Israel: Background Studies, Information Notes, and Statistical Appendix

This paper reviews economic developments in Israel during the 1990s. The paper analyzes tax revenue shortfalls in 1995–97. The factors that have contributed to overruns in the budget deficit relative to its targeted level are examined. The analysis indicates that the main difficulty has stemmed from the revenue side, and details the various areas where problems have occurred. The paper examines whether stable estimating equations for the major tax components can be established. The paper also examines volatility of interest rates in Israel.


This paper reviews economic developments in Israel during the 1990s. The paper analyzes tax revenue shortfalls in 1995–97. The factors that have contributed to overruns in the budget deficit relative to its targeted level are examined. The analysis indicates that the main difficulty has stemmed from the revenue side, and details the various areas where problems have occurred. The paper examines whether stable estimating equations for the major tax components can be established. The paper also examines volatility of interest rates in Israel.


19. Since the early 1990s, one objective of Israel’s economic policies has been the consolidation of the public finances and maintenance of budgetary discipline despite the budgetary strain associated with large inflows of immigrants. However, in recent years, this objective has not always been met. In this study, the factors that have contributed to overruns in the budget deficit relative to its targeted level are examined. The study indicates that the main difficulty has stemmed from the revenue side, and details the various areas where Problems have occurred. Finally, the study examines whether stable estimating equations for the major tax components can be established.

A. Budget Deficit Targets and Outcomes in 1992–97

20. The 1992 Deficit Reduction Law (DRL) attempted to impose control over the domestic component of the state deficit, setting specific targets for each year’s domestic deficit to GDP ratio, with complete elimination of the deficit by 1995.12 However, during the 1994 budget process, the provisions of the DRL were relaxed, mandating only a requirement that future budgets target a domestic deficit that was below that targeted in the previous year, without specifying an end date for a balanced domestic budget.

21. In the first three years following the application of the original DRL, the domestic component of the state budget deficit was reduced from almost 5 percent of GDP in 1992 to 2 percent of GDP in 1994; in fact, the domestic deficits recorded in 1992–94 were smaller than forecast (see table below). However, the situation was reversed in 1995: despite a modest deficit target for 1995 of 2.8 percent of GDP (which was actually larger than the realized deficit in 1994), the domestic deficit was overshot by 0.4 percent of GDP. In 1996, the domestic deficit reached 4.6 percent of GDP, almost twice the targeted level.

Deficit Targets and Outturns, 1992–97 1/

(In percent of GDP)

article image
Source: Data from the Israeli authorities.

The official target was changed from the domestic deficit to the overall deficit in 1997.

22. Prompted by the deficit overruns in 1995 and 1996, the government introduced several new measures in 1997 to improve budget monitoring and control in order to avoid overshooting the deficit target for a third consecutive year. The deficit target was changed from the domestic deficit to the overall deficit, which was set at 2.8 percent of GDP (and is to be reduced to 1.5 percent of GDP by 2001). In addition, the 1997 Budget was adjusted at the beginning of the year to better reflect the actual revenue base from the previous year (see below), and it was agreed that the Finance Ministry would inform the government of budget developments, itemized by component, at the beginning of each quarter.

23. Table II.1 presents a general outline of the state budget balance—programmed and actual—for 1995–97. As can be seen, expenditure during this period was generally kept in line with projections, and was even sometimes below the original target. The Ministry of Finance maintains centralized control over the spending ministries, specifying in detail their annual budgets, and it is common for additional expenditure cuts to be implemented during the year. Table II.1 indicates that the source of the budget difficulties in the last two years lay principally on the revenue side, particularly in the underperformance of tax revenues relative to the budget targets.

Table II.1.

Israel: State Budget Balance, 1995–97

article image
Source: Data provided by the Ministry of Finance.

Excluding fees, which are classified as nontax revenues.

Excluding revenue from asset sales.

Excluding revenue from asset sales.

The deficit target was defined in terms of the domestic balance in 1995 and 1996, and in terms of the total balance in 1997.

24. Table II.2 provides a comparison of budgeted and actual tax revenues at a disaggregated level for 1995–97. As indicated, realized tax revenues in 1995 and 1996 fell short of the budget target by around NIS 3–4 million (around 1.3 percent of final GDP in each case), with the shortfall primarily due to lower-than-anticipated collections of corporate taxes and VAT. Smaller shortfalls in purchase and excise taxes and property taxes have also occurred. A further shortfall in tax revenues on the order of NIS 4 million (equal again to 1.3 percent of estimated final GDP), due mainly to undercollection of VAT, is projected for 1997.

Table II.2

Israel: Tax Revenue, 1995–97

article image
Source: Data provided by the Ministry of Finance.

Excluding fuel excise.

B. Possible Factors Contributing to the Revenue Shortfalls

25. The methodology employed by the Ministry of Finance to forecast tax revenue involves five main steps:

  • Estimating the tax collection in the preceding year: Since the budget for year t is prepared before the end of year t-1, the tax revenue estimate for year t is based on available data in year t-1 at the time of the forecast (typically 8–9 months) and an estimate for the remainder of the year t-1.

  • Adjusting for one-off factors that distort the base: Examples of such factors are exceptionally large tax refunds (such as NIS 500 million in tax refunds deferred from 1995 to 1996, which were added back to the 1996 base when preparing the 1997 forecast) and tax transfers to the Palestinian Authority.13

  • Estimating the effect of legislative changes: The relevant legislative changes are those planned for year t or implemented during year t-1 but which will only have a full-year impact in year t; the effect is computed on the basis of activity and prices in year t-1.

  • Estimating the effect of improved collection efforts. The forecast improvement in actual collections is entered as an assumption; no formal estimation procedure is employed.

  • Estimating revenue growth due to economic activity: The sum of the first four steps yields the “forecast base” of revenue in year t. This forecast base is then multiplied by a real rate of change derived from a set of economic indicators including business sector product, public consumption, private consumption, durables imports, other imports, the number of workers and their real wage, and sales of new dwellings. Finally, the conversion to current prices in year t is made on the basis of the expected increase in the average CPI.

26. To better understand the sources of recent difficulties, Table II.3 provides a breakdown of nominal tax revenue—forecast and actual—over the last three years, categorizing separately the effect on total tax revenue arising from errors in the forecast of the previous year’s tax base, real growth, improvements in collections, legislative changes, and inflation. Both the forecast and actual figures, as well as the breakdown of forecast errors, were obtained from the Ministry of Finance.

Table II.3.

Israel: Tax Forecasts, 1995–97 1/

article image
Sources: Data from the Ministry of Finance, the Bank of Israel, and the Central Bureau of Statistics; and staff estimates.

The data in this table differ marginally from those in Tables 1 and 2 due to the inclusion here of some nontax items.

Due to extraordinary tax refunds and tax clearances to the Palestinian Authority.

Percentage change in average CPI; 1997 figure is annualized nine-month average inflation rate.

Overestimation of previous year’s revenue

27. As noted earlier, at the time of budget preparation, the previous year’s tax revenue (which forms the basis for the budget forecast) is estimated using collections data for part of the year and forecasts for the balance of the year. Table II.3 indicates that this step has been a source of several forecasting errors in recent years. In particular, for the 1995 Budget forecast, the estimated base-year revenue (i.e., the revenue that would be collected in 1994), after adjusting for exceptional factors, was overestimated by 0.3 percent of GDP. For the 1996 forecast, the base-year revenue estimate was again overestimated, this time by 0.8 percent of GDP. In light of these experiences, for the 1997 budget year, the authorities made it a point to adjust their budget forecast at the start of the budget year once more complete data on the preceding year’s revenue were available, which necessitated an expenditure cut of around 0.2 percent of GDP at the beginning of 1997. In consequence, there seems to have been virtually no error in revenue collections arising from this source in 1997.

Overestimation of the revenue increase from economic growth

28. As indicated in Table II.3, the contribution of real economic growth to real tax revenue growth was also a major factor in the revenue errors, accounting for an overestimate of 0.7 percent and 0.4 percent of GDP in 1995 and 1996, and a projected overestimate of 0.7 percent of GDP in 1997. In principle, this could be due to unexpected macroeconomic changes (e.g., if a correct elasticity of tax receipts to real GDP growth was used, but real growth turned out to be lower than anticipated), forecasting error (e.g., if real GDP growth was accurately forecasted but the assumed elasticity of tax receipts to real GDP growth was too high), or both.

29. For 1995 and 1996, the evidence points toward use of incorrect elasticities as the main source of the overestimation. Since real GDP growth was 2 percent higher than expected in 1995 (see table below) and was only ½ percent lower than forecast in 1996, unanticipated macroeconomic slowdowns were not the main reason for the initial overestimation of revenue. Instead, the data shown in Table II.3 indicate that the implied elasticities at the time of the forecast were around 1.5 in 1995 and 1996, which, in retrospect, were too high; the actual elasticities came in on average closer to unity.14

Macroeconomic Indicators, 1995–97

(Percentage change)

article image
Source: Ministry of Finance, State Budget Proposai for Fiscal Year 1995 and 1996; Bank of Israel, Main Israeli economic Data; data obtained through the Bank of Israel; and staff estimates.

Authorities’ estimates as of November 1997.

30. For 1997, however, a sharper-than-anticipated slowdown was likely the main reason behind the overestimation of the economic growth component of real tax revenue growth. Latest estimates by the Central Bureau of Statistics put the 1997 growth rate at 2 percent, significantly below the 4 percent growth envisaged in the budget papers. Hence, despite the improvement in the accuracy of the estimated base year collection and the lower implied elasticity of real tax revenue to GDP growth used (slightly less than unity, which is likely to be close to the actual elasticity), the net contribution of economic growth to tax revenue growth was overestimated by around 0.7 percent of GDP.

Overestimation of revenue increase from improvements in tax collection

31. The projected contribution of improved collection efforts to real growth in tax revenues appears in retrospect to have been optimistic in 1995 and 1996 (and in 1996 was of the wrong sign). Clearly there are inherent difficulties in quantifying this category, and the authorities have been gradually reducing the size of the projected contribution; improved collection efforts were expected to increase real tax revenues by 0.5 percent of GDP in the 1997 Budget, down from 1.1 percent of GDP in the 1995 Budget.

Underestimation of revenue decline due to legislative changes

32. Table II.3 shows that legislative changes were a contributing factor to the tax revenue shortfalls in 1995, when their (negative) effect on revenue growth was underestimated by 0.3 percent of GDP. In this instance, unanticipated legislative changes (i.e., changes that were introduced after the budget had been passed) played a part, including the repeal of the capital gains tax on stock market transactions, reductions in property and real estate purchase taxes, and further reforms in personal income tax, notably, the award of a tax-credit point for married women, and the merging of two tax brackets that took effect in September. The net error in estimating the effect on tax revenue of legislative changes in 1996 and 1997 was minimal.15

Overestimation of revenue increase due to inflation

33. Finally, the assumed effect of inflation on nominal tax revenue was underestimated in 1995 and 1996: since the forecast inflation rates were below the actual rates by 1.1 percentage points in 1995 and 2 percentage points in 1996, nominal tax revenue was actually higher by 0.2 percent and 0.4 percent of GDP in these years. In 1997, however, it appears that the 11.1 percent inflation rate used in the budget forecast will turn out to have been too high, with a consequent reduction in actual nominal revenue due to this source by around 0.8 percent of GDP.

34. In summary, the main reasons for the deviations in total tax revenue in 1995 and 1996 were errors in forecasting the initial tax base and optimistic assumptions on the revenue impact of activity growth and in tax administration efforts. In 1997, however, these areas were addressed by the authorities and they have in consequence played only a minor role. Instead, the unexpected slowdown in actual real GDP and inflation were by far the main reasons for the decline in actual tax revenue relative to the forecasts presented in the 1997 Budget.

C. Behavior of Tax to GDP Ratios Over Time

35. In view of recent difficulties in accurately forecasting tax receipts, this section investigates whether a stable relationship can be found between tax/GDP ratios and cyclical and trend factors over time. Specifically, the tax/GDP ratios were regressed against the output gap (and other factors) using quarterly data from 1988–94, and then the equations were tested for parameter stability when the sample was extended to include data for 1995–96.16 The analysis focused on personal and corporate income taxes and VAT; as can be seen from Table II.2, these three categories are the most important components of tax revenue, making up 34 percent, 11 percent, and 34 percent of the total, respectively. Key features of these taxes are described in Box 1.

Key Features of the Major Taxes

  • Personal income tax (PIT): All individuals, resident or not for tax purposes, are subject to PIT on Israel-source income and income received in Israel. Since 1995, the rate structure has comprised four (inflation-adjustable) brackets of 15, 30, 45, and 50 percent, with a minimum rate of 30 percent applying to passive (i.e., nonwage or nonbusiness) income. Tax is withheld at source on a variety of incomes and payments, including wages and national insurance contributions.

  • Corporate income tax (CIT): All companies, resident or not for tax purposes, are subject to corporate tax on Israel-source income and income received in Israel. The CIT follows the classical system, i.e., profits are taxed at the corporate level and distributed dividends are taxed at the individual level. As of 1996, the regular rate of Company tax is 36 percent.

  • Value-added tax (VAT): A 17 percent VAT is levied on a broad range of goods (imported or domestically produced) and services (domestic only). Exports, unprocessed fruits and vegetables, and some tourism services are zero-rated; certain asset sales are exempt.

36. The economic cycle is measured by the output gap, which is defined as the difference between actual and potential (trend) output as a percentage of potential output. The output gap measure used here is based on potential output as estimated by means of the Hodrick-Prescott (HP) filter.17

37. Each tax/GDP ratio was regressed on up to four quarterly lags of the output gap (GAP) to capture cyclical effects, seasonal dummies, and up to four lags of the dependent variable. Also included were the log of real trend output (In Y*) to capture the effect of real income growth and a time trend to proxy for trend changes in the tax burden.18 The choice of this particular functional form was somewhat arbitrary; there is little theoretical guidance as to the precise form of the relationship between tax receipts and the cycle, and the search for a preferred functional form involved some trial and error.19

38. The coefficient on the output gap represents the change in the tax/GDP ratio associated with a percentage point change in the output gap. A zero coefficient on the output gap would imply that tax receipts move in strict proportion to changes in contemporaneous GDP, with the tax/GDP ratio unaffected by movements over the cycle (i.e., that the elasticity between nominal tax receipts and GDP is unity). A negative coefficient on the output gap implies that tax receipts change less than proportionately with GDP over the cycle, while a positive coefficient implies tax receipts change more than proportionately with GDP over the cycle.

39. The equations were first estimated using data from 1988 to 1994. The results are reported in Table II.4. For each of the taxes, the first column of figures shows the results from the regression with all the initial regressors; the second column shows a parsimonious specification from step-wise deleting insignificant variables from the initial set of regressors; and the third column shows the results after running the preferred specification on data that include 1995–96.

Table II.4.

Summary of Regression Results 1/

article image
Y* = trend output (in 1990 new sheqalim); T = number of observations; k = number of regressors.

t-values in parentheses. Constant and three seasonal dummies included.

F-approximation of the Lagrange Multiplier test for fourth order residual autocorrelation. The number in square brackets denotes the probability of rejecting the null hypothesis of autocorrelation.

Chow test of the form η = [(RSSτ+s−RSSτ)/8]/[RSSτ/(T-k)], where η is approximately distributed as F(8, T-k) on the null hypothesis of no structural change in any parameter between the sample and forecast periods. The number in square brackets denotes the probability of not rejecting the null hypothesis.

40. The results must be interpreted as illustrative only, due to a number of potential shortcomings and simplifications in the data and methodology. For example: the estimated relationship was a simple reduced form rather than a fully developed structural model which includes the determination of the tax base; the sample period used in the regressions was very short; the current GDP base may not have accurately reflected the underlying tax bases due to changes in the latter or to collection lags; and no attempt was made to account for changes in the tax structure or tax administration efficiency over time.20

Regression results

41. As reported in Table II.4, the elasticity of personal income tax receipts with respect to the cycle is smaller than one, as indicated by the negative coefficient on the contemporaneous cycle term. This is consistent with the usual observation that changes in wages and employment tend to lag the cycle, implying that personal income—the tax base—tends to fluctuate less than GDP itself.21 The coefficient on the log of real trend output is significantly positive, indicating evidence of real fiscal drag. The coefficient on the time trend is negative, indicating that holding constant other effects, the PIT/GDP ratio has declined over time, possibly reflecting the effects of the various structural reforms aimed at reducing the tax burden. However, these relationships do not appear to be stable: the equation fails the test for parameter stability when it is re-estimated using data over 1988–96.

42. There is no discernible relation between the CIT/GDP ratio and the output gap. This is unsurprising considering the lags involved in the declaration and collection of corporate taxes: in general, corporate taxes in Israel are calculated on the basis of past-year profits, and are paid “in advance” during the year they are due. Given the lags involved, it is difficult to make any a priori assumptions regarding the relation between the ratio of corporate taxes to (contemporaneous) GDP and the output gap in any given quarter. The coefficient on the log of real trend output is significantly positive, while the coefficient on the time trend is negative, reflecting the effect of the steady reduction in corporate tax rates since 1987. Unlike the case of PIT/GDP, no instability is detected in the preferred specification: the equation passes the test for parameter stability when it is re-run using data over 1988–96.

43. The VAT/GDP ratio is found to vary positively with (the second lag of) the output gap but negatively with real growth in trend output. This latter effect might reflect the change in the structure of the economy toward more export oriented production, which are zero rated under the VAT. The equation also passes the parameter stability test.

44. In summary, the results of this empirical exercise point to the difficulty in establishing a stable pattern to the tax/GDP ratio for personal income taxes which could be usefully exploited to make more accurate revenue projections. This being the case, it is unsurprising for one to encounter fairly large errors in this major revenue term, suggesting the need for prudence in making revenue forecasts in such an uncertain environment. The tentative work here also suggests, however, that the corporate income tax and value-added tax equations presented may provide some scope for developing a stable forecast equation for these items.


Prepared by Ling Hui Tan.


The DRL excluded three types of transactions from the definition of the target deficit: (i) the foreign component of the State budget, which has primarily consisted of foreign grants on the revenue side (foreign revenue averaged about 4.6 percent of GDP and 12 percent of total state budget revenue during 1992–96) and foreign interest and defense imports on the expenditure side (foreign expenditure averaged about 4.7 percent of GDP and 10 percent of total State budget expenditure); (ii) net lending operations of the central government (primarily for housing assistance, the net quantitative impact of which has been minimal since 1994); and (iii) proceeds from the sale of government assets (the magnitude of which has been somewhat erratic from year to year). Furthermore, according to the accounting convention adopted by the State budget, the deficit is in essence an “operational” concept, as it excludes the nominal component of domestic interest payments. See Israel–Selected Issues and Statistical Appendix (SM/96/295, 12/4/96).


As part of the 1994 Paris Agreement, Israel forwards to the Palestinian Authority taxes collected from imports to the autonomous areas and trade with Israeli residents. Forecasts of these transfers have been very inaccurate—the actual final number of NIS 763 million in 1995 was three times higher than the original forecast—due to the paucity of data on the relevant transactions.


Simple regressions of the log of real tax receipts on the log of real GDP over the period 1986–94 yield elasticity estimates of between 1.0 to 1.2.


In 1996, the authorities overestimated the net negative effect of legislative changes (such as reductions in income, corporate, and property tax rates) in that year and in 1997, it appears that they may have underestimated the net positive effect of legislative changes (such as non-adjustment of income tax brackets and increases in the fuel tax and certain sales taxes).


The quarterly data were obtained from Tables F-1 to F-4 of the Bank of Israel’s Main Israeli economic Data; (annual) corporate tax data were obtained from reports of the Revenue Authority, through the Bank of Israel.


See Chapter I of this report for the derivation of this series for Israel.


Statistical tests suggested that the tax/GDP ratios were nonstationary over the sample period. This could have reflected structural breaks in the series, or could have been due to the low power of such tests over short samples.


Other specifications attempted included running the regression in log levels, with log GDP on the right hand side together with the output gap, and running the regression in differences.


With respect to the last point, Israel’s tax structure has evolved substantially over the last ten years. Reforms in personal income tax were implemented in 1987 (rate reductions), in 1990 (introduction of tax credits), and in 1994–95 (lowering of the minimum tax bracket, raising of the maximum bracket, and broadening of the brackets in between). The effective tax rate on undistributed corporate profits was reduced from over 60 percent to 45 percent in 1987, to 42 percent in 1990, and by 1 percent each year thereafter to eventually reach 36 percent in 1996. Finally, the VAT rate was increased from 16 percent to 18 percent in 1992, but subsequently reduced to 17 percent in 1993.


The contemporaneous effect is consistent with the observation that in Israel, income taxes are paid monthly by withholding, rather than filed annually with a one-year lag.