child care allowance, scholarship decile (0: No pension, child care allowance or scholarship)

D: Age groups (5 categories)

E: Royalty income from employer below or above the median? (0: No royalty income)

In total, 792 individuals answered the question “Who files your tax return?” by naming their employers; 7 of them lacked all income data, no APEH data was imputed to them; from among the others we matched APEH data to 288 individuals according to five aspects, to 6 according to 4; to 354 according to 3, to 80 according to two aspects, and finally to 64 individuals based only on the net income decile.

Only 7,136 of the 10,582 CSO households were included in the imputed set; 3,413 households were featured once, 2,043 twice, 950 three times, the data of 381 were used in four places, the data of 166 in 5 places, and 105, 49, 21, 5, 2 and 1 were used in 5, 6, 7, 8, 9, 10 and 11 places respectively. The data of 5,140 of the 22,867 self filled tax forms were included in the imputed set. 4 of these were used five times, 20 four times, 91 three times, 662 twice and 4,363 forms were only used once. The data of 3,771 of the tax forms filled out at the workplace were included in the microsimulation data set. One form was used 13 times, five 12 times, nine eleven times, nine ten times, 19 nine times, 45 eight times, 61 seven times, 119 six times, 141 five times, 217 four times, 297 three times, the data of 604 forms were used in two places and 2,244 were only used once.

The final collection of microsimulation datasets contains ten household sets and ten individual sets, with some variables appearing in both, as described above. The observations in the household sets are the actual households observed in the panel, the observations in the individual sets are the actual persons observed in the Panel. Variables in both types of datasets include actually observed Panel variables and imputed – virtual – variables from the HBS and the tax returns (where applicable). The actually observed data are the same in all ten copies, the imputed data are different in the ten copies.

Various consistency checks between actually observed and imputed data were carried out. These will not be reported here.

Features of the model

Main objective of establishing the TÁRSZIM97 tax-benefit model is to provide an appropriate tool for the analysis of micro-economic developments and social policy in order to support the decision making process. The principal goal is to serve this objective in a user-friendly way. Another basic requirement of the model is the high degree of flexibility which means that the adjustment of parameters should be made possible in an interactive way. The model is static, therefore policy issues with mainly short-term impacts can be covered. Individual behaviour is also assumed to remain stable in the short time periods under consideration. The model aims to analyse the effects of possible tax and benefit regimes and to quantify the differences between regimes by parametrising the ‘new’ tax or social policy regulation and producing output tables for the effects. (Kende et.al., 1997)

These aims had to be taken into accounts in model realisation by selecting proper software tool. Excel has been selected as the software tool for modelling on the microsimulation database. This is a generally accepted and widely used program that displays data in the form of tables, its query language is simple and provides a wide range of easy-to-use tools, it is suitable for the implementation of all microsimulation analyses, and the required hardware tools are also widespread and not too expensive. Its spreadsheet structure offers a convenient way for user input (parameters and options) and its facility to lock and hide cells gives a possibility to avoid users changing part of the model. However, Excel is not capable of managing data sets that were created by multiple imputation procedures, so in addition to creating the microsimulation database we created special procedures within the Excel spreadsheet program that proved to be suitable for the query of the data sets and are just as user friendly as the original functions of Excel are. There is a serious limitation that Excel has a capacity for an individual spreadsheet of 16384 rows and 256 columns. Especially this second condition is far from being generous. However, balance of strengths and weaknesses resulted to choose this package. Also, a special software application was developed to carry out statistical analyses for modelling the multiply imputed datasets. These analyses can be parameterised by filling in dialog boxes, as it is usual in a Windows environment.

The model has three main components:


  • the dataset
  • the core model (special relations among variables)
  • the output system.

In the previous section of this paper, we provided detailed description of the dataset. The core model of TÁRSZIM97 covers three areas:


  • direct taxes
  • indirect taxes (VAT, excise duties)
  • cash benefits.

Direct income tax parameters in the model reflect the real world tax system of Hungary. However, the model also offers several other optional possibilities to create alternative scenarios. The basic characteristics are as follows.


  • Taxpayers can be individuals, couples or households.
  • Taxable income is as detailed as in personal income tax returns and detailed household income is included in HHP (a couple of dozens of income categories).
  • Eligibility criteria to any kind of tax concessions could be all kinds of demographic and income classification criteria the dataset could otherwise provide.
  • The criteria of tax exemptions could also be the number of eligible persons according to various criteria in the household.
  • Various forms of tax exemptions and deductions may be set in the model:
    1. in the case of personal tax: minimum amount of tax exemption, deduction from tax base in fixed amount, or proportional deduction from tax to be paid in fixed amount or proportionally,
    2. in the case of splitting: different tax schedules for individual taxpayers and
    3. income splitting between couples may be set
    4. weighting regime for children can be applied.

The definition of a particular analysis can be handled in two ways:

First, the Tax System Wizard, which leads the user automatically, in the proper order, and extensively through all the required steps for acquiring the desired results. The required steps for setting all parameters are:


  1. Selection of basic taxation model
  2. Scaling the income level
  3. Adjustment of the unique taxation parameters of individual incomes
  4. Adjustment of the taxation parameters of deductions not related to income
  5. Adjustment of the appropriate parameters for the tax schedule or the tax schedule table
  6. Performing the output specification (microsimulation calculations)

Second, selecting topics and adjusting parameters individually. This method gives more flexibility and freedom for the users.

The model offers three basic taxation models for simulating income tax, or to be more accurate, these are the bases for the tax system to be simulated.


I. Personal income tax model
This case is the model of the existing PIT system. Incomes are considered for each individual regardless of the other members of the family (or household). Thus in model I, people are paying their income taxes individually as the current practice is in Hungary.

II. Simple family income tax system
Tax model based on the total household income and applies a single tax schedule. In this case, it is the household which is “producing” the income and paying the taxes for it.

III. Complex family tax system, multiple tax schedules
This is a taxation model based on the total income of the household and the number of children living in the household and uses different tax schedules.

Three kinds of tax schedules or tax schedule systems are assigned to the three basic taxation models. The one which is used automatically by the program depends on the basic taxation model selected. The settings of income limits of tax categories and their (tax) rates can be modified in the tax schedules. Furthermore, items may be added to or deleted from the tax categories.

The income brackets and the rates of the categories thus created may be set by changing the parameters in the personal income tax schedule. (PIT tax schedules menu Parameters menu item). A new bracket can be added and deleted using the Add new tax category and Delete tax category menu items. A unique function is the option to use the wages tax schedule (in 1996 there has been two different schedules for wages and other incomes in Hungary). There are two ways to use the wage tax schedule. One is to use and modify the built-in 1996 PIT tax model. The other way is to change the “Wage Deduction” from Deductions to a specially calculated deduction

The operation and the specification of parameters in the simple family tax schedule is similar to that of the PIT tax schedule. Its usage is different in that the income of the household under total summarisation forms the basis of the tax. Naturally, the options of setting this simple tax schedule come up under the Parameters item of the FIT tax schedules if the basic tax model is model II.

Complex family tax schedule forms the basis of a general family taxation system. All elements of the tax schedule table are tax schedules that behave similarly to the simple family tax schedule. These operate in the following manner. Based on the total income of the household representing the given family and on the number of children, it falls into one of the items of the table consisting of tax schedules. Income limits can be set by the user. This way, each family (household) calculates its taxes from the table using the tax schedule defined by these two properties. The category limit and rate parameters related to any tax schedule element of the table may be set.

The (taxation) attributes of income types may be set separately in the dialog box opened using the Incomes… item in the TaxModel menu. Income types and possible taxation attributes are the following:


Wage-like income from main activity
Costreimbursement from work
Income received for foreign service
Income form governmental indemnification
Income from indemnification received prior to ‘96
Other wage like income
Insurance premium paid by employer
Other non-independent income
Other cost reimbursement
Income from intellectual activity
Entrepreneurial income
Income from small scale agricultural production applying tax exemption
Income from small scale agricultural production applying cost deduction
Rent from land
Other income from private activities
Child care allowance
Income from providing social care
Other income of tax exemption
Income from selling compensation land
Sum deducted due to investments in ‘93-‘94
Sum deducted due to cooperate share purchase in ‘94
Daily allowance in abroad
Other income
Income from abroad
Allowance tax income
Income received from real estate sales
Income received from personal property assets sales
Rent from agricultural land
Income from exchange rate profits
Income from dividends (20% resource tax)
Income from dividends (10% resource tax)
Income from dividends abroad
Assets withdrawn from production
Income form selling share received in kind compensation
Income from privatisation leasing
Other resource tax income

Deductions are items reducing the tax or the basis of assessment that cannot be connected to income and can be used when planning the tax system. Possible deduction types and their possible attributes are the following:


Agricultural land purchase deduction
Tax paid abroad deduction
Tax allowance (for wage incomes)
Deduction for being severely handicapped
Contribution to mutual pension fund
Contribution to mutual health fund
Repayment of housing loan
Tuition fee
Small scale agricultural producer deduction
Insurance premiums paid by the employer in the years ‘90-‘91
Savings for housing
Paid interest of a business loan
Housing acquisition deduction
Employee’s federation membership fee
Unemployment benefit contribution
Social security contribution
Local tax
Donation for domestic fund and community obligations
Investment tax loan (for buying shares and bonds)
Tax refund in low income brackets

The characters of possible tax deductions can be defined in the following manner:

Income (tax basis) decreasing item in percentages
Income (tax basis) decreasing item, constant
Tax decreasing item in percentages
Tax decreasing item, constant
Tax decreasing item, specially calculated

Indirect taxes (VAT and excise duty) can be simulated by prefixed commodity groups changing the tax rate and the ‘behavioural’ option. The calculation of possible price/consumption elasticity can be set in three ways. In the first case it is the gross value of the consumption of the given commodity group is what is fixed by default. If the tax rate in this case is changed, the net value changes in the opposite direction. In the second case, the net value is fixed therefore we assume that the gross value changes parallel with tax rate. In the third case the price elasticity parameter is being defined. This way we may also model that the prices changing due to the tax also change the amount of consumption depending on the elasticity. The calculation of the new consumption in this case is carried out according to the following formula:



Vu, Vr: the new ( = simulated) and olds value of the purchase

e: price elasticity to be specified (elasticity parameter)

Xu, Xv: the old and the new tax rates

Changes in the parameters of the excise tax category are implemented identically in all aspects to when using the VAT model.

The benefit model uses two kinds of modelling methods. In one case, benefit indexing can be set by applying some multipliers only. The second type is more realistic in the sense that the parameters of existing or new (created) benefits – eligibility conditions, levels, the extent of benefits and other major attributes – can be changed, and the effect of this may be examined later on. The property of each benefit can be set using the items in the BenefitModel menu.

The first type includes benefits where historical attributes have definitive significance. These include pensions and unemployment benefits. The variables of the Hungarian Household Panel reflect three types of pensions and three types of unemployment benefits help the user to start off here.


  • Old-age pension on the person’s own right
  • Disability pension
  • Other pension
  • Unemployment benefit
  • Start of career unemployment benefit
  • Retraining benefit

In the case of this method, the specification of parameters is provided along the lines of several dimensions, since for example the age, number of children, present value of the given benefit, employment type variables may differentiate the coefficients.

The following benefits are determined by using the actual microsimulation method of the benefit model:


  • Income support for long-term unemployed
  • Maternity allowance
  • Family allowance
  • Child care allowance
  • Child protection benefit
  • Optional, new benefit type (created by the user)

The definitions of the benefits above require the setting of eligibility conditions of varying complexities. The most important benchmark used in the current Hungarian eligibility system for each benefit is the minimum value of the old-age pension. This value may be adjusted in the model. However, other reference values may also be determined as the bases for eligibility level, either besides or instead of the minimal pension. This is called benchmark level.

The types of benefits available for modelling are the following:

The Income support for long-term unemployed was assessed in the HHP, but can be simulated using the available data set. By default


  • it is received by unemployed individuals who have no unemployment benefits and the net per capita income in the household does not exceed 80 per cent of the minimal old-age pension,
  • and the received sum complements the individual’s income to 80 per cent of the minimal pension but is at least HUF 1,000 per month.

Maternity allowance is also excluded from the assessed database but can be modelled in a simple manner. The default is a lump sum payment when a child is born, which is 1.5 times the minimal pension.

Family allowance is the benefit that has the most complex eligibility criteria among the modelled benefits. The current family allowance scheme is a selection of groups using a series of filters and the forwarding of sums different for each group to one adult member of the family. The assessment of those eligible for family allowance thus is carried out in several stages.


  • Potential beneficiaries are in families that have children under age 20.
  • If there are any, families must be separated to ones that have single parents raising the child(ren) and to ones that do not.
  • The third filter differentiates between households according to the value of their net per capita income (or any sort of equivalent incomes).
  • In the fourth filter, the families may be classified according to the number of children.
  • Finally the sum of allowance is assigned in the above mentioned categories.

Categories of income as well as particular sums to be forwarded may also be set.

Those who are eligible for family allowance are also eligible for child care allowance – thus setting their eligibility parameters can only be modified at this point in an indirect manner. Besides this condition this allowance is also received by those that have children under age 3. The parameters of the received sum, however, can be set in relation to one of the benchmark levels. The default is that this is the same as the minimal pension. (The disabled and those receiving benefits for some exceptional reasons can not be modelled based on the available data set. Similarly, the benefit for permanently sick children can not be fitted into the model because of the lack of information.)

Child protection benefit (introduced in 1997) replaces the earlier regular child bearing assistance. The default procedure is as follows: should the income per capita in the household be under the minimal pension, 20 per cent of the minimal pension per children may be provided for the eligible. The values of the eligibility level and the sum to be forwarded may also be changed in the model.

One of the greatest strengths of the benefit model is that new benefit types may be created that can be built into the microsimulation model.

The value of the new benefit type can be fine tuned on a wide range. Those eligible may receive fix sum, lump sum payments, amounts defined as percent of certain benchmarks or benefits may supplement the family incomes until a certain level. or even with some ratio of the difference. All these benefit values may be deleted in the case of those above the eligibility level or we may forward to those slightly above the limit using parameter specified reductions.

Income thresholds are defined flexibly: the basic eligibility level may depend on any benchmark level, furthermore it can be compared to either the individual’s income or any of the normalised incomes in the household.


The applications described in this section are classical examples for the use of tax-benefit microsimulation models. Two of them are to quantify socio-economic impacts of hypothetical income tax measures, while the third one is to ex ante investigate the distributional effects of a newly designed cash benefit. For each case, a wide range outputs was produced to demonstrate the impact of different hypothetical versions in several socio-economic dimensions, like age, family size, type of settlement, labour market status etc.

Application 1.: Different tax schedules in PIT

The main objective of this application was to compare three different scenarios of the Hungarian Personal Income Tax regime for 1997. (Miski et al., 1997) All three scenarios proposed modifications of different parameters of the tax system and neither of them proposed any change of the system as a whole.

The first scenario was prepared by the Ministry of Finance (case 1). The main features of this proposal were: widening the tax brackets, slightly reducing tax rates in general and simplifying the difficult, complicated subsystem of the tax regulation which was in force that time.

The second scenario, which was proposed by the majority governing party (case 2). This proposal aimed to reduce marginal rates for middle income groups and increase high marginal rates, shifting the tax revenues this way closer to the high income groups of the society.

The third scenario was a government proposal (case 3) to reduce the highest marginal tax rates and increase the middle marginal rates. Therefore the three scenarios were different concerning tax rates, brackets and the level of tax exemption of wages and salaries. There was no difference in any other element of the tax regime like tax deduction, exemption etc.

There was one assumption regarding relative income positions of the different social groups. According to this, these positions remained stable which means that the surveyed structure would be the same in the future. It is obviously a strong assumption which might be not fit to the reality, however in short term these changes on the relative income positions would not be significant.

It should be noted that the coverage of the analyses is limited due to the fact that most social benefits, like pension, are not subject of taxation, therefore most of pensioners are excluded from this investigation.

The output of the model calculations could be grouped into macro and micro effect clusters. The first cluster includes information like total revenue and average tax-rate, while the second one focuses on the socio-economic groups and provides information about tax-rate and other ratios identifying ‘winners and losers’. The main results of investigation were as follows:

One of the main outcomes of the exercise was that total tax revenue and average tax rate were very similar in all cases. Looking at the tax burden trend line, the average tax-rate slightly decreased in all three cases. The average tax-rate was within 0.5 per cent range, the lowest was in case 3, the highest was in case 1, all of them within 13-14 per cent. The total tax revenue was the highest in case 1 resulting 2.5 per cent more income for the government than in case 1. In this respect, the gap between case 1 and 2 was only 0.3 per cent.

The principal aim of a microsimulation model, however, is not to estimate macro figures (there are other tools for that purpose), microsimulation is more suitable for socio-economic impact analyses. In our case, these micro aspects of the results were more appealing. In case 1, the tax rates in low and middle income groups were lower than in case 3 but at high income levels the tax rate is above of that. In case 2, the shape of the difference is similar to case 1 but the cut off point is higher and the magnitude of differences are bigger e.g., the difference of the tax rate between cases 2 and 3 is more than 4 per cent at the highest end of the income distribution.

Even more closer to the micro results are socio-economic differences. The results in the output table shown below compare cases 2 and 3 and the winner-loser approach is used.


Comparison of case 2 to case 3
paying more tax (losers) paying less tax (winners)
occupational status own account workers, low level supervisors, middle and high level managers, intellectuals agricultural workers, skilled and unskilled workers, clerical employees
labour market status employees, entrepreneurs pensioners, inactives, unemployed
type of settlement capital villages, towns
age above 40 under 20, 20-39
educational attainment higher educational degree primary, secondary school
gender male female

Note: groups printed in boldface had a margin over 3 per cent.

Detailed comparison of three cases in age and labour market status dimension can be found in Charts 1 and 2.

Chart 1: Difference of tax rates relative to Case 3 by age groups

Chart 2: Difference of tax rates relative to Case 3 by labour market status

Application 2.: Personal versus family income tax

Since the introduction of the Personal Income Tax system there has been a continuos discussion regarding the appropriate type of taxation, whether the personal or family tax is fair to the society. From a welfare point of view the trade-off between cash benefits and tax exemptions is crucial. The Hungarian tax-system used to allow tax-deduction for dependent children but in the middle 1990s these deductions were terminated. Currently, the discussion is even more heated and therefore TÁRKI contributed to this discussion the microsimulation results. (Mészáros, 1996) The main idea behind these hypothetical versions was to take into account the number of dependent children when the base of tax is calculated. These theoretical versions were not a proper proposal for implementation, the major objective was to quantify the impact of a relatively simple system.

In the first version, a per capita measure was introduced: household total income was divided by the number of individuals and then each individual – adults and children - is subject of taxation. It was a simple but significant step to depart from a system (PIT) where only income recipients, mainly earners, are subject of taxation. Second and third versions are fine tuning of the first one as taking into account the obvious assumption that children consume less than adults. In the second version the economy of scale for assumed household needs is flat, each children are weighted by 0.85, while in the third one, the economy of scale is degressive, first child is 0.95, second child is 0.85 and others are 0.65. These three possible versions are multiplied by two, based on the availability of tax-deductions. All other parameters of taxation like income brackets and rates were identical. This was the major assumption to consider this as a theoretical exercise.

In this application, the unit of analysis, contrary to the previous one, was the household with exclusive emphasis on the number of children. In the first case, when only the effect of income splitting per capita is included, the total tax revenue is higher than in the current system. The results indicated that the current tax exemptions have greater effect on the total sum of tax revenues than the per capita tax-paying regime. In the second, case when per capita tax-paying and tax exemptions parallel were taken into account, the total tax revenue was considerably lower than in the current system, because splitting of income among household members resulted in a lower amount of tax to be paid. Chart 3 indicates that average tax paid by a household is twice as much without exemptions, than with them, in all three versions. The same difference can be seen in Chart 4 among households with different numbers of children.

This exercise demonstrates the demand for a better design, a real alternative system, which equally serves the aim of social justice and the fiscal needs of state. That system would be somewhere between the extremes shown by this applications. The model would be able to meet the requirements of such a governmental project.

Chart 3: Average tax per household, HUF

Chart 4: Average household tax rate by number of children, per cent

Application 3.: Introducing a new social support scheme: the case of Child Protection Benefit

For the request of Ministry of Welfare, TÁRKI was involved in the design of a new transfer, a cash benefit for low income households with children. (Miski, 1997) During the design phase, several alternatives were tested, both from the point of eligibility and the amount given. The basic properties of this benefit were as follows: a family is entitled for this benefit if their per capita income is less than a certain pre-specified threshold. The amount to be received was planned to be regular and fixed. The tested thresholds were 75 and 100 per cent of minimum old-age pension (a widely used ‘poverty’ line in Hungary), and 10, 15, 20 per cent of minimum pension as transfer amount per child.

>There were two assumptions in the model: the first one was regarding the definition of child, we considered all persons under 18 years as children, secondly, we assumed a 100 per cent take up rate. Take up would be a relevant issue indeed, but in Hungary we do not have enough experience so far on that field, due to the fairly recent shift from general benefits to means-tested ones.

Finally, the 100 per cent cut off line, and 20 per cent benefit level were defined. Here, as an example, results of this final version are shown. Applying these conditions, a cost estimate was provided, saying that the total expenditure would be around 2 per cent of cash transfers (including pensions, family allowance etc.) For the coverage of this program, we predict one-third of households with children would be eligible for this benefit. This benefit on the one hand affects mainly households in villages and towns (Chart 5), households with lower educational attainment and households in the first two deciles of income distribution. On the other hand, among beneficiaries, households in Budapest, and households with higher educational attainment receive higher amounts (Chart 6), due to these recipients are seems to be poorer. Overall, these characteristics of the beneficiaries prove that the benefit a relatively well targeted one.

Chart 5: Share of households receiving child protection benefit by type of settlement

Chart 6: Monthly average of child protection benefit by education attainment of the household head, HUF


Kende, Gábor, Miski Zoltán, Rudas Tamás, Tóth István György (1997): TÁRSZIM97 mikroszimulációs modell: dokumentáció Documentation of TÁRSZIM97 microsimulation model, TÁRKI, Budapest

Little, R. J., Rubin, D. B. (1987): Statistical Analysis with Missing Data. Wiley.

Mészáros, József (1996): A családi típusú adózás bevezetésének lehetőségei Feasibility of introduction of family taxation, TÁRKI, Budapest

Miski, Zoltán, Rudas Tamás, Tóth István György (1996): A személy jövedelemadó 1997-es kulcsaira tett javaslatok adóteher megoszlási hatásai, Tax distribution of proposed measures for changing personal income tax system, TÁRKI, Budapest

Miski, Zoltán (1997): Becslés a gyermekvédelmi támogatás hatásairól Cost estimations for child protection benefit, TÁRKI, Budapest

Szabó, Katalin, Szivós Péter (1992): A lakossági jövedelem-változás, réteghatások, szóródás 1992-ben, Changes of household income and its distribution, GKI, Budapest

Szivós, Péter (1993): Experiences of microsimulation applications Paper presented at IARIW Spesial Conference on Microsimulation and Public Policy

Gender Data Archive | Microsimulation | Flexibility | Generational Accounting

© TÁRKI – Last modified: August 21., 2003