Τίτλος Microsimulation for the evaluation of public policy: a tax-benefit model for the Greek economy
Δημιουργός Flevotomou, Maria
Συντελεστής Athens University of Economics and Business, Department of International and European Economic Studies
Matsaganis, Manos
Τύπος Text
Φυσική περιγραφή 391p.
Γλώσσα en
Περίληψη This doctoral thesis develops a tax-benefit model, i.e. a tool for the evaluation of public policy in Greece through microsimulation. Public policy affects incomes, consumption and consequently individual welfare through many routes. Not surprisingly, a large body of economic research has been dedicated to developing methods analyzing its effects. Contributions to the literature fall under two broad strands. Generally speaking, the first strand builds upon the representative agent approach, and is grounded upon the assumption that one may abstract from individual characteristics and preferences and focus on some “average type” to study the impact of policy change. The unit of analysis could therefore well be the average taxpayer, the average pensioner, the typical family etc. Further simplifying assumptions are necessarily made under this approach as such models typically cannot encompass the full details and complex interactions of various policies. For example, the tax system is usually represented in terms of an average tax rate summarizing the overall effect of its various components such as tax allowances, the tax schedule or tax credits. The other, and more recent strand, builds upon the microsimulation approach, which makes use of richer data – at the individual level -on the circumstances of a sample of households that is representative of the population. The application of the microsimulation approach in the economic analysis of public policy makes use of tax-benefit models. Models of this sort study the impact on net incomes of the current or alternative tax-benefit systems by calculating benefit entitlements and tax liabilities for each individual contained in a micro dataset. The representativeness of the sample then ensures that aggregate effects may well be approximated by grossing up - via appropriate weighting techniques - the effects of each individual unit. Our research follows the latter approach aiming to develop a tax-benefit odel for the Greek economy. In particular, we extend and improve upon the basic tax-benefit model currently existing for Greece, EUROMOD, a cross-country comparative model simulating tax and transfer policies in 19 member states of the European Union. The model was designed with a view to provide a consistent comparative framework within which to study the effect of various public policy instruments upon measures of personal income and householdwelfare (Sutherland, 2000)1. The contribution of the current thesis may be framed along two broad dimensions. On a first level, as already mentioned, EUROMOD is an integrated European model. As such, its development primarily focused upon ensuring comparability across countries throughout the model construction process. Inevitably, this necessitated a detachment from a detailed representation of tax benefit systems at the national level. Accordingly, the first direction in which we improve upon the model entails its refinement to better respond to the particular requirements and characteristics of the Greek tax benefit system. On a second, and more innovative, level we extend the model’s scope by incorporating in its structure mechanisms that account for circumstances facing the real– world policy maker: tax evasion, benefit non take up as well as behavioural responses to announced policy measures. In the remaining part of this introductory chapter, we place the thesis in the context of the relatively recent but rapidly evolving literature on microsimulation in economics. The review offers a historical account of developments, highlighting the strengths but also calling attention to the likely pitfalls of specific microsimulation approaches. This sets a reference framework, in which we place our contribution. The introduction closes with a detailed layout of the core chapters of this thesis (Chapters II-VI). In the final chapter (Chapter VII) we discuss our main findings and conclusions. The seminal paper on socio-economic microsimulation modelling dates back to the late 1950s. Orcutt (1957), driven by the aspiration to improve the predictive usefulness of economic models in analyzing the effects of public policy, envisaged a radical departure from standard macro-econometric methods applied at that time. As Wolfson (2000) notes, Orcutt’s vision was grounded upon the empirical observation that “realistic micro-level circumstances and behavioural patterns are sufficiently multivariate and non-linear that they can never, as a matter of mathematical logic, be represented by well-formed tractable aggregate mathematical relationships”. Along these lines, Orcutt’s counter- proposal was a new type of model comprising of elemental units of analysis, such as individuals, families or firms.Information at the unit level would be retrieved from samples representing real populations and would be fed into the model, which would then apply appropriate operating characteristics to produce output. Orcutt further viewed his novel method as a vehicle for broadening the range of socio-economic phenomena that can be studied. In his words, “[then]current models of our socio-economic system have an unduly narrow reach in that they have little to say about such fundamental things as…distributions of individuals, households or firms in single or multivariate classifications” (Orcutt (1957) , page 116). Early developments in the field of microsimulation modelling took place in the 1960sand consisted mainly of the construction of personal income tax models in a number of countries including the US, Sweden, Canada and Norway (see Citro and Hanushek, 1991 andOECD, 1988 for a historical review). The next decade witnessed the expansion of this activity in many European countries, such as the United Kingdom, Austria, Finland, France,Italy and the Netherlands. It was not, however, until the 1980’s that research fully capitalized upon Orcutt’s vision. This was due to the growing availability of rich multivariate microdatasets paralleled by technological advances increasing the power of modern computing.Over the past two decades, progress has been fast and ongoing in the field, as microsimulation has proved to be a very flexible and versatile tool. Important methodological advances have addressed data and modelling issues, while models themselves have be enextended along temporal, geographical and behavioural dimensions.While the role of data availability and technological advances should not be underestimated, the real stimulus for the rapid progress achieved in the field was in fact what Orcutt had precisely expected from the microsimulation approach: predictive usefulness. Public policies are usually designed with a specific purpose in mind (O’Hare and Gupta, 2000). For example, the primary goal of a policy increasing the progressiveness of the tax system may be to reduce income inequality. Income transfer policies may be specifically designed so as to target those in greatest need and ultimately reduce poverty. The effectiveness of public policy in meeting such goals can only be evaluated through the study of the distribution of the induced changes across different segments of the population. This is exactly the “value added” of the microsimulation approach, which has become an increasingly powerful instrument extensively used by governmental institutions for the evaluation of public policy. The qualities of the microsimulation approach flow from its two distinguishing features. First, by operating at the individual level, it takes into account diverse circumstances and characteristics of the population of interest (Citro and Hanushek, 1991). This allows the effects of public policy to be studied along the income distribution and across its various population segments. For example, in evaluating a specific policy measure a microsimulation model may quantify effects upon the incomes of a particular income quintile (e.g. the top 20% of the income distribution) or of a particular socio-economic group (e.g. families with children, employees, etc). It may consequently also provide estimates upon effects on the redistribution of income or on poverty measures. On a related note, it offers users the capacity to identify winners and losers of a proposed reform, thus providing a very useful first approximation of the welfare effects, as well as fiscal and political feasibility, of policy proposals on the political agenda. The second distinguishing feature of microsimulation models is their flexibility. Several desirable properties emerge as a result. On a first level, as already implied in the above discussion, microsimulation models may be used to examine, apart from current, alternative public policy scenarios. Further, alternative scenarios may be specified through changes that involve complicated interactions among more than one government program (Citro and Hanushek, 1991). The analysis of hypothetical reforms stretches from an evaluation of distributional effects, as outlined earlier, to a quantification of their fiscal implications. In other words, microsimulation models enable the accurate evaluation of the aggregate cost or benefit of a reform (Spadaro, 2007). Such aggregate estimates are provided by appropriately weighting and adding up individual changes in net incomes arising from an application of alternative, as opposed to current, rules in a country’s tax and benefit system (Sutherland, 1991). However, it is the numerous directions in which microsimulation modelling has developed over the past years that provide the most obvious and convincing evidence of the flexibility ingrained in this micro-analytic approach. A basic taxonomy of microsimulation models may provide a simple illustration at this point. Generally speaking, we may classify microsimulation models under two broad categories: static and dynamic. Static models typically impute income tax or other liabilities and the receipt of social security and other benefits by applying the rules for eligibility or liability to each of the micro units (Harding, 1993). In replicating current or hypothetical institutional frameworks, static models assume away behavioural responses on the part of micro agents. Their key purpose has hence traditionally been to show the “morning after” impact of a policy change. Given the short- or medium-term time frame under which public policy operates, static models (though early stage achievements in the microsimulation field) have been “a gigantic success story in changing the nature of the political debate” (Caldwell, 1997). On the other hand, theirdisregard of behavioural reactions makes them oblivious to efficiency concerns, their static approach precludes the study of longer-term redistribution effects, while the ex ante nature of the analysis is a potential source of inaccuracy if the rules governing a country’s tax benefit system are not fully adhered to ex post. However, the flexibility embedded in the microsimulation approach has enabled the development of static models along many dimensions, one of which is their extension to include dynamic elements. In fact, the term “dynamic” is often used to describe two different aspects: (i) how behavioural responses are treated in the model, and (ii) how micro data is “aged” to future years2. In the first case, static models have been extended to simulate changes in behaviour induced by a specific policy shock – for example, how a change in the tax system may affect labour supply decisions or consumption patterns. Typically, behavioural parameters are externally estimated and then incorporated into the model. Strictly speaking, however, such models may not be fully dynamic for two reasons. On the one hand, it is practically impossible to model all changes in behaviour induced by a policy reform. On the other hand, they do not consider time issues. This latter aspect is captured by microsimulation models “ageing” individuals in the original micro dataset by simulating major life events (such as death, marriage, divorce, fertility etc) as well as intertemporal behavioural decisions (such as consumption, savings, retirement etc). Such models recalculate the characteristics of each individual year-by-year, stretching to a long-term horizon. Given the increasing concern about the social and economic impact of population ageing, the historical and longitudinal nature of these models has rendered them an invaluable tool in public policy analysis. On the downside, they impose prohibitive – at least, relative to their static counterparts – computing, data and resource requirements. Other directions which microsimulation modelling is taking include the development of cross-country models, a prime example of which is EUROMOD, or the development of models simulating the behaviour of business firms, rather than individuals or households. Another very promising route for microsimulation research is its integration with macroeconomic analysis. A policy change, or the behavioural responses it initiates, may have an effect on the macroeconomy (Sutherland, 1991). For example, a tax reform favouring low paid workers may lower their bargaining power in the next round of wage negotiations and consequently their wages. Further, if the tax reform creates inflationary pressures, the government may increase interest rates. Hence, far from the initial effect of the increase in net incomes of low paid workers, the reform may ultimately lead to lower wages or to lower real living standards if it is inflationary, or even to a change in the nominal incomes of savers and borrowers if interest rates are changed by the government. A detailed account of the various microsimulation models internationally developed over the past decades would be superfluous for the purposes of this thesis. Instead, through the above discussion, we aimed to highlight major developments in microsimulation research with a view to relate them to existing tax-benefit models treating the Greek economy, and ultimately underline the “gaps” we aim to bridge. The microsimulation approach was first used to study the effects of public policy in Greece in the early 1990s. The earliest model (Papapanagos, 1994) was a static one simulating the features of the personal income tax and social security system in Greece. The model has an exclusive focus upon the effects of income taxation, calculating allowances, income deductions, taxable income, tax schedule liability, tax credits and the final taxschedule liability of taxpayers and households in the micro dataset. Social transfers are not examined, neither is indirect taxation. The Greek indirect taxation system is in fact simulated by another microsimulation model (Kaplanoglou, 2000). This is another static model, which is nonetheless endowed with the flexibility to incorporate behavioural responses via assumptions regarding own-price elasticities of consumer goods. As the degree of commodity disaggregation is prohibitively high to allow detailed elasticity estimates, results are contained within a confidence interval with limits that correspond to two extremes of behavioural response: no change in purchased quantities (zero own price elasticities) and no change in expenditure (own-price elasticity of -1). The model is uniquely orientated towards the evaluation of indirect tax policies, and thus does not capture other features of the Greek tax-benefit system. The most complete to date tax-benefit model for the Greek economy is contained as a component of the cross-country comparative model earlier referred to, EUROMOD. EUROMOD is a static microsimulation model. Its component treating the Greek economysimulates rules governing policy on social benefits, social insurance contributions and direct taxation as they operated in 1998, 2001 and 2003. As already stated, EUROMOD acts as a basis for the tax-benefit model developed in the context of the present thesis. Hence, in what follows we provide a selective account of its various attributes and relate them to the improvements and extensions we aim to achieve through its development. On a first level, EUROMOD draws on European Community Household Panel (ECHP) 1996 data, containing information on incomes earned in 1995. This micro sample has been “statically aged”, though not fully, in order to represent more accurately certain population characteristics as they pertained in the simulated years (i.e. 1998, 2001, and 2003). In particular, the static ageing technique applied – uprating – adjusted monetary values for differential income growth and inflation. Other relevant characteristics of the population –mainly socio-demographic in nature – which may undergo dramatic change in a short space of time have been left unadjusted. In other words, weights attached to individuals in the dataset have not been changed to reflect economic and social change, such as for example an increase in unemployment, or in owner occupation or in the incidence of single parenthood. Hence, not all relevant aspects of structural change have been modelled. But even in the unlikely event that this was ever possible, as Sutherland (1991) notes “it should be remembered that grossing-up and updating procedures are not good substitutes for high quality, recent data. The most strongly worded piece of advice to the government was that they should begin collecting new data for a future model as soon as possible”. This is an important driving force behind our intent to improve upon the Greek component of EUROMOD via the incorporation of a more recent data source.Another major motivation for replacing the data source currently used pertains to the type of information contained in the ECHP. As already mentioned, the model construction process of EUROMOD has concentrated upon finding commonality across countries. One aspect of this focus is the model’s engagement in the construction of a database providing comparable microsimulation results across 19 countries. National differences in data collection practice were suppressed to the extent possible or avoided altogether via the use of data sources designed by a common agency. ECHP, designed by Eurostat, is a case in point and is in fact the base dataset for four countries other than Greece (Austria, Denmark, Portugal and Spain). However, this gain in comparability goes hand-in-hand with a loss in the richness of information available at the national level. This may be a severe shortcoming to the extent that the loss in detail inhibits a close enough representation of a specific country’s tax and benefit system. In the case of Greece, the ECHP was indeed restrictive in several aspects and additionally, as discussed above, considerably out of date for a good enough representation of the Greek tax-benefit system. In Chapter II, we summarize our efforts to improve the model by making use of a new, more up-to-date data source which also enables the fine-tuning of simulations to better respond to the particularities and characteristics of the Greek tax benefit system. On a second level, taking for granted detailed data – corrected for as many defects as possible – EUROMOD may still be liable to shortcomings strictly related to the ex ante nature of static microsimulation analysis. Hence, even if the policies simulated represent the current tax-benefit system satisfactorily, the results may not correspond to what is observed ex post. One reason for this is that static models do not account for second order effects induced by behavioural responses to various policies. This is an issue we return to shortly. Another source of inaccuracy could be that rules delineating the tax-benefit system do not adequately describe what happens in fact. This may well be the case if individuals break the rules. For example, EUROMOD is built under the implicit assumption that taxpayers declare all their income to the tax authorities. In reality, however, taxpayers may attempt to evade taxes by withholding information from the tax authorities. Another case in point is that EUROMOD rationally presumes that individuals take up their entitlement to social transfers. However, there are reasons – such as lack of information, fear of social stigma etc – for which people may not in practice claim benefits to which they are entitled. If such deviations from rules are operative, they may have a significant impact upon the validity of the model’s results. On the one hand, aggregate estimates of tax revenue or benefit expenditure will be biased. On the other hand, effects on the incomes of households or individuals will be concealed, thus introducing a bias in the distributional analysis. The above considerations motivated the extension of EUROMOD to include components that explicitly address the phenomena of tax evasion and benefit non take up in Greece. These issues are treated in Chapters III and IV respectively. Finally, the purely static nature of EUROMOD makes it appropriate for the study of the immediate impact of a policy change, before any of the individuals within the model change their behaviour in response to the policy shock. The eventual, however, impact of a policy reform will most probably be different as, in practice, most reforms are not marginal and may even be specifically designed to induce changes in agent behaviour. For example, a shift in the tax regime entailing a lower marginal tax rate increases – ceteris paribus – households’ disposable incomes. This is the first-order effect, typically studied by static microsimulation models. However, income effects as well as changes in the after tax price of labour may also modify labour supply decisions, thus inducing a second order effect upon households’ disposable incomes. Under these circumstances, a microsimulation model may evaluate the eventual efficiency and redistribution performance of public policy only if it can reproduce agents’ behaviour. In Chapter V, we address – at least to some extent – the above concerns by allowing a single aspect of agents’ behaviour, namely labour supply decisions, to vary in response to policy change. The flexibility of EUROMOD enables us to incorporate the behavioural parameters developed externally into the fiscal and distributional analysis performed by the model. On the whole, our research effort has aimed to capitalise upon a valuable potential embedded in a static microsimulation model of the Greek economy. On this basis, and along the dimensions discussed earlier, we developed a tax benefit model which may better inform the evaluation of public policy in Greece. In Chapter VI, we demonstrate the model’s improved potential via an analysis of the fiscal and distributional implications of five different reforms in the Greek tax-benefit system. In the remainder of this chapter, we discuss in some more detail the content of each of the five core chapters of this thesis.
Λέξη κλειδί Income
Ημερομηνία έκδοσης 2009
Άδεια χρήσης https://creativecommons.org/licenses/by/4.0/