AIM-AP. Accurate Income Measurement for the Assessment of Public Policies. Citizens and Governance in a Knowledge-based Society

Transcription

1 Project no: AIM-AP Accurate Income Measurement for the Assessment of Public Policies Specific Targeted Research or Innovation Project Citizens and Governance in a Knowledge-based Society Deliverable 2.16 A user guide to EUROMOD routines for measurement error, tax evasion or targeting errors Due date of deliverable: January 2009 Actual submission date: April 2009 Start date of project: 1 February 2006 Duration: 3 years Lead partner: UESSEX-ISER Revision [draft]

2 This deliverable is made up of 2 EUROMOD recipes, instructing the user in the practical application of EUROMOD to explore the effects of tax evasion and targeting errors (benefit non take-up) respectively. EMRec09_01 A user guide to EUROMOD routines for taking into account tax evasion EMRec09_02 Modelling benefit non take-up with EUROMOD

3 Euromod Recipe: EMRec09_01 A user guide to EUROMOD routines for taking into account tax evasion 1 By Francesco Figari 2, Maria Flevotomou and Manos Matsaganis 3, April 2009 Uses: Estimation of the size of tax evasion Analysis of the characteristics of tax evaders Evaluation of distributional effects of tax evasion Implementation: The underlying assumption is that the respondents of the income survey truthfully and informatively reveal their income to the survey conductors. The difference between the true income collected in the survey and the income reported to the tax authorities is due to tax evasion. In order to take into account tax evasion the following steps must be implemented: - derivation of adjustment factors - incorporation in EUROMOD Derivation of adjustment factors Specific adjustment factors must be included in EUROMOD parameter sheets in order to correct gross income values for under-reporting to tax authorities before simulating tax and benefit instruments. The same factors are used to add into the disposable income the part of income not reported to the tax authorities and hence not included in the simulation of the tax benefit system. The adjustment factors are used to make the income distribution from the survey as close as possible to the income distribution underlying the tax returns data. They can be derived by comparing mean incomes (or aggregate incomes) in the survey data and in the fiscal data. 4 The reference population for which true and reported incomes are compared might be selected according to one of the following possibilities: 1 This recipe draws on Flevotomou and Matsaganis (2009), Implication of tax evasion for taxbenefit models, AIM-AP deliverable. 2 ISER University of Essex. ffigar@essex.ac.uk 3 Athens University of Economics and Business and CERES. maria@aueb.gr, matsaganis@aueb.gr 4 Fiscal data can be in the form of sample of income tax returns or aggregate data. In the former case, check should ensure that the tax returns sample is representative of the population of tax filers, otherwise appropriate weights to the tax return sample must be applied. 1

4 Euromod Recipe: EMRec09_01 - all tax filers - tax filer reporting positive income - tax filers reporting income above the tax-free allowance - tax filers with positive tax liability Demographic information and different income sources may be exploited in order to define sub-groups whose characteristics are informative of tax evader behaviour. Whenever the survey data used as input data for EUROMOD reports only the net values of incomes, the adjustment factors should be considered in the net-to-gross procedure as well, in order to derive a gross value of income that takes into account the tax evasion behaviour of the individuals. Incorporation in EUROMOD The adjustment factors are used to: - correct the gross income variables in order to take into account the tax evasion before running the simulation of the tax-benefit system - include the unreported income in the final disposable income The correction of gross income variables is performed within EUROMOD by including a new policy sheet (called at the beginning of the spine) where the adjustment factors simply multiply the relevant income sources (i.e. this reduces the amount of gross incomes subject to social insurance contributions, income taxes and included in the means of any social assistance schemes). The inclusion of the unreported income in the final disposable income is performed within EUROMOD by including a new policy sheet (called at the end of the spine) where 1 minus the adjustment factors simply multiply the relevant income sources. This amount of unreported income is then included in the final disposable income. EUROMOD flexibility allows for further revisions and updates of the adjustment factors without modifying the structure of the simulation. Warnings: - It is advisable to create a new system (with a new spine and income list set) when tax evasion is taken into account. This allows the user to compare results of the baseline scenario (i.e. without correcting for tax evasion) and the new scenario (i.e. correcting for tax evasion). - A refinement of the above procedure concerns the introduction of stochastic variation in the adjustment factors (i.e. introducing a random noise around the estimated adjustment factors) in order to capture the different behaviour of individuals. 2

5 Euromod Recipe: EMRec09_01 Users and Publications: Flevotomou M. and M. Matsaganis, 2009, Implication of tax evasion for tax-benefit models, AIM-AP deliverable, Mantovani D. and S. Nienadowska, 2008, Tax evasion in Italy, AIM-AP deliverable, Lelkes O. and D. Benedek, 2008, The distributional implications of income tax evasion in Hungary, AIM-AP deliverable, Flevotomou M. and M. Matsaganis, 2008, Estimating tax evasion in Greece, AIM-AP deliverable, 3

6 Euromod Recipe: EMRec09_02 Modelling benefit non take-up with EUROMOD By Manos Matsaganis 1 and Alari Paulus 2, April 2009 These guidelines describe a technical procedure for modelling benefit take-up with EUROMOD. This procedure is flexible in terms of allowing any specification of take-up probabilities (e.g. random model, probit model) and can be used to model simultaneous take-up of multiple benefits at a time. For the clarity reasons, however, the description of the steps involved as follows assumes in the first instance that (a) take-up is modelled for a single benefit only and (b) take-up decision does not vary within a household in case there is more than one benefit unit in the household. Step 1 Produce EUROMOD baseline output, which by default assume full benefit take up. Note that in case there can be more than one benefit unit in the same household, the identifier for the corresponding benefit unit must be added to the output file if the unit differs from household. For instructions on how to define EUROMOD output, consult the EUROMOD Manual EUROMOD Functions (function func_defoutput). Step 2 Set up an alternative scenario in EUROMOD where the benefit in question has been switched off (or set to zero) for everybody. Produce EUROMOD output for that scenario. Again, add the benefit unit identifier to the output if different from household. Step 3 With the help of statistical software package (e.g. Stata, SPSS), set up a model for simulating non take-up behaviour. For example, a simple random model where every benefit unit is assigned a random take-up probability, drawn from a uniform distribution. Alternatively, a more advanced regression type of model could be used (e.g. probit type) where the probability is modelled as dependent on the size of entitlement, other income, claiming costs and/or other personal characteristics. Step 4 Apply the take-up model from Step 3. The estimated take-up rate is usually calibrated to the official estimates or other information from external sources. 1 Athens University of Economics and Business and CERES. matsaganis at aueb.gr 2 ISER University of Essex. apaulus at essex.ac.uk 1

7 Euromod Recipe: EMRec09_02 Step 5 Construct the final dataset with tax-benefit information and disposable incomes by combining information from the two outputs produced in Step 1 and 2. For those units which have been estimated to take up the benefit, use the information from the baseline output (i.e. Step 1). For other units which have been estimated not to take up the benefit, use the information from the alternative scenario (i.e. Step 2). Step 6 Compare the results with the baseline (i.e. under 100% take up) to estimate changes. Step 7 In case the take up model has a stochastic component (e.g. random probability, error term), Steps 4-6 could be repeated a number of times (e.g replications) in order to gain insights into the robustness of the results. Further potential advancements For modelling take-up of more than one instrument at a time, one would have to repeat Steps 2-4 for each combination (e.g. with 2 benefits: full take-up for two benefits, full take-up of benefit 1 and no take up for benefit 2, etc). This will result in more than two output datasets to be used in Step 5. In case there can be more than one benefit unit in the same household and their take-up decisions are modelled independently from each other, one would have to repeat Steps 2-4 for each within household combination (e.g. with 2 units: full take-up for both units, full take-up for the first benefit unit and no take-up for the second benefit unit (and vice versa), no take-up for neither unit). This again will result in more than two output datasets to be used in Step 5. 2