Skip Navigation

Two Papers on Fundamental Tax Reform October 1997

The model utilized for these simulations has been developed jointly with David Altig. The views in this paper do not necessarily reflect those of the Congressional Budget Office. We thank Bob Dennis, Doug Hamilton, Diane Lim Rogers, John Sabelhaus, Frank Sammartino and John Sturrock for their helpful comments. We also thank several participants in the Conference--namely, Eric Engen, Jane Gravelle, and Peter Taylor--whose helpful comments and constructive criticisms during pre-Conference meetings resulted in a vastly improved paper.

Fundamental tax reform would substantially alter the structure of incentives in the US economy. Understanding the consequences of shifting to a flat income or consumption tax therefore requires careful consideration of the changes in microeconomic behavior in order to assess the effects on macroeconomic variables. Our modeling approach accordingly starts with households and firms as the fundamental units of decision making in the economy. All changes in macroeconomic variables are then derived from changes in household labor supply, consumption, and saving decisions. Since the intra generational distribution of income and wealth is important due to the progressive structure of the current tax system, we distinguish households by age and earnings class.

The Joint Committee on Taxation (JCT) asked participants of this conference to examine two basic tax reforms. The first reform involves moving from the current progressive income tax system to a flat (proportional) income tax with an exemption level equal to $10,000 plus $5,000 for each dependent. Such reform would flatten tax rates, remove the double taxation of capital income and eliminate many tax preferences including the housing interest deduction, additional personal itemized deductions, personal tax credits, the deductibility of state income taxes and the favorable (consumption) tax treatment of retirement saving accounts. The second reform involves moving from the current tax income tax system to a consumption tax with uniform tax rates. In this reform, expected capital income is exempted from taxation by a move to full expensing.

Our numerical simulations reveal that the two reforms have very different implications for the economy. Under the income tax, most people earn less; and in the long run, the capital stock declines by 10.5 percent and the production of goods and services falls by 3 percent. Moving to a consumption tax, on the other hand, raises wage rates; the capital stock climbs 32 percent and output expands by 7.5 percent in the long run. Despite the fact that the flat income tax rate has a broader base (income minus deductions) than the flat consumption tax rate (consumption minus deductions), the tax rate in the consumption tax experiment eventually falls to 22.4 percent which is substantially below the long run tax rate of 25.0 percent in the income tax experiment. As the paper will explain, the difference stems from the fact that moving to a consumption tax (unlike moving to a proportional income tax) imposes a lump sum tax on existing wealth and eliminates the taxation of capital income.

The following section discusses the features of the substantially enhanced Auerbach-Kotlikoff life cycle model and our initial calibration. Section 3 explains the main results in light of our modeling approach. Although the model handles a great deal of complexities, it leaves out some portions of reality as reviewed in Section 4. This suggests viewing the model's results cautiously. Nonetheless, the simulation analysis reported herein does paint important brush strokes, even if broad ones.
 

THE MODEL

Our simulation model is based on the Auerbach-Kotlikoff (1987) life cycle model. It features 55 overlapping generations. Each agent lives for 55 years (ages 20 to 75). The model calculates the rational expectations (perfect foresight in our deterministic model) steady states as well as transition paths of factor prices, consumption, labor supply, tax rates, and other economic variables. There are three sectors: households, firms, and the government. The model does not have a monetary sector and all variables are real variables. The simulation results presented in this paper also assume a closed economy.

The model makes a number of important innovations to the Auerbach and Kotlikoff model. First, the model incorporates multiple lifetime income classes. This feature affords an analysis of the intra generational distributional impact of fiscal policy in addition to the intergenerational distributional impact analyzed by Auerbach and Kotlikoff (1987) and Auerbach (1996). Characterizing the intra generational distribution of wealth and income also allows for a more realistic analysis of fiscal policy, since the macro impact of a tax cut may depend on the initial distribution of lifetime income endowments and bequests. Second, the model incorporates an intergenerational bequest motive with bequests distinguished by income cohort. Third, the model includes a tax deduction against wage income. This requires the consumer to solve the lifetime optimization problem with a kinked budget constraint. We handle this complicated problem by assigning virtual marginal tax rates to consumers locating at the kink. Fourth, the model is carefully scaled to dollar units which makes it easy to match the model tax rates (and its Social Security replacement rates) to actual data. Fifth, the model incorporates a more realistic hybrid tax system as well as a more realistic Social Security system with the statutory earnings ceiling and the statutory bend-point formula applied over covered earnings. Sixth, the model incorporates labor-augmenting technological progress.

The Household Sector

Our model is particularly rich in modeling household decision making. Households decide how much to consume and how much to work in each period for given current and future after tax wages and interest rates. Households may--if they desire--not supply any labor at all in a given year and thereby retire or withdraw from the labor force. Following the lead of Fullerton and Rogers (1993), we divide households into 12 lifetime income classes. Classes 1 and 12 reflect the bottom and top 2 percent of lifetime income with classes 2 and 11 making up the remaining 8 percent of the bottom and top lifetime income decile. Classes 3 through 10 represent the intermediate lifetime income deciles. Wages for each lifetime income class grow according to a predictable fixed age-wage profile. We estimated these age-wage profiles from the PSID. Our procedure differs from Fullerton and Rogers (1993) in two main points: First, we control for a "cohort-effect" by including a birth-year indicator in our regression. This removes the effect of wage growth over time. Second, we sort wage profiles by individuals rather than by household wage income.⁽¹⁾ (For this purpose we exclude non-workers.) To see the difference, consider the following example. Suppose a person makes $100,000 per year and is married to someone who makes $20,000. In Fullerton and Rogers (1993), this household would be represented by a single agent who makes $60,000 per year since this is the amount of money available to each spouse if the household wage income is divided equally. In our model, on the other hand, the two agents would be modeled separately. Our procedure increases the dispersion of wage income which, under a progressive tax system, allows for a more accurate calculation of the tax rates faced by rich agents in the economy.

All households maximize a time-separable CES utility function with an intratemporal elasticity of substitution of 0.8 and an intertemporal elasticity of substitution of 0.25. The first parameter determines to what extent households are willing to substitute consumption for leisure in any given period while the second value determines how easily households substitute consumption (leisure) today for consumption (leisure) tomorrow. We also assume that households have a pure rate of time preference--that is the value at which future utility from future consumption and leisure is discounted--of 1.5 percent. These figures are the same as used by Auerbach and Kotlikoff (1987) who also review the relevant empirical literature. We incorporate an income class specific utility weight for bequests in our model in order to reflect the substantial differences in bequests across income classes. In particular we calibrate bequests in the initial steady state to reproduce the empirically observable size of bequests (Fullerton and Rogers [1993], p.99) by income class relative to mean income in the economy. Population growth is exogenous and set equal to 1 percent.

Production

Firms are perfectly competitive and employ labor and capital such that profits are maximized. There is only one production sector and therefore only a single good that can alternatively be used for investment and consumption. Firms in our economy produce according to a Cobb-Douglas technology. Since the production function is defined net of depreciation, we choose a capital share of 25 percent which accords well with most empirical research using this specification (see Auerbach and Kotlikoff [1987]). Technology is labor augmenting and grows at 1 percent per year.

Government

The government collects revenues for its spending on goods, services, transfers and interest payments via consumption taxes, wage taxes, income taxes, and capital income taxes. Each of these taxes may be modeled as proportional or progressive via a quadratic tax function. In addition, the government levies a payroll tax on wages to finance transfers to the elderly via Social Security and Medicare.

The Current Hybrid Tax System. We approximate the hybrid nature of the current US tax system by splitting the federal income tax into a progressive wage tax, a flat capital income tax, and a flat consumption tax. Following Auerbach (1996), capital income is taxed at a flat rate of 20 percent (a weighted average of the effective marginal tax rates on housing and non-housing capital) and we allow firms to expense 20 percent of new investment in order to reflect the accelerated deprecation allowance under current law. Together, these assumptions imply an effective marginal tax rate on capital income equal to 16 percent. Since contributions to pension funds under current law are part of labor compensation but receive consumption tax treatment, we levy a 2.5 percent tax on consumption and reduce taxes on wages accordingly. (The reader is referred to Auerbach [1996] for the derivation of these numbers.) An ordinary least squares regression is used to approximate the statutory wage tax schedule for individual filers with a quadratic function. The regression achieves a very good fit with an R² equal to 0.998. This function is then applied to wage income above the federal personal exemptions and standard deduction which, in total, equals $9662.⁽²⁾ To account for itemized deductions, we used the IRS Statistics of Income to compute how itemized deductions (not including the mortgage interest deduction which was already factored into the 20 percent tax rate on capital income) rises with income. We find that itemized deductions increase by $0.0755 for every dollar of income above the combined standard deduction and exemption level. That estimate is derived from a regression with an R² of 0.99.

Our calibration also includes state and local taxes as well as the remaining federal taxes. State taxes are represented with a flat income tax of 3.7 percent. Consistent with NIPA values, we collect an additional consumption tax of 8.8 percent which reflects indirect business taxes and excise taxes. Thus the total tax on consumption under current law is assumed to be 11.3 percent. We ignore property taxes following the view that property taxes equal benefits received at the local level.

Tax Evasion. Because we are using a very close approximation to the statutory code to parameterize our tax functions, we have, up to this point, ignored the possibility of evasion. Without any correction for tax evasion, government revenues in our model would be higher than those found in the NIPA accounts--and indeed exceed their NIPA values by almost exactly the amount of tax evasion estimated by Slemrod and Bakija (1996). We corrected for evasion with a negative proportional income tax rate of 2.6 in the initial steady state which reduces the average and marginal income tax rates for all agents in the economy. (The marginal and average tax rates are still positive for all agents however since, at a minimum, everyone faces a flat state income tax of 3.7 percent.) For both tax reform experiments, we assume that evasion reduces the taxable base (income net of standard deductions and exemptions) by the same percentage before and after the reform.⁽³⁾

Social Security, Medicare, and Other Transfers. We calculate the OASI replacement rates for covered earnings using the statutory bend point formulas. Benefits are also scaled in order to reflect survivor benefits. The endogenous OASI payroll tax necessary to finance these benefits equals 9.8 percent which is close to the 1995 value of 10.52 net of the trust fund contributions of 0.7 percent of payroll.⁽⁴⁾ Trust fund contributions are included in other non social security-related wage taxes. We assume that payroll taxes are only partly distortionary up to the maximum taxable earnings of $61,700. In particular, we set the perceived link between the present value of taxes and the present value of future benefits to 25 percent. Thus, agents consider only 75 percent of the payroll tax as a tax on labor.⁽⁵⁾ Agents with labor supply above the maximum taxable earnings face a marginal payroll tax of zero, thus the payroll tax is non-distorting for them. We model Medicare (HI) as a non-earnings related transfer to agents age 65 and older and disability insurance (DI) benefits as a lump-sum transfer to agents below age 65. These benefits are financed through payroll taxes of 2.9 percent (HI) and 1.9 percent (DI) which equal their current statutory values. In contrast to the DI tax and the OASI tax, the HI tax is not subject to the earnings ceiling. In addition to modeling the social insurance system, we rebate about 1.8 percent of national income to agents as a (wage-indexed) lump sum transfer. This transfer accounts for other transfer programs as AFDC and Medicaid.

Government Debt Service. Finally, we select the level of government debt in the initial steady state to set the real interest payments on government debt equal to 1.5 percent of national income, its 1995 level. Targeting interest payments correctly is important in order to accurately reflect any gains from lower debt service should interest rates fall after tax reform. Our model does not explain the equity premium and therefore has only a single interest rate, the (real) net rate of return to capital, which is substantially higher than the real rate of return to government bonds. The ratio of debt to national income is consequently about half of that observed for the US economy. The results would be the same if we used a lower interest rate for government debt that moved one-for-one with the real rate on capital.

Description of Initial Steady State

The model does a good job at generating endogenous values of variables which match their real-world counterparts. The generated economy lines up well with the actual economy even though most of the model's parameters are picked either according to the estimates in the literature or according to statutory code.⁽⁶⁾

In terms of aggregate values, we obtain an economy-wide average marginal tax rate on wage income equal to 21.5 percent which is close to the TAXSIM calculations reported in Auerbach (1996) while our economy-wide average tax rate on wage income equals 13.3 percent. Total government revenue net of payroll taxes, is 24.4 percent of NI (national income), matching the value found in the 1995 NIPA accounts less property tax revenue. The model generates a pre-tax interest rate equal to 9.6 percent. The net saving rate equals 5.3 percent. The capital-NI ratio is 2.6 which is close to the 2.8 value derived from the 1994 balance sheets published by the Federal Reserve Bank.⁽⁷⁾ Simulated consumption comprises 73.2 percent of NI (whereas the actual value was 74.3 percent for 1995 in the NIPA accounts), net investment equals the saving rate of 5.3 percent (which equals its actual value at an economy-wide depreciation rate equal to 5.0 percent), and government spending on goods and services accounts for the remaining 21.5 percent which is equal to government revenues net of interest payments on the debt and the lump sum transfer noted earlier.

The wage rate of an individual at age 45, which corresponds to the peak earning age for all of the income classes, is $4.00, $14.70 and $79.53 for income class 1, income class 6, and income class 12, respectively. Class 1 represents the lifetime poorest, and classes 6 and 12 reflect median earners and the lifetime richest group. Annual labor income--endogenously derived from leisure choices--ranges between $9,700 and $160,000 at around age 45. The model generates a net national income per capita of all agents between ages 21 and 75 equal to around $39,000, which is very close to its empirical value of about $38,500 derived by dividing 1995 national income by the sum of the labor force and retired individuals.

The average tax rate on wage income, averaged across individuals of all ages in income class 12, equals 20.4 percent, while the average marginal tax rate equals 29.5 percent with top marginal tax rates of 35 percent. For income class 6, these average and average marginal rates are 11.3 percent and 19.9 percent, respectively, while for class 1, the rates are 0 percent and 2.9 percent.⁽⁸⁾ The proportion of income derived from wage and interest income across these groups also match the SOI data rather closely.
 

THE IMPACT OF FUNDAMENTAL TAX REFORM ON MACROECONOMIC VARIABLES

Unified Flat Income Tax

In accordance with the specifications given by the JCT, the first experiment replaces the progressive federal income tax with a flat income tax on wage income and capital income. The total of personal exemption and deduction is raised to $16,101.⁽⁹⁾ Itemized deductions are eliminated. The OASDI and HI programs remain the same. Any positive or negative budget savings from changes in interest rates are reflected in replacement tax rates. Replacement tax rates are set to finance the same amount of government spending for goods and services in the new tax regime as in the initial steady state. The consumption tax is reduced from 11.3 percent to 8.8 percent to reflect the loss of the consumption tax treatment of retirement saving accounts. The expensing rate for investment remains at 20 percent. The state and local income tax rate is increased from 3.7 percent to 4.4 percent to reflect the loss of deductibility of state and local income taxes from the federal income tax base.⁽¹⁰⁾ As described before, we assume that tax evasion as percent of the taxable base stays the same. Table 1 presents the effects of this tax reform on major macroeconomic variables. We assume that the reform is implemented on January 1, 1997. 1996 therefore represents the pre-reform economy.
 

What accounts for these results? Essentially, the decline in labor supply and saving stems from the increase in marginal tax rates for a number of income classes under the reform. Although the reform repeals a number of tax preferences in the current system, that is not sufficient to finance the increase in the personal tax exemption. Only classes 1 and 2 will be totally exempt from taxes due to the higher deductions; the rest will face higher marginal tax rates on labor income over at least part of their life. In addition, the effective marginal tax rate on all capital (including housing) under the reform increases from 16 percent to 20.0 percent.⁽¹¹⁾ As a result of higher marginal tax rates, this reform proposal reduces the incentives to work and save which slows economic activity and causes the tax base to contract. That contraction of the capital stock and labor supply reinforces itself along the transition path of the economy as the shrinking tax base increases the marginal tax rates needed to finance government spending at its original level, leading to further contraction of the capital stock and labor supply.

Flat Consumption Tax

The second experiment involves moving from the current tax system to a consumption tax. Technically, the experiment is identical to the one described above except that firms can fully expense new capital investment. The results of this experiment are shown in Table 2.
 

Our model shows that households react to the drop in their wealth by reducing their consumption of both goods and leisure. Put differently, they increase both their saving and their labor supply to make up for the lower value of their wealth. In addition to the wealth effect, the labor supply and saving responses can be explained as a result of intertemporal substitution. Since after-tax interest rates rise shortly after tax reform and then fall subsequently, people face incentives to work and save more assets shortly after the reform.⁽¹³⁾ On the other hand, higher marginal tax rates on many people's labor income offset some of the positive effects on labor supply.

In the simulations, labor supply initially increases by 2.5 percent, which boosts output by 1.2 percent. At the same time, the removal of the capital income tax and the drop in the value of existing assets leads to an increase in the saving rate from 5.3 percent to 9.0 percent in the short run. Labor supply decreases slowly over time and its level in the long run barely exceeds its original level. In the medium and long run, the growth in output is mostly driven by capital accumulation. Four years after the reform, the capital stock is 7.2 percent larger, and by year 9, it has increased 14 percent above baseline. Eventually, the capital stock exceeds its initial level by 31.5 percent. Accordingly, output is 2.4 percent larger than baseline in year 4 and 4.0 percent larger in year 9. In the long run, output increases by 7.5 percent.⁽¹⁴⁾ In response to the changes in long run factor supply, the interest rate falls from 9.6 percent to 7.8 percent and (before-tax) wages increase by 7.1 percent above the baseline.

Our findings indicate that the proportional tax rate on consumption would have to be initially 25.8 percent.⁽¹⁵⁾ Note that this rate is tax inclusive. It is higher than the initial tax rate under the uniform flat income tax experiment reported earlier (because the consumption base is smaller than the income base) but is about 5 percentage points lower than what the tax rate would have been without a lump-sum tax on existing wealth.⁽¹⁶⁾ Because the reform increases output, the consumption tax rate drops to 26.3 percent after 9 years and 22.4 percent in the long run. After around 30 years, the revenue from higher output overcomes the initial base narrowing and thus the replacement tax rate falls below that found in the flat income tax experiment. The economy grows above the baseline despite the increase in marginal tax rates on the labor income for many households. That result follows from the fact that a consumption tax removes the tax on capital income which stimulates saving and investment.
 

SHORTCOMINGS OF THE MODEL

The model incorporates many complex details of the real economy and relies on only a few exogenous so-called "deep" parameters specifying the utility and the production function. We think that these features are especially important for assessing the effects of fundamental tax reform. However, as in any model, our model abstracts in some ways from reality. As a result, we urge a cautious interpretation of our exact quantitative results. We outline below some of the omissions we consider most important.

No Differentiation of Production Sectors

The model assumes a single production sector which produces both consumption and investment goods. Thus, the model cannot distinguish among different sectors. Other models used in this conference, specifically those of Fullerton and Rogers (1993) and Jorgenson and Wilcoxen (1997), feature multiple sectors and can therefore capture the substitution between housing and non-housing capital as well as the effect of reducing the tax-differential between corporate and non-corporate activities. We may therefore underestimate possible efficiency gains in the flat income tax experiment which eliminates distortions among different types of capital.

No Borrowing Constraints

The overlapping generations model underlying our simulations allows consumers to borrow against future resources without constraint. Some empirical evidence, however, suggests that as much as 20 percent of the population faces binding borrowing constraints (e.g. Hayashi [1987], Mariger [1987]). Introducing binding borrowing constraints would not alter our results much because only the bottom 2 percent of earners in our model occasionally hold negative contemporaneous asset (net worth) positions.⁽¹⁷⁾ Since this group accounts only for a tiny part of accumulated wealth, incorporating borrowing constraints would not significantly alter the results presented above.

The empirical evidence could also be interpreted as evidence for rule-of-thumb savers who consume a certain percentage of their income independent of future wages and interest rates. Incorporating such behavior would likely dampen the savings response to tax reform. It is unclear, however, how to derive such a rule-of-thumb. Additionally, such consumers may change their rule after a fundamental tax reform in unpredictable ways. Finally, most of our short-run results in the consumption tax experiments are driven by the labor supply response to the lump-sum tax on existing wealth rather than by a saving response. For example, in another experiment we performed (not reported) which partly shielded existing assets against taxation, gains to aggregate variables were significantly attenuated.⁽¹⁸⁾

No Uncertainty Regarding Wage Income and Longevity

Our model employs fixed wage efficiency profiles for each earnings class and does not incorporate wage income uncertainty. Since uncertainty about future wages may induce agents to build up a stock of precautionary wealth ("buffer stock") our model may over-predict the post-reform saving response because precautionary saving are not sensitive to changes in the interest rate (see, e.g., the contribution to this Conference by Eric Engen [1997]).

Our assumption of a certain lifespan until age 75 implies the availability of actuarially fair annuities (Blanchard [1985]; Feldstein [1988]). In reality those annuities do not exist and the lack of annuities should give rise to additional saving against longevity uncertainty which is not reflected in our model. However, Social Security and Medicare are both annuity programs. Since they replace a large part of income for the median earner, the lack of an annuity market may not distort our results to a major extent. Nonetheless, a realistic incorporation of longevity uncertainty would be a very useful extension to our model.⁽¹⁹⁾

No Explanation of the Equity Premium

It is theoretically unclear whether a model without aggregate level uncertainty should target the risk free rate of return or the rate of return to capital since the latter reflects a premium for uncertainty which itself has not been explained sufficiently (i.e., the so-called "equity premium puzzle"). Smetters (1996) shows that responses to fundamental tax reform depends crucially on which rate of return is targeted. Models, such as ours, which target the equity rate tend to produce larger responses than those targeting the risk free rate. However, without jointly explaining the risk free interest rate and the equity rate, this issue cannot be resolved.
 

WHAT ACCOUNTS FOR THE DIFFERENCES IN RESULTS AMONG MODELS?

Although most of the models in this conference produce similar qualitative results, the models differ--sometimes significantly--in their quantitative conclusions, especially regarding the effects of the consumption tax. This section outlines what we believe are the key reasons why the simulation results of our model differ from both those of the other intertemporal models and from the reduced-form equation models.

Intertemporal Models

Jorgenson-Wilcoxen Model. The key difference between the J-W model and our model is the initial effects of the consumption tax on labor supply. The J-W model produces a 7 percent increase while our model produces a 3 percent increase. This, in turn, explains why the J-W produces a first-year increase in output equal to 4 percent, whereas our model produces only an 1 percent increase. The J-W model produces a large labor supply response because they assume that the utility of their representative agent can be represented by the natural log of full consumption, which implies an intertemporal substitution elasticity (IES) that is four times larger than the value that we use. Higher values for the IES increase the responsiveness of labor supply to changes in the after-tax rate of return to capital. This result occurs because agents choose to work more today in order to increase saving in an attempt to take advantage of the increase in the after-tax rate of return to capital.⁽²⁰⁾
 

Fullerton and Rogers Model. Our model is most similar to the model of Fullerton and Rogers but differs in four important ways. Those differences explain why the long-run gains in output from a consumption tax are smaller in the F-R model, even when the F-R model assumes a value for the intertemporal substitution elasticity that is twice the size of ours. First, every agent in the F-R model faces the same marginal tax rate regardless of income. By contrast, our model has a progressive income tax system and some of the gains to output come flattening the current tax structure. Second, intergenerational bequests are fixed in the F-R model--and so they do not respond to changes in relative prices--whereas bequests in our model are endogenous and respond to those chances. That feature reduces the sensitivity of total saving to the rate of return in the F-R model but enhances that sensitivity in our model. Third, the F-R model utilizes a Stone-Geary utility function that requires a minimum purchase of certain commodities in each period. Using a Stone-Geary utility function tends to reduce the long-run saving response somewhat since the consumer's choice is narrowed down to that between leisure and discretionary spending.⁽²¹⁾ Fourth, our model, but not the F-R model, incorporates an initial level of government debt. A shift to a consumption tax produces lower interest rates that, in turn, reduces the government's cost of debt service. Because the policy changes are assumed to be revenue neutral, the government's debt level is unaffected by tax reform and so a reduction in debt service lowers replacement tax rates, which increases output in the long run.

Engen's Model. In Engen's model, precautionary saving accounts for about 60 percent of total wealth whereas it is non-existent in our model. As a result, saving in our model tends to be more sensitive to the after-tax interest rate in our model than in Engen's model.

Our model, however, includes saving for an intentional bequest that is distributed to one's heirs, whereas the Engen model incorporates an accidental bequest that is distributed evenly throughout the economy. The intentional bequest feature of our model may not be particularly important for most poor households and for some middle income class households but it is important for modeling the behavior of the wealthy top 10 percent of households who own almost 70 percent of the capital stock (Survey of Consumer Finances [1992]) and for whom precautionary saving may or may not be an important saving motive. The incorporation of an intentional bequest motive increases the sensitivity of saving to the after-tax interest rate and the distribution of bequests to one's own family members--rather than to society at large--would reduce saving for precautionary reasons.

The importance of both intentional bequests and precautionary saving has received great scrutiny in the literature and many economists come down on each side of both issues. Indeed, some economists believe that neither motive for saving is very important. But most economists would probably agree that an ideal model would include both of these saving motives. Incorporating both features, however, is very difficult to do at this point. Yet both models render very similar qualitative--though not identical quantitative results--regarding tax reform.

Reduced-Form Equation Models

An advantage of the reduced-form equation models presented in this conference is their reliance on simple equations instead of complex mathematical programming problems, like those in our model. But that simplicity makes their models vulnerable to a significant criticism.

Elasticity Driven Models: Coopers and Lybrand, Gravelle, and Robbins and Robbins. The simulation results of the Coopers and Lybrand (C-L), Gravelle and Robbins and Robbins (R-R) models are primarily driven by the values they choose for the saving and labor supply elasticities. The authors gather those elasticities from the literature and from their own estimation. These elasticities are used instead of the intertemporal substitution elasticities and intratemporal elasticities in our model.

The critical assumption made by the C-L, Gravelle and R-R models is that these elasticities will remain constant after a policy change. But those saving and labor supply elasticities are derived parameters that combine both consumers' preferences and policy. In principle, those parameters will vary as policy is changed. (This criticism is called the "Lucas critique.") By contrast, the parameters in the intertemporal models--the intertemporal substitution elasticities and intratemporal elasticities--are "deep" parameters that describe household preferences and are not affected by changes in policy.

Simulation analysis suggests that the derived parameters can vary radically from one policy change to the next. For example, when we use our model to simulate the growth in pay-as-you-go Social Security over the past several decades, we find a negative relationship between aggregate consumption and wealth and a negative relationship between aggregate consumption and contemporaneous interest rates. The reason is that an increase in pay-as-you-go Social Security reduces wealth which, in turn, increases interest rates in our simulated economy. But the growth in Social Security also increases aggregate consumption as resources are transferred from higher saving workers to lower saving retirees. Clearly, saving elasticities whose values are estimated from historical variation in wealth and interest rates could be seriously biased. It is not surprising therefore that Joel Prakken found little historical relationship between labor supply and wealth, as he reported in this symposium. The major sources of historical variation would tend to produce correlations between consumption, labor supply and aggregate economic variables that are weaker than that we expect to observe under tax reform.

For the purpose of obtaining accurate parameters, what matters is not so much the length of sampling period or the size of the variations in aggregate variables in the historic times series but the source(s) of these variations. But there has not been anything close to fundamental tax reform in US history like the ones being considered for the JCT symposium.

An additional problem with the C-L, Gravelle and R-R models is that they ignore the lump-sum tax on existing wealth of a consumption tax. (Recall that this lump-sum tax was a significant contributor to the short-run increase in saving and labor supply in our model.) By ignoring this lump-sum tax, the predictions of those models about the responsiveness of the capital stock and output to tax reform will be biased downward.

The R-R model produces much larger increases in the capital stock and output in both the short and long run after a switch to a consumption tax than the other models. The reason is that the R-R model assumes that a move to a consumption tax will be met with a large net capital inflow that will equalize the after-tax interest rate across countries. But, as Jane Gravelle (1996) points out, the response of the capital inflow in the R-R model would be smaller if the R-R model had allowed for the fact that foreigners normally do not pay U.S. taxes on interest, dividends and capital gains.

The Large Macroeconomic Models: DRI and MA. These two models are subject to the same criticisms as the above reduced-form equation models. Like the above models, both models assume that the parameters in their reduced-form equations would remain constant after a switch to a consumption tax. (They differ somewhat from the C-L, Gravelle, and R-R models in that their supply elasticities are chosen so their models track the historical performance of the economy; the other models choose their elasticities from the empirical literature.)

The consumption function used in the MA model is probably the most realistic of the two models because it is a reduced-form equation of the life cycle model. The consumption function used in the DRI model is atheoretic. Nonetheless, even the MA model fails to satisfy the Lucas critique. The reason is that the model's reduced-form parameters are a function of the distribution of assets across different age cohorts. The MA model assumes that this distribution remains constant after a switch to a consumption tax whereas, in fact, the distribution changes significantly in all of the intertemporal life cycle models discussed earlier, including our own. This assumption tends to bias the results of the MA model downward.

While both models attempt to include some measure of the lump-sum tax on existing wealth of moving to a consumption tax, neither model is able to properly capture the intergenerational redistributive aspects of moving to a consumption tax. That redistribution is a driving force in the intertemporal models. As a result, both the DRI and MA models predict gains to moving to a consumption tax that are smaller than those reported using our model.
 

CONCLUSION

This paper evaluates two tax reform proposals in a extended and improved version of the Auerbach-Kotlikoff life cycle model with 55 overlapping generations. We find that replacing the current tax code with a flat income tax that does not exempt housing capital is likely to reduce the levels of the capital stock, output and wages in the long run. However, moving to a flat tax that exempts capital income from taxation would substantially raise output and wages. Our model exhibits relatively large increases in the labor supply due to the reduction of progressivity in the tax schedules and, most importantly, due to the lump-sum tax on existing wealth when moving to full expensing. Taxation of existing wealth reduces the consumption of all normal goods including leisure. Those increases in labor supply are likely to be greatly diminished if a consumption tax is combined with some kind of transition relief to ease the burden on people who hold existing assets.

Our model includes a large number of complex processes by explicitly solving for the exact transition path of an economy with 55 utility maximizing overlapping generations, each divided into 12 income classes. The model incorporates many important and complex aspects of reality but, as any model, excludes some parts of reality. Like the results of any simulation model therefore the results should be viewed cautiously. Nonetheless, the model seems to capture some important effects of tax reform due to its systematic footing in consumer optimization. Most notably, the model shows that a tax on existing assets can significantly increase labor supply and output in the short run, and demonstrates the importance of analyzing transition provisions in any proposal to reform the tax code.

2. We add the standard $4,000 deduction, personal exemptions of $2,550 and exemptions of $3110 for the 1.2 children of an average agent in the model (consistent with the 1 percent population growth rate) to arrive at this figure. The computation of the marginal tax rate applied to wage income includes interest income.

3. Since the taxable base shrinks due to higher exemptions in both experiments we set our downward adjustment of flat income taxes due to evasion to 1.4 percent after the tax reform. Note that this may be an optimistic approach if evasion occurs mostly among agents with higher income.

4. See Social Security Administration (1995).

5. See Auerbach and Kotlikoff (1987), Chapter 10, for more detail on this issue.

6. The only real unobservable "free parameter" in the model is the weighting parameter placed on consumption versus leisure. We choose the value of this parameter--as is traditionally done--in order to generate a reasonable average 35 hour work week. All of the other utility parameter choices follow Auerbach and Kotlikoff (1987) who discuss their empirical foundation.

7. Total reproducible assets equaled $15.6 trillion in the Fed's 1994 balance sheets. This included $5.8 trillion of residential structures, $2.5 trillion of consumer durables, $6.1 trillion of nonresidential fixed private capital and $1.2 trillion in inventories. National income equaled $5.5 trillion in 1994.

8. Lifetime poorest face a positive average marginal tax rate across their lifetime because of a few periods where those agents are at the 'kink' and face positive shadow marginal taxes.

9. This is consistent with a non-refundable exemption of $10,000 plus $5,000 per dependent. Each agent has (1.01)²⁰ 1.22 dependents and so the exemption equals $10,000 + $5,000(1.01)²⁰ = $16,100.95.

10. Since the model has a single unified government sector it is irrelevant for the results that these additional revenues would in reality accrue at the federal level.

11. Our model has only a single production sector and, thus, does not explicitly capture any substitution between housing capital and non-housing capital that tax reform might induce.

12. In a model with a realistic level of adjustment costs, the initial labor supply response is attenuated because existing wealth would be partly (possibly fully) shielded from the lump-sum tax. See Auerbach (1996).

13. Experiments which reduce the loss in wealth with some transition relief as, for example, a lump sum tax rebate let us believe that the wealth effect dominates the labor supply response.

14. Our long run marginal tax rate including the additional consumption tax of 2.5 percent rate is roughly midway between the marginal rates found by Auerbach (1996) for the Armey-Shelby Plan and the Hall-Rabushka Plan experiments. Our predicted output growth also falls between the figures reported by Auerbach for his simulations.

15. This is the statutory tax rate before taking evasion into account.

16. The consumption tax base is about 75 percent of the size of that under the income tax base. Without the tax on existing wealth therefore the replacement tax rate would have to be about (3/4)^-1 times 23.5 percent (the initial tax rate under a uniform income tax), or about 31 percent.

17. Much of the borrowing in the actual economy is for homes and education. Home borrowing however does not lead to a negative net worth position and so many households in our economy with positive assets can be interpreted as having borrowed for a home. Borrowing which leads to a negative net worth involves mortgaging one's own future human capital. This typically takes the form of education loans which our model does not incorporate and so our model is presumably underestimating borrowing of this type.

18. This implies that intertemporal substitution alone cannot explain the relatively strong responses in our model. Recall that we use an intertemporal substitution elasticity over full consumption (consumption and leisure) equal to only 0.25. Whereas Robert Hall (1978), for example, found that the intertemporal substitution elasticity over consumption alone is very small (around 0.10)--although subsequent work found larger values--Dale Jorgenson and Peter Wilcoxen (this conference), for example, use a value of 1.0, Eric Engen (this conference) uses 0.33 based on his own estimates and Diane Lim Rogers (this conference) considers both 0.15 and 0.50.

19. Models featuring lifetime uncertainty need to explicitly solve the dynamic programming problem of children who anticipate a bequest which is not perfectly deterministic in either size or timing. We are not aware of any model that does so. If bequests are distributed lump sum across the entire population (as in some models) or confiscated by the government (as in some other models), longevity uncertainty only augments the rate of time preference.

20. This effect is partly offset by the fact that the J-W model assumes away any intergenerational redistribution of wealth from old retirees to young workers. In our model, the intergenerational redistribution accounts for much of the labor supply response. The J-W model cannot reflect such redistributions because the entire household sector is represented as a single infinitely lived agent.

21. Enforcing a minimum consumption level is important in the F-R model because the Inada condition--i.e., that marginal utility tends to infinity as consumption approaches zero--is not applied to each consumption item. It is unclear how important such a constraint would be in our model which enforces the Inada condition over total consumption. In our simulations, the Inada condition already prevents consumption from becoming very small in any given period.
 

References

Auerbach, Alan J., 1996. "Tax Reform, Capital Allocation, Efficiency, and Growth," in: Aaron, Henry and William Gale (eds.), Economic Effects of Fundamental Tax Reform, The Brookings Institution, Washington, D.C., 29-81.

Auerbach, Alan J. and Laurence J. Kotlikoff , 1987. Dynamic Fiscal Policy, Cambridge University Press, Cambridge, England.

Blanchard, Olivier J., 1985. "Debt, Deficits, and Finite Horizons," Journal of Political Economy, 93: 223-247.

Browning, Martin and Annamaria Lusardi, 1996. "Housing Saving: Micro Theories and Micro Facts." Journal of Economic Literature, 44 (4): 1797 - 1855.

Caballero, Ricardo J., 1991. "Earnings Uncertainty and Aggregate Wealth Accumulation," American Economic Review, 81: 859-71.

Engen, Eric, 1997. Paper prepared for this conference.

Feldstein, Martin S., 1988. "The Effects of Fiscal Policies when Incomes Are Uncertain: A Contradiction to Ricardian Equivalence," American Economic Review, 78: 14-23.

Fullerton, Don and Diane Lim Rogers, 1993. Who Bears the Lifetime Tax Burden, The Brookings Institution, Washington, D.C.

Gravelle, Jane G, 1996. "Comparison of Models." Mimeo. Congressional Research Service.

Hall, Robert, 1978. "Stochastic Implications of the Life Cycle-Permanent Income Hypothesis: Theory and Evidence," Journal of Political Economy, 86: 971-987.
 

Hall, Robert and Alvin Rabushka, 1983. The Flat Tax, Hoover Institution Press: Stanford.

----------, 1995. The Flat Tax, 2nd Edition, Hoover Institution Press: Stanford.

Hayashi, Fumio, 1987. "Tests for Liquidity Constraints: A Critical Survey and Some New Observations," in: Bewley, Truman F. (ed.) , Advances in Econometrics II, Fifth World Congress, Cambridge University Press, Cambridge, England, 91-120.

Jorgenson, Dale, 1996. Paper prepared for this conference.

Mariger, Randall P., 1987. "A Life-cycle Consumption Model with Liquidity Constraints: Theory and Empirical Results," Econometrica, 55: 533-557.

Poterba, James M. and Andrew A. Samwick, 1995. "Stock Ownership Patterns, Stock Market Fluctuations, and Consumption," in: Brainard, William C. and George L. Perry (eds.), Brookings Papers on Economic Activity, 2: 295-358.

Slemrod, Joel and Jon Bakija, 1996. Taxing Ourselves: A Citizen's Guide to the Great Debate over Tax Reform. MIT Press: Cambridge, Mass.

Smetters, Kent A., 1996. "How Important are the Parameter Assumptions in the Analysis of Capital Income Taxation?", mimeo. The Congressional Budget Office.

Social Security Administration, 1995. Annual Statistical Supplement to the Social Security Bulletin. U.S. Government Printing Office: Washington, D.C.

---