7. Government Sector
- International Monetary Fund
- Published Date:
- November 2005
a. Some basic concepts1
The government’s fiscal operations have a major impact on aggregate expenditure and output as well as on the allocation of resources in an economy. In view of the central macroeconomic impact of the budget, the forecasting of government transactions is an intrinsic part of the design of stabilization and adjustment programs. Forecasting fiscal aggregates is also a key part of the process of fiscal policy formulation because only on the basis of a set of budget projections can a government assess whether fiscal adjustment is needed and design appropriate measures to promote such adjustment.
- Fiscal forecasting must be based on a consistent set of overall macroeconomic assumptions, for example, for output, inflation and the balance of payments. The overall assumptions have to include, as well, those relating to the external environment facing the economy, for example prices of oil and other key primary commodities, international interest rates, and economic developments in major trading partner countries. Therefore, fiscal forecasting demands prior coordination and information-sharing among the various economic agencies of the government and a central decision concerning an appropriate and realistic forecast of the broad macroeconomic aggregates over the budget planning period. Tax revenue and expenditure forecasts should be based on the same set of macroeconomic assumptions even though they may be prepared by different departments or ministries. It is also important to ensure that the budget forecast and the macroeconomic assumptions (for example, GDP and inflation) refer to the same period. This may require adjusting the assumptions if the budget period is different from a calendar year, for which most macroeconomic forecasts are prepared.
- Fiscal revenues and expenditures affect, and are affected by, the overall macroeconomic situation, in particular by changes in economic activity, household behavior, the rate of inflation, and developments in the external environment, including the exchange rate. For example, budget expenditure on unemployment assistance depends on economic activity, which itself is determined in part by the size of government expenditure. In order to ensure consistency between the forecasts for the fiscal aggregates and those for the broader macroeconomic aggregates, an iterative process of estimation may be followed. This means that the provisional macroeconomic and fiscal forecasts are adjusted to bring them into line with each other.
- Ideally, the forecasting process should allow for feedback at the level of individual tax and revenue categories. For example, a large tax on cigarettes might reduce the volume of their consumption, reducing the tax base from what it would have been in the absence of the tax. It is also useful to keep in mind the timing of tax changes, in particular the lag structure of the impact of a tax change, although this can be hard to quantify.
- Every tax system defines in law the associated tax base, the rate of tax, and exemptions, if any. The legal tax base is usually too complex to be useful for economic analysis and forecasting. Therefore an alternative, or proxy, tax base is selected by considering economic criteria. For example, the base of the personal income tax is usually household income, and the law specifies ranges of incomes to which various tax rates apply. Data for aggregate personal income by taxable income range are, however, not usually available. A proxy variable may be sought within the national accounts. Aggregate disposable income would be a candidate proxy variable for forecasting. If disposable income data are not available, even broader aggregates, such as GDP, could be used. In a typical case, import duties are payable at different rates on a number of categories of foreign goods. The revenue from import tariffs could be related to an aggregate variable from the external sector for purposes of analysis or forecasting—in this case, to the value of all imports. In both of these cases, it is reasonable to assume that changes in the proxy base will tend to move in line with changes in the items on which tax is actually due. An advantage of proxy bases for forecasting tax revenue is that values of the proxies—say GDP or imports—may already have been forecast for the future period for which revenue forecasts are desired. Box 7.1 below provides some possible proxy bases that can be used to forecast revenue from different tax categories. A more detailed table is provided in Appendix I.
Box 7.1.Suggested Proxy Bases for Tax Revenues
|Tax Revenue Source||Suggested Proxy Base|
|1.||Taxes on income, profits, and capital gains||GDP at current prices|
|2.||Sales tax||Private consumption at current prices|
|3.||Excise duties||Private consumption at current prices|
|4.||Import duties||Value or volume of imports|
- In preparing fiscal data for revenue forecasting, it is necessary to decide on the appropriate level of disaggregation of the tax system. Since a tax system generally consists of a large number of taxes, selectivity is needed in order to avoid excessive detail. As a rule of thumb, it is useful to disaggregate tax categories that contribute at least 5 percent of tax revenue on average over a sample period. Also, a group of taxes should be disaggregated if their bases are quite dissimilar. A balance needs to be struck between the greater information provided by a finer disaggregation and the cost of making the exercise unwieldy. A higher degree of disaggregation does not necessarily enhance the accuracy of a forecast.
- A significant part of fiscal expenditure depends heavily on the discretion of policymakers. This mixture of discretionary and automatic elements renders the task of forecasting expenditures difficult in the absence of knowledge about the policy measures planned by the government. In practice therefore, and in this chapter, there is relatively more emphasis on forecasting revenue than on forecasting expenditure.
b. Forecasting revenue
The discussion below sets out the main approaches to forecasting tax revenue: a model-based approach, the effective tax rate approach, the elasticity-based approach, and a time-series approach. Use of the assumption of unitary buoyancy in the context of high inflation is presented as a further alternative. A final section mentions forecasting nontax revenue.
(1) The model-based approach
Some countries use general-equilibrium forecasting models to project tax revenue. Such models have the advantage of being able to take into account the interdependence of the revenue system and the macroeconomy: higher tax rates on personal income tend to raise revenues, but at the same time reduce incentives to work, to earn income, and to report income to the tax authorities; general equilibrium models can, in principle, allow separately for the positive and negative effects on revenue.
A second type of model for forecasting tax revenue uses detailed data from a large sample of actual taxpayer reports (tax “returns”). Estimated revenue is derived from projections of the tax bases for categories of taxpayers and the detailed provisions of the tax laws for each category and then aggregated across categories. A well-known example of this approach is the income tax model for the United States developed at the Brookings Institution, which served as a prototype for models in OECD countries.2 Similar models for the corporate profits tax are in use in the United States, the United Kingdom, and other industrial countries.
A third example of this approach is to construct a model consisting of a number of estimatable equations for major categories of revenue and expenditure. Models of this type are useful in analysis as well as forecasting—for example, in analyzing the impact of a given change in the tax structure or expenditure composition. Such model-based techniques require a heavy investment of resources and a large amount of statistical information.
(2) The effective tax rate approach
This approach relies on calculated values for the implicit average (or “effective”) tax rate of a tax category, defined as recorded tax revenue divided by the tax base. The effective tax rate may differ from the statutory tax rate or the legal tax rate schedule if there are exemptions or deductions, illegal or privileged tax-free transactions, recording errors, or changes within a period. Moreover, typically the effective-rate method is applied to aggregates (such as all luxuries, all imports, or all sales) to avoid taking account of different rates of taxation on individual items or sub-categories. The revenue forecast is calculated by multiplying the forecast of the tax base by the observed effective tax rate in the preceding period. The attractiveness of the approach is its simplicity and that it does not require a knowledge of the detailed tax system and provisions of the tax laws, such as exemptions.
For example, even if a general sales tax has a complex structure of rates and exemptions, one can simply use the recorded values of sales-tax revenue and private consumption expenditures for a recent period, and calculate the ratio of the former to the latter as the effective rate of the sales tax. This calculation can be done without knowledge of allowable exemptions or of the statutory tax-rate structure. Then, on the basis of a forecast of private consumption for a particular future period, one can project revenue from the sales tax for the same period using the calculated effective rate from a past period. Moreover, if one does not have a forecast of private consumption, but only one for GDP, it is possible to calculate the effective tax rate with respect to GDP simply on the basis of the observed proportion of private consumption in GDP during a recent period.
The simplicity of the effective tax rate approach explains its widespread use in forecasting in many countries. In the context of the transition economies, however, the approach has to be applied with caution because the approach rests on three important assumptions that are unlikely to hold in these economies. The assumptions do not hold perfectly well for non-transition economies, either. The assumptions are:
- Unchanged structure of the tax base. If the composition of the tax base changes from one period to the next, the effective tax rate approach may give misleading revenue forecasts. In the above example, if there is a significant change in the composition of private consumption between taxed and tax-exempt items, then the approach would not give reliable forecasts.
- Unchanged tax system. If there are changes in the rate structure and rate level, relying on the effective tax rate would be insufficient by itself and would need to be complemented by other information and adjustments.
- Unchanged compliance ratio. If there is a change in the rate at which taxpayers comply with the tax laws, the effective tax rate would lead to under- or over-estimation of revenues. In periods of high inflation, general economic instability, or growth of an “underground” (gray-market) economy, tax compliance ratios frequently fall quite dramatically.
(3) The approach based on tax elasticity
A widely used method of revenue forecasting exploits the stable relationship between the growth of receipts from a given tax and the growth in the tax base. Thus, a change in revenue from a tax is decomposed into two parts, one corresponding to a change in the tax base and its impact on revenues, and one corresponding to a change in the tax system (including changes in the tax rate, the tax structure, the coverage of the tax, and so forth). It is important to note that the term “tax system” is used here in a broad sense, encompassing the system of tax administration, which influences the compliance ratio. Recall that the elasticity of tax revenue is defined as the ratio of the percentage change in revenue to the percentage change in the base under the condition of no change in the tax system during the period. Given an estimate of the elasticity of the tax in question and a forecast of the growth rate of the tax base, a forecast of the change in revenue can be obtained by simply multiplying the growth rate in the tax base by the elasticity (see Figure 7.1).
The main advantage of the elasticity-based approach is that, although it is difficult to obtain precise estimates of elasticities, it is often possible to estimate a range of likely values, especially for taxes on income and profits, based on the past experience of the same country or the experiences of countries that are comparable in levels of income and economic structure. Observed elasticities for major tax categories for many economies typically fall in a relatively narrow range. Judgmental estimates of elasticities are useful in cases in which historical data for statistical estimation are generally not available.
Logically, the elasticity of revenue with respect to the base will be unity in the case of proportional taxes (VAT, sales tax) and greater than unity in the case of a personal income tax system with increasing rates on higher income levels. Exemptions and deductions may result in observed elasticities that are lower than these theoretical values. Elasticities can decline further in the presence of high inflation for the following reasons:
- A lag between the time tax liability is generated and the time when it is paid (unless tax liabilities are indexed to the price level).
- A specific (fixed in money terms), as opposed to an ad valorem (proportion of value), basis for the levy of indirect taxes.
- Caps on the taxable base (for example, social security contributions that apply to wages only up to a certain level).
- Problems with tax administration and compliance.
Taxes on property and on land generally have an elasticity with respect to nominal GDP of significantly less than unity because of lags in the reassessment of property values.
Figure 7.1.Tax System, Tax Base, and Tax Revenue
(4) The proportional adjustment method for estimating elasticity
The main drawback of the elasticity approach is difficulty in estimating the correct current value of the elasticity. Simple estimation based on observed values of revenue and tax base over a sequence of years will not take into account legislative changes in the tax system, those that are planned for the future or those that occurred in the past.
When changes in the tax system are frequent and major, the elasticity of a tax can in principle be estimated by adjusting past revenue data to remove the impact of discretionary changes in the tax system.3 This adjustment can be done based on an exogenous estimate of the impact of each change in the system on tax revenues. If a change in the tax system in a given year changes revenue by a certain proportion, then tax revenues in previous years can be adjusted by the same proportion to put them on a comparable basis with the later figures. This approach assumes that the changes in the tax system have a stable proportional impact on the tax yield.
For example, assume that T1, T2, …, T5 are tax revenues over a five-year period and year five is chosen as the reference year. Assume also that discretionary changes in the tax system occurred at the beginning of years two and four, with estimated yearly impacts of D2 and D4 respectively on revenue in those years. Then, the expression
Apart from data for actual revenue collections, the information required to apply the proportional adjustment method includes estimates of the revenue impact of discretionary changes in the tax system. Sometimes these estimates are prepared by the fiscal authorities and are available along with the budget estimates. When a discretionary change takes place part way through the year, it is necessary to adjust the tax base taking into account the timing and the impact of the change in the tax system on a full-year basis over two successive years.
The proportional adjustment method results in a time series of adjusted tax revenue. The series for adjusted revenue and the tax base, or proxy, can be combined to yield a regression estimate of the underlying elasticity in an equation of the form,
where B is the tax base, b is the elasticity, and In refers to the natural logarithm. This regression relationship can be used to obtain a forecast of tax revenue by substituting a forecast value of B in the equation and solving for the corresponding value of AT for the period t + 1. A forecast obtained in this manner should be adjusted for any changes in the tax system that may be envisaged for the forecast year.
(5) The time-series approach
The elasticity approach is feasible if there have been no changes in the tax system (in rates, exemptions, and compliance) during a sufficiently long period to permit estimation of its value. Alternatively, tax elasticity can be estimated if the impacts of past and prospective tax changes can be estimated independently, year by year, and used to construct an adjusted data series that shows what revenues would have been in the absence of changes. It is quite likely, however, that neither of these conditions is satisfied: the tax authorities of the country in question change tax rates and deductions frequently, and the data required to estimate the revenue impacts of the individual changes are not readily available.
If we apply the methods of Sections c and d without allowing for tax changes, regression results will measure tax buoyancy; they will include both the combined effects of changes in tax bases (for example the growth of household income) and also the effects of changes in tax laws (adjusting tax brackets for inflation, increasing tax rates to raise revenue, or strengthening tax administration and collections). The slope coefficient may appropriately be interpreted as an average for the historical period. However, the tendency of parliament to change tax laws is a political process, and not a stable parameter of economic behavior. Since the estimate of the slope coefficient represents buoyancy and not elasticity, it is unclear whether it adds significantly to the accuracy of the forecast relative to simpler, non-regression methods.
An alternative is to discard the concepts of tax elasticity and buoyancy and the economic basis of the revenue equations in general, and use regression analysis to exploit trends and correlations in the series of data for revenue and the proxy base, including autocorrelation in the revenue series. This approach is in the spirit of the “time series analysis” branch of econometrics.4 It does not involve the assumption of an absence of tax changes, and it requires modest types and quantities of data.
The time-series approach does not rest on economic theory, but on the assumption that the future will be like the past, without specifying why that might be true. The time series approach will provide accurate forecasts if, in fact, there are stable trends and correlations at work behind the scenes that will continue into the forecast period.
Suppose that nominal income and other tax bases tend to increase steadily, and that tax revenues also increase steadily—whether because parliament gradually changes tax laws to adjust for inflation, or because the country has a unitary-elastic system, or some combination of these two cases. Under these circumstances, the time series method suggests estimating an equation of the form,
where, as above, T is actual revenue and B is the tax base. In practice the specification would probably be confined to one or two current and lagged terms on the right-hand side because these variables tend to be highly collinear, and the addition of further lags would fail to achieve any significant decrease in the standard error of the regression.
(6) The effect of inflation on tax revenue
In the presence of continuing high inflation, the authorities may be forced to modify tax rates more or less continuously, or modify the tax system, in order to keep buoyancy close to unity so that revenue does not grow or shrink excessively. To illustrate this point, imagine a hypothetical economy with zero inflation and a revenue system based solely on specific taxes. If there is no inflation the elasticity of the tax system will be unity for real increases in output (or income). The ratio of revenue to GDP will remain constant. In the opposite case, in a country with high inflation but no output growth, the revenue received from specific taxes will be constant in nominal terms, and will fall in relation to GDP as the price level rises. In this second case, if no changes are made in tax rates, the buying power of total revenue will shrink and become negligible as time passes; elasticity is zero.
If a country’s tax system contains specific taxes, and inflation occurs, tax rates must be raised continuously to keep revenue growing in line with nominal GDP; if all taxes are proportional (and there are no exemptions), then revenue will keep pace with the value of output without legislative changes in the tax structure. A progressive tax—for example, higher rates of tax on personal income falling in higher tax brackets—results in the opposite problem; revenue will increase relative to income when there is high inflation, even with little increase in real income, until virtually all income is taxed at the highest marginal rate, and the ratio of income tax to income for the average household is at a value originally intended only for the rich. This point is made diagrammatically in Appendix IV.
Whether elasticity is low or high, if the economy experiences a changing tax share with high inflation, the legislature will be pressured to act to offset the effects of inflation on the buying power of government revenue and household income. It can raise or lower specific rates repeatedly, increase or decrease personal tax brackets frequently (“indexing”), or move the system toward one of proportional tax rates so that inflation adjustment is automatic. (For example, law-makers could choose a constant VAT rate on all expenditures, and/or a single-rate system of personal income taxation with no deductions or exemptions.) If such changes in the nature of the tax system occur, not only will buoyancy tend to equal unity, but elasticity will approach unity as well.
Unitary buoyancy may not be precisely achieved in every period. Parliamentary adjustments may under- or overcompensate for actual inflation in any particular year or group of years. Correcting a tax system for inflation may be a complex exercise, and changes may go too far or not far enough in the short run. Alternatively, parliament may intend to raise or lower the ratio of revenue to the base, which will also influence the average relation between revenue and the base over time. In such cases, the assumption of unitary elasticity, that the ratio of revenue to base will remain unchanged, will not be correct. However, the higher the inflation in the economy, and the more entrenched, the likelier it is that the ratio of revenue to GDP will remain unchanged in the short run as a first approximation. Forecasts based on this deduction may be as accurate as those based on either of the regression approaches unless there is a change in tax policy.
(7) Forecasting nontax revenue
Nontax revenue includes receipts from property income, fees and charges (wedding and fishing licenses, bridge tolls, receipt of funds for the right to extract mineral ores on public land, the right to operate a radio transmitting station), fines, and net profits of financial and nonfinancial public enterprises including the central bank. The operating surpluses of most departmental enterprises (the cafeteria in the finance ministry) are also counted as nontax revenue. If the government has a monopoly on the import and sale of a particular commodity (tobacco, sugar, salt, petroleum), and charges a high price relative to the world market, either to earn revenue or to discourage consumption, the resulting revenues are considered to resemble those from excise taxes and are included as tax revenue. In the case of other public enterprises (railroads and the national airline, the postal service, telecommunications), the operating surplus, if any, is considered nontax revenue if transferred to the central government. (Social security contributions would normally be counted as tax revenue.)
The flow of aggregate nontax revenue is difficult to relate to macroeconomic developments according to theoretical behavioral patterns. Profits of the central bank may be low temporarily because it has intervened in the foreign exchange market to support the exchange rate, or because it attempted to sterilize a capital inflow by selling bonds to the domestic public at prices lower than those at which it acquired the securities originally. Whether the railroads and the telephone system make a larger or smaller profit depends partly on whether they are permitted to adjust their fees and charges upwards as their costs rise or are obliged to wait until after an upcoming election to do so.
Because of the absence of compelling economic explanations of these kinds of behavior, it is reasonable to resort to trend and autocorrelation exclusively for nontax revenues. Specific knowledge about likely developments in the forecast year, including the timing of increases of the prices of public services, may significantly enhance the accuracy of forecasts.
c. Forecasting expenditure
Compared with revenue, there is significantly less scope for forecasting the level of government expenditure through reliance on economic relationships. This is largely because of the political nature of the process by which many decisions on public expenditure are made, thereby making a significant portion a part of discretionary policy. Broadly speaking, expenditure can be divided into two categories: those that are compressible in the short-run and therefore discretionary, and those that are not. The latter category would clearly include interest payments, which are determined by the size of the public debt and the prevailing interest rate structure, leaving little discretion for the government. Other economic categories of expenditure that are partly endogenous and partly discretionary are social insurance payments, including unemployment and pension benefits, and expenditure on wages and salaries. For example, the wage and salary bill would reflect in part the size of the civil service, existing pay-scales, and labor market conditions in general. It would also be significantly affected by government policies regarding changes in civil service employment and its wage and incomes policies. Ultimately, total noninterest government spending is compressible; from a macroeconomic point of view, it is a policy instrument and not an endogenous variable.
(1) Wages and salaries
Expenditure on wages and salaries, a major category of current expenditure, is likely to be influenced by institutional reforms aimed at restructuring the government sector. These would be reflected in government policies regarding civil service employment and public sector wages. The latter would also reflect the ongoing rate of inflation and developments of wages in the private sector. In economies with a large civil service, policies reducing the number of government employees may be put in place. A forecast of the wage bill should take account of the government’s intentions about the size of the civil service and the estimated change in average wages.
(2) Subsidies and transfers
A forecast of expenditure on subsidies and transfers, in the absence of new initiatives and a strong political will for change, will normally reflect previous trends. However, in an economy undergoing significant reform, reductions in subsidies and/or transfers may be both desirable and possible. Change in the desired role of the state, government down-sizing for whatever reason, the liberalization of prices, the restructuring and privatizing of state enterprises, reducing or limiting unemployment benefits, and national pension reform are changes that may have significant quantitative implications for expenditure on subsidies and transfers.
(3) Expenditure on goods and services
Aside from the wage bill, these expenditures are the main operating expenses of the government. They typically share the brunt of expenditure cuts. However, for government to continue operating in an efficient way there must, as a first approximation, be rough proportionality between the size of the civil service and expenditure on goods and services.
(4) Interest expenditure
In principle, information on the (i) size, (ii) domestic or foreign composition, (iii) maturity profile, and (iv) interest rates applicable to different maturities of public debt would be obtained in order to forecast interest payments. The breakdown of the debt into portions carrying fixed interest rates or contracted at floating rates is also relevant. As a first approximation, one can estimate the change in interest payments on government debt from an estimate of the interest rate and the change in the average level of indebtedness. The estimate of the change in indebtedness usually requires an estimate of the fiscal balance, which in turn, requires an estimate of the interest payments themselves. A simplified method is illustrated below, in Section d(4).
Interest rates applicable to domestic debt are affected by government policy regarding the financing of the deficit and the stance of monetary policy. These policies should be taken into account in projecting the average interest rate applicable to government borrowing.
(5) Capital expenditure
In comparison with current expenditure, capital expenditure is generally considered to be more subject to adjustment and frequently bears the brunt of fiscal tightening. In practice, however, capital projects are not so easy to turn on and off, especially when they depend on external concessional sources of funding. Indeed, capital spending in most countries in any year is set in the context of a multi-year investment program subject to annual changes in the light of resource constraints and changing fiscal priorities. In projecting capital expenditures, one needs to talk careful account of spending already in the pipeline that is difficult to reverse. Moreover some, though not all, forms of government investment are crucial complements to private output and a higher growth rate of GDP and rising living standards.
(6) Financing of the budget balance
Financing is derived from forecasts of fiscal revenues and expenditures. The forecaster would, however, need to assess the breakdown of financing between domestic and external sources, and bank and nonbank categories, in the light of information about the amount of external financing that is available and the scope for the private nonbank sector to absorb additional government debt. Forecasting would also need to take into account of the impact of exchange rate changes on the domestic currency value of external financing, the range of domestic financial instruments available for financing, and the potential consequences for domestic interest rates of the size of nonbank financing.
d. Guidelines for forecasting the fiscal accounts of Turkey
Forecasting the fiscal sector in Turkey is complicated because it is difficult to forecast the operations of the extrabudgetary funds (EBFs) and local governments. One approach that can be used is to forecast the various components of central government (“consolidated budget”)5 revenue and then forecast revenue for the EBFs and local governments as a percentage of general government revenue. Data for ratios in recent years are given in Box 7.5, below. On the expenditure side of the budget, the forecasting problem is less difficult.
(1) Tax revenue
Table 7.3 provides a breakdown of central government tax revenues. A defining characteristic of recent fiscal history in Turkey is that the tax system—including tax rates, the number of exemptions, and types of tax—has been changed frequently, partly because of the long sequence of years of high but variable inflation. The Turkish authorities have not systematically reported estimates of the changes in revenue resulting from these discretionary measures. Thus, the parameters of regression equations should be interpreted as buoyancies rather than elasticities. Two other features of the results are worth noting. First, the changes in tax laws designed to address fiscal imbalance mean that the revenue yields for one year may be responsive to the previous years’ collections. National elections may cause similar patterns in revenue data. Such reactions can lead to autocorrelation in the revenue series, and most regressions of revenue data on unlagged structural variables reported below appear to involve autocorrelation. Second, the sample period (1985-95) is quite short, so coefficient estimates are less stable, in principle, than if longer samples were available.6
For each category of tax revenue considered below, the first regression presented relates revenue directly to a proxy for the tax base, as in equation (7.1). Since the form of the equation is exponential (the variables are in logarithms), the slope coefficient is a direct estimate of buoyancy. As discussed, these values are likely to be fairly close to unity.
The second equation for each revenue category is also motivated by considerations of economic structure. Compared to the first equation, allowance is made in the second for three factors: (1) A part of the taxes due in year t may not be collected until t+1. (2) Tax revenue may change more (less) than proportionately in relation to the proxy base when GDP is high or low relative to its trend. (For example, corporate profits taxes may rise more than GDP in a boom.) (3) Inclusion of a lagged value of the endogenous variable as a regressor may be taken as a proxy for a longer-term lag structure.
The third regression, calculated for some tax categories, is motivated by time series considerations rather than economic structure alone. The data are approached from the point of view of exploiting whatever patterns exist in the past as a guide to projecting revenue collections in coming periods. In the time series versions of the revenue equations, current and lagged values of the proxy base and lagged values of the revenue series are inserted on the right-hand side of the equation to the extent that the econometric results are strengthened. (See equation (7.2) above.)
Regression coefficients and statistics are reported below. Actual and fitted values of revenue for each equation are given at the end of Section d(3), below, in Table 7.1. Technical aspects of the regression results are discussed in Appendix V.
Tax revenues of the Central Government
Personal income taxes
Withholding taxes on wages and salaries constitute the bulk of personal income tax revenues. Self-employed persons—including farmers, tradespersons, artisans, and certain professionals—are explicitly exempt from income taxation. Capital gains are not taxed. Interest earned on government bonds is also tax-exempt; interest earned on private sector debt instruments need not be reported although taxes may be withheld. Household expenditure on certain specified basic consumer goods is deducted from personal income in computing taxable income. Since 1993 efforts have been underway to reduce tax evasion among independent professionals by introducing taxpayer identification numbers and limiting the type of businesses subject to lump-sum taxation, but these measures have not yet had a noticeable impact on revenue collection. (Further information on the tax system is presented in Appendix III to this Chapter.)7
Data for aggregate household income are not available so GDP has been used as a proxy for the base of the personal income tax. In the regression results tabulated below, GCRPT stands for central government revenue from personal income taxation; an initial “L” signifies that the natural logarithm has been used; and the values of right-hand-side variables are contemporaneous with those of the endogenous variable unless a time subscript is attached to indicate a lag (“t-1”,“t-2”, and so forth). The time series for the regressions presented in this section, with additional identifying information, are shown in the data appendix at the end of the volume. Variables used in these equations are in the same units as are specified in the data appendix. For example, both GCRPT and GDP are in billions of liras. The Schwartz and Akaike information criteria (SIC, AIC below) are described in Appendix V along with alternative measures of autocorrelated residuals and interpretation of the coefficients when a trend variable is included.
In equation (7.3), the buoyancy is estimated to be close to unity, 1.05,8 roughly the expected magnitude. The interpretation of the standard error of estimate is that future values of revenue will lie within plus or minus 11 percent of the forecast value about two thirds of the time and exceed eleven percent in the other third of cases.9 The Durbin-Watson statistic indicates that forecasting could be improved by taking into account the autocorrelation in the residuals or respecifying the model. The value of R2 is very high despite the short sample of annual data and the relatively large standard error of estimate. This is a case of “spurious correlation.” The high rates of inflation from year to year throughout the sample tend to affect nominal GDP and taxes on personal income similarly; the fact that the inflation rate is common to both nominal series tends to insure a high value of R2 irrespective of any causal relation between them. To a greater or lesser extent, these characteristics of the regression results from personal income taxes also apply to the other revenue categories discussed below.
In equation (7.4) and subsequent equations, the variable LTRENGDP contains trend values of the log of nominal GDP. A value for average, or cyclically neutral, buoyancy can be inferred by considering the hypothetical case in which GDP is equal to its trend. In equation (7.4), buoyancy thus is given by the sum of the coefficients on GDP and trend GDP (see the discussion in Appendix V), which in turn is equal to 0.4972 + 0.5929 = 1.09, close to the same value as in equation (7.3). For taxes paid out of the deviation of personal income about trend, buoyancy equals 0.6, which is small relative to taxes paid out of trend GDP. This is the opposite result of what would be expected for the respective elasticities since the personal income tax system is progressive (see Apppendix III). Because large increases in nominal income in Turkey in this period are due much more to inflation than to real growth, the result suggests that parliament does more to adjust the income tax structure for the effects of inflation in years in which inflation is higher than average. Such tendencies to overadjust the tax structure may well not be stable parameters of economic behavior; the result is nothing more than a statistical average for the given sample. Thus, it is debatable whether it should be incorporated in projections of future tax collections.
One can infer an estimate of long-run buoyancy from equation (7.5) by imposing steady-state conditions—GDPt-1, GDPt-2, and trend GDP are equal, and GCRPTt is equal to GCRPTt-1. Because of the presence of the lagged endogenous variable on the right-hand side of (7.5), one may employ the interpretation that long-run values of the other coefficients are given by 1/(1 - 0.5765) times the regression estimates.10 Long-run buoyancy is therefore estimated as the sum of the coefficients on GDP and trend GDP divided by one minus the coefficient on the lagged dependent variable, (0.7020 - 0.9719 + 0.6936)/(l - 0.5765) = 1.00. An equation such as (7.5) that is meant to reflect time series considerations is not fully consistent, logically, with the usual interpretation of a distributed lag, but it is of interest to note that the buoyancy estimate implied by this interpretation is of similar size to the estimates resulting from equations (7.3) and (7.4). In equation (7.5), the coefficient on deviations of GDP from its trend is not only less than long-run buoyancy, but is negative (0.7020 - 0.9719 = -0.27). This is consistent with the interpretation of equation (7.4) of overadjustment in high-inflation years—the inflation adjustment by parliament is made more liberal pre-emptively, to restore income to households that would be taken away by the tendency of inflation to push taxpayers into higher brackets.
Corporate income taxes
The yield of the tax on corporate profits in Turkey is about 1 percent of GDP, which is surprisingly low. The low yield reflects weak tax administration, inadequate inflation adjustments to the tax base, and generous investment incentives that erode the corporate income tax base. Moreover, preliminary estimated tax liabilities for the current year are based on nominal profits earned in the preceding year. In the presence of high inflation, this practice implies that estimated profits will tend to be far too small, and therefore there will be a long lag before tax liabilities are paid. As inflation continues at a high rate, the lag ensures that the real value of the liabilities is reduced (the so-called Tanzi effect).
Data for aggregate profits of enterprises are not available. A series for aggregate operating surpluses, including private as well as public companies, is also not available on a comprehensive basis. Even if available, it might not be easy to forecast either of these magnitudes for a future year. GDP is therefore used as a proxy for the base of the tax on corporate profits in the equations below.
In equation (7.6), the buoyancy of the tax on corporate income is 0.9—less than, but not too different from, unity. The implication is that corporate tax payments grow somewhat less fast than GDP. In interpreting this result, one may keep in mind that profits of corporations are not necessarily proportional to GDP; in a recession a firm may have a zero profit-tax liability. It is also possible that tax laws have been changed progressively to favor corporations, although that is not necessarily demonstrated by this regression result and is contradicted for recent years by the ratios in Box 7.3.
Profits and profit tax liabilities are likely to be strongly procyclical rather than simply proportional to GDP. In equation (7.7), this conjecture is borne out, with a larger coefficient on deviations of GDP from trend than on trend GDP. Long-run buoyancy is estimated in this equation to be (1.0663 – 0.8651)/ (1 – 0.8209) = 1.12. In the case of corporate taxes, the more complete structural equation and the time-series specification resulted in the same econometric equation so only two regressions are reported for corporate tax revenue.
As an alternative to the regression estimates, forecasts of direct (corporate and personal) taxes may be based on the ratios given in Box 7.3.
Box 7.3.Turkey: Selected Tax Ratios for Direct Taxes in the Consolidated Budget
|(Percent of GDP)|
|Personal income tax||5.3||5.5||5.4||4.7||4.4|
|Corporate profits tax||1.1||0.9||1.0||1.1||1.4|
Indirect taxes on goods and services
Collection of indirect taxes increased steadily from 6 percent of GDP in 1991 to 8½ percent in 1995 (Tables 2.3 and 7.3). This increase mainly reflected increases in the various VAT rates and changes in the distribution of the petroleum consumption tax. The share of the petroleum tax earmarked for extrabudgetary institutions and local governments was reduced in 1994 and the share for the central government was thus raised; this can be seen in Box 7.5.11 Collections of the supplementary excise taxes on tobacco and alcoholic beverages have been negligible in recent years due mainly to difficulties in collecting this tax from the state monopoly responsible for these products. Customs duties on imported goods remained fairly stable, averaging 0.7 percent of GDP between 1991 and 1995 despite wide fluctuations in imports. Duties are expected to decline to 0.4 percent in 1996 with the reduction in tariff rates upon Turkey’s entry into customs union with the European Union.
Revenue from taxes on domestically produced goods and services has been divided into two categories for forecasting. One is the VAT; the data series is referred to below as GCRDVAT. For the other category, revenue from taxes on imports has been added to non-VAT domestic indirect tax revenue, and the resulting series is called GCROT below; this series includes revenue from the so-called supplemental VAT (cigarettes, beverages, and playing cards only, an excise-like tax), taxes on petroleum products, and taxes on financial transactions and stamp duties. Nominal private consumption expenditure is taken as a proxy for the base of the VAT. For OT, imports in liras was used at first as the proxy base since this revenue grouping includes import duties, but GDP yields superior econometric results, which are presented.
For non-VAT indirect taxes, GCROT, buoyancy is estimated in equation (7.8) to be about 1.06. Parallel interpretations for the other two equations for this category give long-run buoyancy estimates of 1.24 (equation (7.9)) and 1.12 (7.10). The lag between spending and revenue for this category of taxation is captured reasonably well by including a lagged endogenous variable, in equation (7.9), and even better by a lagged (two years) value of GDP in equation (7.10). Sticking with the interpretation of parliamentary adjustment, one would say in the case of equation (7.10) that initially there is underadjustment of tax rates compared with inflation, partly corrected after two years. However, petroleum and supplementary VAT rates are proportional. In principle such taxes have elasticities equal to unity, requiring no adjustment for inflation. The implication that revenues grow somewhat faster than GDP is more likely caused by changes in rates and categories not related to inflation, to shifts in the composition of spending among categories taxed at different rates, and/or to changes in the percentage of revenues shared with local governments and extrabudgetary funds.
In equation (7.11), the estimate of the proportional change in VAT revenue relative to private consumption is 1.09. Since the average proportional change in consumption relative to GDP over the same period is approximately 1.01, the buoyancy of VAT revenue is about 1.08.12 In equation (7.12) buoyancy is estimated to be 1.31. The divergence of estimated buoyancy from unity can be attributed to shifts in the composition of expenditure among categories taxed at different rates and to parlimentary changes in the number and contents of the various categories and the rates applicable to them. In practice the Turkish VAT is significantly progressive, higher rates being applied to categories of goods that are usually considered luxuries. (See Appendix III.) An alternative interpretation of the buoyancy estimates, for both categories of indirect tax revenue, is a (modest) tendency on the part of parliament to increase revenue from indirect taxes to narrow the fiscal deficit.
As an alternative to the regressions above, forecasts based on the tax ratios shown in Box 7.4 can be used. Note that import duties are expected to decline by an estimated 0.3 percent of GDP, about 1.1 percent in lira-price terms, from the beginning of 1996 when Turkey completes tariff harmonization prior to joining the European Union.
Box 7.4.Turkey: Selected Tax Ratios for Taxes on Goods and Services
|Taxes on domestic goods and services|
|Of which: Domestic VAT||3.4||3.6||3.8||4.3||3.9|
|(As percent of imports, c.i.f.)|
|Total taxes on imports|
|Of which: VAT||9.4||9.6||9.6||9.6||8.7|
|(As percent of GDP)|
|Total taxes on imports||2.0||2.1||2.3||2.3||2.6|
|Of which: VAT||1.3||1.4||1.6||1.7||1.9|
Tax revenues of local governments and extrabudgetary funds
Tax revenues of local governments and extrabudgetary funds (EBFs) are difficult to forecast independently. Much of the revenues of each group represents agreed portions of revenue collected by the central government, with the percentage that is allocated to local governments and to individual EBFs changing from year to year. One approach to forecasting these items is to estimate each as a percentage of the revenue from direct or indirect taxes, respectively, in the central government budget. Box 7.5 shows the percentages of direct and indirect revenue of local governments and EBF’s in the revenues received by general government during the years 1991-95.
During the 1991-95 period the ratio of tax revenues of the EBFs and local governments to general government revenues varied somewhat. The sharp decline in the ratio of indirect tax revenues in 1994 reflects the decision to raise the percentage of petroleum consumption tax revenues going to the central government from 50 to 90 percent, with a corresponding decline for the share sent to EBFs and local governments. Future revenue developments for these units of government will depend not only on revenue-sharing agreements with the central government but also on developments in the tax bases and rates of taxes levied directly by local governments, such as the property tax (see Chapter 2, Section d).
Box 7.5.Turkey: Tax Revenues of Local Governments and Extra budgetary Funds, 1991-95
(2) Nontax revenue of general government
Regression equations are presented below for nontax revenue (GGNR) of general government. Nominal output (GDP) is used as a proxy for the base. Given the dissimilar components of nontax revenue, GDP can only be a rough indicator of likely changes. However, alternative regressions, including an election year dummy (for profits of state-owned enterprises13) and the growth of the lira money stock (as an indicator of central bank seigniorage), did not yield strong results. There is also a dummy variable for nontax revenue that is required because of a break in the comparability of the data series between 1992 and 1993; social security contributions have been included starting in 1993.
Buoyancy in this case is estimated at well below unity in both equations, unlike the results above for various categories of tax revenue. Nontax revenue is a quarter of total revenue, so a forecast of total revenue that includes nontax revenue based on either of the equations above will tend to show revenues shrinking relative to GDP. Especially because we do not know whether other breaks in the continuity of this series may have occurred in earlier years (there is little change in value from 1991 to 1992), it would be sensible to forecast nontax revenue based on the ratio of this series to nominal GDP in recent years instead of using equation (7.13) or (7.14).
(3) Total revenue of general government
It is reasonable to use GDP as a proxy for the base of all tax revenue taken together, which is done in this section; among series that are readily available and likely to be forecast for a future year, there is no single indicator that is more representative of developments of tax bases in general in the economy. However, GDP is a rough proxy at best. To cite two examples, corporate profits vary more than proportionately with GDP, and excise taxes are related to consumption rather than to GDP.
The equations below give estimates of long-run buoyancy of 1.04–1.07. The dummy variables for elections and for the break in the GGNR series are both significant. The latter increases the regression forecast of total revenue by about one third in recent years. A lag of one year in the tax-base proxy, GDP, is indicated by the data. The long-run coefficient on deviations of GDP from trend, 0.6152/(1 - 0.3299) = 0.9181, is somewhat smaller than the long-run buoyancy, (0.6152 + 0.0848)/(l - 0.3299) = 1.045, which may be interpreted as more than proportional parliamentary tax relief in high inflation years, as discussed above. The standard error of estimate for equation (7.17), 1.1 percent, indicates that the aggregate revenue equation is likely to be reasonably accurate as a forecasting tool if factors influencing revenue remain as during the sample period.
Table 7.4 summarizes the major components of general government expenditure in 1991-95, including their ratios to GDP. Some of these ratios could provide a basis for forecasting expenditure in 1996, especially for the case in which policy is unchanged. In general, however, it is more sensible to regard government spending as determined by policy rather than by trend.
To link expenditure forecasts to the stance of policy, the simple table presented in Box 7.6 will be a useful device: Expenditure is divided into categories; then, for each category, the forecast is decomposed into “quantity” and “price” components. For some categories, “price” changes (changes in the implicit deflators) will depend on developments in the economy in general, such as the GDP deflator and import prices. However, for civil service pay in Turkey, special calculations may be needed depending on the terms of settlement of a strike or the government’s announced objectives regarding real pay levels in the coming year. Given the stance of policy, “volume” changes for various categories may be forecast judgmentally relative to the rate of growth of output projected for the economy as a whole.
Compensation of public sector employees
If the forecasts are to reflect unchanged fiscal policy, it may be decided that A, in Box 7.6, should equal the forecast of real growth of GDP.14 This amounts to assuming that the output of government services, such as education and health and law enforcement, will grow in line with output in general. If fiscal policy is to be tightened, then the value of A may be less than the rate of growth of output.
|Actual||Fitted GCRPT||Actual||Fitted GCRCT||Actual||Fitted GCROT|
|Year||GCRPT||Eq. (3)||Eq (4)||Eq. (5)||GCRCT||Eq. (6)||Eq. (7)||GCROT||Eq. (8)||Eq. (9)||Eq. (10)|
|1995||1995||212119||227015||214499||518037||486528||516387||1899903||1913878||1893294||1880670|Box 7.6.Forecasting Categories of Government Expenditure
|change, %||change, %|
|Compensation of employees||A||B|
|Purchases of goods and services from enterprises||C||D|
|Interest payments on domestic and foreign debt||n.a.||n.a.|
|Purchase of stocks of SEEs||H||D|
In case of a sharp cutback in the growth of government spending, it may be decided to limit the number of civil servants to the level of the last pre-forecast year. This is called a “workforce ceiling” or employee ceiling. It means that no agency or ministry can hire additional workers except to replace those who resign or retire. In this case, the “volume” change in employee compensation would equal zero.15
An even more severe measure is to ban any new hiring (for some specified temporary period), a “hiring freeze”. In this case, the volume change in employee compensation will depend on the rate at which existing employees leave the payroll due to retirement or job-switching. Suppose that normal attrition of the civil service in Turkey is two percent per year.16 If a freeze is imposed, there will be a gradual decline from the day the policy goes into effect that will reduce the number of workers by 2 percent by the end of one year.17
The “price” of employee compensation will depend on civil service pay policy. If the policy is to hold real government wages constant, B in Box 7.6 will approximately equal the rate of change of the CPI. In the early 1980s real government sector wages fell in Turkey, and during the latter 1980s there was a period of increases, associated with a relaxation of controls on labor organizations representing civil servants (Chapter 2, Section d). Elections also seem to have an effect on government pay—a comprehensive increase scheduled for early 1996 was brought forward to the latter months of 1995 in anticipation of the national election scheduled for that time. (The cumulative increase for the two year period, 1995-96, was not to be affected.)
Expenditure on goods and services
This category of spending is likely to vary approximately in line with the number of civil servants. Goods and services include uniforms for soldiers and policemen, typewriters and paper for government clerks, and cars and trucks. The average increase in price of these goods and services will tend to reflect trends in the average price of domestic output and imports. Note that the increase in the “prices” of other categories of government spending are also likely to be in line with the change in the CPI (and/or the change in the GDP deflator and import price index). Thus, all of the price changes are noted “D” in Box 7.6 except the one for personnel costs. If additional information is available, the framework in Box 7.6 can accommodate different deflators for the different categories of spending.
Other non-interest expenditure
If the forecasts are based on an assumption of unchanged policies, then the volume increases in other types of spending (letters E through I in Box 7.6) may equal that of real GDP, or reflect past trends in real government spending. If there is a significant stabilization package or some other substantial change in policies, then the real increases that are forecast for transfers, subsidies, net lending, and so forth would appropriately reflect the direction of the new policies (even though the actual percentage changes proposed as forecasts may be based to an extent on judgment). If transfers in real terms are forecast to grow less rapidly than population, clearly real transfers per head are intended to decrease under the new policies.
For the case of Turkey, it is useful to forecast foreign and domestic interest payments separately to make allowance in a systematic way for both exchange-rate and interest-rate effects. The general idea is to project changes in foreign and domestic debt and interest rates, and then multiply and aggregate to calculate the total interest payment of general government.
It will be useful to have in mind several developments that have characterized government financing in Turkey in recent years. One is that foreign borrowing by government turned negative in 1994-95 as the Turkish government repaid more debt than it borrowed from foreigners in these two years.18 This development is not typical of prior years, and is related to the foreign exchange crisis that occurred in 1994. Whether access to foreign markets is restored in the future will depend on whether lenders perceive that the policies and developments that contributed to the crisis in Turkey have been changed.
A second development is that government debt sold publicly is all short-term. In a high-inflation environment, lenders are unwilling to hold longer term debt instruments except at interest rates incorporating a very large risk premium. The fact that government debt is short term implies that one may use current interest rates in forecasting interest payments without allowing for a portion of longer-term debt borrowed at different rates in the past.
The various levels of government in Turkey borrow from a number of sources. Advances from the central bank to the central government have traditionally borne an interest rate of 4 percent, which is strongly negative in real terms for the rates of price increase that have occurred during the 1980s and 1990s. The maximum amount that may be borrowed in this form is limited to a proportion of the change from year to year in central government expenditure (12 percent for 1995, 10 percent for 1996), but in an inflationary environment the limit is not necessarily binding. On the other hand, because of the effects of such borrowing on overall liquidity, the central government may not use this source of funds to the maximum extent.19
Besides advances, the central government may borrow from the central bank in another way. When state-owned enterprises experience losses as a direct consequence of carrying out official policy, they may be compensated in the form of nontransferable treasury securities.20 The interest on such debt may itself be paid in the form of securities, instead of cash, at the goverment’s discretion. (For this reason, the change in the stock of government debt may grow each year by more or less than the size of the deficit “on cash basis,” as discussed below.) This debt may grow in discrete jumps rather than smoothly, the timing of “borrowing” linked to episodes of “debt consolidation,” which is a kind of cosmetic accounting exercise to improve the apparent health of the balance sheets of official financial institutions. Since the interest on this debt may not (under present practices) appear as a cash expenditure (because it is securitized), it is treated as interest at a rate of zero in this workshop. This debt finances a disguised form of expenditure, called “quasi-fiscal expenditure,” which, in a more transparent accounting system would appear as subsidies in the accounts of the central government.21
The government also borrows in the form of marketable treasury securities (with a maturity of three months up to one year) and bonds (a maturity of one year). As in other economies, these may be held by banks, enterprises, households, and nonresidents. Turkish commercial banks are obliged to hold treasury bills equal to 30 percent of their deposit liabilities (see Chapter 4, Section d), in addition to their reserve requirement. In a high-inflation economy, this so-called liquidity requirement creates an enormous “demand” for government securities. Nevertheless, with the unavailability of foreign finance in 1994-95 the government has sought to increase its domestic borrowing from banks and nonbanks. As a result, the nominal rate of return on marketable government securities has been high, even in real terms, including a variable, and occasionally very large, risk premium.
A method for forecasting interest expenditure is indicated in Box 7.7, below, and illustrated with data for 1995 in Box 7.8. The interest expenditure figures presented there apply to general government. The first and second lines in Box 7.7 contain historical data for foreign debt in U.S. dollars and the average foreign interest rate on borrowing by Turkey. These figures are taken from Table 7.5 in this chapter and from Table 8.11 in the following chapter. The third line shows the resulting interest payment due on foreign debt of general government.22 For this calculation, the debt stock is the arithmetic mean of the end-period debt stocks from years t and t-1.23
Total interest expenditure by general government is given in line 4, taken from Table 7.2. Interest expenditure on domestic debt, line 5, for the historical period is calculated by subtracting the estimated foreign interest payment, line 3, from the total in line 4.
The stock of domestic debt in line 6 is from Table 7.5. Line 8 is calculated by dividing the estimated interest payment on domestic debt, line 5, by the average stock of domestic debt, line 7, and multiplying by 100. The difference between this implicit average rate of interest and the market rate for treasury bills (line 9)24 is interpreted as arising from the proportion of general government domestic borrowing that takes place at non-market interest rates, assumed to be at a rate of virtually zero. The percentages shown in line 8 are actually affected by factors other than brrowing terms, such as change from year to year in the average maturity of government debt, but information on this aspect of borrowing is not available in a systematic form.
Use of the method in Box 7.7 for forecasting the interest payment of general government requires four magnitudes. One of these, the interest rate paid by Turkey on foreign debt, can be projected based on figures in Chapter 8. Besides data in Table 8.11 on the average interest rate on foreign debt, Table 8.13 gives a forecast of foreign interest rates for 1996. The change in the latter series can be used to forecast the level of the former for 1996.
Box 7.7.Turkey: Analysis of General Government Interest Expenditure
(In trillions of liras except as noted)
|1. Foreign debt of general government|
|(in billions of U.S. dollars)||28.5||30.8||34.1||34.4|
|2. Implicit average interest rate on|
|government foreign debt (percent)||8.0||7.3||7.2||7.5|
|3. Estimated interest payment on the average|
|stock of government foreign debt|
|(in trillions of liras)||15.5||23.8||69.1||117.8|
|4. Total interest expenditure of general|
|5. Interest expenditure on domestic debt||29.8||102.2||257.3||535.5|
|6. Stock of domestic debt||222.1||384.9||846.7||1,557.8|
|7. Average stock of domestic debt||166.0||303.5||615.8||202.2|
|8. Implicit average interest rate on average|
|stock of domestic debt||18.0||33,7||41.8||44.5|
|9. Market interest rate on three-month|
|treasury bills, yearly average (percent)||94.2||85.8||156.2||123.0|
|10. Implicit proportion of funds borrowed|
|at market rate (percent)||19.1||39.3||26.8||36.2|
|11. Memorandum item:|
|Liras/dollar, period average||6,872||10,985||29,609||45,845|
The other three needed values are as follows: (1) the market interest rate on government securities, (2) the proportion of finance acquired by general government at market terms, and (3) the general government balance (overall surplus or deficit) for the forecast year. The first two of these inputs will be determined by the assumed stance of policy. The proportion of financing to be obtained from abroad must be forecast judgmentally in the absence of budgetary or other official information. The value of the government balance will result from the other fiscal forecasts (revenue and non-interest expenditure).
The fiscal balance depends on interest expenditure, and at the same time interest expenditure depends on the amount of borrowing needed to finance the projected fiscal balance. Box 7.8 demonstrates how to treat this simultaneous relationship. The general procedure is as follows: (1) Write an algebraic equation for the budget balance that includes forecast revenues and expenditures plus the unknown value of interest expenditure, I. (2) Write a second algebraic equation for interest expenditure on the average stock of debt, disaggregated into foreign and domestic components and including borrowing in the forecast year (which depends on I). (3) Then solve the two equations simultaneously.
The method is illustrated in Box 7.8 using data for 1995 (as though it were a forecast year). Lines 1 through 4 show “forecasts” of government revenue and expenditure, with unknown interest expenditure represented by the symbol I. (Privatization receipts are added to revenue since they reduce the amount of the deficit to be financed through borrowing.) Foreign debt and interest on it is calculated in lines 5 through 11. The portion of the forecast-year deficit to be financed abroad is determined (according to an assessment of attitudes of foreign lenders and/or the policies of government), the average foreign debt stock for the forecast year is calculated, and the interest on it, and the interest payment is converted to liras. Domestic debt is the amount of the deficit not financed abroad, calculated in lines 12 through 16 of Box 7.8.25 The value in line 15 is from Box 7.7, line 8. In line 17, the sum of the interest expenditure on foreign and domestic debt is set equal to the unknown, I; the equation can be solved for the forecast value of I, which will yield a forecast of the fiscal balance for the forecast year.
Since actual data for 1995 have been used in Box 7.8, instead of forecast data, one can check the resulting interest expenditure forecast against the actual figure in Table 7.2. It is obviously too low. Note that the discrepancy cannot be due to forecast error since lines 8 and 10 in Box 7.7, implied by data for 1995, are used to forecast 1995 in Box 7.8. The reason for the discrepancy is related to unrecorded quasi-fiscal expenditures. The increase in the debt of general government given in the fiscal tables of this chapter, for recent years, does not match the borrowing that would have occurred to finance the budget deficits that occurred in those years.
A part of the discrepancy between the change in government debt and the fiscal deficit can be explained by change in the lira value of foreign debt caused by changes in the nominal exchange rate. A second part (quite small) is due to privatization receipts, treated below the line in the Turkey fiscal accounts. But there is a residual discrepancy that is occasionally large. This discrepancy reflects payments made by government to reimburse state banks (including the central bank) for financial losses sustained by those institutions because they carried out government policy in a way that was inconsistent with financial viability.
Box 7.9 shows how to adjust the forecast of interest expenditure to allow for quasi-fiscal flows not included in the fiscal accounts. The first line contains data for the change in the stock of government domestic debt, calculated on the basis of data in Table 7.5. Changes in foreign debt are shown in line 2 in dollars and in line 3 in liras; line 3 contains figures from line 2 that have been converted at the average-period exchange rate. Notice that this treatment of foreign debt systematically excludes an estimate of the valuation adjustment (as discussed in Chapter 4); since the lira is depreciating continuously against the U.S. dollar, the value of Turkey’s foreign debt will increase as measured in liras even when no new borrowing occurs. The calculation in line 3 is designed to exclude the part due solely to exchange rate changes, and to include the part due to net new borrowing. The calculations are, however, only an estimate, and no allowance is made for the tendency of Turkey’s foreign debt stock to increase in terms of dollars when the dollar weakens against the German mark.
The conventional, recorded deficit of general government is shown in line 5 of Box 7.9 using data from Table 7.2. The fiscal balance is adjusted by subtracting privatization receipts (which are small). The difference between the deficit, so adjusted, and the change in total debt excluding valuation effects, is given in line 8. The discrepancy is attributed to quasi-fiscal expenditure. In the memorandum items at the bottom of Box 7.9, these quasi-fiscal expenditures are reported in percent of GDP, along with measures of the conventional deficit, so one can easily compute the fiscal balance that would be recorded if the quasi-fiscal expenditures were moved onto the fiscal accounts explicitly.
Box 7.8.Turkey: Interest Forecasting Example for 1995 for General Government
|Trillions of liras|
|1.||Total revenue plus privatization receipts (forecast1)||1,925,1|
|2.||Non-interest expenditure (forecast1)||1,774.8|
|3.||Interest expenditure, foreign plus domestic (unknown)||I|
|4.||Fiscal balance plus privatization receipts,|
|1,925.1 - 1,774.8-I||150.3-I|
|5.||Foreign debt at end-1994 (Table 7.5, billions of $)||34.1|
|6.||Foreign financing during 1995 (forecast)||0.3|
|7.||Average stock of foreign debt||34.25|
|8.||Average interest on foreign debt (forecast)||7.5|
|9.||Turkish liras/dollar, average period (forecast)||45,845|
|10.||Foreign interest expenditure, converted to trillions|
|11.||External borrowing during 1995 converted to|
|12.||Domestic debt at end-1994 (Table 7.5)||846.7|
|13.||Domestic debt at end-1995: end-1994 debt minus|
|foreign borrowing, minus fiscal balance, plus|
|privatization receipts, 846.7 - 13.7- 150.3+I||682.7+I|
|14.||Average stock of domestic debt||764.7 + I/2|
|15.||Market interest rate in percent (forecast1), times|
|proportion of funds borrowed at this rate (forecast1)||44.5|
|16.||Domestic interest expenditure, (0.445)-(764.7 +I/2)||340.3+0.223I|
|17.||Total interest; foreign interest plus domestic|
|interest,I = 117.8 + 340.3 + 0.223 I||589.6|
|18.||Unrecorded quasi-fiscal expenditure (forecast, Box 7.9)||221.7|
|19.||Revised average stock of domestic debt,|
|(764.7+I/2) + 221.7/2||875.6+I/2|
|20.||Revised domestic interest expenditure,|
|(0.445)(875.6 +I/2)||389.6 + 0.223 7|
|21.||Revised total interest, I = 117.8 + 389.6 + 0,223 I||653.1|
|22.||General government balance plus privatization/|
|recipts, line (4)||-502.8|
In order to make a forecast of interest expenditure for 1996, a projection must be made of unrecorded quasi-fiscal expenditure (line 8). To some extent the quasi-fiscal expenditure is attributable to recapitalization of the central bank and certain state banks, as already described. If it is judged that the compensation, or recapitalization, of various banks will continue in the next year, then one can use the value for 1995, in percent of GDP, as a forecast. If, on the other hand, one believes that these expenditures constituted an episode of extraordinary financing, linked to the financial crisis of 1994, then it would be reasonable to take a smaller percentage, or zero, as the expected value in 1996.
Box 7.9.Turkey: Unrecorded Net Quasi-Fiscal Expenditure
(In trillions of liras, except as noted)
|1.||Change in domestic debt||461.8||711.1|
|2.||Change in foreign debt ($ millions)||3,269||297|
|3.||Change in foreign debt in liras||96.8||13.6|
|4.||Change in total debt excluding VA||558.6||724.7|
|5.||General government deficit (Table 7.2)||305.0||528.2|
|6.||Privatization receipts (Table 7.2)||14.5||25.2|
|7.||Deficit net of privatization receipts||290.5||503.0|
|expenditure, (4) - (7)||268.1||221.7|
|(In percent of GDP)|
|Deficit of general government1||7.9||7.0|
|Unrecorded quasi-fiscal expenditure||6.9||2.9|
Conventional definition (excluding unrecorded quasi-fiscal expenditure).
Conventional definition (excluding unrecorded quasi-fiscal expenditure).
e. Exercises and issues for discussion
- (1) Forecast central government tax revenue by major categories for 1996 using the format of Table 7.3. In preparing the forecast, use the forecasts of nominal GDP and private consumption from the projections in the preceding workshop, as appropriate. No forecasting equations exist for “additional taxes” or “taxes on wealth.” These small categories can be forecast judgmentally.
- (2) Using the disaggregated forecasts of central government revenue from exercise (1), and the information presented in this chapter, forecast general government revenue, including both tax and nontax revenue. Explain how you calculated the main elements of revenue for local governments and the extrabudgetary funds.
- (3) Prepare a forecast of general government and public sector operations in the format of Table 7.2. Use your sectoral projections of output and prices from the preceding workshop to forecast expenditures in the format of Table 7.4. For this forecast of general government and public sector operations, use the following assumptions:
- The stance of fiscal policy is unchanged from 1995. (Discuss and reach a decision about what this means for Turkey in 1996.)
- Overall, the state economic enterprises (SEEs) record a deficit equal to two percent of GDP.
- Net external borrowing is negative one half of one percent of GDP (with unchanged policies).
- Import duties decrease on January 1, 1996. Tax receipts from imports fall by 0.3 percent of GDP because of these tariff reductions, enacted in preparation for membership in the EC.
- The average interest rate on marketable domestic government debt remains unchanged in real terms from 1995.
- Privatization receipts increase in line with nominal GDP.
- (4)Forecast total revenue using one of the aggregate equations (20 through 22). How does the result compare with the forecast from exercise (2), above? In principle, which is more accurate?
- (5)What value for total revenue would be forecast using the assumption of unitary buoyancy? Is this figure larger or smaller than the forecasts obtained in exercises (2) and (4), above, and why? Under what assumptions would the forecast of unitary buoyancy be the most appropriate?
- (6)Relative to your projections, review the likely effect of the following factors on the development of the main components of revenues and expenditures in 1996:
- lower output growth;
- higher inflation;
- a more depreciated exchange rate;
- higher international and domestic interest rates;
- less consumer expenditure;
- higher growth of imports.
- (7)Comment on changes in the structure of revenues and expenditures that appear desirable in the medium term.
- (8)What interpretation should be attached to the “slope” coefficients of the “structural” regression equations in this workshop (the coefficients of the tax-base proxies)? What are the implications of such a coefficient being greater or less than unity? Can one claim that use of these individual equations to forecast revenue will give superior results to the assumption of unitary buoyancy for total revenue? Would it be sensible to assume buoyancy much greater or less than unity over the medium term?
- (9)If national pension (social security) payments by workers are included in nontax revenue, as stated in the text, what do the regression results suggest about the financial soundness of this program?
|(In billions of Turkish liras)|
|Total expenditures and net lending||762,764||1,339,629||2,428,083|
|Of which: social security||108,940||200,398||376,882|
|Quasi-fiscal outlays and|
|extrabudgetary transfers 1/||64,107||57,079||81,453|
|Other current expenditure||268,704||451,221||801,638|
|Wages and salaries||217,192||346,490||592,796|
|Purchases of goods and services||45,236||70,181||164,432|
|Net lending 1/||677||425||(2,943)|
|Balance (revenue - expenditure)||-244,541||-305,041||-528,180|
|State economic enterprises (SEEs)|
|Total public sector borrowing|
Budgetary data have been modified to include certain quasi-fiscal outlays and off-budget transfers to entities outside the general government.
Budgetary data have been modified to include certain quasi-fiscal outlays and off-budget transfers to entities outside the general government.
|(In billions of Turkish liras)|
|Taxes on income||125,793||278,074||435,999|
|Personal income tax||106,661||181,884||329,795|
|Corporate income tax||19,132||43,976||103,241|
|Taxes on wealth||2,531||5,659||24.438|
|Taxes on domestic goods and services||89,447||214,353||429,232|
|VAT on domestic transactions||50,892||110,918||212,119|
|Supplementary VAT (excises)||388||8,029||16,937|
|Petroleum consumption tax||12,791||46,625||103,180|
|Financial transactions tax||7,129||16,467||25,340|
|Other indirect taxes||10,276||18,637||42,459|
|Taxes on imports||46,213||89,596||194,609|
|VAT on imports||30,985||65,824||142,861|
|Other duties and levies||2,057||1,930||3,315|
|Abolished indirect taxes||289||24||33|
|Total tax revenues||264,273||587,706||1,084,311|
|(In percent of GDP)|
|Direct taxes Of which||6.5||7.3||6.1|
|Personal income tax||5.4||4.7||4.4|
|Corporate income tax||1.0||1.1||1.4|
|VAT (domestic and import)||4.1||4.6||4.7|
|Import duties (excluding VAT)||0.8||0.6||0.7|
|Total tax revenues||13.3||15.2||14.4|
|(In billions of Turkish liras)|
|(In billions of Turkish liras)|
|Current transfers except social security||95,838||158,017||227,105|
|Quasi-fiscal outlays and|
|Goods and services||45,236||70,181||164,432|
|(In percent of GDP)|
|Current transfers, except social security||4.8||4.1||3.0|
|Other goods and services||2.3||1.8||2.2|
|(In billions of Turkish liras)|
|(In billions of Turkish liras)|
|Nonsecuritized debt to the Central Bank||102,354||255,695||217,940|
|Other government (excluding SEE’s)||27,577||47,432||196.806|
|(In millions of U.S. dollars)|
|(In percent of GDP)2/|
|Domestic debt, central government||18.0||20.7||18.0|
|Nonsecuritized debt to the Central Bank||5.2||6.6||2.9|
|Foreign debt, central government||15.7||23.3||18.9|
|(In billions of Turkish liras)||1,981,867||3,868,429||7,554,758|
|(In millions of U.S. dollars)||180.422||130,652||164,789|
|Liras per dollar, end of period||14.473||38,726||59,650|
|Liras per dollar, period average||10,985||29,609||45,845|
End-of-period stock divided by GDP.
End-of-period stock divided by GDP.
|(In billions of Turkish liras)|
|Total current revenues||140,393||246,738||410,564||924,033||1,583,537|
|Sales of goods and services||133,013||232,330||386,679||866,734||1,474,001|
|Other sales revenues||1,296||1,874||3,103||1,427||4,341|
|Total current expenses||-166,519||-293,111||471,931||-1,027,111||-1,582,690|
|Cost of goods and services sold||-88,365||-155,774||-255,743||-541,831||-899,261|
|Provisions for exchange rate differences||-10,346||-13,288||-15,145||-67,738||-45,592|
|Operating surplus (+) or loss (-)||-26,126||-46,373||-61,367||-103,078||847|
|Direct tax obligations||-1,644||-2,361||-3,758||-6,239||-26,824|
|Operating surplus/loss after direct||-27,770||-48,734||-65,125||-109,317||-25,977|
|After-tax available income||-27,742||-48,714||-65,093||-109,269||-25,977|
|Wages and salaries||-33,151||-57,486||-99,576||-156,848||-222,904|
|Duty losses accrued on goods sold||5,695||15,422||11,624||18,591||16,341|
|Operating surplus adjusted for accrued|
|Operating surplus adjusted for accrued|
|revenues including accrued duty losses||-14.0||-11.8||-11.8||-9.0||1.1|
|(In percent of GNP)|
|Wages and salaries||5.2||5.2||5.0||4.0||2.8|
|(In billions of Turkish liras)|
|Total social security system|
|(In percent of GNP)|
|Total social security system|
Some Tax Revenue Sources and Their Proxy Bases1
|Tax Revenue Sources||Tax Bases, Proxies|
|A. Taxes on net income and profits|
|1. Corporate||1. Profits of enterprises derived from the national accounts; operating surpluses of enterprises; GDP|
|2. Individuals||2. Wages and salaries and other private income from the national accounts; national income; GDP|
|B. Social insurance taxes or contributions|
|3. Employer||3. Wages and salaries; GDP|
|4. Employee||4. Wages and salaries and nonwage income of self-employed persons|
|C. Taxes on property|
|5. Real estate||5. Rents; aggregate value of real estate; aggregate wealth|
|6. Personal wealth||6. Property income; national income; wealth|
|7. Death and gift taxes||7. Demographic indices of deaths and number of persons in the highest income brackets; GDP|
|8. Property transfers||8. Value of urban properties and number of transactions|
|D. Taxes on production, consumption, and sales|
|9. Sales, turnover, or value-added tax||9. Private consumption expenditure; retail sales|
|10. Selective excises||10. Production or consumption figures on major excisable commodities; private consumption|
|11. Profits of fiscal monopolies||11. Value or volume of consumption of fiscal monopoly commodities; GDP|
|E. Taxes on international trade and transactions|
|12. Import duties||12. Value and volume of imports|
|13. Export duties||13. Value and volume of exports|
|14. Export marketing boards||14. Estimated profits on the exports handled by boards; total exports|
|15. Exchange profits||15. Value of exchange transactions|
|F. Other taxes|
|16. Business and professional licenses||16. Number of licenses issued; nominal value added in reatiling; GDP|
|17. Poll taxes or personal taxes||17. Demographic data for relevant age group|
|18. Stamp taxes||18. Value of important categories of transactions involved and estimated frequency of transactions|
|19. All other taxes on income, property, production, and trade||19. National income; GDP|
The table is adapted from Sheetal Chand, “Some Procedures for Forecasting Tax Revenue in Developing Countries,” DM/75/91, IMF, October 1975.
Proportional Adjustment Method, The General Case1
Assume the following series of actual revenue collection in years 1 through n:
T1, T2, …, Tn-1, Tn
where n is the last or the current year.
Let the estimated revenue effect of discretionary changes in the years in which they occurred be as follows:2
D2, …, Dn-1, Dn
Let us now construct an adjusted series of revenue equal to those revenues that would pertain if the current year’s tax structure had been in operation throughout the period under analysis:
AT1, AT2, …, ATn-1, ATn
Since the current year (n) is the reference year, the actual and adjusted revenue for that year would be the same.
ATn = Tn
However, actual revenue collection in years n-1, n-2, 2, 1, must be corrected for discretionary changes in subsequent years. Under the proportional adjustment hypothesis, the series of adjusted revenue, with n as the reference year, becomes as follows:
which can be expressed by the general formula:
It may be helpful to provide a numerical illustration of the method. Let the actual revenue collections in years 1 through 5 be as follows:
T1 = 100, T2 = 140, T3= 170, T4= 250, T5 = 320
Assume that discretionary changes occurred only in years 2 and 4, with
D2 = 20, D3 = 0, D4= 30, D5 = 0
Let us now construct and adjusted revenue series with year 5 as reference year. The effect of the discretionary action in year 4 was to raise revenue collections in that year by 13.64 percent:
If this discretionary action had been taken at the beginning of the data period, the proportional adjustment hypothesis would imply a corresponding proportionate increase in revenue collections in all years preceding year 4. Similarly, the effect of the discretionary action in year 2 was to raise revenue in that year by 16.67 percent:
Only revenue in year 1 needs to be corrected for the change in year 2. Thus, if both discretionary actions had been taken at the beginning of the data period, the revenue series would have become as follows under the proportional adjustment hypothesis:
AT5 = T5 = 320
AT4 = T4 = 250
Turkey: Tax Summary (as of May 1996)
|Personal Income Tax||Progressive rates from 25 percent up to 55 percent|
|Corporate Tax||25 percent.||Net accounting income minus tax exemptions plus non-tax-deductible expenses.|
|Value-Added Tax||1 percent.||a. Supply of selected agricultural products (excluding supplies to final consumers).||a. Supply of agricultural products and services by income-tax-exempt small farmers and by farmers taxed on lump-sum basis.||a. Supplies of agricultural products to the consumers are taxed at 15 percent general rate, whereas applicable rate for basic staples is 8 percent in all cases.|
|b. Supply of sendees and capital goods subject to the financial leasing law.||b. Excluding the leasing of passenger cars.|
|c. Supply of books, newspapers and periodical magazines.|
|d. Supply of second-hand passenger cars.|
|8 percent.||a. Supply of basic staples (for example, general stock animals, fish, poultry, milk, cereals, bread and fresh vegetables and fruits).|
|b. Supply of services by private schools and colleges.||b. Educational services supplied by nonschool entities are subjected to the general rate|
|c. Admission fares to theaters, movie, operas, operettas, and ballets.|
|d. Supply of papers used by the printing press.|
|e. Supply of natural gas and cash-register machines.|
|15 percent.||Ail remaining supplies of goods and services within the country and importation of such goods except luxury items (below).||1. Exportation of goods (zero rating).|
|2. Banking and insurance services.|
|3. Supply of securities, money and foreign currency, gold bullions, postage stamps, fiscal stamps, and other similar stamps.|
|4. Supply of sea, air, and railroad vehicles including maintenance and repair services rendered in sea- and airports.|
|5. International transportation services.|
|6. Diplomatic supplies.|
|7. Certain social, cultural and military goods and services supplied by bodies governed by public law.|
|8. Transportation of foreign petroleum products through domestic pipelines.|
|23 percent.||a. Supply of luxury items (for example, caviar, furs, cosmetic products).|
|b. Supply of passenger cars, motorbikes, boats, aircraft, and durables (audio video, sets, and so forth.).||b. Passenger cars modified for disabled persons are taxed at 15 percent. |
Larger ships and durables for industrial use are taxed at 15 percent.
|c. Supply of plastic packaging materials.||c. Plastic garbage bags and plastics for industrial use are taxed at 15 percent|
|d. Rentals of passenger cars, motorbikes, boats, and aircraft.|
|e. Supply of fire arms (handguns and rifles).|
|f. Supply of slot machines and other gambling devices.|
|g. Supply of cable-TV and telephone message services.|
|h. Supply of services in bars, discos, night clubs and in golf courses, private clubs and saunas.|
|40 percent||a. Supply of cars with engine capacity above 2000 cc.|
|b. The financial leasing of cars.|
|Petroleum Consumption1 Tax||Percent|
Heating oil 45
Fuel oil 30
Liquid petroleum gas 20
|C.I.F. value of imports plus all taxes, funds, charges, and fees.||None.||There are currently two other levies on petroleum products besides the Petroleum Consumption Tax (customs duty, and Petroleum Price Stabilization Fund).|
The upper limits of PCT are increased with recent legislation to compensate revenue losses due to the elimination of some of the existing levies following entry into the E.U. Customs Union.
|Customs Tax||Levied at different rates||C.l.F. price of importing goods.||The goods specified in Article 7 of the Customs Tax Law.|
The goods imported for physical and legal persons that are specified in Article 8 of the Customs Tax Law.
|Diplomatic exemptions.||The goods specified in Article 9 of the Customs Tax Law.|
|Personnel and household goods.||Personal and household goods under the provisions determined by the customs administration are exempted.|
|Relief for tourists.||Cars and other belongings of tourists who visit Turkey can obtain these reliefs.|
|Advantages for people living on border regions.|
|Temporary exemption.||According to Article 119 of the Customs Tax Law, the customs duties for the goods entered temporarily into the territory of Turkey are subject to guarantee-deposit under the conditions determined by the Ministry.|
|Supplementary VAT||Taxable base for the Supplementary VAT (SVAT) is same as that for the VAT. SVAT is levied on selected goods imported, or “supplied domestically by their manufacturers.”|
|100 percent.||a. Cigarette and tobacco products.|
|100 percent.||b. All kinds of spirits||b. All kinds of wines, except sparkling wine.|
|15 percent.||c. Wines and beer.|
|10 percent.||d. Nonalcoholic beverages.||d. Plain and fruit sodas, fruit juices.|
|50 percent.||e. Methylated spirit.|
|60 percent.||f. Playing cards.||f. Toy playing cards intended for children.|
|60 percent.||g. X-ray films.|
|Motor Vehicles Tax1||Specific rates varying with age, net weight and luxury category of the vehicle (there are four luxury categories—the rates shown in the tariff have to be multiplied by the factor 3 for first category, 2 for second category, 1.5 for third category, and I for fourth category vehicles).||Automobiles, vans, panel, and land vehicles.||As of 1995, an automatic indexation procedure is in force, based on end-of-period inflation (WPI) during the previous year.|
|Specific rates varying with age and carrying capacity of vehicle.||Vans, buses, trolley buses, trucks, lorries, towing vehicles.|
|For lorries, trucks, and towing vehicles, carrying capacity is determined according to their maximum merchandise and loading weight.|
|For vans, buses, and trolley buses by the maximum passenger capacity.|
|Specific rates varying with age and type of vehicles.||Personal motorcycles (including motorized bicycles) foreign and passenger transport motorcycles, triporters.|
|Specific rates varying with age and engine power for yachts, motor boats, and all kinds of private boats.||Tractors, private yachts, private boat, and all minds of motor boats.|
|Specific rates proportional to engine horsepower.||Tractors.||Tractors are taxed one time only, at the time of purchase.|
|Specific rates varying with age and maximum take-off weight of the vehicles.||Airplanes and helicopters.||Aircraft belonging to Turukuzu and Tuurhene Murumu.|
|Financial Transactions Tax||5 percent.||Bank charges and insurance premium.|
|Motor Vehicles Purchase Tax1||6 percent (ad valorem—standard rate). |
There are other rates depending on the classification of the vehicle.
|Purchase price of any motor vehicle inclusive of VAT.|
This tax will be abolished once the Special Consumption Tax comes into effect.
This tax will be abolished once the Special Consumption Tax comes into effect.
Buoyancy and Elasticity with High Inflation
The relation between buoyancy and elasticity is illustrated for three cases in the set of diagrams in Figure 7.A1. In all panels, total tax revenue is shown on the vertical axis, and the proxy for the tax base, assumed to be GDP, is on the horizontal axis; all values are nominal. Rays through the origins (dotted lines) show unchanging values of the ratio of total revenue to GDP; revenue values that lie along a ray imply unitary buoyancy and conversely. Actual tax collections are shown by the solid lines. Growth in real terms is assumed to be zero (or, real change in output is small relative to changes in the price level), so that changes in GDP reflect primarily inflation.
In the top panel, the tax system has zero elasticity. There is no tendency for revenues to increase as nominal GDP rises over time. (This would be the case if all taxes were specific.) In order to fund the government sector, parliament is forced by continuing high inflation to raise tax amounts, causing the horizontal tax line to shift upwards repeatedly. In principle, by adjusting the amounts of specific taxation each year in line with inflation, the legislature can increase revenues in step with nominal GDP so that the ratio of one to the other remains constant, as in the diagram. Elasticity is zero, but buoyancy is unity.
In the second panel, the underlying tax system has an elasticity higher than unity—for a given percentage change in nominal GDP, the accompanying percentage change in collected revenues is larger than the GDP change (with no changes in tax rates or exemptions). This would be the case, for example, if all tax revenues arose from progressive systems of taxation of personal income and/or corporate profits. These tax systems specify higher rates of taxation on upper “brackets” (portions) of income, reflecting the notion that individuals or companies paying those high rates on part of their income are unusually prosperous and can afford to bear a somewhat higher tax share. But it is high inflation rather than increases in real income that is at work in Figure 7.2, and the tax burden is increasing in real terms for taxpayers in general. To offset this unintended increase, the legislature each year will increase the tax brackets in line with the inflation that is occurring. Thus, the tax line in the middle panel shifts to the right, avoiding the increase in taxes that would otherwise occur. If the adjustments in tax brackets are the correct size, they will just offset the effects of the high elasticity in the presence of high inflation, and the economy will experience unitary buoyancy.
In the bottom panel, the tax system itself has unitary elasticity-for example, a VAT on expenditure with a unified rate and no exemptions. In this case, the ratio of revenue to GDP stays constant with no adjusting legislation from parliament. The buoyancy of the tax system is therefore equal to its elasticity. Tax systems in practice typically combine the three types of relation between elasticity and buoyancy. Proportional taxes tend to result in a tax line lying along the ray in the bottom panel, while any specific taxes, exemptions and deductions would reduce the slope of the tax line, and a progressive income tax system would increase this slope. The higher the inflation rate, the stronger are the pressures on the legislature to raise specific-tax rates, to index brackets, or to modify the tax system by increasing the role of proportional taxes.
Figure 7.A.1Tax Buoyancy and Elasticity: Analytical Boundaries
Technical Aspects of Estimates of the Revenue Forecasting Equations
Regression results for forecasting revenue by tax category are presented in Section d of this Chapter. Notes on certain aspects of those statistics are given in this appendix, covering especially (1) evidence of autocorrelation in the presence of lagged endogenous variables, (2) interpreting deviations from trend GDP, (3) the election year dummy variable, and (4) model selection criteria.
The Durbin-Watson statistic is not an unbiased measure of autocorrelation in the context of an equation containing lagged endogenous variables. It may retain usefulness as a general indicator, and it is presented for all of the regressions for that purpose. A better measure is the t-value of the appropriate lag term of a vector autoregression of the estimated residuals, for example from the equation,
The values of ρ1 and the t-value of ρ1, estimated simultaneously with the revenue equation, are reported when inclusion of lagged dependent variables may bias the Durbin-Watson statistic. Acceptance of the null hypothesis that ρ1 is not significantly different from zero means a finding of low probability of (first-order) autocorrelation.
The estimate of ρ1 is also biased downward in small samples, frustrating the intention of the use of this substitute statistic. However, for t-values of ρ1 that are small, implying little probability of autocorrelation, the bias will also be small. The formula for computing the bias in the estimate of ρ1 is -(1 + 3ρ1)T where T is sample size. Thus, if autocorrelation is apparently slight, for example estimated ρ1 – 0.2, the bias is around 15 percent (for T – 11), but for estimated ρ1 = 0.75, bias is 30 percent, and the implied unbiased value of ρ1 is therefore 1.0.1
To allow for cyclical effects on tax revenues, trend values of the proxy base (for example GDP) have been included as well as actual values in the second and third versions of the regression equation. For those cases in which the trend is found to be significant, the following is an interpretation: Let Y* represent the natural logarithm of trend GDP, and Y be the log of actual GDP. If R is the log of revenue from a particular tax, one might use the simple specification,
so that b is the effect on revenue of the portion of GDP that is higher or lower than trend, and c is average buoyancy.2 Ignoring terms after c.Y*, and rewriting the above expression, one obtains
so that, in a regression of Rt on Yt, and Y*t, the coefficient on Yt, will be an estimator of b, the coefficient on Y* will be an estimator of c - b; and the sum of the two coefficients will be an estimator of c.
In some cases a dummy variable for election years is included on the right-hand side of the revenue equations.3 As discussed in a preceding section, it may be inferred that parliament is under pressure to adjust the tax system regularly because of continuing high inflation unless the tax system has an average elasticity close to unity. It may be politically difficult to carry out adjustments to preserve buoyancy in cases of taxes that must be increased when legislators face an election in the fall. (Personal-income and corporate-profits taxes are progressive and are therefore reduced when adjusted.) If the political effect on tax adjustment occurs with each election, and if in each election year the adjustment is roughly the same size (in proportional terms), then allowing for it in the regression equation will increase the explanatory power of the regression estimates. Election-year effects are found to be significant and are reported in the body of the chapter for non-VAT (tariffs, excises) indirect taxes and total revenue.4
Values of SIC and AIC shown with regression results below are logarithms of the Schwartz and Akaike Information Criteria. Both of these model selection criteria involve the product of the mean of the sum of squared residuals and a penalty factor based on the number of estimated coefficients relative to sample size. (Thus, SIC and AIC are in some respects similar to R-square-bar.) They are intended to guard against the case in which the addition of a regressor would reduce the sum of squared residuals without being likely to improve one-step-ahead forecasts. In general, smaller values of SIC and AIC (for logarithms, a larger negative number) are preferable to larger values comparing across several equations that model a given relationship. (In this sense, the criteria are like standard errors.)5
See Peter Kennedy, A Guide to Econometrics (Cambridge, Mass.: The MIT Press, 4th ed., 1998), pp. 149-150.
That is, c measures the proportional increase in revenue “associated with” a proportional increase in the cyclically neutral level of GDP.
In this case the dummy variable (“dichotomous” variable) takes the value of one in years of elections of the members of parliament and zero in other years. Elections are usually held once every four years, late in the year.
The effect of elections on revenue collection appears to be a factor in recent years.
The Schwartz Criterion has been shown to be consistent for large samples; when the true model is among the specifications considered, the probability of selecting the true model from among others approaches unity as the sample size increases. The Akaike Criterion is not consistent but is asymptotically efficient; in a sequence of models increasing in complexity with sample size, it favors those whose forecast-error variances approach that of the true model at least as fast as those selected by any other criterion. This description of SIC and AIC is taken from Francis X. Diebold, Elements of Forecasting (Cincinnati, Ohio: South-Western College Publishing, 1998), pp. 85-91.
Nuri Erbas, Neil Ericsson, Caryl McNeilly, Abebe Selassie, Abdelhak Senhadji, Marinus Verhoeven, and Oleg Zamouline contributed helpful comments on this chapter. The discussion in Sections a and b draws on “Forecasting Budgetary Aggregates” by Frederick Ribe, IMF, Course on Fiscal Adjustment in Eastern and Southern Asia, September, 1994, as well as “Some Procedures for Forecasting Tax Revenue in Developing Countries,” by Sheetal K. Chand, DM/75/91 (Washington: International Monetary Fund, October, 1975).
See Organization for Economic Cooperation and Development, “A Comparative Study of Personal Income Tax Models,” Paris, 1988. Currently the OECD offers a course on Taxation Modeling at the Joint Vienna Institute for participants from transition economies.
If such changes in the tax system are relatively few, they can be taken into account in the framework of a regression-based estimating procedure by the insertion of dummy variables into the specification of the equation.
See, for example, Francis X. Diebold, Elements of Forecasting (Cincinnati, Ohio: South-Western College Publishing, 1998), and D.F. Hendry, Dynamic Econometrics (Oxford, England: Oxford University Press, 1995). A helpful recent textbook is Walter Enders, Applied Econometric Time Series (New York: John Wiley, 1995).
In the case of Turkey, the “consolidated” budget is the budget of the central government. Budgets of certain autonomous government agencies that appear in annexes of the central government budget document are included in its totals, but this budget does not include extra-budgetary funds. See Chapter 2, Section d. In the present chapter, the terms “consolidated budget” and “central government budget” are used interchangeably.
If structural change occurs, use of a longer data sample does not in general improve regression estimates from the point of view of forecasting accuracy. This would seem to apply especially to the early 1980s in Turkey, a period of significant change. Moreover, in 1983 the fiscal year was changed to correspond to the calendar year. (In 1982, the fiscal year was nine months in length, March-December, and in earlier years it was March-February.)
The point of the regressions is to construct quantitative indicators of future revenue. Therefore it is important to estimate parameters that are as representative as possible of the current situation. Revenue data from the early 1980s are omitted from the data samples, even in cases in which they are available, to avoid including a sub-period of possibly rapid structural change. However, one or two revenue observations from before 1985 have been included as needed to accommodate the addition of lags to the equation. In cases in which pre-1985 revenue data are not readily available, regression estimates containing lags are based on one or two fewer observations than the structural estimates.
Descriptive material about individual tax categories presented in this section has been taken primarily from International Monetary Fund, Turkey—Recent Economic Developments, IMF Staff Country Report No. 96/122 (Washington: IMF, November, 1996).
By inferring the standard error of the slope coefficient from the reported statistics, one can show that this estimate is significantly different from unity at the 95 percent confidence level. That is, on average during the past eleven years, parliament has under-adjusted the personal income tax system for inflation or it has raised tax rates; since this system is progressive, revenue has therefore shown a definite though small tendency to increase relative to nominal GDP.
Standard errors of the estimates (SEEs) are given in the same units as the dependent variables, natural logarithms, and therefore may be interpreted as proportions (decimal fractions) of levels of revenue.
This refers to the structure of a Koyck distributed lag (a lag structure with geometrically declining weights for terms further in the past). See, for example, Harry H. Kalegian and Wallace E. Oates, Introduction to Econometrics: Principles and Applications (New York: Harper and Row, 3rd ed., 1989), Chapters.
A dummy variable for this shift was not significant in the regression equation for GCROT presented below. It is not known whether there were other shifts in the disposition of revenue from this tax in the sample years before 1991.
For LCP (log of nominal private consumption) regressed on LGDP, the slope coefficient is 1.0079.
It was hypothesized that if SEEs suppressed price increases prior to an election, they would as a consequence realize lower net profits in those years and therefore provide less nontax revenue to the accounts of the central government.
This is not the only conceivable interpretation of unchanged policy regarding government expenditure. Suppose, for example, government outlays had grown one percentage point faster than GDP, in real terms, on average over the past ten years. In that case it might be sensible to interpret unchanged policy as implying a continuing slight increase in the role of government.
Statistical carryover will usually result in a small year-on-year increase even after the workforce ceiling goes into effect. Suppose, for example, government employment grew by five percent during year t, and the ceiling was imposed starting with the first day of year t+1. It follows that the average level of employment in the latter year will be 21/2 percent greater than the average level in the former year, despite the ceiling. If the ceiling were to remain in effect during year t+2, the growth of employment in that year would be zero comparing average levels.
This is approximately the rate of population growth in Turkey in the 1990s.
That means that the average volume change for the first full year of the program is -1 percent compared to the level of jobs just before the freeze went into effect.
While foreign financing was negative in 1994-95 (shown in Table 7.2), nevertheless foreign debt denominated in U.S. dollars increased slightly in 1995 (Table 7.5) because of exchange rate changes between the U.S. dollar and the German mark. A large share of Turkish foreign debt is denominated in marks.
For example, suppose nominal GDP grows by 95 percent, as in 1995, and central government spending remains at 23 percent of GDP, as it was in 1994. Then, the ceiling of 12 percent of the increase in nominal expenditure implies that central bank financing could be as high as (0.12)(0.95)(0.23) = 0.026 or 2.6 percent of GDP. In 1995, the central government used about half of this potential source of financing.
The central bank may sustain losses, for example, in the foreign exchange market by intervening to support the lira even when the market expects, and ultimately benefits from, a devaluation. Similarly, the agricultural bank may be asked to make low-interest loans to farmers and to forgive part of the principle of the loans if the harvest is poor due to bad weather,
Debt consolidation exercises have sometimes included short-term advances of the central bank to the central government, especially when the outstanding stock of advances appeared to approach the legal limit.
This assumes that the average rate of interest on government foreign borrowing is equal to the rate of nongovernment borrowing from abroad since the average interest figures apply to all foreign debt.
Thus, there is an implicit assumption in Box 7 that interest due on the value of debt held on average in period t is paid in period t. Since there may be a brief lag in practice, this assumption is an approximation to the actual situation.
Note that the simple arithmetic mean of end-period values is used to represent the average debt stock. Had the averaging method from Chapter 6, Appendix II been used, the debt stock would be smaller and the implicit interest rate larger. This would not change the forecast of government interest payments substantially since the effects offset each other, but some intermediate calculations (the figures in line 10) would be different.
From International Monetary Fund, Turkey—Recent Economic Developments and Selected Issues, IMF Staff Country Report No. 98/104 (Washington: IMF, September 1998), p, 172.
In line 12, note that subtracting the “fiscal balance” is the same as adding the deficit.