Tax Reform of the Century – the Swedish Experiment

What can changes in tax structure accomplish? The Swedish tax reform of 1991 is the most far-reaching reform in any industrialized country in the post-war period. It represents a thorough application of a strategy of rate cuts cum base broadening, and it has affected a myriad of economic incentives in a more or less substantial way. This article reviews the lessons from a major evaluation effort, sponsored by the Swedish government, and involving a large number of researchers.

Introduction

Sweden might be best known as the home of film director Ingmar Bergman, and – for good or for worse – as the prototype welfare state. What might be less well known is that Sweden recently implemented the most far-reaching tax reform in any western industrialized country. Although Sweden was a latecomer to the bandwagon of world-wide tax reforms of the 1980s, with the U.S. Tax Reform Act of 1986 (TRA86) as a celebrated example, the architects of the Swedish tax reform of 1991 (TR91) applied the strategy of rate cuts cum base broadening in an unusually thorough manner. Under the catchy slogan ’tax reform of the century’, marginal income taxes were dramatically lowered, and various tax shelters eliminated. According to pre-reform estimates, the rate cuts entailed a revenue loss on the order of six percent of GDP. Measured in this way, TRA86 stands out as a relatively modest endeavour, with a projected revenue loss of 1–2 percent of GDP due to rate cuts.

It goes without saying that sharply reduced income tax rates is a sensitive political issue, particularly so in a country where tax policy used to center on the idea that a steeply progressive income tax is an efficient way of transferring resources from the rich to the poor. To understand why TR91 gathered wide political support, both external and internal considerations are important. The integration of world capital markets during the 1980s implied that it became more difficult to tax capital income at rates that differed very much from those applicable elsewhere. Also, the major changes in the tax structures of a number of other countries provided an impulse. However, it would be wrong to view TR91 as a mechanical response to these developments. After all, Ronald Reagan’s tax policy is not a very natural source of inspiration for Sweden’s Social Democrats. While many of the arguments of the proponents of TR91 bear a striking resemblance to those put forth in the context of TRA86,1 their origin should be traced primarily to the domestic debate.

As early as 1978, Nobel Laureate Gunnar Myrdal complained that high marginal tax rates had turned Sweden into a ’nation of wanglers’. According to Myrdal (1978), the progressive income tax had created such strong incentives for high income individuals to exploit various tax avoidance schemes (including outright tax fraud), that the Swedish tax system no longer redistributed income. Myrdal’s view carried particular weight since he belonged to the political left. Parties more to the right, and quite a few economists, had for long warned that the income tax created large disincentive effects. But Myrdal seemed to suggest that also those in favor of egalitarian outcomes, and concerned less about efficiency, had reason to reconsider the role of the progressive income tax. Although Myrdal offered no hard facts to substantiate his claim, the perception that the rich could avoid their fair share of the tax burden was probably instrumental in softening the Social Democrat’s traditional resistance to proposals involving lower marginal tax rates.

But there were also other influential arguments. Since the late 1970s, when inflation reached double-digit levels, there was widespread concern that the tax system promoted the wrong kind of investments. Investments in noncorporate assets, and housing in particular, were given preferential tax treatment. Many Swedish economists argued, as did many of their collegues in the U.S., that the uneven playing field created substantial efficiency losses. The nonuniform treatment of the returns on different assets also created considerable scope for a number of straightforward tax arbitrage operations, more often than not involving purchases of low-taxed assets with borrowed money. This spelled bad news for the fisc – for several years households claimed tax deductions to such an extent that net revenue from taxes on personal capital income was negative.

The corporate income tax attracted similar criticism. A variety of tax allowances, and a high statutory tax rate, was once part of a deliberate strategy of stimulating firms to plough back profits into the business. The idea was that large and expanding firms were good for growth. During the 1980s there was a shift in emphasis. According to the new view, the corporate tax breaks were an obstacle to efficient capital allocation. The high rates of profit retention required to take advantage of the various tax allowances created a capital lock-in effect, which prevented necessary structural readjustments. Finally, towards the mid 1980s, an overheated economy, and concerns about an acute labor shortage, seemed to give added weight to the argument that there was an urgent need to strengthen labor supply incentives.

TR91 was presented as a way of remedying all these problem in one giant stroke. According to its proponents, the reform would avoid the classical goal conflict between efficiency and income distribution. In spite of drastic marginal tax cuts, high-income earners were not supposed to gain relative to other groups. As a result of a generally more efficient economy, all strata in society should gain. Moreover, total tax revenue was not supposed to decrease – the Swedish tax take should remain the highest in the world. The revenue gains from broader tax bases were to make up for the losses from lower tax rates.

What can changes in tax structure accomplish? Did the proverbial free lunch materialize? This paper reviews the lessons from a major evaluation effort, commissioned by the Swedish government, and involving a large number of foreign and Swedish researchers.2 The next section gives a brief outline of TR91. We proceed by discussing the considerable difficulty in evaluating a tax reform in the midst of a very sharp recession. We review the evidence on behavioral responses, and we seek to identify areas where TR91 mattered the most. We also present an assessment of the overall effects on efficiency and equity. Our concluding remarks draw some lessons for tax reform more generally.

What TR91 implied

Pre-reform estimates presented by the Ministry of Finance indicated that the rate cuts for the personal income tax alone, together with additional outlays for housing and child allowances to cushion the distributional effects of the reform, would reduce revenues by an amount equivalent to between six and seven percent of GDP. Nearly 40 percent of this loss was to be recouped through a new system of taxing capital income. A broadening of the VAT (of 23 percent) to include goods and services previously exempted, or granted lower rates, would yield additional revenue on the order of 30 percent of the budget loss, while almost 15 percent was to be financed by elimination of loopholes and preferential rules for taxing earned income. Enlarged tax bases due to a generally more efficient economy would offset roughly five percent of the revenue loss. The reform hence brought a major reallocation of the total tax burden away from earned income to consumption and to individual capital income (including the return to housing).

A noteworthy feature of TR91 was the move away from the principle of global income taxation towards a dual income tax, by introducing separate tax schedules for earned income and capital income.3 The new taxation of earned income meant that almost 85 percent of the income earners would pay only local income tax. In 1991 the countrywide average of the local income tax was 31 percent. A national income tax of 20 percent was imposed for incomes exceeding 185,000 kronor (equivalent to US $ 33,500, at the 1991 exchange rate) which meant that the top marginal tax rate on earned income was set at 51 percent. TR91 implied that the marginal rate would be reduced by between 24 and 27 percentage points for large groups of full-time employees, see Figure 1.

The new proportional capital income tax was set at 30 percent, and levied on dividends, interest income and both long and short-term realized nominal capital gains. As before, interest on all kinds of debt would be fully deductible. A stated purpose of the separate capital income tax was to reduce the value of interest deductions and to limit the scope for tax avoidance in various forms.

Although the initial ambition of TR91 was to levy a uniform VAT (of an unchanged 23 percent) on all commercial turnover of goods and services, several areas have remained tax exempt. These include various cultural and social services, and housing rents. Housing costs have risen as a result of the tax reform, however, partly because the VAT was broadened to include real estate maintenance, heating and electricity. Owner-occupiers were further affected by the inclusion of expenditures on housing investment in the new VAT. Before TR91, the tax was 13 percent.

The changes in the corporate income tax were no less dramatic. The statutory tax rate was reduced from 57 to 30 percent. In order to maintain an unchanged level of revenue from the corporate tax, the rate reduction was combined with substantial broadening of the base. The possibility to undervalue inventories for tax purposes was eliminated, and the investment funds system, introduced in the mid 1950’s as the main tool of a counter-cyclical fiscal policy, was discontinued.

Figure 1

Even a casual comparison reveals obvious similarities between TRA86 and TR91. Both reforms were far-reaching with the same intent of reducing various behavioral distortions from the tax systems. The approach to tax reform was similar, combining substantial tax cuts with a broadening of the base. TR91 was much larger in scope, both in terms of revenue effects from the rate cuts, and by covering a wider range of tax instruments. Obviously, these differences partly reflect the vast difference between Sweden and the U.S. in the financing requirements for the public sector. As a proportion of GDP, the total tax yield was almost 56 percent in Sweden in 1989, compared to 30 percent for the U.S. However, the relative importance of different taxes, summarized in Table 1, is broadly similar. In both countries, there is a heavy reliance on receipts from the personal income tax and from social security contributions. Taxes on goods and services are more important in Sweden, whereas corporate and property taxes contribute a larger share of tax revenue in the US. The reallocation of the total tax burden brought by TR91, away from personal income to consumption, is clearly visible.

Table 1. Sources of Tax Revenue in Sweden and the U.S.

Source: Revenue Statistics of OECD Member Countries 1965–94, (OECD, Paris, 1995).

A natural experiment?

A major reshuffling of the tax structure may seem like that rare opportunity to sharpen the estimates of the behavioral elasticities that would interest a public finance economist. However, the implementation of TR91 was accompanied by the most severe economic downturn since the 1930s. Between 1991 and 93 GDP fell by more than five percent, open (i.e. excluding those enrolled in various labour market programs) unemployment rose from less than two percent to more than eight percent, asset prices tumbled, and residential construction activity came to a virtual standstill. At the same time, income inequality increased, and the government deficit reached record heights. In this environment, it is clearly hard to sort out cause and effect with much precision.4 Macroeconomic time series for the years surrounding TR91 contain very little of the kind of independent information needed to discriminate between alternative hypotheses about behavioral responses. Microeconomic panel data, allowing the analyst to control for individual variation over time, may seem like a safer bet, but there is still an obvious risk that the macroeconomic noise pollutes the microeconometric analysis.

The Swedish tax evaluation effort includes several contributions which try to make do with data sets including pre- and post-reform observations. But the macroeconomic turmoil has led a number of researchers to confine attention to the information that can be gathered from an ex ante evaluation of TR91. These evaluations have two ingredients in common. The first is an assessment – based on previous findings reported in the literature, or on new findings based on pre-reform data sets – of the reasonable value of some relevant behavioral parameter. The second ingredient is a careful assessment of the change in the incentive structure implied by TR91. This is a far from trivial exercise, since TR91 contained several provisions with counteracting influences on the relevant incentive margin. Combining these ingredients, the ex ante evaluation produces a set of predicted responses, which serve the useful purpose of scrutinizing the political arguments. Do recent research, and refined analysis of the incentive structure, produce the kind of behavioral responses promised by the architects of TR91?

An additional complication stems from the fact that the sharp drop in the activity level cannot be treated as independent of TR91. Towards the end of the 1980s, the Swedish economy showed many signs of overheating.5 Due to a high demand pressure, wage costs increased substantially, which led to eroded competitiveness in world markets. At the same time, and in the wake of financial deregulation, the indebtedness of households and firms reached very high levels. This left the Swedish economy in a vulnerable position when the international economy slowed down in the early 1990s. While the international recession may explain part of the rise in unemployment, a more important reason for the severity of the recession is probably that macroeconomic policy was firmly devoted to nonaccommodation. As had been the case previously in many other Western European countries, fighting inflation and defending an overvalued exchange rate now became the top priority for Swedish economic policymaking. As a result manufacturing output fell sharply.

These developments, certainly unrelated to TR91, explain a major part of the severe economic recession. But TR91 may help to explain why the slump in the exporting sectors spread to the sheltered parts of the economy. By lowering the value of household interest deductions, TR91 increased real after-tax borrowing costs, and stimulated indebted households to sell off assets. Also, higher borrowing costs, in conjunction with certain tax hikes aimed specifically at the housing sector, contributed to the collapse of the construction sector. These contractionary impulses amplified the downturn, and contributed to a general weakening of aggregate demand.6 James Tobin once cautioned that it takes a heap of Harberger triangles to fill an Okun gap. Whatever its long-term merits, it is easy to conclude that the timing of TR91 was unfortunate. While part of this can be blamed on bad luck, TR91 was designed to be implemented in a situation with fair macroeconomic wind. The official inquiry which preceded TR91 contains a wealth of material, but there was hardly any discussion of the macroeconomic aspects.

Behavioral responses

As suggested by Slemrod (1992), the evidence about the responses to the major U.S. tax reforms of the 1980s can be interpreted as an indication of a behavioral hierarchy. The most responsive decisions, at the top of the hierarchy, are those involving the timing of transactions, followed by a variety of financial, accounting and evasion responses. The least responsive decisions, at the bottom, concerns the ”real” ones, including labor supply, savings and investment. Let us review the Swedish evidence.

E. Labor supply

In the public debate surrounding the reform, the potential impact on labor supply played a major role, although many economists cautioned against overly optimistic supply projections. The development of the wedge between labor costs to the employer (including wage taxes) and the after-tax remuneration of blue- and white-collar employees (accounting for income taxes, as well as VAT and other indirect taxes) is depicted in figure 2. We see that TR91 marks a sharp break of a long term trend, where by the end of the 1970s the marginal take-home pay of the average blue-collar worker had fallen to less than 30 percent of pre-tax labor costs. In relative terms the increase from 1989 to 1991 was 23 percent for the blue-collar worker and as much as 76 percent for the upper-tier white-collar worker.

Figure 2

Although these numbers clearly demonstrate the magnitude of the reform one should not be misled into generalizing to all groups in society. In fact, looking across a representative sample of all individuals it appears that around a quarter saw an increase in the tax wedge.20 The explanation is that a large fraction of Swedish wage-earners work less than full-time at relatively modest marginal-tax rates that were not much affected by the reform, and that many households in this category are entitled to income-dependent housing allowances that were increased as part of the reform. A similar pattern with a mixture of increased and decreased marginal tax rates holds for TRA86; see Hausman and Poterba (1987). In fact, as is emphasized by Auerbach and Slemrod (1996), it lies in the nature of a reform aimed at maintaining the degree of redistribution that it is very difficult to lower marginal taxes for everybody.

Various labor supply studies, conducted before TR91 and as part of the evaluation effort, tend to confirm the ”elasticity pessimism” underlying Slemrod’s response hierarchy, with labor supply at the bottom. A representative estimate of the compensated wage elasticity of hours worked among Swedish prime-age males is on the order of magnitude of 0.1,21 but the estimates are so imprecise that the predicted hours response to the reform of the representative white-collar worker of fig 2 has a typical confidence interval ranging from 1.5 to 15 percent. Unfortunately, there is a paucity of ex post studies of the impact of the reform, but the panel study of Klevmarken (1996) finds that changes in marginal wages between 1985 and 1992 were associated with statistically significant increases of the supply of hours worked, both for men and for women.22

Although a very large amount of research has focused on hours worked, it is possible that other margins of labor supply response may prove more important in the longer run. It is noteworthy that TR91 implied a large change in the incentive to undertake investment in education. TR91 lowered tax rates at higher (post investment) income levels, but left tax rates more or less unchanged at lower (pre investment) income levels. Edin and Holmlund (1995) calculate internal rates of return to investing in a four-year university program. In the early 1980s the return was four to five percent before tax, and one to three percent after tax, implying effective tax rates between 40 and 90 percent depending on the year chosen as a basis for the calculations. After the reform the effective tax rate fell to 25 percent. The impact of this reduction is not confined to formal education, but applies more broadly to the choice between careers with different earnings profiles.

This fraction applies to a comparison between 1985 and 1992 for a sample of households from the HUS-panel; see Klevmarken et al. (1995). It accounts for the broadening of the VAT coverage and other factors affecting the real wage.

See studies by Blomquist and Hansson-Brusewitz (1990), Ackum Agell and Meghir (1995) and Aronsson and Palme (1994).

The estimated effects are quite large. Women who experience a tax decrease by more than 20 percentage points are estimated to increase yearly labor supply by 400 hours, i.e. by roughly 30 per cent of average work hours. Corresponding estimates for men are 200 hours and 10 per cent.

Income distribution

In the political process of selling TR91 to various interest groups and the electorate in general a key element was the claim that the reform would be distributionally neutral. This was interpreted in a bookkeeping sense to mean an unchanged relation between an exogenously given distribution of pre-tax factor income and the distribution of income after taxes and allowances. Against this background it was natural that a mechanical evaluation of the distributional impact along these lines was one element of the evaluation effort, although it comes natural for an economist to point to the limitations of such an exercise.

During the 1980s a growing number of critics came to doubt whether the Swedish tax system achieved much in terms of redistribution. There were three ingredients to the critique. First, various loopholes and tax arbitrage activities created a wedge between ”true” income and taxable income. Second, substitution between market labor and leisure and household production created a discrepancy between taxable income (which only derives from market activities) and potential income (which also includes the value of leisure and home production). Third, taxation based on yearly income redistributes income across different phases in an individual’s life-cycle, i.e. from more toward less productive ages, but it is less clear how much redistribution that actually is achieved across households with different lifetime income.28

The recent study by Björklund, Palme and Svensson (1995), however, gives little support to the critics of the old system. According to this analysis the old Swedish tax system certainly achieved a substantial amount of redistribution of yearly incomes.29 The amount of redistribution is not much affected by going from actual income to a measure of full income. Also, when the authors follow a panel of individuals over the period 1974–91, and take the sum of discounted income over this period as a measure of lifetime income, the conclusion is that the tax system redistributes almost as much in terms of lifetime income as in terms of yearly income. Of course, this only says something about the tax system as it evolved over this 18-year period, and gives no conclusive evidence on the long-run redistributive properties of the system of the late 1980s. Nevertheless it leads us to be somewhat skeptical about some of the more popular views of the old tax system as being void of any redistributive effect.

Björklund et al. also examine differences in pre- and post-tax Gini coefficients under the old and new tax rules. Looking at the aggregate amount of redistribution across all groups of households the differences appear minuscule, i.e. at least in the short term (up to 1992) TR91 seems to have lived up to the promise of ”neutrality” with respect to the income distribution. This conclusion, however, conceals important differences in the structure of the tax and subsidy system. Broadly speaking, there are three counteracting differences between the old and new tax system. First, the taxation of earned income is clearly less redistributive under TR91. Second, with the new flat tax on capital income, tax payments become more proportional to actual capital income than under the old system, where it was possible to avoid capital taxes altogether through tax arbitrage. Since capital income is concentrated at the top of the income distribution this tends to make the new system more redistributive than the old one. Third, child and housing allowances play a larger role after TR91 than before. Since they largely redistribute income from households without children to families with children TR91 represents a shift of emphasis towards more redistribution across various phases of the life cycle rather than between households with different lifetime incomes.

By simply comparing Gini coefficients before and after taxes one takes an unduly narrow view on income distribution. In particular one takes the pre-tax distribution of factor income for granted, thereby glossing over the strong assumptions about tax incidence implicit in such an exercise. Whereas it may be reasonable for a small open economy like the Swedish one to assume that the pre-tax return to capital is determined in international markets and unaffected by Swedish tax policy, the assumptions about the incidence on wages merit more attention than they are commonly given by income distribution analysts. In fact there has been a recent trend in Sweden, as elsewhere, towards more inequality of factor incomes, and one may ask if this development has been induced by tax reform to any extent. This should depend on the relative supply responses of high- and low-skilled workers, and on the degree of complementarity in production between different types of labor. However, given the generally small labor supply responses, it appears unlikely that the changing wage structure has primarily been induced by changes in tax structure. In the longer run the strengthened incentives to invest in human capital should be more important for the wage distribution. It is hard, though, to have a very definite opinion on the implications for the distributional analysis of TR91.30

Efficiency – a double dividend?

A main argument of the proponents of TR91 was that economic efficiency should improve. For many politicians, efficiency was here interpreted as a synonym for various easy-to-observe responses, like increased labor supply, and higher savings. In some quarters, there was also a hope that TR91 should deliver an easily detectable growth bonus. For an economist, however, efficiency is defined in terms of not directly observable areas between compensated demand and supply curves, and there is no simple relation between the implied tax distortions and the magnitude of behavioral responses. Much the same goes for economic growth. According to the standard neoclassical growth model, a badly designed tax system may create important negative level effects, without the long-term rate of growth being affected. As we will see shortly, TR91 even contained provisions which tended to reduce registered GDP growth, but to increase consumption opportunities.

From an excess burden perspective, the most important aspect of TR91 is that it implied a shift in the tax burden from highly taxed labor income, to lightly taxed housing capital. As a consequence, the relative prices of leisure and housing consumption increased. Although there is room for disagreement on the exact magnitudes of some key behavioral elasticities, there is little reason to question the soundness of this strategy. Table 2 shows the results from an attempt at quantifying the marginal efficiency cost due to labor supply distortions of the Swedish tax system.31 The compensated elasticities of labor supply are chosen to reflect the range of findings in recent empirical studies.32 The marginal tax wedges (differences between productivity and real take home pay) for some categories of employees, before and after TR91, account for income, payroll and indirect taxes.

Table 2. Marginal excess burden per krona of tax revenue (in percent)

Source: Agell, Englund and Södersten (1995).

A key point is that a small behavioral elasticity matters, if the marginal tax wedge is sufficiently high. The logic behind the marginal welfare cost per dollar of tax revenue implies that there is a region of tax wedges at which the efficiency cost starts to increase rapidly. At some marginal tax wedge, an additional tax hike creates additional excess burdens, but no extra tax revenue. Although the table brings out the sensitivity of results to alternative behavioral assumptions, one can hardly rule out the possibility that pre-reform tax wedges were close to that level. The point estimates of labor supply elasticities, not to mention the confidence intervals, reported in recent studies are consistent with the view that the pre-TR91 tax system had marked negative incentive effects.

Of course, microeconomic estimates suggesting high marginal welfare costs of taxation do not necessarily imply that tax cuts are in order. When there is a binding revenue constraint, lower tax wedges on labor income make sense only if other taxes can be raised in a less distortionary way. The remarkable aspect of the Swedish situation was that there was scope for a ”double dividend”. Higher taxes on housing generated a substantial part of the revenue required to finance the tax cut on labor income, but they also reduced intersectoral investment distortions. Due to the generous tax and subsidy rules, discussed above, housing investment was given a considerable advantage over investments in other sectors. Available estimates indicate (for realistic values of the debt-equity mix, inflation, etc) that new investments in owner-occupied housing could reap a net marginal subsidy – the (financial) cost of capital for the prospective homeowner was well below the real rate of interest. TR91 did much to promote a less inefficient allocation of investment resources.

The case of housing illustrates that the short-term response of an aggregate production measure, like GDP, may provide a poor indicator of the welfare effects of tax reform. Before TR91, household savings was channelled into an activity where the marginal productivity of capital was considerably below the opportunity cost, which in an open economy can be approximated by the world real rate of interest. At the same time, the housing sector gave a substantial contribution to Swedish GDP. In any year, housing investment, valued from the production side, added to aggregate investments. TR91 gave households strong incentives to redirect their savings to other uses, including net purchases of foreign assets.33 During the adjustment phase (when the housing sector shrinks, and the net foreign asset position improves) GDP growth tends to slacken. In spite of this negative production effect, aggregate consumption possibilities tend to increase – every krona’s worth of savings transferred from the housing sector to international asset markets implies that Swedish national income increases with the difference between the world real rate of interest and the marginal productivity of housing capital.

Undoubtedly, TR91 has affected economic efficiency along a number of margins in addition to those just discussed. The reduced progressivity of the income tax has enhanced educational incentives. The new corporate income tax brought about a more uniform treatment of investment projects within the corporate sector. The broader base for the value added tax implied higher tax wedges on the ’white’ consumer service sector, which competes with do-it-yourself activities and services produced in the underground economy. However, in these cases there is scant evidence on the behavioral response.

Finally, an important objective of TR91 was to simplify the tax code, and there is reason to believe that TR91 did much to reduce the transaction complexity of the tax system. TR91 made it much less profitable to invest resources in a variety of tax avoidance activities. Survey evidence indicates that households’ time spent on tax compliance declined substantially in the years after TR91.34 There is also evidence suggesting that the tax authorities got an easier work-load.

Concluding remarks

What is the verdict on TR91 five years after its implementation? Have we seen the behavioral responses that the reform architects expected? Did the reform contribute to a more efficient economy? Although we should keep our fingers crossed in view of the severe crisis in the Swedish economy in the early 90s, we concur with Auerbach and Slemrod (1996), who argued in their survey of TRA86 that there has been a hierarchy of responses. A number of financial activities related to tax planning were rendered meaningless and have been virtually wiped out. We have also seen large and expected effects on portfolio composition in general, with a shrinking of both sides of private sector balance sheets. Households have been induced to shift out of owner-occupied housing, resulting in falling house prices and a standstill of new construction. In the longer run, this will result in a more efficient allocation of the capital stock.

At the other end of the spectrum, major real activities like labor supply and savings appear quite insensitive, at least as far as can be inferred from short-run behavior. Since real activities can be expected to be more important from a welfare perspective, it might be conjectured that the reform only has made a small contribution to increase the efficiency of the Swedish economy, like Auerbach and Slemrod (1996) conclude for TRA86. While such a conclusion may be warranted at U.S. tax rates, it is hardly correct when starting out from marginal tax rates of 70 percent and more. Further, at such levels – close to the top of the Laffer curve – the marginal excess burden is highly non-linear in tax rates, implying that accounting for heterogeneity and uncertainty about the correct elasticity values is very important; the expected aggregate excess burden is much larger than the marginal excess burden for the average taxpayer evaluated at point estimates of elasticities.

Although a standard efficiency calculation – comparing hypothetical equilibria before and after the reform – unambiguously shows that TR91 was efficiency improving, these benefits were not without costs. In fact, the reform may be viewed as an investment with quite visible short-run costs that have to be weighed against less visible, and perhaps also less certain, long-run benefits. The short-run costs of TR91 were primarily of two types. First, by shifting savings out of real assets like housing and consumer durables into financial assets the reform implied a reduction of effective demand. Since the reform was implemented in a recession when output was arguably demand determined, we conclude that it led to a further deepening of the recession with accompanying short-term production losses. Second, the rapid implementation of the reform led to sizeable capital losses in the housing sector, with an ensuing arbitrary horizontal redistribution across households. These observations serve as a reminder of the point emphasized, e.g. by Feldstein (1976), that one has to distinguish between tax reform and de novo tax design. While a comparison of the pre and post-91 tax systems comes out in favor of the post-91 system, an evaluation of the reform has to weigh the long-run benefits against the short-term costs.

The costs of the reform were not unavoidable. The government took a rather careless attitude to the transition problem. A more gradual phase-in of the reform certainly would have dampened the short-run costs considerably without giving up any of the long-run benefits. While this would be a rather obvious recommendation to a benevolent dictator, things are more complex in a parliamentary democracy where one cannot tie the hands of future governments. Indeed, bringing about political consensus about a reform as far reaching as TR91 involved a rather delicate balance of the gains and losses of different interest groups in society. If parts of the reform would have been implemented gradually or with a lag, they would also have been more susceptible to future political pressures.

Even a ”tax reform of the century” implemented with such force as TR91 did not stay unaffected for very long. Already after three years, in 1994 one could make a list of some 75 tax changes involving minor and major deviations from the original reform. Such a count of changes obviously is a cheap argument; after all the world changes continuously, and one would hope that the tax law adapts. Nevertheless, some of the changes represent reversals of the guiding principles of TR91. One example is the numerous changes in different VAT rates. Another is the increase of the top marginal tax rate on earned income from 50 to 55 per cent. The latter change was presented as a temporary measure and part of a package to cut the growing budget deficits after 1992. While these deficits are largely attributable to the recession, they partly reflect an underfinancing of the reform by 2 to 3 percent of GDP. It is not surprising that it was easier to sell an underfunded reform package, where all groups could be portrayed as winners, than a fully financed reform where some groups would appear as losers, at least in a short-run accounting sense.

In what direction should we expect the Swedish tax system to evolve in the future? The 1980s was a decade of tax reforms aimed at a more uniform tax structure, conspicuously ignoring much of the development in public finance since the 1970s emphasizing the role of differentiated taxation in funding government expenditure. In the 1990s, the pendulum of tax reform discussion has swung in the opposite direction, with a renewed emphasis on differentation, e.g. for environmental reasons or as a means of fighting unemployment. When these factors now are being considered in Sweden, future reforms can build on a tax structure that has fewer counterproductive asymmetries than the pre-TR91 system.

The Swedish public sector ranks among the largest in the world. As long as this remains, Swedish tax rates will inevitably be high. Our discussion suggests that, especially with a high aggregate tax rate, the structure of taxation matters. TR91 has reduced the aggregate excess burden. However, the ever growing integration with international capital and labor markets will undoubtedly put increasing pressure on the Swedish tax system in the future.

Jonas Agell, Peter Englund and Jan Södersten

Författarna är professorer i nationalekonomi vid Uppsala universitet.