Flat Taxes and Labor Supply in. Central and Eastern Europe

Transcription

1 Flat Taxes and Labor Supply in Central and Eastern Europe Kathleen F. Easterbrook May 12, 2008 Department of Economics Stanford University Stanford, CA Advisor: Professor Robert E. Hall Abstract This study examines how flat taxes have affected labor supply in the Central and Eastern European countries where they have been adopted. I study a structural relationship among key variables, including tax rates and labor supply. According to the model, work incentives and labor supply increased with the adoption of the flat tax in Lithuania, Russia, Slovakia, Ukraine, Romania, and Georgia. In these countries, flat taxes reduced the marginal tax rate on income and thus increased incentives to work. In contrast, the model indicates that Estonia s labor supply decreased because the flat tax actually increased the marginal tax rate. Latvia experienced no change in the marginal tax rate with the adoption of the flat tax. A surprising finding is that the marginal tax rates are in the range of 60% to 75% in six of the eight countries even after the adoption of flat tax reforms. Actual labor supply statistics suggest the model I use estimates labor supply well for five countries but does not explain all labor supply shifts in emerging economies. I thank Bob Hall for his insight and humor, as well as for challenging me to understand the nuances of the model I use in this study. Without his guidance, this project would not have been possible. I also thank Geoffrey Rothwell for his invaluable advise throughout the research process. I am incredibly grateful to Greg Rosston and Josie Smith for their critical feedback. Finally, thank you to my friends and family for supporting my efforts, both while I was writing my thesis and over the past four years. Data sets used in this study are available at 1

2 Easterbrook 1 Contents 1 Introduction 3 2 Economics of Flat Taxes Flat Tax Reforms in Central and Eastern Europe Efficiency and Equity of Flat Taxes Literature Review Flat Tax Studies Flat Tax Simulations Studies on Taxation and Labor Supply Labor Supply Model Comparative Statics A More Detailed Analysis of the Prescott Model Calibration Average Marginal Tax Rates Estimating the Proportion of People in Each Tax Bracket Estimating the Effect of Flat Taxes on Work Incentives and Labor Supply Over Time Limitations of the Model Data Tax Code Data National Statistics Labor Supply Data Results 37

3 Easterbrook Implications of Conditioning on c y Flat Taxes Effects on Incentives to Work Goodness of Fit: Estimated Versus Actual Changes in Labor Supply Estonia Lithuania Latvia Russia Slovakia Ukraine Romania Georgia Conclusion 54 A The Dynamic Form of the Prescott Model 62 B Estimating the Income Distribution 64 C Additional Charts of Results 68 C.1 Addendum to Flat Taxes Effects on Incentives to Work C.2 Addendum to Goodness of Fit D Tax Code Data 71 E Income Distribution Data 74

4 Easterbrook 3 1 Introduction In recent years, fourteen countries in Central and Eastern Europe have adopted flat tax reforms. During that same time, the countries economies grew as they transitioned from centrally-planned to market-based economies. Since 2000, countries in Central and Eastern Europe have averaged over 5% growth annually, a higher growth rate than Western Europe. Both flat taxes and the economic growth of the region have drawn international attention. Yet few studies have examined whether flat taxes have contributed to the economic growth of these countries. My study attempts to fill this gap in our understanding of flat taxes. I examine whether flat taxes have contributed to economic growth by increasing individuals incentives to work and their supply of labor. No matter their relation to regional growth, flat taxes have been part of Eastern and Central European countries transitions to market economies. With the collapse of the Soviet Union in 1991, Central and Eastern Europe began to transform politically and economically. Some countries in my study were formerly part of the Soviet Union and gained independence. Others were part of the Eastern-block previously under Soviet control. Economically, the countries began to decentralize their economies and experience higher rates of economic growth. Among the economic reforms, the flat tax has been associated with less corruption and smaller amounts of government intervention in these economies. In 1994, Estonia became the first country in Central and Eastern Europe to implement a flat tax. In 1992, Estonians elected a pro-reform government. In 1993, the government approved the flat tax of 26%. The flat tax replaced a progressive tax system with rates of 16%, 24%, and 33%. The other Baltic states, Lithuania and Latvia, soon followed their neighbor, implementing flat tax reforms in 1994 and Russia began the second wave of flat tax adoption when it replaced its personal income tax brackets with a 13% flat tax in Slovakia, Ukraine, Romania, and Georgia implemented flat taxes in 2004 and More recently, Macedonia, Montenegro, Albania, the Czech Republic, and Bulgaria have

5 Easterbrook 4 adopted flat tax reforms. 1 For proponents of the flat tax, the reform is emblematic of the economic progress of these countries. The goal of this study is to understand one mechanism through which flat taxes could have encouraged economic growth. I examine how flat taxes have affected individuals incentives to work and their supply of labor. I focus on labor supply because flat taxes apply to personal income taxes in all countries. Furthermore, I limit my study to countries that adopted the flat tax prior to 2006 due to data constraints. The countries in this study are: Estonia, Lithuania, Latvia, Russia, Slovakia, Ukraine, Romania, and Georgia. Table 1 summarizes the changes in personal income tax rates with the adoption of flat taxes in the eight countries. Table 1: Flat Taxes: Will they increase work incentives and labor supply? Country Year Flat Tax Tax Flat Tax Implemented Brackets Rate Estonia %, 24%, 33% 26% Lithuania %, 18%, 24%, 28%, 33% 33% Latvia %, 10% 25% Russia %, 20%, 30% 13% Slovakia %, 20%, 28%, 35%, 38% 19% Ukraine %, 15%, 20%, 30%, 40% 13% Romania %, 23%, 28%, 34%, 40% 16% Georgia %, 15%, 17%, 20% 12% The results indicate that individuals incentives to work have increased under the flat tax in six of the eight countries: Lithuania, Russia, Slovakia, Ukraine, Romania, and Georgia. I use a structural equation to estimate the average number of hours people work per week before and after the flat tax was implemented in each country. The equation describes the 1 Serbia is often included in the list of countries that have adopted flat taxes. I exclude Serbia because it effectively has two tax rates on income. Serbia applies one rate to all labor income and another to income from all sources that is above a threshold.

6 Easterbrook 5 relationship among hours worked, the marginal tax rate, and the consumption-to-output ratio, and it was used in Prescott (2004). Given a specified value of the consumption-tooutput ratio, a lower marginal tax rate implies a greater supply of labor. I interpret an estimated increase in labor supply as increased incentives to work. The model estimates that labor supply increases in six countries after they implemented flat taxes because the reforms decreased the marginal tax rates on labor. The marginal tax rates did not decrease in Estonia or Latvia. The model suggests that the incentives to work and labor supply did not increase in those countries. I compare the model s estimations to actual labor supply data to understand the appropriateness of the model and the extent to which flat taxes explain labor supply shifts. The model s estimations approximate actual labor supply statistics in five of the seven countries for which sufficient data are available. These results suggest that flat taxes have often affected work incentives as the model indicates, but that the model does not explain all labor supply shifts in emerging markets. The model does not account for labor market regulations and implies high wage elasticity. Both have been the targets of some objections to the model and are likely sources of error in the estimations. The rest of this paper is organized in the following fashion. Section 2 discusses the flat tax reforms in more detail and the economic implications of flat taxes. In Section 3, I review previous studies on flat taxes. Section 4 describes the labor supply model I use to understand individuals work incentives. In Section 5, I describe the data used in the study. Section 6 elaborates on the findings of this study, and Section 7 concludes. 2 Economics of Flat Taxes I define a flat tax as a personal income tax with one marginal rate. Since flat taxes often have personal exemptions, the flat tax usually has two marginal rates in practice. A personal exemption is a specified amount of income that the taxpayer earns which the government

7 Easterbrook 6 does not tax. Income below the level of the personal exemption is effectively taxed at a 0% rate. As defined here, flat taxes do not necessarily have broad tax bases or low marginal rates. Additionally, flat taxes are not necessarily consumption taxes nor must they apply to corporate income. I draw these distinctions because my definition of flat tax differs from the Hall-Rabushka flat tax (Hall and Rabushka 1983). The Hall-Rabushka flat tax is a consumption tax with a broad base and a marginal rate of 19% on both personal and corporate income. Under the Hall-Rabushka flat tax, most tax deductions are eliminated, and saving is not taxed. My definition is different than the Hall-Rabushka flat tax since the flat tax reforms in the countries I examine do not mirror the Hall-Rabushka reform. I choose to define flat taxes as having one marginal rate because the definition describes the common feature of the tax codes of countries widely considered to have flat taxes. I will elaborate on the actual flat tax reforms below. I also distinguish actual flat tax reforms from the Hall-Rabushka proposal because the economics benefits often associated with flat taxes depend on the structure of the flat tax. Hall and Rabushka (1995) anticipate labor supply will increase because lower tax rates will improve incentives to work. They expect people will save a higher proportion of their incomes than they currently do because saving is not taxed. However, the actual flat tax reforms do not mirror the Hall-Rabushka proposal. It is unclear that we should expect the countries to experience the benefits of Hall-Rabushka flat tax. The goal of this study to examine whether the countries experience one particular benefit associated with the flat tax increased work incentives and labor supply. When I discuss the benefits of flat taxes, it is important to clarify that a flat tax is a tax reform and not necessarily a tax cut. A tax reform could describe many changes to the tax code, including simplification, changes in tax deductions or the number of brackets, and adjustments of rates. A tax cut implies that the marginal tax rates are reduced. A

8 Easterbrook 7 flat tax will only qualify as a tax cut if the marginal rate is reduced as part of the reform. The benefits from simplifying the tax code can be thought of as benefits of a flat tax reform regardless of the tax rate. The benefits of lower marginal tax rates are benefits associated with tax cuts irrespective of the flatness of the tax reform. This study focuses on the benefits of the lower rates. Although the flatness of the tax code is not directly related to my results, I choose to examine flat tax reforms because of the attention they have drawn and lack of empirical work on actual flat tax reforms. 2.1 Flat Tax Reforms in Central and Eastern Europe The structures of the flat tax reforms that countries have adopted differ greatly from one another, except for the common feature that they impose one marginal tax rate on personal income. Personal income tax (PIT) rates range from 12% in Georgia and 13% in Russia and Ukraine to 33% in Lithuania. Unlike other countries in this study, Georgia implemented a proportional tax on personal income as it abolished the personal exemption. Some countries also reformed taxation of corporate income and dividends as part of their flat taxes. Estonia, Slovakia, and Romania all set corporate income tax (CIT) rates to be equal to PIT rates, and dividends in Estonia and Slovakia are exempt from taxation. Slovakia also set its VAT at the same rate as its PIT and CIT rate. It is worth noting that Slovakia s flat tax reform is considered to be most similar to the Hall-Rabushka flat tax. Table 2 summarizes the differences in the countries flat taxes. 2.2 Efficiency and Equity of Flat Taxes The particular structures of the flat tax reforms determine the extent to which the countries benefit from flat taxes. Clearly, countries that implement flat taxes with high rates are less likely to realize the benefits of a low tax rate. If a country s flat tax still has many deductions, one will not expect the country to benefit greatly from simplicity. Since many of the countries have implemented flat taxes with lower rates than the PITs of developed

9 Easterbrook 8 Table 2: Flat Taxes Reforms (by Year of Adoption) Country PIT Personal CIT Dividends Tax Rate Deduction Rate Estonia 26% Yes 26% Exempt Lithuania 33% Yes 29% No withholding system Latvia 25% Yes 25% Exempt for Latvian cos. Russia 13% Yes 35% 30% withholding Slovakia 19% Yes 19% Exempt Ukraine 13% Yes 25% 13% final withholding Romania 16% Yes 16% 10% final withholding Georgia 12% No 20% 10% final withholding nations, asking how flat taxes have affected work incentives is a natural question. Based on the structure of the Hall-Rabushka flat tax, it is possible that a country that adopts a flat tax can benefit in three ways. First, the country can benefit from the simplicity of the flat tax. Second, it can benefit from the lower marginal rate. Third, the country can benefit from the fact that the flat tax signals to other countries that it is becoming more fiscally conservative and market-oriented. I will elaborate on each of these potential benefits as I address how the flat tax can increase economic efficiency. I will briefly discuss the trade-off between efficiency and equity that accompanies the flat tax reform. Because my study focuses on potential efficiency gains from the flat tax, I will not describe the equity issues in detail. Theory suggests that the simplicity of the flat tax increases compliance with the tax code. The flat tax decreases loopholes and tax avoidance. The low marginal rate also encourages individuals to comply with the tax code since a low rate decreases the benefits of avoiding taxes. If all types of income are taxed at the same rate, individuals have no incentive to shift income to sources that are taxed at lower rates. Additionally, a simple tax code is less burdensome for the government to administer than a complex tax system. People and the

10 Easterbrook 9 government devote fewer resources to paying, collecting, and avoiding taxes. A simple tax code also creates fewer distortions of economic behavior than a more complex system. If individuals earnings are taxed at the same rate regardless of the how the income is earned or spent, the tax will have a much smaller effect on individuals decisions. Consider a mortgage deduction, which lowers the cost of buying a house relative to buying other goods. The artificially low price of housing encourages more people to buy homes than would in the absence of the mortgage deduction. Perhaps an equilibrium with many people owning homes is desirable. The point is that the mortgage deduction affects economic behavior. Flat taxes distort behavior less because of uniform rates and fewer deductions. Flat taxes can also encourage economic growth via the low marginal tax rate. A low tax rate on salary and earnings encourages people to work more, provided the uncompensated wage effect dominates the income effect. Labor supply increases. I will discuss this point further in section 4. A low rate on corporate income encourages people to start businesses and foreign multinationals to expand into the country. The increase in economic activity boosts output. If the flat tax is a consumption tax, such that dividends, interest, and capital gains are not taxed, people will have incentives to save more. Higher saving leads to more investment and higher output. Finally, the label flat tax has value because of the signaling effect. When a country adopts a flat tax, it often signals to other countries and its citizens that the country is becoming more market-oriented. Freer markets generally encourage people to conduct business in the country, which has the potential to lead to economic growth. Flat taxes also signal that the government is attempting to decrease tax avoidance and increase certainty in the economy. Less uncertainty associated with working and conducting business encourages economic activity. The signaling effect works because flat taxes are often associated with simplicity and lower marginal tax rates. The countries in Central and Eastern Europe are positioned well to reap the benefits of

11 Easterbrook 10 flat taxes. Corruption was higher in these emerging economies than in advanced economies, so it is likely that they benefit more from increased tax compliance and certainty. Also, as emerging markets, the governments do not have the abundant resources to devote to tax collection. Since their countries were formerly part of the Soviet Union or Soviet satellites, flat taxes send strong signals of freer markets and reform. Like many fiscal policy issues, the flat tax often contains the trade-off between efficiency and equity. Theory suggests that the flat tax shifts the tax burden to the middle class. Many actual and hypothetical flat tax reforms are designed such that the government initially collects the same amount in tax revenue as would be collected under the prior tax system. Under the flat tax, the wealthy pay lower taxes. A higher personal exemption ensures that low-income individuals do not bear higher tax burdens. Thus, an increased tax burden falls on the middle class. A broad tax base mitigates the middle class s increased tax burden. With few tax shelters and deductions for income, the tax burden on wealthier individuals does not decrease as much as their marginal rate. A flat tax that is a consumption tax also affects the distribution of the tax burden. Consumption taxes place higher burdens on low-income individuals compared to wealthier individuals. Low-income individuals have fewer means to save and consume a higher proportion of their earnings. Further discussion of the distributional effects of the flat tax is beyond the scope of this paper. For a longer discussion, see Keen, Klemm and Varsano (2006) and Gonzalez-Torrabadella and Pijoan-Mas (2006). The former article s discussion includes an explanation of the progressivity of the flat tax. 3 Literature Review An extensive review of the academic literature on flat taxes reveals that few papers attempt to quantify the effects of the flat tax reforms that have been implemented in various countries. Many studies thoroughly discuss the reforms in a particular country or set of countries in

12 Easterbrook 11 Central and Eastern Europe, but do not attempt to quantify the reforms effects. These studies qualitatively describe how the reforms change work incentives. With the exception of Ivanova, Keen and Klemm s (2005) study of Russia s flat tax, these studies do not attempt to measure the effects on the labor market. Studies that attempt to quantify the effects of flat tax reforms do not focus on actual flat tax reforms. Rather, these studies are policy experiments with hypothetical flat tax reforms. My study is most similar to the studies on hypothetical flat tax reforms. Both the authors of hypothetical flat tax studies and I use equilibrium models to compute the effects of flat tax reforms. The notable difference and the gap in the literature I hope to help fill is that I attempt to quantify the effects of actual flat tax reforms. I will elaborate on both the flat tax studies and simulations in this section. I draw my methodology from the body of literature which simulates how taxes affect labor supply. The difference between this literature and my study is that I examine the effects of flat tax reforms. Most studies in this literature focus on U.S. or European economies or are theoretical. Yet, these studies ask the same fundamental question. As such, I can draw my methodology from studies that assess the U.S. or European economies. I will discuss the papers from which I draw my methodology separately. 3.1 Flat Tax Studies Despite the large amount of media attention that flat taxes have received, there are still few academic articles on actual flat tax reforms. Of the academic articles that I reviewed, most do not perform rigorous quantitative analyses of how flat taxes affect the labor market. The articles stick to academic discussion of the potential effects of flat taxes on many economic indicators, including labor supply, economic growth, compliance, and tax burden across different income groups. Keen et al. (2006) of the IMF provide an overview of the studies on flat taxes. They too conclude that empirical evidence is limited. Like many flat tax studies, the paper discusses how flat taxes affect work incentives in theory. Keen et al. (2006) suggest

13 Easterbrook 12 that the empirical literature on how tax reforms affect labor supply could be drawn on to simulate the likely labor supply effect of adopting a flat tax (Keen et al. 2006, 27). This simulation is precisely what I do in my study. Keen et al. (2006) also stress that the flat taxes that have been implemented differ greatly. This fact is apparent as one reviews papers that compare the flat tax reforms. One useful paper that compares the different reforms is Jenn (2006), which provides details on the tax codes over time and discusses the economic arguments of the flat tax. One notable empirical study in the flat tax literature, Ivanova, Keen and Klemm (2005), evaluates Russia s experience with the flat tax in its first year. It is also the only empirical study of which Keen et al. (2006) are aware. Ivanova et al. (2005) ask why government revenues increased in that year, if compliance increased, and how labor supply was affected. Using national account data and national statistics, Ivanova et al. (2005) find that personal income taxes were not the cause of increased tax receipts and point to higher oil and gas prices as an explanation. Their results indicate that compliance increased. A smaller proportion of wealthier individuals, who faced large reductions in their marginal tax rates with the flat tax, evaded taxes. Directly relevant to my study, Ivanova et al. (2005) find the amount people work did not significantly change when Russia implemented flat taxes. They use panel data from the Russian Longitudinal Monitoring Survey. In a differences-in-differences econometric model, they estimate how much more individuals who faced large reductions in the marginal tax rate worked when compared to individuals who faced virtually no change in their marginal tax rate. According to their findings, the difference between the two groups of people is insignificant. There are two reasons to reexamine their findings. The first is that their study only examines labor supply in the first year of the flat tax. If the labor market does not respond to the tax change in the first year, but in the second or third year, such a change is not captured in Ivanova et al. (2005). Second, data constraints only allow them to include people

14 Easterbrook 13 who pay taxes in their study. They cannot include the individual who faced a 12% marginal tax rate prior to the reform and now faces a 0% rate due to the increase of the personal exemption. This individual s incentive to work has changed. One would want to include him in the study of how flat taxes affect the labor supply. Moore (2005), Brook and Leibfritz (2005), and Golias and Kicina (2005) focus on Slovakia s flat tax reform, which was implemented in Interest in Slovakia partially stems from the comprehensive nature of its flat tax. In addition to setting a 19% marginal rate on personal income, Slovakia set a 19% corporate tax rate and VAT. The country also eliminated the double taxation of capital gains and interest. The studies examine the implications of the flat tax for economic growth, tax burdens across income groups, foreign direct investment, and labor supply. All the studies predict labor supply will increase because of lower marginal tax rates and less deadweight loss in the labor market. Moore (2005) and Brook and Leibfritz (2005) calculate tax wedges to support their predictions, but also credit other reforms for increased work incentives. Moore mentions that welfare reforms may be a larger cause than the flat tax of increased incentives to work. Brook and Leibfritz argue that labor market reforms will increase the total labor supply. Such labor market reforms include increasing the number of hours in the working week and cutting severance pay obligations. Neither Moore nor Brook and Leibfritz formally model the labor supply. Other articles that discuss particular tax reforms include Chua (2003), Chalupka (2004), and Burger and Sklenar (2006). Chua (2003) and Chalupka (2004) are qualitative analyses of Russia and Slovakia s flat tax reforms, respectively. Burger and Sklenar (2006) conduct an empirical study of how Slovakia s flat tax reform affected corporations tax compliance. They use a self-designed survey of corporations to evaluate their level of tax evasion. As corporations compliance is beyond the scope of my study, I do not elaborate on this study. The timing of flat tax reforms partially explains the lack of empirical studies. With the exception of the Baltic States, countries have adopted flat taxes after Barring Russia,

15 Easterbrook 14 countries have only implemented flat taxes since Thus, studies involving long time series of data are not currently possible. Despite the lack of rigorous quantitative analysis, these articles meticulously evaluate the flat tax reforms and provide the economic theory behind future empirical work, including my study. 3.2 Flat Tax Simulations The other part of the literature on flat taxes consists of studies that examine the effects of hypothetical flat tax reforms. Studies of hypothetical reforms are policy experiments, in which the authors intend to shed light on how proposed reforms would affect the economy. These studies use equilibrium models to compute the effects of the flat tax. Most studies conclude that a flat tax has economic benefits and increases the labor supply. Many of the studies of hypothetical tax reforms have focused on the U.S. economy. Hall and Rabushka (1983) propose an overhaul of the U.S. tax code. Under their proposal, the government would impose a consumption tax on individuals. Personal income from wages and salaries above the personal deduction level would be taxed at 19%. Business profits that are not reinvested in land, capital, and structures also would be taxed at 19% (Hall and Rabushka 1995). Hall and Rabushka argue the lower marginal rates will increase incentives to work, especially among married women and people close to retirement. These individuals have high labor supply elasticities. They estimate that increased labor supply could increase GDP by 3% (Hall and Rabushka 1995). With increased incentives to save and take business risks, Hall and Rabushka (1995) estimate that GDP would increase 6%. Both Ventura (1999) and Cassou and Lansing (2004) explore the effects of a Hall- Rabushka tax on the U.S. economy. Ventura (1999) uses a general equilibrium model to compute the effects of a Hall-Rabushka flat tax. Ventura (1999) indicates that, on average, the number of hours that people work stays the same. However, wealthier individuals, who experience large decreases in marginal tax rates and are presumably the most productive

16 Easterbrook 15 members of society, increase the hours they work. Less productive individuals decrease the hours they work. Cassou and Lansing (2004) estimate the effects of the Hall-Rabushka flat tax in both a standard neoclassical growth model and a human-capital based endogenous growth model. According to their study, the flat tax on personal income, with the flattening of the tax brackets and lower marginal rates, is the biggest contributor to economic growth. Several other studies simulate how other flat tax reforms would affect the U.S. economy. Altig, Auerbach, Kotlikoff, Smetters and Walliser (2001) use a life-cycle model to simulate the effects of proportional and flat tax reforms in the U.S. Their model indicates that a flat tax reform increases the labor supply and economic growth. Boyd and Seldon (1996) evaluate the flat tax proposed by Representative Dick Armey. They find that savings, production in most sectors, and disposable income in every income bracket increases. I will not elaborate on these studies as I do not draw my model from them. Their results on how flat taxes affect the U.S. economy are not comparable to the small, emerging economies of Central and Eastern Europe. Several studies explore the effects of flat tax reforms in European countries. Aaberge, Colombino and Strom (2000) simulate how the labor markets in Italy, Norway, and Sweden respond to flat tax reforms. They model individual decisions to work based on heterogeneous preferences, productivity levels, and discrete job options. Aaberge et al. (2000) find that labor supply increases in all countries, although incomes increase by more in Norway than in Italy and Sweden. In Italy, labor market regulations limit the job options, and in Sweden, many women currently work and thus exhibit lower labor supply elasticities. As a result, labor supply increases are smaller in Italy and Sweden. Aaberge et al. (2000) also finds that higher income households experience larger income increases than lower income households. De Mooij, Folmer and Jacobs (2007) simulate the effects of a flat tax in the Netherlands using MIMIC, a general equilibrium model. 2 Their results imply that a flat tax can increase 2 MIMIC stands for Micro Macro model to analyze Institutional Context (De Mooij et al. 2007).

17 Easterbrook 16 labor supply and income inequality. Their model allows them to measure distributional effects of tax reforms, and it includes union bargaining and decisions about human capital accumulation in wage determination. However, the model does not include capital in the production function. Despite its usefulness for assessing certain reforms, I choose not use the MIMIC model because it does not include capital. Gonzalez-Torrabadella and Pijoan-Mas (2006) use a general equilibrium model to predict how a flat tax will affect Spain s economy. Their model assumes that the government taxes income from all sources wages, salary, capital at one rate. They calibrate their model with data from the European Community Household Panel. Their results indicate that labor supply, capital stock, and output increase in response to a flat tax reform. Like the other studies, they find that income inequality also increases. Two studies of which I am aware examine the effects of hypothetical flat tax reforms in countries in Central and Eastern Europe that have not adopted flat taxes. The studies focus on the efficiency-equity tradeoff of flat taxes. Benedek and Lelkes (2007) simulate the effects of a flat tax on Hungary. They find that a flat tax benefits wealthier households while less-wealthy households are worse-off. Cajner, Grobovsek and Kozamernik (2006) conduct microsimulations to examine the effects of flat tax reforms in Slovenia. Cajner et al. (2006) find that flat taxes often increase labor supply and GDP, but also increase income inequality. Cajner et al. (2006) also find that if labor supply is inelastic, flat taxes have the potential to decrease labor supply. Cajner et al. (2006) say that an inelastic labor supply is unrealistic. The dependence of the labor supply response on elasticity is of interest because a high elasticity contributes to my results of increased work incentives. My study is most similar in design to the studies on hypothetical tax reforms. The authors of these studies and I use equilibrium models to compute the effects of the flat tax. The models vary depending on the specifications of the tax reforms being evaluated. My study contributes to this body of literature in that I evaluate the effects of actual flat tax

18 Easterbrook 17 reforms rather than hypothetical ones. 3.3 Studies on Taxation and Labor Supply The literature that simulates flat tax reforms is a small part of the larger body of literature on how taxes affect labor supply. The model I use to evaluate flat tax reforms comes from this literature. Specifically, I base my model on one developed by Edward Prescott. Prescott and Richard Rogerson conduct numerous studies on how taxes on income affect individuals decisions to supply labor. Many of Prescott and Rogerson s studies focus on how taxes affect labor supply in Europe and the United States (Prescott 2004, Rogerson 2006, Rogerson 2007a, Rogerson 2007b). They do not evaluate flat tax reforms in their studies. However, their models shed light on how the labor supply responds to taxes, the same issue I explore. To apply Prescott s (2004) model, I adjust it so that I can simulate a flat tax reform. The model that serves as a basis for my own comes Prescott s (2004) study Why Do Americans Work So Much More Than Europeans? Prescott (2004) argues that differences in marginal tax rates on income help explain why Americans work more hours than Europeans. The partial equilibrium model he uses makes many simplifying assumptions about how individuals choose the number of hours to work. For example, it assumes that all individuals have the same preferences for leisure and consumption. The main variables that determine the average number of hours are the marginal tax rate on income, the consumption-to-output ratio, a preference parameter, and the labor factor share. Despite its simplicity, Prescott (2004) shows that the model accurately estimates the average number of hours that people work. The model s estimate of average hours worked is very close to the actual average number of average hours worked in Germany, France, Italy, Canada, the United Kingdom, Japan, and the United States in two time periods, and I use this model because of its applicability, accuracy, and simplicity. The model I use is just one of many models Prescott or Rogerson has used to evaluate how

19 Easterbrook 18 taxes affect labor supply. Prescott, Rogerson and Wallenius (2007) develops more theoretical life-cycle models that estimate what proportion of his life an individual will choose to work. Like Prescott (2004), Rogerson (2006) and Rogerson (2007a) examine why people on different sides of the Atlantic work different hours. The models in Rogerson (2006) and Rogerson (2007a) indicate that the size of the government, or the overall tax burden, affects the hours that individuals work. Rogerson (2007b) argues that one must include how the government spends its revenues when assessing the impact of taxes on the labor market. I intend my study to contribute to this group of literature by extending the applications of Prescott s (2004) model. It is worth noting that the latter two groups of articles I described simulations of hypothetical flat taxes and Prescott s and Rogersons studies are subsections of a larger body of literature on optimal taxation and tax reform. This literature is too extensive to cover in this review. Auerbach (1985) and Auerbach and Hines (2002) have more detailed discussions. Rebelo and Stokey (1995) is also often cited on the topic of tax reform. 4 Labor Supply Model To understand how the flat tax affected labor supply in each country in Central and Eastern Europe, I use the model developed in Prescott (2004) to estimate labor supply changes. The model aggregates each individual s labor supply decisions in a single time period. Examining one time period is appropriate since multiple time periods do not affect the resulting structural relationship among taxes, labor supply, consumption, and output. See Appendix A for a description of the model in a dynamic economy. Although the model estimates labor supply at a macro level, the individual s decision to work is the model s foundation. I assume that an individual maximizes his utility and chooses consumption and leisure accordingly. I express the flow utility of an individual in

20 Easterbrook 19 one period as u(c, h) = log c + α log(1 h), (1) where c represents consumption and h represents weekly hours worked normalized such that the total number of available hours is 1. The parameter α > 0 represents the individual s value of leisure relative to consumption. Like most theories about labor supply, mine assumes that work decreases a person s utility holding consumption fixed, even though, in reality, individuals might derive satisfaction from their jobs. People work so that they can consume. The utility function in equation (1) can be derived from the constant relative risk aversion utility function, u(c, h) = ( ) ( ) 1 α c 1 σ + (1 h) 1 γ. (2) 1 σ 1 γ The limit of equation (2) is equation (1) as the parameters σ and γ approach 1. By choosing values for σ and γ, I could use a different utility functions in my model. I use equation (1) since it is the specification used in Prescott (2004). I assume an individual s budget constraint is (1 + τ c )c (1 τ h )wh + b. (3) τ c and τ h are the tax rates on consumption and labor, w is the real wage rate, and b is the fixed amount of non-labor income. b represents income from savings or inheritance that the individual has accumulated and government transfers that do not depend on earnings. The budget constraint assumes that the individual can only earn income by supplying labor. Maximizing utility with respect to consumption and using the Lagrange multiplier yields the first-order condition, u c = λ(1 + τ c ). (4) Equation (4) is the Frisch consumption demand equation, which describes marginal utility in

21 Easterbrook 20 terms of consumption. 3 Maximizing equation (1) with respect to h and using the Lagrange multiplier results in the Frisch labor-supply equation, u h = λ(1 τ h )w. (5) The Frisch labor-supply equation describes labor supply in terms of wage and marginal utility of earnings. The marginal rate of substitution between leisure and consumption is the ratio of the equations (4) and (5); u h u c = (1 τ h)w 1 + τ c. (6) The marginal rate of substitution is equal to the ratio of the marginal costs. The Frisch labor-supply equation shows that the marginal cost of leisure is (1 τ h )w, and the Frisch consumption demand equation shows that the marginal cost of consumption is 1 + τ c. The equality in equation (6) will hold as long as the individual has the ability to maximize his utility. 4.1 Comparative Statics I use the marginal rate of substitution to illustrate how the income and uncompensated wage effects impact the number of hours a person chooses to work. If a person s b increases, the person will consume more and work less. u c decreases due to diminishing marginal utility of consumption. u h also decreases, and the first-order condition holds. This demonstrates the income effect; an increase in a person s income decreases the amount the person chooses to work. The uncompensated wage effect applies to the effect that marginal earnings have on the labor supply. If (1 τ h)w 1+τ c increases, a person will be willing to work and consume more. u h increases, and u c decreases. Under the uncompensated wage effect, an increase in w and 3 Equation (4) is also known as the Arrow-Borch condition. However, the Arrow-Borch condition often refers to the equation when it is used to represent optimal risk allocation between two parties in insurance markets.

22 Easterbrook 21 decreases in the marginal tax rates τ h and τ c increase the number of hours an individual will work. A tax reform that lowers tax rates on income has a theoretically ambiguous effect on the amount of labor an individual will supply. The effect of a lower tax rate is the same as the effect of a higher wage. The uncompensated wage effect suggests the person will work more hours since he faces a lower marginal tax rate. However, the income effect suggests the person has less incentive to work. Those who argue that reducing tax rates will increase the labor supply believe that the uncompensated wage effect dominates the income effect. In the case where the tax reform keeps revenues the same, the reform has no income effect at a macro level. In switching to a flat tax, some individuals faced lower marginal and average tax rates. These people are at the lower and upper ends of the income distribution. At the lower end are those who were in the first income tax bracket and paid a low marginal rate before the flat tax. When the flat tax was adopted and the personal exemption increased, these individuals paid no income taxes since they earned less than amount of the personal exemption. People at the upper end of the income distribution also saw their marginal and average tax rates fall. The way the flat tax affected these individuals work incentives is theoretically ambiguous. The flat tax reforms in Central and Eastern Europe did not uniformly lower average and marginal tax rates for all individuals. Some people saw no change in their marginal tax rates, since many governments chose a flat tax rate that was the same as a marginal rate in one of the previously used tax brackets. This happened in Lithuania, Latvia, and Georgia. Individuals whose marginal tax rates did not change may have experienced a change in their average tax rates. In theory, the income effect explains how these individuals adjusted the number of hours they work. Individuals fall into several other categories: (1) those who faced changes in their marginal tax rates but no change in the average tax rate, (2) those who experienced increases in

23 Easterbrook 22 their marginal rates, but decreases in their average tax rates, (3) those who experienced increases in their marginal and average rates, and (4) those who experienced decreases in their marginal rates, but increases in their average rates. Figure 1 summarizes how theory suggests individuals in each group adjust their labor supply. Figure 1: Theoretical Changes in Labor Supply: The arrows indicate the direction that labor supply shifts in theory. The uncompensated wage effect will determine how one expects the flat tax to affect people in category (1) as I describe above. These individuals experience no change in total income. Changes in marginal wage are the only factor affecting their labor supply. It is theoretically ambiguous how people in category (3) adjusted the number of hours they work. The income effect suggests they would increase the number of hours they work; the uncompensated wage effect suggests the opposite. In practice, individuals in category (3) are middle class in their countries. In countries that chose high flat tax rates, middle class individuals may have experienced increases in their marginal and average tax rates. This likely happened in Lithuania, Estonia, and Latvia. Their governments implemented flat tax rates of 33%, 26%, and 25%, respectively. These rates are high when compared to the previous rates in the income tax brackets. For people who fall into categories (2) and (4), it is theoretically clear how labor supply should change. Those who experience increases in their marginal rates and decreases in their average rates should reduce the number of hours they work. These individuals are lower-

24 Easterbrook 23 middle class in society. For them, increases in the personal deductions more than offset the increases in the marginal rates. Thus, their average tax rates fell while their marginal rates increased. In contrast, theory suggests that individuals who experienced decreases in marginal tax rates and increases in average tax rates should increase their labor supply. A broadening of the tax base or an increase in compliance could explain why some individuals marginal rates fell while their average tax rates increased. 4.2 A More Detailed Analysis of the Prescott Model Qualitative theory alone cannot explain how work incentives and labor supply changed under flat taxes. I use a structural relationship to estimate how labor supply responded to flat taxes in each of the eight countries in Central and Eastern Europe that adopted the reform. The model describes the relationship among the following variables: average number of hours people work per week, the marginal tax rate on income, people s preferences for leisure, their consumption decisions, and output. I compare the model s estimates of average hours people work per week in two time periods in Estonia, Lithuania, Latvia, Russia, Slovakia, Ukraine, Romania, and Georgia. Specifically, I examine how the average number of hours worked changes in the year the flat tax is implemented from the previous year. I interpret changes in labor supply as altered work incentives. I also compare the model s estimations of labor supply to actual labor supply figures to understand the extent to which flat taxes affected work incentives and explain actual labor supply shifts. Because of my interest in the net effects on labor supply and data constraints, I do not examine the separate effect of flat taxes on each group I describe above. In measuring changes in work incentives, I make no direct assumptions about how the government uses tax revenues. To estimate how flat taxes affected work incentives and the labor supply, I continue to examine the choices of the utility-maximizer in a single time period. I combine the tax

25 Easterbrook 24 variables in equation (6) such that In equation (7), τ is equal to u h u c = (1 τ h)w 1 + τ c = (1 τ)w. (7) τ = τ c + τ h 1 + τ c. (8) For convenience, I will refer to τ as the marginal tax rate on income and τ h as the marginal tax rate on labor. Taking the derivatives from equation (1), I find that u h u c = The first-order condition of the utility-maximizer is thus αc 1 h αc 1 h. (9) = (1 τ)w. (10) I assume that the market is competitive and the wage rate w is determined by the worker s marginal productivity. I assume all workers have the same productivity and calculate the marginal productivity at a macroeconomic level. The function governing the country s output is y = Ak θ h 1 θ. (11) The variable k is capital stock, and h is the labor supply. The parameter 0 < θ < 1 is the capital cost share, making 0 < 1 θ < 1 the labor cost share of production. A is the total factor productivity for the economy, but the parameter is not emphasized in the model. From equation (11), I calculate the marginal product of labor to be y h = (1 θ)ak θ h θ = (1 θ) y h. (12) Setting the wage rate w equal to the marginal product of labor (1 θ) y h in the first-order condition and solving for hours worked h results in the equation h = 1 θ (1 θ) + c y α. (13) (1 τ)

26 Easterbrook 25 This equation describes the structural relationship between the hours that utility-maximizers work per week on average and the marginal tax rate, conditional on c. I use equation (13) y to examine the effect of flat taxes on work incentives and labor supply. I often refer to the structural relationship in equation (13) as the model because equation (13) is the expression of the equations and assumptions of the Prescott model. Sections 4.3 and 4.4 describe how I use the relationship to evaluate the effects of flat taxes on work incentives and how I test the goodness of fit of the model. 4.3 Calibration I hypothesize that the structural relationship shown in equation (13) holds for all countries in my study and over multiple time periods. My use of the structural relationship is a three-step process: (1) calibration, (2) measurement of changes in work incentives, and (3) comparison to actual labor supply figures. In this section, I explain how I calibrate the model and how I calculate τ. I describe how I evaluate the effects of flat taxes on incentives to work and the goodness of fit in Section 4.4. I calibrate the model separately for each country using data from the year prior to the flat tax reform. 4 Specifically, I solve for each country s α using the values τ, θ, c, and h from y the year prior to the implementation of the flat tax. I will explain my calculation of τ below. With regard to the other parameters, I assume that the capital share of output θ is This implies that labor accounts for 0.67 of the cost share of production. These figures are standard factor shares for economies. For each country s values of c y and h, I use the actual indicators from the year before the flat tax reform. It is worth noting that h is a decimal in the equation. Following Prescott, I divide the actual hours worked by 100 to determine the value of h for calibration. Conversely, when I estimate h in the model, I multiply the figure by 100 to determine the model s estimation of average hours worked per week. 4 Due to data constraints, I calibrate the model with 1997 data for Lithuania and 1999 data for Russia.

27 Easterbrook Average Marginal Tax Rates In the structural relationship, τ represents the marginal tax rate on income. Marginal tax rates, as opposed to average tax rates, are appropriate to use in the analysis of labor supply because marginal tax rates affect individuals decisions to supply their next hour of labor. Prescott (2004) uses marginal tax rates to evaluate labor supply. However, my approach to calculating marginal tax rates differs from the one Prescott (2004) uses. Prescott bases his calculations of marginal rates on average tax rates. He calculates the average tax rate or tax burden on personal income and multiplies the rate by 1.6. He adds the product to his measure of marginal rate from social security taxes. 5 I follow Barro and Sahasakul (1983, 1986) in using statutory tax rates to calculate the marginal tax rate. Barro and Sahasakul (1983) argue that statutory tax rates are the appropriate measure of marginal tax rates; statutory tax rates describe the marginal tax on all wages and salaries that are not spent in a way that qualifies for a deduction. Before I explain how I calculate τ from statutory tax rates, recall the relationship among particular tax rates shown in equation (8): τ = τ h+τ c 1+τ c. τ c represents the tax on consumption, so I use each country s VAT for the τ c. τ h represents the marginal tax on labor. I use personal income tax rates τ w and payroll tax rates τ p to calculate the marginal tax on labor. It is useful to think of τ h as the sum of τ w and τ p. See equation (14) for the exact relationship among the variables. Although both personal income and payroll taxes are taxes on labor, governments distinguish between the two because the taxes fund different programs. Personal income taxes usually contribute to general government revenues, whereas payroll taxes often fund social insurance programs. Also, personal income taxes often apply to capital income an income source ignored in this study. Both employees and employers pay a percentage of wages and salaries in payroll taxes. I calculate payroll tax rate τ p as the sum of the employer and 5 The marginal rate from social security taxes is equal to social security tax revenues (GDP- indirect tax revenues)(1 θ).

28 Easterbrook 27 employee contribution rates. I include both employer and employee because labor is taxed no matter who has the legal obligation to pay. In using statutory tax rates for the values of τ w and τ p, I follow the principles of Barro and Sahasakul s (1983, 1986) method for estimating the average marginal tax rate. 6 Barro and Sahasakul (1983) uses statutory tax rates to determine the average marginal tax rate on personal income, τ w. Like Barro and Sahasakul, I use the weighted average of statutory tax rates to determine the marginal tax rate on labor τ h. See Appendix D for information about the statutory tax rates I use. Equation (14) describes the calculation of each country s average marginal personal income tax rate. n n τ h = τ hi p τi = (τ wi + τ pi )p τi. (14) i=1 i=1 τ hi represents the marginal tax rates associated with different income levels, and p τi represents the weighting for each marginal tax rate. i denotes the different tax brackets. The various values of τ hi are the sum of τ pi and τ wi. The weights p τi are the proportion of people that face particular marginal tax rates. Each country has several marginal tax rates that an individual might face due to tax brackets on personal income or different payroll tax rates for different levels of income. For example, there are two marginal PIT rates under the flat tax, 0% and the flat tax rate. A household that faces a 0% tax rate earns less than the amount of the personal exemption. 7 Similarly, when payroll taxes change at certain income thresholds or have a cap, I treat the different payroll tax rates in the same way I treat tax brackets. 6 There is a slight discrepancy between my method and that of Barro and Sahasakul (1986) in how we account for employers contributions to payroll taxes. However, Barro and Sahasakul (1986) bases estimates of marginal social security tax rates on the statutory tax rates. 7 Georgia is the exception; there is no personal exemption.

29 Easterbrook Estimating the Proportion of People in Each Tax Bracket I calculate p τi, or the proportion of people in each tax bracket, by imposing an income distribution on the country. I assume the country has a lognormal income distribution because the lognormal provides a reasonable approximation of the income distribution and is simple. Numerous studies on developing regions such as Africa and South Korea have used the lognormal distribution (Boccanfuso, Decaluwe and Savard 2003). The lognormal tends to approximate the actual income distribution especially well when the economy has a relatively large number of individuals at the lower end of the distribution. The Pareto distribution is another commonly-used income distribution. It is better than the lognormal at approximating actual income distributions when there are relatively more individuals at the higher ends of the income distribution (Boccanfuso et al. 2003). Since emerging economies tend to have greater number of less-wealthy individuals, I use the lognormal distribution. It is possible that the lognormal distribution underestimates the proportion of people in higher income brackets. Any underestimation in the income distribution leads to an underestimation of the average marginal tax rate. I do not consider this a significant problem in my study because I am most interested in the changes in the estimated labor supply in one year. A lognormal distribution is defined by two parameters, µ, or the mean value of the natural logarithm of income, ln i, and the standard deviation, σ. I estimate these two parameters from income distribution data, annual statistics of gross national income, and population figures. The following equation describes the calculation of µ: µ = n q=1 ln i q n [ ] vq GNI where ln i q = ln. (15) r The variable v q is the percentage of national income attributed to each quantile q, and r

30 Easterbrook 29 is the number of people in each quantile. There are a total of n quantiles. 8 I calculate σ based on the five or ten data points of ln i. After determining σ and µ, I can calculate the proportion of people in each tax bracket. Appendix B describes the lognormal approximation in more detail. Having obtained the average marginal rate on labor according to equation (14), I compute the average marginal tax rate on income as shown in equation (8) for each country and relevant year. When evaluating work incentives, I am interested in the average marginal rates of two tax regimes: (1) the progressive one that existed before the implementation of the flat tax and (2) the flat tax regime. The calculation of p τi differs slightly depending on whether I am evaluating work incentives or testing the goodness of fit of the model. 4.4 Estimating the Effect of Flat Taxes on Work Incentives and Labor Supply Over Time Having calibrated the model, I estimate how flat taxes affected work incentives and how well the model fits actual labor supply data. To understand how flat taxes affected work incentives, I calculate the hours worked after the adoption of the flat tax in each of the eight countries by adjusting the tax rate τ to its appropriate level. From the calibration, I know the value of α, and I solve the equation for h. I compare the estimated labor supply in the first year of the flat tax to the labor supply in the year prior to the flat tax. An increase (decrease) in hours worked under the flat tax implies that the flat tax increased (decreased) incentives to work. When evaluating changes in work incentives, I hold all parameters constant except for the tax rates and levels of income to which they apply. All other inputs income distribution, GNI, population are held constant to understand how flat taxes affect the existing labor force. Perhaps most consequentially, I hold c y constant such that the estimated labor supply is 8 When evaluating incentives to work, the income distribution statistics have income share divided by quintiles for Estonia, Lithuania, Russia, Ukraine, and Romania, and deciles for Latvia, Slovakia, and Georgia. Each quantile, by definition, has the same number of people.

31 Easterbrook 30 conditional on c y. Conditioning on c y has two implications for the model. The first implication is that the estimated change in hours worked is similar to the compensated wage effect of a tax reform. Second, a constant c y means that individuals believe that they can continue consuming at the same amount relative to output as they did before the tax reform. The continued consumption levels are the result of the income effect. In practice, the government may be returning large portions of tax revenues to the taxpayers in lump sums. I will elaborate on the implications of conditioning on c y in section 6.1. I also use the structural relationship in equation (13) to estimate labor supply for multiple years for each country in order to test the appropriateness of the Prescott model. When estimating h for numerous years, I allow all variables to fluctuate to their respective annual levels: v q, GNI, r, c, τ y hi, 9 and τ c. For each year, I determine the value of τ, adjust c, and y use the calibrated α to solve equation (13) for h. I compare the estimated values of h to the actual values of labor supply. Section 6 describes the results of both exercises. 4.5 Limitations of the Model Critics of Prescott s model usually argue that the model overestimates the effect of taxes on labor supply and fails to capture many determinants of labor supply. First, the model does not account for other policies that affect the labor market. Ljungqvist and Sargent (2005) argue that unemployment benefits should be included in an explanation of labor supply, particularly one that attempts to explain why Americans work more than Europeans. The Precott model does not incorporate unemployment benefits. According to Ljungqvist and Sargent (2005), the exclusion of unemployment benefits overstates the effect of taxes and oversimplifies the model. A second type of criticism focuses on the high labor supply elasticity of the Prescott model. By assuming individuals preferences have a logarithmic form, the model inherently suggests that labor supply is quite elastic. Additionally, holding c y constant reduces the 9 τ hi = τ pi + τ wi.

32 Easterbrook 31 income effect associated with tax reforms. The uncompensated wage effect has a relatively larger effect on labor supply such that people work more with lower tax rates. For instance, a lower tax rate may mean individuals receive smaller transfers. They work more as a result of lower taxes and smaller transfers. Alesina, Glaeser and Sacerdote (2005) describes the discrepancy between the labor supply elasticity that Prescott s model implies and the elasticities found in micro-studies; the elasticities in micro-studies tend to be much smaller. Alesina et al. (2005) acknowledge that the micro-studies might underestimate labor supply elasticity because the calculations do not account for the indirect effect that taxation and higher government transfers have on labor supply. Nonetheless, Alesina et al. generally do not support Prescott s model, arguing that the labor supply elasticity is too high to be realistic. Prescott responds to these criticisms by arguing that empirical evidence corroborates the model. He argues that most labor supply differences between Europe and the U.S. can be explained by the higher tax rates in Europe. According to Prescott, the high elasticity of his model a labor supply elasticity of nearly 1 is reflection of reality: The differences in marginal tax rates and labor supply are large: Canada, Japan, and the United States have rates near 0.40, and France, Germany, and Italy have rates near The prediction based on a substitution elasticity of one is that Western Europeans will work one-third less than North Americans and Japanese. This prediction is confirmed by observations. Added evidence for this substitution elasticity is that it explains why labor supplies in France and Germany were nearly 40 percent greater during the period than they are today. These countries increased their marginal effective tax rate from 40 percent in the early 1970s to 60 percent today (Prescott 2006, 225). For a summary of labor supply elasticities found in other studies, see Alesina et al. (2005).

33 Easterbrook 32 Another limitation of the model relates to its simplicity. Prescott assumes individuals have homogenous preferences and productivity. He assumes perfectly competitive markets prevail and wages equal marginal productivity. The model does not incorporate labor regulations that affect employment rates and the number of hours employed people work. Such regulations include hiring and firing policies, minimum wage laws, and laws governing vacation time and unions rights. The simplifications may come under especial scrutiny when evaluating countries transitioning from centrally-planned economies. Prescott (2004), however, finds the model appropriate to describe labor supply in Western Europe despite substantial labor market regulations. Thus, I use the model to estimate labor supply in Central and Eastern Europe. Of course, the simplifications partially explain why the model s estimations are not perfectly accurate. 5 Data The data I use in my study can be divided into three groups: (1) tax code data, (2) national statistics, and (3) labor supply data. I use tax code data to determine the average marginal tax rates and national statistics primarily to compute the proportion of people in each tax bracket. I use labor supply data to calibrate the model and compare the estimated hours worked to actual data. I will elaborate on the sources and statistics for each subset of data accordingly. Although the sources of data I use are reliable, data from Central and Eastern Europe are subject to more error in estimation than data from advanced countries. Since the collapse of the Soviet Union in 1991, the countries in Central and Eastern Europe have undergone political and economic restructuring. It is difficult to obtain precise data from these countries when the data points are rapidly changing. Furthermore, the governments of these countries have fewer resources than the governments of developed countries to devote to precise calculation of economic indicators.

34 Easterbrook Tax Code Data PricewaterhouseCoopers annual publication Individual Taxes, Worldwide Summaries is my primary source of tax code descriptions. It contains four-to-five page summaries of the individual taxes levied in over one hundred countries. PricewaterhouseCoopers published an edition of the series nearly every year that I require tax data. (There is no 2006 edition; I extend the rates for 2005 when estimating In some cases, another source indicates tax code changes in I incorporate any changes.) For my study, I use PricewaterhouseCoopers information on personal income taxes and levels of the tax brackets, personal exemptions, and payroll taxes, including the legal tax obligations of the employer and employee. Individual Taxes, Worldwide Summaries also describes how personal income taxes are computed and whether individuals can deduct payroll taxes when calculating their taxable income. I use this information when determining the income levels that serve as upper bounds for each marginal tax rate. Lastly, the publication provides descriptions of the tax base, tax credits, property taxes, and tax administration. My study does not incorporate this data, although more complex analyses may require such detail. I supplement data from Individual Taxes, Worldwide Summaries with data from national sources for four countries: Estonia, Russia, Slovakia, and Georgia. No edition of Individual Taxes, Worldwide Summaries contains information on Estonia s 1993 tax code. Data for Estonia come from the Estonian Embassy in the U.S. Personal income tax data for Russia comes from Ivanova et al. (2005) and the Russian Embassy to Canada. The tax data I use for Slovakia come from the Slovak Ministry of Finance. Individual Taxes, Worldwide Summaries does not contain information on Georgia s tax codes. Instead, I use tax codes descriptions from Keen et al. (2006) and Alvin Rabushka s comments on Georgia s adoption of the flat tax (Rabushka 2005). In general, I obtain information on value-added tax rates from Keen et al. (2006). Appendix D describes the statutory tax information that I use in my study.

35 Easterbrook National Statistics In addition to tax code data, I use several economic indicators to calibrate the model ( c y ) and calculate the proportion of the population in each tax bracket (GNI, population, and income distribution). The structural relationship in equation (13) depends on the ratio consumption-to-output. The variable c y is the household final consumption expenditure as a percentage of GDP from the World Bank s World Development Indicators database. The database is a compilation of statistics from the World Bank s annual reports World Development Indicators and is the World Bank s largest database. World Development Indicators contains information on national accounts, social indicators, the environment, balance of payments, and finance. The World Bank defines final household consumption as the market value of all goods and services purchased by households, including durable goods and rent, but excluding real estate purchases. Since c y is a ratio, it is comparable to data denominated in local currency units at current prices. To calculate the proportion of people in each tax bracket, I use data on income distribution, gross national income, and population. The income distribution statistics are from the World Income Inequality Database 2 (WIID2), a compilation of data by the World Institute for Development Economics at the United Nations University. WIID2 combines income distribution data from multiple sources, including the World Bank, the Luxembourg Income Study, Transmonee Data from UNICEF/ICDC, and Central Statistics Offices. The data series collected includes Gini coefficients, quintile and deciles shares of income and consumption, and mean incomes. Unfortunately, WIID2 does not report all types of data for every year. Ideally, I would use a data series for the last year prior to the flat tax that had quantile distributions of people s gross incomes for each country. However, quantile shares of income are not available for every year. When WIID2 publishes quantile shares in applicable years, the shares often refer to shares of disposable income rather than gross income. In the cases of Ukraine and Romania,

36 Easterbrook 35 the quantiles I use refer to shares of consumption, as income quantiles were not available in relevant years. Despite these limitations, WIID2 contains income/ consumption shares for quantiles in relevant years for all countries in my study. Where statistics are unavailable in particular years, I either hold income distribution constant or linearize the income shares from years before and after the year of missing data. See Appendix E for more detail. Based on the quantile income shares, I approximate a lognormal income distribution for each country using gross national income and population. I obtain data on gross national income and population in the relevant years for each country from the World Bank s World Development Indicators database. I obtain statistics for gross national income denominated in current local currency units. These are the units I need to estimate the proportion of people in each tax bracket since tax bracket levels apply to current local currency figures. I make two adjustments to the gross national income figures. First, I multiply all of Romania s GNI figures by 10,000. In July 2005, Romania adopted a new leu; one new leu is equal to 10,000 old lei. Romania had previously suffered from high inflation, and the revaluation was intended to align the leu s value more closely with the euro and dollar. The World Bank s GNI figures do not reflect this currency change. Since tax bracket levels apply to old leu levels, I adjust GNI. Second, Russia s ruble depreciated substantially when Russia defaulted on sovereign debt in Russia adjusted its tax bracket levels to appropriate levels after the devaluation. Gross national income figures from the UN reflect the devaluation of Russia s currency better than those from the World Bank. I use the UN s figures of gross national income at market prices for Russia. I check the World Bank s GNI figures against those from the UN. Except for the two cases mentioned, all figures from the two sources have very similar magnitudes.

37 Easterbrook Labor Supply Data For each country, I calibrate the model using the actual average number of hours individuals worked per week in the year prior to the flat tax. For several other years before and after a country implemented a flat tax reform, I compare the model s estimations of average hours worked to the actual average number of hours worked. I obtain data on actual hours worked from the International Labor Office s online database LABORSTA. The International Labor Office is an agency of the United Nations. The LABORSTA database contains annual statistics on the economically active proportion of the population, employment, unemployment, wages, consumer prices, and hours worked. When necessary, I supplement this data with statistics from national sources. I use LABORSTA s statistics on hours actually worked and the proportion of the population that is economically active and employed to calculate average hours worked. Using the same statistic as Prescott (2004), I calculate average hours h worked as h = h e ea, (16) where h e represents that average hours that an employed person works and e represents the employment rate of people in the labor force. e represents the proportion of working age people that are economically active. For the majority of country statistics, working age people refers to all individuals who are older than 15 years. In statistics for Estonia, working age people are those between the ages 15 and 69; for Ukraine, they are between the ages 15 and 70, and for Latvia and Romania, they are between ages 15 and 74. I calculate the values for e (ea) by dividing the number of economically active (employed) people by the total population of the respective age range. LABORSTA s values of h e come from national surveys that each country s official statistics office conducts. In statistics from Estonia, Russia, Slovakia, and Ukraine, hours worked refers to hours that employees worked, thus excluding self-employed individuals. Although

38 Easterbrook 37 not ideal, this fact does not affect intertemporal comparisons within these countries. I also use national sources for statistics on Estonia, Lithuania, and Latvia. The values of ea I use for Estonia come from Statistics Estonia, a public database sponsored by the Estonian government. For Latvia s ea statistics, I refer to the Central Statistical Bureau of Latvia. Additionally, I use the Statistical Yearbook of Lithuania to obtain values of both h e and ea for Lithuania. The Department of Statistics of the Government of the Republic of Lithuania publishes the Statistical Yearbook. I use the online database of population and social indicators from Lithuania s Department of Statistics as a second source. Unfortunately, average hours worked in Lithuania are only measured since Thus, I use 1997 data to calibrate the model for Lithuania. 6 Results The structural relationship in equation (13) suggests that as countries lower their marginal tax rates on income, labor supply increases, conditional on c. In adopting flat tax reforms, y six of the eight countries Lithuania, Russia, Slovakia, Ukraine, Romania, and Georgia lowered their marginal tax rates on income. The structural relationship implies that individuals in these countries experienced increases in incentives to work. Since c y is constant when evaluating work incentives, the model implies that the uncompensated wage effect dominates the income effect at a macro level of analysis. One of the more interesting results is that marginal tax rates on income remained relatively high even after the implementation of the flat tax. With the exceptions of Russia and Georgia, marginal tax rates on income ranged from approximately 60% to 75% after the eight countries adopted flat tax reforms. At first glance, it is also surprising that flat taxes in Estonia and Slovakia did not improve incentives to work appreciably. I find that work incentives did not increase because Estonia s average marginal tax rate actually increased with the adoption of the flat tax, and Slovakia s

39 Easterbrook 38 marginal tax rate fell by only 1.67 percentage points. The decrease in Slovakia s marginal tax rate is relatively small when compared to the decreases in other countries. This finding may be initially surprising because the flat taxes in Estonia and Slovakia have been high-profile reforms. Actual labor supply statistics suggests the model estimates labor supply with mixed accuracy. The model s estimations are reasonable approximations of labor supply in five of the seven countries for which enough data exist to make a comparison: Estonia, Lithuania, Latvia, Russia, and Slovakia. In four countries, Estonia, Lithuania, Latvia, and Slovakia, the model s estimations are, on average, within 1.5 hours of actual labor supply statistics. The difference between actual and estimated labor supplies is larger for Russia, but the model approximates the trends in the actual data. When comparing the estimated and actual labor supplies, I also find that the model often overestimates labor supply changes. Table 3 summarizes my findings. Table 3: Results Country τ τ h Is the model Avg. error 1st year of flat tax (% points) (hours) a decent fit? (hours) Estonia 64.1% yes 1.4 Lithuania 65.6% yes 1.0 Latvia 68.5% yes 1.1 Russia 55.9% yes, follows trend 2.2 Slovakia 70.3% yes 0.5 Ukraine 60.8% unclear 1.0 Romania 76.1% no 4.4 Georgia 42.4% na 10 na 10 Not enough labor supply figures from Georgia are available to comment on how well the model fits the actual data.

40 Easterbrook Implications of Conditioning on c y The Prescott model considers h conditional on c. To understand how flat taxes changed y work incentives, I hold c y (and the income distribution) constant and adjust the tax rate τ. As I have mentioned, the model s estimation of labor supply is similar to the compensated effect of a tax reform. In contrast, I adjust c y (and the marginal tax rate) to its appropriate annual statistic when I evaluate how well the model s estimations of labor supply fit actual labor supply data. It is worth discussing how the use of c y affects the model s estimations because it is a nonstandard way to construct a labor supply model. Labor supply models usually focus on wages and non-labor income as determinants of labor supply. Output is introduced into the Prescott model under the assumption that wage is equal to the marginal product of labor and competitive labor markets prevail. Figure 2 illustrates the relationship between c and h in the model. A lower c implies y y a greater number of hours worked and larger labor supply. Each curve represents the relationship between c y and h for a specified tax rate. In Figure 2, the upward shift of the curve represents a reduction in the marginal tax rate on income. To approximate how incentives to work changed with flat tax reforms, I calculate the magnitude of the shift for a given c. y Shifts along the curve represent any changes in the economy that affect the consumptionto-output ratio. Changes in saving and investment behavior could be the reason the economy moves from one point to another on a particular curve. With a higher capital stock, labor is more productive and earns higher wages. Higher wages will likely affect consumption behavior via income and uncompensated wage effects, but whether consumption increases or decreases is ambiguous. Output also increases under the scenario of higher capital stock, and thus, it is likely that c y changes. Since c y captures changes in investment and output, movements along the curve can represent business cycle effects. I allow for movements along the curve when I evaluate the model s goodness of fit over

41 Easterbrook 40 Figure 2: The Relationship Among c, h, and τ y time. I account for changes in consumption and output by adjusting c y to its appropriate annual statistic. Figure 3 illustrates the simultaneous shifts of τ and c. The adjustment y of c y in the year of the flat tax reform also allows for a more varied income effect. When estimating changes in work incentives, I hold c y constant. However, taxpayers may not continue consuming at previous levels after the tax reform. By allowing c y to adjust in each year, I am able to capture changes in consumption behavior. 6.2 Flat Taxes Effects on Incentives to Work The structural relationship among h, τ, and c y suggests that the marginal tax rate on income decreases and the average number of hours that individuals work per week increases in six countries, conditional on c y. I interpret the estimated increase in labor supply to mean incentives to work increased. Estonia and Latvia are the exceptions; work incentives did not improve in those countries because marginal tax rates did not decrease under the flat tax. Across Central and Eastern Europe, a decrease in the tax rate of one percentage point causes individuals to work, on average, approximately 0.47 more hours per week according to the model. Figure 4 shows the relationship between the change in marginal tax rate and

42 Easterbrook 41 Figure 3: The Relationship Among c y, h, and τ When c y Changes estimated labor supply. Table 3 has the figures for changes in τ and h shown in Figure 4. See Appendix C for the marginal tax rates and estimations of hours worked before and after the implementation of flat taxes. Figure 4: The Relationship Between τ and h Deviations from the estimated increase of 0.47 hours per one percentage point decrease in the marginal tax rate are primarily due to different labor/ leisure preferences and different values of c. Countries that have a lower α have larger labor supply gains when tax rates y decrease, ceterus paribus. A lower α implies that the country s residents are less averse to working. Most countries have an α are between one and two, which is right in line with 1.54,

43 Easterbrook 42 the value of α Prescott (2004) uses. Figure 5 shows a practically perfect linear relationship between α( c ) and the magnitude of the labor supply increase when tax rate decreases by y one percentage point. Table 4 shows the values of α, the percent changes in τ and h with the flat tax, as well as the labor supply elasticity with respect to wages. Figure 5: The Effect of α and c y on Labor Responsiveness Table 4: Preferences, Percentage Changes, and Labor Supply Elasticity α % Decrease in τ % Increase in h Labor Supply Elasticity (% h/% τ) Estonia Lithuania Latvia Russia Slovakia Ukraine Romania Georgia It is worth noting that marginal tax rates in Eastern and Central Europe remained high even after the adoption of the flat tax. Prior to adopting flat taxes, these countries marginal tax rates on income ranged from approximately 60% to 85%. Since they adopted flat taxes, marginal tax rates ranged from 60% to 75%, except for Russia and Georgia, which had rates

44 Easterbrook 43 of 56% and 42%. The high tax rates on income are primarily due to high payroll taxes, which are over 30% in all countries but Georgia, 11 and VATs between 17% and 20% in all countries. Contributing to high τ s, the Baltic states did not implement flat taxes with particularly low tax rates. The remainder of this section will discuss how the flat tax affected each country s marginal tax rate and its residents incentives to work. The model suggests that residents of Romania experienced the greatest increases in work incentives under the flat tax. This is due to a highly elastic labor supply a 1% decrease in the tax rate increases labor supply by 3.78% and a sizable decrease in the marginal tax rate of over 8 percentage points. Romania has the lowest value of α, 0.51, indicating that the countries residents do not strongly dislike work. As a result, the model suggests Romania experienced a 36% increase in labor supply with the adoption of the flat tax. After Romania, Georgia s residents experienced the largest increase in work incentives due to the flat tax. Georgia s marginal tax rate decreased by over 17 percentage points, and the structural relationship estimates that labor supply increases by 6.3 hours. The changes amount to a 29% decrease in the marginal tax on labor and 32% increase in labor supply. The decline in the marginal tax rate is the result of changes to PIT and payroll taxes. First, the marginal PIT rate decreased by 5 or 8 percentage points for most people when Georgia adopted a flat tax of 12%. Prior to the flat tax, 87% of the population faced PIT rates of 17% or 20%. Second, payroll taxes decreased from 33% to 20% in If only PIT rates changed and payroll tax rates were constant, labor supply would only increase by 12%. According to the model, the substantial decrease in marginal tax rates boosts labor supply proportionately. Unfortunately, no actual labor supply data corroborate the estimated increases of Romania and Georgia. Actual labor supply statistics for Georgia are simply unavailable for years 11 Georgia had a 20% payroll tax in The highest payroll tax rates are over 30% in all other countries. Slovakia, Ukraine, and Romania have caps on certain payroll taxes such that the payroll tax rate is less than 30% for some levels of income. Russia has a regressive payroll tax, so higher levels of income have a marginal rate less than 30%.

45 Easterbrook 44 after the implementation of the flat tax. Actual labor supply statistics for Romania indicate that labor supply did not increase in I will elaborate on the discrepancy in the next section. One explanation for the discrepancy is that the model s parameters are incorrect. Romania s labor supply is highly elastic, even when compared to other countries elasticities in the Prescott model. As I discussed, the Prescott model assumes a highly elastic labor supply. It is questionable whether the elasticity and magnitude of Romania s expected labor supply increase is realistic. Initially, it seems surprising that Lithuania s residents should experience one of the largest increases in work incentives. The model estimates a labor supply increase of 5.0 hours or 32%. Of the eight countries, Lithuania implemented the flat tax with the highest rate of 33%, and 33% was the rate of the highest tax bracket in Lithuania prior to the flat tax. However, the amount of the personal exemption increased more than three-fold under the flat tax. Hence, the marginal tax rate decreased. Additionally, Lithuania s labor supply elasticity with respect to wages is This elasticity may not be realistic, and unfortunately actual labor supply data only exist for years after Lithuania implemented the flat tax such that I cannot compare the estimated and actual labor supply changes from 1993 to Although the magnitude is uncertain, the large decrease in the marginal tax rate suggests that work incentives increased. Work incentives also improved in Russia under the flat tax. Both the reductions in PIT and payroll tax rates lowered the marginal rate and thus increased labor supply. In addition to adopting a 13% flat tax on personal income, Russia implemented a regressive payroll tax in If only PIT rates had changed, the labor supply would have increased by 3.6%. Since the payroll taxes also decreased, the model estimates labor supply will increase by 14.5%. Reductions in PIT rates are responsible for the increased work incentives and estimated labor supply increase in Ukraine. 90% of the population experienced a decrease in the PIT

46 Easterbrook 45 rate from 20% to 13%. The marginal tax rate on income fell 6 percentage points. The model estimates labor supply will increase 2.5 hours or by 14.5% as a result. Slovakia experienced a small increase in work incentives under the flat tax. The marginal tax rate fell only slightly such that the model estimates only a small increase in labor supply of 0.8 hours or 5%. Prior to the flat tax, approximately 25% of the population faced a 10% PIT rate, 43% faced a 20% rate, and 27% faced a 28% rate. Thus, the implementation of a 19% flat tax did not change the average marginal tax rate appreciably. This finding might be initially surprising since Slovakia s flat tax has drawn much attention and is considered to be the closest to the Hall-Rabuskha flat tax. The model suggests that work incentives did not increase in Latvia and Estonia when they implemented flat taxes. Latvia experienced practically no change in tax rate with the adoption of the flat tax. Prior to adopting the flat tax Latvia had a regressive tax system, in which an estimated 99% of the population fell into the tax bracket with a 25% rate. A flat tax of 25% did not change the marginal tax rate for most taxpayers. The model suggests that labor supply will also remain nearly constant in Latvia. Finally, the results imply that work incentives in Estonia decreased under the flat tax. Estonia s marginal tax rate on labor increased by over 3 percentage points with the adoption of the flat tax. Prior to the flat tax, 59% of the population faced the marginal PIT rate of 16%. When Estonia adopted the flat tax of 26%, 54% of the population experienced an increase of 10 percentage points in its marginal rate. (5% of the population saw its marginal rate decrease to 0%.) The increase in PIT for 54% of the population outweighs the decrease in PIT from 33% to 26% that 20% of people experience. Thus the model estimates a decrease in labor supply of 6.3%.

47 Easterbrook Goodness of Fit: Estimated Versus Actual Changes in Labor Supply The model accounts for labor supply shifts reasonably well for five of the seven countries for which there are enough data to make a comparison. Since the model approximates actual labor supply for five countries, it is likely to be useful to understand how flat taxes affected work incentives and labor supply in certain cases. The conditions under which the model provides reasonable estimations of labor supply are uncertain. However, I hypothesize that the economic stability of the country and quality of the actual statistics affect the model s ability to estimate labor supply for that country. I use two criteria to evaluate the goodness of fit: (1) how close the model s estimations are to actual labor supply data and (2) how well the trends in the estimations mirror trends in actual data. The average error, or average difference between the model s estimations of hours worked and actual labor supply statistics, is less than 1.5 hours for Estonia, Lithuania, Latvia, and Slovakia. I consider the model to estimate labor supply well for those countries. Ukraine s error is also less than 1.5 hours; however the goodness of fit is unclear because the model does a poor job of approximating trends. Although the difference between actual and estimated hours exceeds two for Russia, the model accurately estimates labor supply trends such that I consider the model a decent fit. The model does not estimate the labor supply well for Romania. I will compare the model s estimations to actual labor supply figures below. Table 5 summarizes how well the model fits the actual labor supply data. The model overestimates the labor supply changes after the adoption of flat taxes in the five countries. (See Table 5.) The overestimation is particularly interesting since one criticism of the Prescott model is that it implies the labor supply is more elastic than many studies suggest. Figure 6 shows the absolute values of the labor supply elasticities implied 12 The error is the average absolute value of the difference between actual and estimated labor supply data. I include all years for which I have labor supply data in the calculation, excluding only the year for which I calibrate the model. For instance, I exclude the error for 1993 from the calculation of Estonia s average error. By design, the error for 1993 is 0.

48 Easterbrook 47 Table 5: Goodness of Fit Country Is the model Error Error Does the model a decent fit? (hours) 12 (% of labor supply) overestimate h? Estonia yes % yes Lithuania yes % yes Latvia yes % yes Russia yes, follows trend % yes Slovakia yes % not consistently Ukraine unclear % in 2004 Romania no % yes by the Prescott model. The tendency of the model to overestimate labor supply changes provides some support for this criticism. Figure 6: Labor Supply Elasticities Implied by the Prescott Model Errors in estimation also reflect the fact that the model does not incorporate all factors affecting labor supply. Unless consumption and output capture the full economic effects of a currency change, the model does not show the effects of Romania s introduction of a new leu in The model may not describe the full economic effects of Russia s sovereign debt default in Nor does it show the full economic impact of Georgia s Rose Revolution in 2003 and its major economic reforms. The absence of economic reforms in the model may be particularly important as I am examining transition economies. Unfortunately, labor supply data are not available for Lithuania and Georgia in periods