Tax revenue volatility and
the government budget in
the Netherlands
Master Thesis Economic Policy
Author:
Jenny
Boelens
Student number: 1210815
1
stsupervisor:
prof.dr.mr. C.A. de Kam
2
ndsupervisor :
dr. G.H. Kuper
Abstract
The paper looks into possible explanations of tax revenue volatility in the Netherlands and discusses some of its consequences. In particular, volatile tax revenues can pose a problem because
unexpected revenue changes threaten the process of balanced budget setting and may lead to political turmoil. A cross-country analysis of potential causes of tax revenue volatility, using data for 23 European countries over the 1995-2004 period, shows that the make-up of national tax mixes explains some of the observed volatility. Specific features of tax systems, such as the deductibility of mortgage interest paid and loss compensation for firms do not contribute to the explanation of the observed variation in tax revenue volatility. Furthermore, data for 1972-2005 suggest that the revenue of the personal income tax and the corporation income tax, respectively, moves with the business cycle and the interest rate. It is concluded that the results reported in this paper offer no building blocks for improving the tax revenue estimation procedures currently employed by the Dutch Ministry of Finance. Although there seems to be some scope for refining the current procedures, there is a problem with the availability of the data required.
Table of Contents
1. Introduction...4
2. Volatility and predictability...5
2.1 Definitions...5
2.2 Measures of volatility and predictability...6
2.3 Consequences ...7
2.4 Solutions...8
2.5 The relationship between volatility and predictability ...11
3. The causes of volatility: a cross country analysis ...13
3.1 Introduction...13
3.2 A note on methodology ...18
3.3 Results...19
3.4 Conclusion...23
4. The causes of volatility: an analysis for the Netherlands ...24
5. Budget estimation...28
5.1 Current estimation methodology...28
5.2 Area’s for improvement ...29
5.3 Improvements in the current methodology ...31
5.4 Corporate Income Tax...34
5.5 Alternative models...37
6. Conclusion ...44
References ...46
Data sources ...48
1. Introduction
According to two recent publications1 drafted by staff of the International Monetary Fund (IMF), the average volatility of Dutch tax revenue has increased over 1995-2004 in comparison with previous periods. This increase in volatility has occurred while on average tax revenue volatility declined in other European countries. This could mean that the Dutch tax system has become more sensitive to income fluctuations and other possible shocks affecting tax revenue.
As long as fluctuations in tax revenue are predictable, tax revenue volatility does not necessarily constitute a problem. Tax revenue fluctuations over the business cycle may be desirable as they work as automatic stabilizers of the economy. Because aggregate income fluctuates over the business cycle, the tax-base and tax revenue will also move. During economic booms tax revenue increases relative to economic growth, which dampens the economic upswing. During recessions, the economy is stimulated because tax revenue decreases relative to economic growth. Thus, fluctuations in tax revenue stabilize the economy as a whole.
Unfortunately, the Dutch government frequently experienced large windfalls and setbacks in tax revenue. Most recent setbacks occurred in 2002 and 2003, large windfalls occurred in 1997, 1999 and 2000. This implies that fluctuations in tax revenue have not been easy to predict. Larger fluctuations or volatility in tax revenue may increase unpredictability, with significant consequences for the government budget and political stability. Therefore, this thesis will discuss volatility and the unpredictability of tax revenue.
The next chapter starts with the definitions and measures of tax revenue volatility and unpredictability. Subsequently, I will discuss their consequences and possible solutions. In the last section of chapter 2 I will study the relationship between tax revenue volatility and unpredictability. Chapter 3 analyses factors that could explain the differences in tax revenue volatility between countries of the European Union. Chapter 4 studies variables that could explain volatility in the Netherlands. Chapter 5 is devoted to an analysis of current tax revenue estimation methods of the Dutch Ministry of Finance. Additionally, it discusses improvements of the current methodology and finally it introduces two alternative models for tax revenue estimation purposes. Chapter 6 concludes.
2. Volatility and Predictability
2.1 Definitions
The term volatility is often used in the financial sector and most often refers to the standard deviation of fluctuations in the value of a financial instrument with a specific time horizon. This measure is used to quantify the risk of a particular financial instrument. With respect to tax revenue, volatility measures the fluctuations of tax revenue over time.
Not all fluctuations in tax revenue are considered as volatility. Fluctuations can be divided into four categories, namely those caused by business cycles, gradual increases/decreases in the take-up of fiscal provisions, economic shocks and discretionary measures.
First, the level of tax revenue fluctuates over the business cycle. With the economy booming, tax revenue is higher because employment rises, companies make more profits and consumption, especially of luxury goods, will increase. During recessions these effects are opposite, which creates a pattern of fluctuations over the business cycle.
The second category is fluctuations caused by gradual increases/decreases in the take-up of fiscal provisions, such as tax deductions and exemptions. Increases and decreases in the take-up of such provisions may lead to fluctuations in tax revenue and increase tax revenue volatility. An example is the gradual increase in private homeownership since 1970, which has increased the take-up of mortgage interest deduction and thus decreased personal income tax (PIT) revenue.
Economic shocks may also cause tax revenue fluctuations. Shocks, such as stock market crashes or oil shocks, may affect tax revenue both directly and indirectly. For example, a stock market crash can have a direct effect on PIT, as a wealth tax on the deemed return on stocks and other wealth is a part of this tax. The shock might also affect other economic factors, like consumer and producer confidence, hurting economic growth and thus tax revenue. Both direct and indirect effects may increase tax revenue volatility.
The last category consists of fluctuations due to the effects of discretionary measures. Discretionary measures are the result of fiscal policy and typically include changes in tax rates or tax-bases. Therefore, fluctuations caused by such measures do contribute to volatility.
2.2 Measures of volatility and predictability
Tax revenue volatility and predictability can be measured by different indicators. To measure volatility one could use the macro progression factor (MEP) and the Coefficient of Variation (CV). Predictability can be measured by different versions of the inequality coefficient (IC).
The MEP measures the change in tax revenue when the Gross Domestic Product (GDP) changes with one percent. It indicates the sensitivity of tax revenue with respect to income changes. The values of short and long run progression factors may differ considerably from each other. Over a long period the average value of the progression factor equals one, in the short run the progression factor can fluctuate significantly. In periods of low economic activity the progression factor is generally lower than one. Due to possible shifts from highly taxed consumption commodities to lowly taxed commodities, Value Added Tax (VAT) revenue decreases. Also, direct taxes tend to decrease relatively more than GDP in such periods. In times of high economic activity these effects are opposite, so the progression factor is often higher than one is such periods.
The standard deviation of tax revenue, the second indicator of volatility, is often used in the financial sector to quantify the risk of financial instruments. It is not possible to compare the standard deviation across different tax categories or different countries. Therefore, it is often standardized into the CV, which is the standard deviation of a variable divided by the mean of this variable. High values of the CV imply high fluctuations in tax revenue and thus high volatility.
The MEP is the most useful indicator if one is interested in the effects of economic growth on tax revenue. If one is more interested in the total volatility of tax revenue, the CV is more suitable. Therefore, which indicator is most useful depends on the purpose of one’s analysis. I will use the CV to measure volatility in this and the third chapter because I am interested in total tax revenue volatility. The CV should be interpreted with some carefulness. Normally, a high CV implies high volatility. However, there are situations in which the interpretation of this coefficient is slightly more complicated. Both the staff of the IMF and I calculate the CV with tax revenue series divided by GDP. This is not uncommon in economic literature as this method corrects for economic growth. However, the use of such series may create situations in which tax revenue volatility is low, while the CV indicates otherwise. A situation in which tax revenue is stable and GDP fluctuates produces a variable tax/GDP ratio and a high value of the CV. The value of the CV may even be higher than in a situation where tax revenue and GDP are both volatile, as the tax/GDP ratio is less variable. Although the CV indicates a higher volatility, the first situation may be more desirable. Fortunately, in practice tax revenue fluctuates more than GDP. Therefore, it seems justified to use the CV.
Pt is the prediction of a variable in year t, Rt is the realized value of that variable in year t and P*t is the
alternative prediction of the variable in year t. This coefficient calculates predictability, whereby large errors have a relatively high influence on the level of the coefficient because squared errors are used. The estimation errors are standardized by the use of an extra alternative estimation term in the denominator. Standardization is useful because it allows for the fact that variables with a stable development are easier to predict than those with more unstable developments.
The most used form of this coefficient is based on using the realization values. This is the Theil2
coefficient, whereby the alternative estimation term disappears from equation 1. Basically, this coefficient calculates the normal prediction error over a period relative to realization values of the tax category in question.
In the second version of this coefficient the alternative prediction is the value of the previous year. It considers differences in the stability of the development of tax revenue, which influences the predictability of tax categories. Therefore, this coefficient is a better indicator of predictability in comparison with the Theil coefficient. I will use this second version of the IC to measure predictability in section 2.5.
2.3 Consequences
Tax revenue volatility is a likely cause for tax revenue unpredictability. A government will experience larger financial windfalls and setbacks if the unpredictability of tax revenue increases. Windfalls may cause problems, because they lead to increased political pressure to increase government expenditure or decrease the tax burden. Because governments have to decide if and how they want to spend the extra money, the pressure and the potential changes in expenditure or tax burden may lead to political and administrative unrest. This creates uncertainty for the financial situations of citizens and companies, which might influence their consumption pattern or other economic behavior and affect the economy negatively.
Setbacks may cause problems for the budget because they create lower surpluses or higher deficits. A single setback will have minor budgetary consequences, if a government has sufficient reserves. When a government does not have substantial reserves, it may need to cut expenditures or increase tax burdens. Politicians will be hesitant to execute such measures because they will face a lot of resistance and opposition from the public and opposite politicians. Also, it will decrease the popularity of the politicians in question, so especially during or close to election times these measures are not likely to be executed.
The government budgets of small economies are usually more vulnerable to fluctuations in tax revenue than government budgets of larger economies. Their budgets are more inflexible and less diversified than those of larger economies. These economies are especially vulnerable for large setbacks because some small economies do not have enough resources to compensate. This might actually endanger the level of services offered by governments, so the consequences of volatility and unpredictability may be very large for these economies. For this reason, most literature on tax revenue
volatility is focused on small economies like small island states and the states of the United States of America (USA).
McAleer et. al. (2005) and Purfield (2005) studied the revenue volatility of the small island states Maldives and Kiribati. The tax revenue of these economies is almost completely dependent on one single source, respectively tourist tax and fishing license fees. The underlying tax-bases are narrow and very volatile, so tax revenue for these economies is very sensitive to shocks like natural disasters. A recent example is the tsunami of December 2004, which caused severe damage to the tourist based economy of the Maldives and reduced the number of tourists visiting in the post tsunami period significantly.
Although less than those of the Maldives and Kiribati, the budgets of the USA state and local governments are also vulnerable to fluctuations in tax revenue. Most expenditure of these governments is inflexible3, and their revenue possibilities are limited. Serious budgetary crises could arise if financial windfalls are not used to create reserves to compensate for financial setbacks. This could threaten the services offered by the state and local government. These services are of great importance to the society, therefore a stable government budget is important.
Articles concerning the issue of volatility and the stability of government budgets for specific states are, among others, Vasche and Williams (2005) (California) and Matthews (2005) (Georgia). Examples of more general articles on tax revenue volatility are the articles by Sobel and Holcombe (1996) and Groves and Kahn (1952). Both articles focus on the effects of instability or variability of tax revenue with respect to income or tax-base and not on volatility in general4.
As mentioned earlier, budgets of central and federal governments of larger economies are less vulnerable to fluctuations in tax revenue. Generally, these governments have access to capital markets or possess reserves to compensate for occasional setbacks to secure the level of government services. However, structural overestimation of tax revenue will lead to severe budgetary problems. In this situation, cutbacks in expenditures or an increase in the tax burden will be inevitable. Underestimation of tax revenue may also cause problems. Occasional windfalls will lead to the political and administrative unrest described above. If windfalls happen frequently, the risk exists that they will be considered to be structural. In this case governments might feel pressured to increase government spending structural or decrease the tax burden. This will have large negative consequences for the budget because these measures are hard to reverse.
2.4 Solutions
So the consequences of volatility and the unpredictability of tax revenue may be substantial for both large and small economies. However, they differ, so they require different solutions. One solution to decrease volatility is to choose a tax mix with taxes which are the most stable and the least sensitive for income changes. Garrett (2006) proposes a portfolio approach to find a tax mix that minimizes volatility. This approach originates from portfolio theory, a theory used in the financial sector that
3In most states the expenditures for services provided by tax funds, like schools police forces and fire departments, are
inflexible as these kinds of services are labour intensive, creating costs that continue from year to year
4Most literature on tax volatility for other countries like the Netherlands or other European countries also focus only on
proposes how rational investors will use diversification to optimize their portfolios. The author found that the tax mix of some states in the USA is not optimal because these portfolios rely too much on volatile taxes. A shift from more volatile taxes to more stable tax categories might help to increase the stability of government budgets.
The portfolio approach is not appropriate for small island economies like Kiribati and Maldives as these economies do not have much taxable commodities to choose from. If possible, it would be wise to diversify their tax mix, so they do not rely on a single source of revenue. However, the creation of some reserves would probably be a more realistic solution.
The portfolio approach might work better for state and local economies like the USA states. These governments have more available taxable commodities, so they are able to choose less volatile tax categories and create a stable tax mix. The portfolio approach could also work for central or federal governments of countries. However, the creation of an optimal tax mix can have large side effects. In most cases, it would imply a shift from direct to indirect taxes because, in general, indirect taxes are less volatile. Also, because progressive tax systems are more volatile, the portfolio approach might imply a reduction of the progressiveness of the tax system. From the perspective of public finance theory these measures and thus the portfolio approach may not be desirable.
Musgrave (1959) develops a multiple theory of the public household in which he identifies three different objectives of budget policy. These objectives are defined as the use of fiscal instruments to (1) secure adjustments in the allocation of resources, (2) secure adjustments in the distribution of income and wealth and (3) secure economic stabilization.
The first objective is concerned with achieving an optimal allocation of resources. Although the pricing mechanism of the market will achieve a large part of this allocation, there are market areas where this mechanism fails. In these areas, government intervention may be desirable. This could be achieved by intervention in the form of regulation, for example with respect to competition policy, or in providing services such as public transport or social security. Most expenditure to establish these interventions will not be very flexible. Therefore, governments need a stable source of resources. From the perspective of this objective, the portfolio approach is desirable.
economic stability. The least disruptive way to do this is to let the automatic stabilizers function5. There
are several articles6 written on the effectiveness of the automatic stabilizers in the Netherlands and other Organization for Economic Co-operation and Development (OECD) countries. The effectiveness of the automatic stabilizers in the Netherlands varies between 0.11 and 0.36 in the different studies, which means that between 11 and 36 percent of consumption shocks is absorbed. The progressiveness of a tax system and reliance on PIT and CIT are main causes for automatic fiscal stabilization. If the government would apply the portfolio approach to decrease volatility, they would also decrease the power of the automatic fiscal stabilizers. Therefore, the portfolio approach is also in conflict with this third objective of budget policy.
According to the analysis above, the portfolio approach is in conflict with two out of three government objectives identified by Musgrave (1959). Whether this approach should be implemented depends on how much governments value these objectives. A government with the main objective to keep public finances stable (American state and local governments), would probably arrange their tax systems so that volatility is minimized. For such governments the portfolio approach might be suitable. The Dutch central government is more interested in the use automatic fiscal stabilizers and/or a proper distribution of income and wealth reached through the tax system. Therefore, the Dutch government will not actively control for volatility by making shifts in the tax system at the expense of other objectives. Instead, it has to cope with volatility and unpredictability of tax revenue. This can be done in three ways.
First, governments can secure the level of services provided against fluctuations in tax revenue by creating reserves. These reserves can be created during economic booms, when the level of tax revenue is higher, to compensate for the lower level of tax revenue during recessions. This allows government revenue to fluctuate, without endangering the level of government expenditure. The standards for budget deficits/surpluses from the coalition agreements of 2003 and 2007 were respectively -0.5% GDP in 2007 and 1% GDP in 2011. The current government tries to create such reserves and/or pay back public debt.
Second, strict budgetary rules can help to manage windfalls and setbacks. The Netherlands already introduced a so-called trend-based budget policy in 1994, which helps managing windfalls and setbacks. This policy separates expenditures and tax revenue and maximizes government expenditure by using an expenditure framework with strict expenditures ceilings during a government term. This method, also called the Zalm7 standard, has decreased some of the political and
administrative unrest. Also, the standard decreases the political pressure to change the expenditure pattern, because windfalls in tax revenue can not be used to increase expenditures and setbacks do not have to be compensated with cutbacks in government expenditure8.
5 Stabilizing can also be achieved by manual increases in government by increasing/decreasing expenditure in bad/good times.
According to the Keynesian multiplier Kahn (1931) government spending can stimulate the economy in downturns of the business cycle and dampen the economy in booms. However automatic stabilization is less interruptive.
6 Examples are Van den Noord(2000), Barrell and Pina(2003), Brunila et al.(2003) and Scharnagl&Tödter(2004) 7 This standard is named after the former minister of Finance, Gerrit Zalm.
8
The last option is to study volatility and unpredictability. Because setbacks and windfalls occur only if the fluctuations in tax revenue are not expected, improving the estimates of tax revenue may decrease the occurrence of such windfalls and setbacks. Understanding the causes of volatility and unpredictability might help to improve the estimates.
2.5 The relationship between volatility and predictability
Tax revenue volatility is a serious problem if it decreases the predictability of tax revenue. To figure out to what extent the predictability of tax revenue is actually affected by volatility, I will study the relationship between tax revenue volatility and the predictability of taxes in this section.
Period Variable Total Tax VAT Duties CIT PIT
71-05 CV9 6 8 9 18 25 IC10 77 95 132 286 122 71-75 CV 5 2 6 8 9 IC 27 23 45 347 46 76-80 CV 1 6 6 7 1 IC 114 76 313 1711 273 81-85 CV 5 3 2 7 11 IC 68 137 205 199 191 86-90 CV 3 3 3 9 3 IC 53 155 66 469 159 91-95 CV 5 6 8 7 13 IC 46 68 131 453 79 96-00 CV 1 3 4 7 12 IC 67 89 132 272 118 01-05 CV 1 3 3 11 7 IC 105 120 122 227 75
Table 1: CV and IC of tax revenue over the period 1971-2005 Source: own calculations
Table 1 shows the coefficient of variation and the inequality coefficient of total tax revenue and four individual tax categories calculated over the period 1971-2005. The results in the first two rows of table 1 show that the most volatile taxes also possess the highest IC’s, with the exception of PIT. According to its CV, PIT is the most volatile tax category; however it’s IC is much lower than that of CIT. The volatility in PIT revenue can be explained by large fluctuations in the tax level due to large taxation reforms in 2001 and in the early 1990s. These reforms changed the tax/GDP ratio significantly, which is reflected in a high volatility coefficient in these periods and over the total period. Table 1 does not show a clear relationship between the volatility and predictability of tax revenue over time. To be able to study this relationship more thoroughly, I plotted the annual relative change in tax revenue and the relative prediction error11 for total tax revenue and the four tax categories included in
table 1. These graphs are in Appendix A, figures 1 through 5. I used the relative change and prediction error instead of the coefficients used in table 1, because these measures are more comparable to
9
The coefficient of variation is calculated with Tax/GDP ratio’s corrected for the effects of discretionary measures
10
The uncertainty coefficient is calculated with realization data and the one year estimates of the Budget, as presented in September in the year before (Miljoenennota)
each other in size. As table 1 shows, the values of the IC’s are much larger than the value of CV’s which makes it difficult to make clear graphs.
The first figure in appendix A shows the plot for total tax revenue. It shows that until 1982 the relationship between volatility and unpredictability was weak. After 1983, this relationship became stronger; in most cases when large upward fluctuations are experienced, tax revenue is underestimated. The opposite is also true.
Figure 2 and 3 in appendix A show the plots for respectively VAT and duties. Figure 2 shows that until approximately 1993 the relationship between unpredictability and volatility has been weak. After 1993 this relation became much stronger, but declined again after 2001. The plot for duties in figure 3 shows an entire different picture. The development of the duties is relatively stable; a change over 5 percent is almost never experienced. Nevertheless, the deviations are much larger and change from year to year. Although the development of the revenue of duties is stable, its revenue is hard to forecast.
The plots of corporate and personal income tax in figures 4 and 5 show that the relationship between volatility and the prediction error over time is weak. Both taxes are difficult to estimate because of their volatility, but the predictability is not necessarily different in times of high volatility. The plot for PIT shows some clear outliers in estimation deviations before 2000, after 2000 both change and deviations seem to be rather stable. The plot of CIT shows that the development of CIT revenue is volatile, however the prediction errors are even larger. These errors are almost twice as great as the changes in CIT. A possible explanation for this is that the development of CIT is not explained easily from the development of most economic factors.
3. The causes of volatility: a cross country analysis
3.1 Introduction
In their most recent publication12, the staff of the IMF observed an increase in tax revenue volatility in the Netherlands over the period 1995-2004. This while, on average, volatility decreased in the European Union as a whole. The staff of the IMF attributes this divergent development to greater swings of the business cycle in the Netherlands and to certain properties of the Dutch tax system, notably the deductibility of premiums paid into pension plans and of mortgage interest paid by home owners. It is possible that factors influencing these deductions have become more volatile over the past ten years, thereby increasing tax revenue volatility.
However, the Netherlands is not the only country that allows such deductions or experienced a marked business cycle. Therefore, in this chapter I will trace the impact of the business cycle and some relevant properties of the tax system on tax revenue volatility for 23 European countries over the period 1995-2004. The aim of this analysis is to find out whether countries with certain specific provisions in their tax system experience greater tax revenue volatility than other countries. The variables selected for this analysis are mortgage interest deductibility, loss compensation under CIT, deduction for premiums paid into pension plans, composition of the tax mix and output gap volatility. The countries studied are listed in appendix B. For some variables it was not possible to include all countries, because not enough data was available. This is also specified in appendix B. Data sources of this and other chapters are listed at the end of this paper. In the following I will first discuss each of the variables mentioned above and specify the regression equations, next I will discuss some econometric issues and finally I will present the results.
3.1.1 Mortgage interest deduction
Home owners in the Netherlands may deduct their mortgage interest paid for PIT purposes13.
Although several other countries allow for similar deductions, only Denmark, Finland, Greece (until 2002) and Norway also allow an unlimited deduction. Most other countries only grant the deduction up to a certain maximum amount or do not allow the deduction at all.
Mortgage interest deduction can have two different effects on the volatility of the tax-base of PIT and thus on PIT itself. First, it can have a destabilizing effect as changes or shocks in interest rates and mortgage volume affect the aggregate amount deducted and therefore PIT revenue. Also, gradual increases or decreases in the take-up of the deduction may impact PIT revenue. These effects are expected to be larger for countries that allow an almost unlimited deduction than for other countries. Second, the deduction might have a stabilizing effect on PIT revenue as the deduction dampens the effect of increases and decreases in the taxable wage sum. To find out whether countries with mortgage interest deduction experience a different PIT volatility, I will estimate the following equation:
12 IMF (2006b) 13
CVPITi,t = Ci,t + GAPi,t + ITRLABi,t +h0i,t + h1i,t + h2i,t + εi,t i = 1, …, N t = 1,…,T (2)
C = constant CVPIT = CV of PIT revenue
GAP = output gap
ITRLAB = Implicit Tax Rate (ITR) of labour
h0 = dummy variable that takes value 1 if a country has no mortgage interest deduction and 0
otherwise
h1 = dummy variable that takes value 1 if a country has limited mortgage interest deduction and 0
otherwise14
h0 = dummy variable that takes value 1 if a country has (almost) unlimited mortgage interest
deduction and 0 otherwise ε = error term
The index i represents the country and index t represents time. The CV is calculated as explained in chapter 2. However, instead of using the standard deviation over a period, I used yearly deviations. I used both the output gap15 and ITR as control variables. I have chosen to include the output gap as a
control variable because the position of the business cycle can influence the volatility of tax revenue. Furthermore, it was not possible to split fluctuations in tax revenue into those caused by discretionary measures and other causes. Data series should be corrected for such changes, as the effects of discretionary measures are not considered volatility. Unfortunately, for most countries, data series including such corrections were not available. Therefore, I included the ITR to control for the effects of discretionary measures, because the ITR reflects mostly changes due to such measures. I used the output gap and the ITR as control variables in all regression equations in this chapter.
To avoid exact collinearity, I have to omit one of the dummies. I have chosen to omit h0. This dummy
serves as a reference for the other two dummies, because the values of their parameter estimates are relative to the omitted dummy. Therefore, regression equation becomes as follows:
CVPITi,t = Ci,t +GAPi,t + ITRLABi,t +h1i,t +h2i,t + εi,t (3)
Where relevant, I have followed the above procedure in the rest of this paper.
3.1.2 Loss Compensation
All countries considered in this analysis allow tax compensation for past losses. Most countries allow companies to carry forward their losses to future years, which reduce the profits and the resulting CIT revenue of these years. The period over which losses may be carried forward varies between countries; some allow it for only 5 years, in other countries companies can carry forward their losses indefinitely. In a few countries a carry-back is also allowed, whereby losses may be compensated with profits from previous years and (a part of) the CIT paid in previous years is refunded.
14
A limited mortgage interest deduction means that the deduction is somehow limited to a certain fixed amount, a certain percentage of the total amount of paid interest or by a percentage of net income or taxes to be paid.
Loss compensation can have two opposite effects on the volatility of CIT revenue, depending on the symmetry of business cycles. Loss compensation will not be claimed much during recessions and downturns. Instead, losses which can be used for future compensation are built up. When the economy recovers, companies start to use cumulated losses for tax compensation, thereby slowing down the growth of CIT revenue. When the business cycle is asymmetric and good times last longer then bad economic times, the effect of loss compensation on CIT revenue will be destabilizing. During these upswings companies can run out of losses, which causes relatively large increases in CIT revenue compared to previous years. This increases volatility. However, when business cycles are symmetric loss compensation is likely to be stabilizing. Companies are not likely to run out of losses as fast as in the first situation, so the loss compensation dampens the effects of profit growth.
The Netherlands allow companies an unlimited carry forward and a carry back of three years. With that, the Netherlands have the most extensive loss compensation of the countries considered. Therefore, I included this variable in my analysis. Although the expected effect of loss compensation on volatility is not clear, it is still interesting to find out whether this feature of the Dutch tax system has a negative or positive effect on the volatility of CIT revenue.
To test for the potential effect of loss compensation on tax revenue the following regression equation will be estimated
CVCITi,t = Ci,t + GAPi,t + ITRCITi,t + l1i,t + l2i,t + l3i,t + εi,t (4)
CVCIT = CV of CIT revenue
ITRCIT = ITR of CIT
l1 = dummy variable that has value 1 if a country allows a limited carry forward and no carry back
and 0 otherwise
l2 = dummy variable that has value 1 if a country allows an carry forward of 10 years or longer
but no carry back and 0 otherwise
l3 = dummy variable that has value 1 if a country allows an unlimited carry forward and a carry
back and 0 otherwise
To avoid exact collinearity l3 is omitted.
3.1.3 Pension premium deduction
Next to mortgage interest deduction, the Netherlands also allow for the deduction of contributions paid into pension plans by both employers and employees. This deduction can possibly increase the volatility of the PIT and CIT tax-base, and thereby PIT and CIT revenue, because both fluctuations in the contributions paid into pension’s plans and the number of participants in pension plans effect the aggregate amount.
pension plan with individual accounts. The benefits are solely based on the amount contributed to the account, plus income gains and minus expenses and losses allocated to the account. A DB plan is a plan that defines the benefit for an employee upon that employee’s retirement. The benefit is based on the former wage of the employee. Within a DB plan pension premiums are more likely to fluctuate. For example, when returns on investment are lower than expected, pension premiums will be raised to be able to pay out the guaranteed pre-specified benefits when participants retire. Within a DC plan, contributions will not necessarily be raised in such a case, because the level of benefits is not guaranteed.
The difference between DB and DC systems will be even larger if pension funds invest in risky assets, like equities. The return of these assets is more variable, which increases the risk on unexpected losses. The crash of the stock markets a few years ago decreased the capital of Dutch pension funds (mostly DB) severely; as a result pension premiums were increased.
Because of the differences in pension systems and therefore their different effect on the volatility of pension premiums, it is not enough just to indicate whether a country allows the deduction or not. Instead of using a simple dummy variable as I did with the mortgage interest deduction I constructed an indicator, which allows me to take most of the factors mentioned into account. Besides the factors already mentioned, I also included the importance16 of pensions for each country. A graphical
representation of the composition of the dummy can be found in appendix C. Most of the information to construct this data was only available for the period 2001-2004, so I included only four years per country for this analysis. I constructed separate indicators for CIT and PIT, as the deductibility of contributions paid into employees’ pension plans is different for companies and individuals. The other three factors in appendix C are the same for both indicators.
To test the impact of the deductibility of pension premiums on the volatility of CIT and PIT, I will estimate the following regression equations
Personal Income Tax revenue
CVPITi,t = GAPi,t + ITRLABi,t + PIi,t + εi,t (5)
PI = Pension indicator PIT
Corporate Income Tax revenue
CVCITi,t = GAPi,t + ITRCITi,t + PICi,t + εi,t (6)
PIC = Pension indicator CIT
3.1.4 Tax mix
There are large differences in the tax mix of the 23 selected European countries. The sensitivity of taxes to for example fluctuations in GDP differs between tax categories. Van den Noord (2000) calculates the tax elasticity’s for three tax categories for several countries. The results of the article and my results in chapter 2 show that CIT and PIT are more volatile than indirect taxes in the Netherlands. However, this tax sensitivity differs per country. For example, in Italy and Denmark the tax elasticity of indirect taxes is as high as or even higher than CIT. To study whether a higher proportion of certain taxes in the total tax mix influence volatility, I will estimate the effect of the proportion of the three tax categories mentioned above on the volatility of total tax revenue.
The following regression equation is estimated
CVTOTi,t = GAPi,t + ITRLABi,t + ITRCAPi,t + ITRCONi,t + sITi,t + sPITi,t + sCITi,t + εi,t (7)
CVTOT = CV of Total Tax revenue
ITRCON = ITR of consumption
sIT = proportion of IT in total taxation
sPIT = proportion of PIT of total taxation
sCIT = proportion of CIT of total taxation
In this equation I included the ITR’s of consumption, capital and labour because total taxation is influenced by changes in all these categories.
3.1.5 Output gap volatility
Tax revenue of most tax categories are sensitive to fluctuations in economic growth. It follows that a country with significant swings in the level of economic activity should experience larger tax revenue volatility. One way to measure the intensity of business cycles is by studying output gap volatility. Business cycles affect almost all tax types. In good (bad) times the tax-bases for both indirect and direct taxes tend to increase (decrease), thereby increasing (decreasing) tax revenue. Therefore, the impact of output gap volatility on total tax revenue, IT revenue, PIT revenue and CIT revenue will be looked into using the following regression equations
Total tax revenue
CVTOTi,t = GAPi,t + ITRLABi,t + ITRCAPi,t + ITRCONi,t + outputgapi,t + εi,t (8)
outputgap = output gap volatility, measured as the absolute deviation from the mean output gap
Indirect tax revenue
CVIT = CV of IT
Personal income tax revenue
CVPITi,t = GAPi,t + ITRLABi,t + outputgapi,t + εi,t (10)
Corporate income tax revenue
CVCITi,t = GAPi,t + ITRCITi,t + outputgapi,t + εti,t (11)
3.2 A note on methodology
The analysis concerns a data-panel of 23 countries for the period 1995-2004. A data panel combines time series data of a cross-section of countries. By combining time series and cross-sectional data, one obtains a larger set of observations which improves the econometric estimates. I used a so-called fixed-effects model, which allows the intercept to vary between countries to account for the effects of unobserved heterogeneity.
A problem of working with panel data can be heteroskedasticity. One of the classical assumptions of using least squares regressions is the assumption of homoskedasticity, meaning that the variances of the disturbance term are constant. When an equation has heteroskedasticity in its error term, the OLS estimators are still unbiased but they will no longer be of minimum variance. The presence of heteroskedasticity will make the ‘t’ and the ‘F’ tests inaccurate and thereby the standard errors of the regression coefficients. It is unlikely that the variance between cross-sections is homoskedastic. Therefore, I will use White’s heteroskedasticity consistent covariance matrix estimator17 which
provides correct estimates of the coefficient covariance in the presence of heteroskedasticity of unknown form.
All regression equations in this chapter contain two or more independent variables, so it is possible that multicollinearity exist between the independent variables. In the case of perfect multicollinearity each independent variable is an exact combination of all other independent variables. In practice, this extreme case is unlikely to occur. However, there is often imperfect multicollinearity, which means that there is correlation between the independent variables. The usefulness of the regression may decline if the imperfect multicollinearity is high. It will lead to higher standard errors, which may cause over-acceptance of the null hypothesis. Furthermore, removing or adding a variable or adding or deleting a small number of observations may lead to drastic swings in parameters estimates. I used two methods to test whether there are linear relationships between the independent variables of my series, namely studying pair wise correlation coefficients and looking at auxiliary regressions. The first method calculates the correlation coefficients between two independent variables. When these coefficients exceed 0.818 there is a linear relationship that could harm the regression results. I listed the results for the correlation coefficients for all regression equations in appendix D1 tables 1 through 9. These
tables show that none of the coefficients exceeds 0.8, so there are no pair-wise correlation relationships that might cause problems. With auxiliary regressions one independent variable is regressed against the other independent variables. With this method it is possible to detect correlation relationships existing between two or more variables, something that can not be discovered by looking at pair wise correlation coefficients. If the value of the adjusted R-squared of the auxiliary regressions exceeds 0.8, there are correlation relationships that could harm the regression results. The results of these regressions are included in table D2, tables 1 through 9. Again none of the independent variables shows higher values than 0.8, so there is no multicollinearity that will harm the regression results.
Time series have to be stationary to be able to work with them. When series are not stationary there is the danger of obtaining apparently significant regression results from unrelated data. Normally, one could test whether series are stationary by applying the Augmented Dickey Fuller (ADF) test19. However, the period considered is relatively short, only ten years. It is too short to find trends in the series and tests like ADF will not be reliable for these short samples. It is not likely that my series are non-stationary, as most of my series are growth rates, dummies or ratios. Such series are likely to be stationary.
Last, the problem of autocorrelation can exist. This means that the residuals are correlated with their own lagged values. It leads towards an upward bias in the estimates of the statistical significance of coefficients estimates, so it is important to check for autocorrelation. The most common test for autocorrelation is the Durbin Watson test. The statistic of this test gives values around two when autocorrelation does not exist. When the statistic is much lower than 2 there is negative autocorrelation, when it is higher than two it implies the existence of positive autocorrelation. I will discuss the results of the Durbin-Watson test, when there are signs of autocorrelation, together with the regression results in the next section. Autocorrelation can be modeled by adding an autoregressive term to the equation. This term considers the autocorrelation, thereby solving the problem. However when such a term is used, one creates a dynamic panel, which makes the fixed-effects estimation inconsistent. This can be solved by using Generalized Method of Moments (GMM) estimation. This estimation method is consistent in the presence of autocorrelation. So when autocorrelation is present, I will correct this by including an autoregressive term and estimate the model with GMM.
3.3 Results
In this section the results of the regression analyses are presented and discussed. For each variable I show the main results in the text, the complete results are in Appendix D3.
3.3.1 Mortgage interest deduction
Table 2 shows that only one of the control variables, namely the output gap, is significant. The ITR does not have a significant influence on the volatility of PIT revenue. I included the ITR and the output
gap as control variables, so their coefficients are not relevant for my analysis. Therefore I will only mention the sign and significance of these variables but not analyze them further.
I included an first order autoregressive term (AR(1)) in the equation because the Durbin Watson statistic of the regression equation showed significant autocorrelation. The term is very significant, so its use is justified. Because this term had to be included, the equation has been estimated by using GMM.
Variable Coefficient Probability GAP 0.096 0.007 ITRLAB 0.336 0.692
H1 -0.051 0.641
H2 -0.137 0.395
AR(1) 0.930 0.000
Table 2: regression results mortgage interest deduction
The coefficients of both dummy variables are negative. These coefficients are relative to the dummy I left out, namely to the group without the mortgage interest deduction. The negative signs of the coefficients suggest that countries that allow mortgage interest deduction experience lower volatility than countries without the deduction. This implies that the stabilizing effect of the mortgage interest deduction dominates. However, both coefficients are not significant, so mortgage interest deduction does not significantly influences PIT revenue volatility. Therefore, I did not find evidence that countries without mortgage interest deduction experience different tax revenue volatility than countries that do allow this deduction.
3.3.2 Loss compensation
Variable Coefficient Probability GAP 0.003 0.753 ITRCIT -0.619 0.260
l1 -0.093 0.483
l2 -0.057 0.656
Table 3: regression results loss compensation
3.3.3 Pension premium deductions
Table 4: regression results pension premium deduction PIT
Table 5: regression results pension premium deduction CIT
Table 4 and 5 show the results of the regressions with the pension premium deduction on respectively PIT and CIT. Due to the short time period it was not possible estimate a fixed-effect model, so I included a common constant in this regression. The initial regression results for both variables showed the existence of autocorrelation. Therefore, I included an autoregressive term in both equations. The results show that both autoregressive terms are significant.
The results for PIT show that only the constant and the implicit rate of labour are significant. The variable of interest, that is pension premium deduction, does not seem to have a significant influence on volatility. Consequently, I did not find evidence that countries with unlimited pension premiums deductions for employees such as the Netherlands experience a greater PIT revenue volatility than countries with no or a more limited deduction.
The results for CIT show that only the constant is significant. Again the variable of interest, the pension indicator is not significant. Therefore, countries with unlimited pension premiums deductions for employers do not experience a different CIT revenue volatility than countries that allow the deductions in lesser extent.
However, I used indicators to test the influence of pension premium deduction on tax volatility of countries. With these indicators I tried to account for several aspects of a country’s pension system such as the size of the deduction, the pensions systems of countries and their financial risk. Nevertheless, it is possible that pension systems that work the same on paper, have different effects in practice, due to legislation or other factors. Pension systems are very complicated and differ a lot across countries, so it is difficult to quantify the possible effects of a pension system on tax revenue. Especially the Dutch system is unique, due its pillar system and the participation rate. Therefore, the indicators are only an approximation to which extent pension premiums can influence PIT revenue.
Variable Coefficient Probability
C 1.043 0.000
GAP -0.001 0.855 ITRLAB -2.487 0.000
PI -0.015 0.625 AR(1) 0.909 0.000
Variable Coefficient Probability
C 0.379 0.009
GAP 0.036 0.196 ITRCIT -0.450 0.263
3.3.4 Tax mix
Variable Coefficient Probability GAP -0.002 0.026 ITRLAB -0.272 0.034 ITRCON -0.065 0.070 ITRCAP 0.178 0.406 SIND 0.276 0.041 SPIT 0.133 0.250 SCIT 0.254 0.068
Table 6: regression results tax mix
Table 6 presents the results of the tax mix analysis. The results show that countries with a tax mix with higher proportions of IT and CIT experience higher volatility in total tax revenue. Both coefficients are positive and significant. The results for CIT fit the situation for the Netherlands. The parameter is positive and significant, which implies that countries with a higher proportion of CIT in their total tax mix, experience a greater volatility. The parameter for IT is also positive and significant, so countries with higher proportions of indirect tax in their tax mix experience greater volatility. This is not what one should expect based on the elasticity results for the Netherlands, however for some other countries the volatility of indirect taxes is high. The parameter for PIT is positive but not significant, so there is no evidence that countries with different proportions of PIT in the total tax mix experience different tax revenue volatility.
Output gap volatility
Variable IT PIT CIT Total tax GAP -0.001 0.000 0.000 -0.001*** ITRLAB -5.406 -0.340* ITRCON 3.021 0.276*** ITRCAP -0.139 -0.064*** Outputgap -0.008 -0.005 0.001 0.001 AR(1) 0.442* 0.436*
Table 7: regression results output gap volatility *Significant on a one percent level **Significant on a five percent level
Table 7 presents the results of the four output gap volatility regressions. In the second and third column of the table I listed the result of the regressions on IT and PIT. I included an AR(1) term in both regressions, because of the presence of autocorrelation and estimated these regressions with GMM. None of the variables is significant. Therefore, I did not find evidence that countries with marked business cycles experience a different volatility in PIT and IT revenue than other countries.
The fourth column shows the results of the regression of CIT revenue. Here, none of the variables are significant so I did not find evidence that CIT revenue volatility differs between countries with less or more marked business cycles.
not significant, so countries experiencing heftier business cycles do not necessarily have to experience a higher total tax revenue volatility.
3.4 Conclusion
This section studied whether the increased tax revenue volatility in the Netherlands could be explained by certain properties of the Dutch tax system and the fluctuations of the business cycle. With the exception of some aspects in the tax mix, I did not find factors that could explain differences in tax revenue volatility between European countries. Countries with marked business cycles or certain provisions in their tax system do not experience different tax revenue volatility then countries without these features.
One explanation for these results is that there might be another important factor that could explain a large part of differences in tax revenue volatility. This factor is the progressiveness of a tax system. From a theoretical point of view the progressiveness of a tax system should influence tax revenue volatility. As explained in chapter two, tax revenue fluctuate more in progressive tax systems, which leads to greater volatility. It was not possible to test the influence of progressiveness on tax revenue volatility because I was not able to find reliable data for the MEP for all countries and years considered in this analysis. It would have been very interesting to test whether the progressiveness in tax systems can explain the observed differences in tax revenue volatility by the IMF, as progressiveness is likely to have an influence on tax revenue volatility.
4. The causes for volatility: an analysis for the Netherlands
In this chapter I will study whether certain determinants of the properties of tax systems discussed in the last section can explain tax revenue volatility in the Netherlands. I will estimate the relationship between fluctuations of these factors and the fluctuations of tax revenue. The factors I included for PIT are total mortgage debt, long term interest and pension premiums paid by employees, for CIT I will test its relationship with the pension premiums paid by employers. In addition, I will test the relationship between the fluctuations of CIT, PIT and IT and business cycle fluctuations.
I will use data series over the period 1972-2005. One exception is the regression with the contributions paid into pension plans by employees because these data series are only available as from 1981. In the last chapter I was not able to use tax series corrected for discretionary measures, as those series were not available for all countries in my sample. However, I was able to construct such series for the Netherlands. These series reflect only endogenous growth and can be found in appendix I.
I will estimate the following regression equations
PITt = Ct + GAPt + DEBTt + INTERESTt + εt t = 1,…, T (12)
PITt = Ct +GAPt + PPEEt + εt (13)
CITt = Ct + GAPt + PPERt + εt (14)
ITt = Ct + GAPt + εt (15)
PIT, CIT and IT are respectively, personal income tax, corporate income tax and indirect tax. C is a constant, GAP is the output gap, PPEE are the contributions paid into pension plans by employers,
DEBT is the total Dutch mortgage debt, INTEREST is the long term interest rate and PPER are the
pension premiums paid into pension plans by employees. The index t represent time. All series, except the output gap, are used as the change relative to last year’s value. I estimated the influence of pension premiums paid by employees on PIT revenue separately because this data series is shorter than those for mortgage debt and interest. The descriptive statistics of my data series are in table 1 of appendix E.
The output gap is included both as variable of interest and control variable. The business cycle is a likely cause for tax revenue fluctuations. Therefore, including it as a control variable corrects for the fluctuations caused by the business cycle. However, is it also interesting to find out to what extent the business cycle solely influences tax revenue.
methods discussed in section 3.2 to test for multicollinearity. The results of both tests are in appendix E tables 3 through 6. Tables 3 through 5 show the results for the pair wise correlations. None of the coefficients exceeds 0.8, so there are no signs that there is multicollinearity that could harm the results. For equation 15 it was not necessary to look at correlation coefficients as there is only one independent variable. The auxiliary regressions for equation 12 are in table 6. None of the results indicate high values of imperfect multicollinearity. I did not run such regressions for the other equations, as none of them contain more than two variables.
To test the specification of my regression equation I will test my equations for autocorrelation, heteroskedasticity and normality. The first two issues have already been discussed in section 3.2. I will test for the existence of autocorrelation using the Durbin Watson test, the presence of heteroskedasticity will be tested using the White Heteroskedasticity Test20 and the Lagrange Multiplier (LM) test for Autoregressive Conditional Heteroskedasticity (ARCH)21. The first test is a test with the null hypothesis of no heteroskedasticity against heteroskedasticity of some unknown general form, the second test tests for a specific form of heteroskedasticity. I will present the results of these three tests together with the regression results. One of the basic assumptions of the least squares method is that the residuals are normal distributed. Therefore, I will use the so called Jarque-Bera test, which has the null hypothesis that the residuals are normally distributed. I will present this test statistic together with the regression results in tables 7 through 10.
Table 7 shows that changes in the volume of total mortgage debt do not have an influence on PIT revenue, as it coefficient is not significant. The output gap and the long term interest do have a significant influence on PIT revenue. The coefficient for the output gap is positive. This implies that in times of economic recovery, when the output gap becomes less negative of more positive, PIT revenue will grow. When the output gap becomes more negative or less positive, PIT revenue will decrease. This is exactly what one should expect. When the economic growth is increasing, more people get employed and start to earn higher salaries, thereby increasing tax revenue. The opposite is true when economic growth declines.
The long term interest rate has a negative effect on PIT revenue. When the interest rate increases, PIT revenue decreases. A raise in the long term interest will increase the interest payable on mortgages, thereby increasing the mortgage interest available for the deduction. However, the estimated coefficient is very small. This could be explained by two other factors. One would expect the number and value of new mortgages to decline when the interest rate has been raised. Next, mortgage interest rates are often set for several years and do not change immediately after a change in the interest rate. Therefore, one would not expect that the deducted interest to increase immediately.
I included an autoregressive term in the equation, because the original equation showed signs of autocorrelation. Adding a first order autoregressive term did not solve this problem, therefore I included a second order term. In table 7 I present the results of the Breusch-Godfrey22 serial autocorrelation test. This test is consistent when an autoregressive term is added, as opposed to the
Durbin-Watson statistic. The test results show that the null hypothesis of no autocorrelation is not rejected, therefore including the second order autoregressive term solved the problem of autocorrelation. The other test results in table 7 show that the residuals are normal distributed and that there are no signs of heteroskedasticity.
Table 8 shows the results of regression equation 13. The results confirm the positive significant influence of the business cycle; again the coefficient is positive and significant. Pension premiums paid by employees do not seem to have an influence on PIT revenue. The staff of the IMF found the same results regarding the influence of the business cycle; however, with respect to pension premiums their results are opposite. They found that contributions paid into pension plans together with business cycle fluctuations explain a significant part of the fluctuations in PIT revenue. These different outcomes can be explained by the use of different models and different data series. The IMF used an Autoregressive Conditional Heteroskedasticity (ARCH)23 model. These models are designed to model the conditional variance of a variable and are mostly used in the financial sector to model volatility of financial products. As the residuals of my regression did not exhibit such heteroskedasticity, it made no sense to use this model. Furthermore, the data series of my analysis and that of the IMF differ in three ways. First, the IMF used total pension premiums paid, when I only used the premiums paid into pension plans by employees. It seems more sensible to use just premiums paid by employees as these contributions are the ones deductible for PIT purposes. Second, as opposed to the IMF, I used data series corrected for the effect of discretionary measures and therefore only reflecting endogenous growth. Because the fluctuations caused by discretionary measures are not considered as volatility, one should correct for such effects. Third and last, the IMF used quarterly data. This creates more observations and improves the quality of the econometric estimates. Unfortunately, quarterly tax series corrected for discretionary measures are not available, therefore I used yearly data. Which results are better is difficult to say. The differences between the models are the result of the properties of the data; in both cases the model specification fits the data. The IMF did use a larger data set. However, no correction for discretionary measures and including pension premiums paid by employees can influence the results significantly.
Tables 9 and 10 show the results of the last two regressions. The business cycle does not seem to have an influence on both CIT and IT revenue. With respect to CIT this could be explained by the effects of loss compensation, which dampens the effect of profit growth. The development of IT revenue might be relatively stable over the business cycle, as consumption always exist. The contributions paid into pension plans by employers do not influence CIT revenue significantly. It is possible that these premiums are just a small part of the total costs for companies. Therefore, a change in these premiums would hardly affect profits and thereby CIT revenue.
In both chapter 3 and 4 I did not find much information about the sources of the increased volatility found by the IMF in the Netherlands. I found that countries with higher proportions of CIT and IT in their tax mix experience higher volatility and that the business cycle and the long term interest rate
23 This model has first been introduced by Engle (1982) and generalised as Generalised ARCH (GARCH) by
influence PIT revenue in the Netherlands. The other factors I considered do not explain the differences in volatility between countries or explain the Dutch tax revenue volatility.
5. Budget Estimation
5.1 Current estimation methodology
In 2001, the tax forecasters at the Dutch Ministry of Finance developed a new model to estimate tax revenue. This model was first used to estimate tax revenue for the Budget of 2002, presented in September 2001. The model estimates tax revenue by determining the endogenous growth and the discretionary measures of all individual taxes. These changes are added to the last year value of tax revenue of the individual tax categories. The estimation equation per tax category is the following
Taxt,i = Taxt-1,i + DMt,i + EGt,i (16)
Where Taxt,i = tax revenue of tax type i in year t
DMt,i = discretionary measures influencing tax type i in year t
EGt,i = endogenous change of tax type i in year t
Discretionary measures are policy changes which can affect the tax-base or the tax rate of one or more tax categories. Such measures may change the tax burden of individuals and companies. When such policy changes occur, the estimated tax revenue will be corrected for the estimated direct effects of these discretionary measures. Additional to the direct effects of discretionary measures, indirect effects may occur. For example, a raise in the duty of tobacco will increase tax revenue on tobacco due to higher revenue per unit. This is the direct effect. However, the price raise is also likely to decrease the consumption of tobacco, thereby decreasing tax revenue. This indirect effect is caused by the discretionary measure(s), but will be reflected in the endogenous tax growth via changes in the consumption pattern. Therefore, only the direct effects are considered in estimating the value of discretionary measures.
The endogenous growth of the individual tax categories is estimated using relevant economic indicators/variables, which explain the change in tax revenue. With the help of regression analysis an estimation equation for endogenous growth is constructed. In general, these equations have the following form
EGt = Taxt * ( β1 + β2 * var1t+ β3* var2t+ … + βk+1 * varkt) (17)
Var1t through varkt represent the economic variables in year t, the coefficients β1 through βk+1 are
determined by regression analysis. A separate equation is estimated for each tax category. Most of the variables used are estimated by the Netherlands Bureau for Economic Policy Analysis (CPB). The tax forecasters try to estimate the amount of tax revenue that is received in a given year, because due to the Government Accounts Law24, tax revenue has to be reported on a cash flow basis. For most tax categories, the tax payments are actually received in the year they are levied on. However, the payments for three categories, namely PIT, VAT and CIT, take a while longer. The payment
periods for PIT and VAT are on average 2 years, for CIT, it may take 5 years before all payments are received. For these tax categories tax revenue on cash basis differs considerably from revenue on accrual basis. Therefore, the revenue for these categories are first estimated on an accrual basis and then divided over the years in which payments will be received.
5.2 Area’s for improvements
Period 71-75 76-80 81-85 86-90 91-95 96-00 01-05 71-05 Indirect Taxes 3% 7% 9% 5% 5% 7% 4% 6% Value Added Tax 5% 4% 9% 5% 4% 7% 4% 6% Duties 3% 12% 11% 4% 7% 6% 4% 7% Corporate Income Tax 14% 37% 18% 33% 25% 29% 20% 25%
Total Tax 4% 4% 3% 4% 2% 4% 5% 4% Personal Income Tax 3% 7% 9% 5% 5% 7% 4% 9%
Table 8: relative deviations between actual and estimated tax revenue
Table 8 shows that the relative deviations25 between actual tax revenue and its estimated value can be large. These deviations are largest for CIT and PIT, with a average prediction error of respectively 25 percent and 9 percent over the period 1970-2005. The average deviations for duties and VAT are 7 and 6 and 4 percent for total tax revenue. Although the current model could use improvement, one should keep in mind that estimations can only approach reality. Table 9 shows the relative deviations over 2001-2005 between the actual and estimated values of three economic variables estimated by the CPB with more extensive models estimated by the staff of the Ministry of Finance. The average deviations of these variables are around 4 percent, which is close to the 5 percent of total taxation. Considering that tax revenue is estimated with a relatively simple model 26, the estimation results for total tax revenue are not bad at all. That said, as table 8 showed, the model introduced in 2001 has apparently not contributed to improved estimation results for total tax revenue,
2001 2002 2003 2004 2005 01-05 GDP 3.3 2.3 1.6 4.4 6.2 3.5 Private Consumption 3.6 2.8 2.5 4.3 6.9 4.0 Total Consumption 3.7 3.4 3.2 3.8 5.2 3.9
Table 9: relative deviations between actual and estimated values of selected economic variables source: own calculations and MEV 1999-2007
Unfortunately, the estimates for individual taxes deviate more from their actual values than for total tax revenue. Therefore, it appears that the estimates of all individual tax categories could use to be improved. To figure out where the improvements should be made, I will analyze which parts of the estimation process are likely to contribute the largest prediction errors.
The first element which may create prediction errors is the value of last year’s tax revenue. In September of every year the budget for next year is presented. A few weeks before, tax revenue is
25
Calculated as the difference between actual and estimated tax revenue, divided by actual tax revenue
26 However, one should note that the tax revenue estimation method uses variables of the CPB. Therefore it uses the more
