[Economic Report of the President (2004)]
[Administration of George W. Bush]
[Online through the Government Printing Office, www.gpo.gov]





 
CHAPTER 5

Dynamic Revenue and Budget Estimation


A central conclusion of the study of taxation is that taxes affect
behavior and distort the choices of firms, workers, and investors. In
particular, a higher tax on an activity tends to discourage that
activity relative to others. These behavioral responses to a tax change
can, among other things, alter the revenue effect of the tax change,
a topic that is the focus of this chapter. Revenue estimation is called
dynamic if it incorporates the revenue implications of behavioral
responses to tax changes and static if it does not incorporate these
revenue implications. Like changes in taxes, changes in government
spending can encourage or discourage certain behavior; budget estimates
are dynamic if they incorporate the budgetary implications of these
behavioral responses.

If policy makers are to make informed decisions about policy
changes, all significant effects should ideally be included in
estimates of the policy's budgetary implications. Several obstacles
have prevented macroeconomic behavioral responses from being
incorporated in such estimates. This chapter discusses the ongoing
efforts to provide a greater role for fully dynamic revenue and budget
estimation in the analysis of major tax and spending proposals.

The key points in this chapter are:
Currently, official revenue estimates of proposed tax
changes are not fully static because they incorporate
the revenue effects of many microeconomic behavioral
responses. These estimates are not fully dynamic,
however, because they exclude the effects of
macroeconomic behavioral responses.
Changes in taxes and spending generally alter
incentives for work, investment, and other productive
activity. These macroeconomic behavioral responses
have revenue and budgetary implications.
Steps have recently been taken to provide more
information about the revenue effects of macroeconomic
behavioral responses. At least in the near term, it
may not be practical for macroeconomic effects to be
incorporated in official estimates. But estimates of
these effects should be provided as supplementary
information for major tax and spending proposals.
Dynamic estimation of policy changes should
distinguish aggregate demand effects from aggregate
supply effects, include long-run effects, apply to
spending as well as tax changes, reflect the differing
effects of various policy changes, account for the
need to finance policy changes, and use a variety of
models.


Revenue Estimation and
Microeconomic Behavioral Responses

To frame the issues, it is useful to begin with a simple example of
how a tax change can affect behavior and how the behavioral response
then alters the revenue impact of the tax change.

An Example of Revenue Implications of
Microeconomic Behavioral Responses

Consider an excise tax on apples (similar to that discussed in
Chapter 4, Tax Incidence: Who Bears the Tax Burden?). If the current
tax rate is 25 cents per apple and 1,000 apples are produced and
consumed at this tax rate, tax revenue is $250. Now, suppose the tax
rate is cut to 20 cents per apple. If apple output and consumption
don't change, total tax revenue falls to $200, a decrease of $50.
Therefore, a purely static estimate of the revenue loss would be $50.

The actual change in tax revenue is likely to be different, however,
because consumers and producers respond to the tax rate change. The
tax rate drives a wedge between the price paid by consumers (including
the tax) and the price that producers receive (net of the tax). When
the tax rate is reduced, this wedge is reduced, meaning that
consumers are likely to pay a lower price and producers are likely
to receive a higher price. For example, at the 25-cent tax rate,
consumers might pay $1.13 per apple and producers might receive
88 cents per apple; at the 20-cent tax rate, consumers might pay $1.10
and producers might receive 90 cents. The lower price paid by
consumers induces them to consume more apples and the higher price
received by producers induces them to produce more apples. As
explained in Chapter 4, the changes in the two prices must be such
that consumers' desired increase in consumption equals producers'
desired increase in production.

Suppose that 1,100 apples are produced and consumed at the lower
tax rate. (The actual increase in the quantity of apples depends on
how responsive consumers and producers are to their respective prices;
the increase is larger when both groups are more responsive.) Tax
revenue is then $220 (20 cents per apple times 1,100 apples), not
$200. Thus, the tax cut lowers revenue by $30, not $50. Of the $50
static revenue loss, $20 is ``paid for'' by the increase in apple
production and consumption caused by the tax cut. In other words,
40 percent of the tax cut ``pays for itself'' through this revenue
feedback.

Conversely, increasing the tax from 25 cents to 30 cents yields
$50 of additional revenue if the quantity of apples remains at 1,000.
However, the quantity of apples is likely to fall as the tax rate
increases. With the higher tax, consumers pay a higher price and
producers receive a lower price, prompting a decline in the desired
levels of apple production and consumption. If the quantity of apples
falls from 1,000 to 900, revenue rises from $250 to $270, so that the
revenue gain from the tax rate increase is $20 rather than the $50
that would occur with no behavioral response.

The actual revenue effects of such a tax change may be more complex
than this discussion suggests. As the quantity of apples changes,
the quantities of other items produced and consumed also change. If
those items are also subject to taxes, changes in those quantities
also impact revenue. In any event, behavioral responses to a tax
change can alter its revenue impact.


Incorporation of Microeconomic Behavioral Responses
in Revenue Estimation

The insight that microeconomic behavioral responses to tax changes
affect revenue has been incorporated into the official
revenue-estimation process.

The staff of the Joint Committee on Taxation (JCT) prepares the
official revenue estimates for thousands of proposed tax changes
submitted by members of Congress each year. Similarly, the Department
of the Treasury prepares official revenue estimates for tax changes
proposed by the President and some changes considered by the Congress.
Official estimation of the revenue effect of a tax change is commonly
called scoring. Each revenue estimate presents the estimated change in
revenues in the current fiscal year and up to 10 subsequent fiscal
years, a period referred to as the ``10-year window.''

In preparing their estimates, JCT and Treasury economists routinely
include the effects of microeconomic behavioral responses to tax
changes. For example, when excise taxes change, JCT and Treasury
estimates reflect how much sales of the taxed item are expected to
change. So, official revenue estimates for the hypothetical apple tax
change described above would reflect an estimate of the change in the
quantity of apples. For changes in the tax treatment of a particular
type of business investment, the revenue estimates reflect shifts
between that type of investment and other types.

Changes in the capital gains tax rate provide another example of
how behavioral changes play a prominent role in the scoring process.
Economic theory and statistical studies have established that capital
gains taxes deter realization of capital gains--the sale of assets
that have risen in value. A cut in the capital gains tax rate,
therefore, is likely to spur an increase in capital gains
realizations. Put simply, investors are likely to sell their assets
to take advantage of the lower tax rate on any gains they have already
accrued. This increase in realizations will mean that the capital gains
tax will be applied to a larger tax base, partially offsetting the
cut in the tax rate itself. Indeed, depending on the timing and
structure of the rate cut, it may actually raise revenue immediately
after enactment. JCT and Treasury estimates recognize these effects.

Economists' understanding of--and data on--human behavior is
incomplete. This makes it difficult to determine the exact magnitude
of behavioral responses to tax changes and their exact impact on tax
revenues. Nevertheless, a revenue estimate that ignores such
behavioral responses will be inaccurate. By taking account of
microeconomic behavioral responses, the JCT and Treasury produce
estimates that are likely to be more accurate than strictly static
estimates.


Macroeconomic Behavioral Responses
to Policy Changes

Despite advances in making revenue estimates more dynamic, the
incorporation of behavioral responses has been subject to one
fundamental limitation. The official revenue estimates assume
that macroeconomic aggregates, such as total investment, total labor
supply, and GDP, are not affected by tax and spending changes. Because
the estimates ignore these potentially important effects, they are not
fully dynamic.

Lowering taxes on labor and capital income strengthens incentives
to work and invest and is likely to spur increases in these activities.
Additional work and investment boosts national income, which increases
the tax base and thus partially offsets the revenue loss from lower
tax rates.

As an example, suppose that the current income tax rate is 25
percent and that total national income is $1,000. Total tax revenue
is $250. Now, suppose the tax rate is cut to 20 percent. If total
income did not change, total tax revenue would be $200. The lower
tax rate, however, is likely to encourage work and saving, boosting
total income. If income rises to $1,100, total tax revenue will be
$220, not $200. Thus, the tax cut lowers total tax revenue by $30,
not $50. In other words, 40 percent of the tax cut ($20 of the $50
static revenue loss) ``pays for itself.''

Popular attention is often focused on the possibility that an
income tax rate cut could stimulate so much additional income that
it would fully pay for itself. Most economists believe that, starting
from current U.S. tax rates, such an outcome is unlikely for a
broad-based income tax change. It is important to realize, however,
that any behavioral response alters the size of the revenue loss from
a tax cut, even if it does not transform the loss into a revenue gain.

Official scoring of income tax rate changes already includes a
number of microeconomic behavioral responses. The scoring takes into
account a variety of ways in which the rate cut may raise taxable
income, such as a shift from tax-exempt fringe benefits to taxable
wages. However, the estimates do not recognize that lower income
taxes can encourage greater labor supply and capital accumulation,
and thereby raise total income in the economy.

The exclusion of macroeconomic behavioral responses from official
revenue estimation is not due to ignorance of these responses or
disagreement about their existence. Instead, it reflects the
judgment that accurately including these effects is impractical,
due to the controversy about their magnitudes and the complexity
of modeling them. Uncertainty about the correct model of the economy
and the size of behavioral responses to tax changes, and disagreement
about the appropriate time frame for revenue projections has made
consensus difficult to achieve.

Ultimately, it may be possible for macroeconomic effects to become
part of the official scoring process for those tax and spending
proposals that are likely to have significant macroeconomic effects.
However, given the time and resource constraints facing revenue
estimators and the lack of consensus about these issues, that goal
is not likely to be feasible in the near future. To promote informed
policy making, though, it is essential that fully dynamic revenue
estimates (incorporating macroeconomic as well as microeconomic
effects) be presented as supplementary information for major tax and
spending proposals.

Recently, estimators have taken major steps in precisely this
direction. In November 1997, the JCT compiled and published estimates
of the macroeconomic effects of fundamental tax reform prepared by
nine sets of economists using nine different economic models. The
JCT subsequently began developing its own macroeconomic models and
formed a blue-ribbon panel of academic and private-sector economists
to further explore dynamic revenue estimation. In January 2003, the
House of Representatives adopted Rule XIII.3(h)(2), which requires
the JCT to prepare analyses of the macroeconomic effects of major tax
bills before such bills can be considered by the House. In May 2003,
the JCT prepared such an analysis of the Ways and Means Committee's
version of the Jobs and Growth tax bill. In December 2003, the JCT
published a description of the methodology it used for this analysis.
The Congressional Budget Office (CBO) provided a similar analysis of
the President's 2004 Budget in March 2003 and provided a more
detailed description of its analysis in a July 2003 technical
document. Private organizations have also prepared dynamic analyses
of proposed tax changes that reflect macroeconomic behavioral
responses

User's Guide to Dynamic Revenue
and Budget Estimation

Recent work suggests six guidelines for dynamic revenue and budget
estimation.

Guideline 1: Dynamic Estimation Should Distinguish
Aggregate Demand Effects and Aggregate Supply Effects

Tax cuts can affect output through two different channels, by
changing aggregate demand and by changing aggregate supply. Any
aggregate demand effects are likely to be concentrated in the first
few years. Aggregate supply effects are likely to occur over a longer
time period.

Changes in Aggregate Demand

In the short run, tax cuts may push an underperforming economy back
toward its potential by raising consumers' disposable incomes and,
thus, their demand for goods and services. Tax cuts may also increase
firms' demand for investment goods. These effects increase the
aggregate demand for goods and services.

The extent of the increase in aggregate demand depends upon how the
tax cut is financed. If the tax cut is accompanied by reductions in
government spending, little or no stimulus to aggregate demand is
likely to occur. If the tax cut is financed by borrowing, then
aggregate demand is more likely to be stimulated.

The net effect of a tax cut on aggregate demand also depends upon
the reaction of the Federal Reserve. If taxes are cut in an
under-performing economy, the Federal Reserve may perceive less need
for interest-rate reductions. In such a case, the boost to aggregate
demand from a tax cut would, at least in part, be offset by the
reduced stimulus provided by the Federal Reserve.

The Federal Reserve is less likely to offset the aggregate demand
stimulus from tax cuts, however, in a low-interest-rate economy,
because interest rates cannot go below zero. Under these
circumstances, the fiscal stimulus provided by a tax cut may
reinforce, rather than replace, monetary stimulus. This case
seems relevant for the 2003 tax cut; the Federal Reserve's target
for the Federal funds interest rate was 1.25 percent from November
6, 2002, to June 25, 2003, and has since remained at 1 percent.

Aggregate demand effects are primarily relevant in the short
run. These effects tend to fade over time as prices and wage rates
adjust and the economy returns to its normal level of output. The
bulk of the aggregate demand stimulus from a policy change is likely
to be felt within a few years.

Changes in Aggregate Supply

Tax cuts can raise after-tax returns to work and investment,
encouraging both activities. This effect increases aggregate supply
because it increases the amount of goods and services that the
economy is capable of producing.

Tax changes can also improve the long-run allocation of resources,
allowing greater output to be produced with a given set of resources.
One way to do this is to make tax rates more uniform across different
types of income, as exemplified by the reduction in dividend and
capital gains tax rates adopted in the Jobs and Growth Tax Relief
Reconciliation Act of 2003. This provision reduced the tax burden on
investment in the corporate sector, which was more heavily taxed than
investment in the noncorporate sector. Over time, this tax reduction is
expected to make the allocation of resources more efficient, leading the
economy to allocate more resources to the corporate sector.

Because some commercially available forecasting models tend to
emphasize short-run aggregate demand effects, it may be necessary to
develop models that place greater emphasis on long-run aggregate supply
effects. In their recent dynamic analyses, the CBO and the JCT used a
mix of models with varying emphases on short-run and long-run effects.
The time frame over which tax revenues are estimated should be long
enough to fully capture the longer-run, supply-side effects, which
leads us to the second guideline.

Guideline 2: Dynamic Estimation Should Include
Long-Run Effects

While official revenue scoring is confined to a 10-year ``window,''
it is important that dynamic revenue estimation provide some
information for a longer horizon. Presenting the dynamic revenue
estimates as supplementary information, rather than as part of the
official revenue estimate, facilitates the use of a longer horizon.

The longer horizon is necessary because exclusive use of the
10-year horizon skews the emphasis given to different macroeconomic
effects. As discussed in guideline 1, aggregate demand effects are
likely to be fully realized within the 10-year window but have little
long-run importance. In contrast, aggregate supply effects may not
fully materialize within the 10-year window: Although changes in
labor supply may occur relatively quickly, changes in the capital
stock occur more slowly.

Economic analysis indicates that, in a closed economy, such
capital accumulation takes place over a period of decades. Consider
an example in which the government cuts taxes slightly on capital
income, starting from a 25 percent marginal tax rate. If standard
parameter values are assumed for a leading model of economic growth
(the Ramsey growth model used in Chapter 4 and described in the
Appendix to this chapter), only 42 percent of the long-run increase
in the capital stock is put in place within the first 10 years. That
is, more than half of the increased capital stock accumulates outside
the conventional 10-year window. In fact, only two-thirds of the
increase takes place within 20 years. In an open economy,
international capital flows may allow the capital stock to adjust
somewhat more quickly. Still, this analysis indicates the importance
of considering a long time horizon when estimating a tax cut's effect
on aggregate supply.

A longer time horizon would give adequate emphasis to the tax
cut's aggregate supply effects. It would also permit policy makers
to accurately compare the fundamental consequences of different types
of tax and spending changes. This leads to the third and fourth
guidelines.

Guideline 3: Dynamic Estimation Should Be Applied
to Spending Changes as well as Tax Changes

The logic behind dynamic estimation applies to spending as well as
tax changes. As discussed more fully in guideline 4, spending programs
can also affect aggregate demand and aggregate supply. To be sure,
dynamic budget estimation for spending changes can be even more
difficult, and has been less common, than dynamic revenue estimation
for tax changes. Nevertheless, including macroeconomic effects only
for tax changes would lead to an unbalanced and misleading comparison
of policies.

Guideline 4: Dynamic Estimation Should Reflect the
Differing Macroeconomic Effects of Various Tax and
Spending Changes

Not all types of taxes or spending programs would be expected to
have the same effects on the economy. Moreover, the policies that may
have the most beneficial effects in the short run (because they
provide a powerful boost to aggregate demand) can, in some cases, be
the least beneficial, or even harmful, in the long run (because they
fail to boost aggregate supply by increasing work and investment).

In the short run, the most immediate stimulus may be provided by
an increase in government purchases of goods and services, an
increase in transfer payments, or by tax cuts designed to boost
consumer spending. These policies, however, are generally not the
best ways to boost aggregate supply. Although some spending programs,
such as infrastructure construction and education, may increase
economic growth, others, particularly some transfer payments, would
be expected to reduce aggregate supply by weakening incentives to
work and invest.

In the long run, the strongest boost to aggregate supply is likely
to come from tax cuts designed to boost investment. Tax cuts on
capital income are most likely to have this effect. Their short-run
revenue feedback may be small (because their aggregate demand effects
may be limited), but their long-run revenue feedback may be large.

Consider again the example discussed above, in which the government
cuts capital taxes slightly, starting from a 25 percent marginal tax
rate. Using the same assumptions as before (as detailed in the
Appendix to this chapter), increased growth resulting from the tax
cut significantly moderates its revenue loss. This is particularly
true in the very long run (after 50 years or so), as the economy
settles back toward equilibrium. At that point, the reduction in
tax revenues is about half of what conventional scoring would
indicate. The estimated revenue feedback is so large for two
reasons: the policy being considered is a reduction in the capital
tax rate and the estimate refers to the very long run.

It is also possible for some tax cuts to reduce the incentive to
work and save. In this case, the revenue loss from the tax cut is
larger than it would have been without macroeconomic behavioral
changes. A prime example is an increase in a tax credit or deduction
that is phased out as income rises. Such a phase-out is another form
of a higher marginal tax rate on income. For example, married couples
filing jointly lose $5 of tax credits per child for each $100 of
additional income above $110,000. For a couple with two children,
this phase-out increases their effective marginal income tax rate by
10 percentage points. Given the existence of the phase-out, any
increase in the size of the credit lengthens the interval of income
to which this higher marginal tax rate applies. As with any increase
in marginal income tax rates, parents in this income bracket have
less incentive to work, save, or otherwise increase their income.
This outcome does not mean that increasing the child credit is a bad
idea. The President proposed such increases in 2001 and 2003 and
advocates making the increases permanent. As long as the income
phase-out remains in place, however, revenue estimates for increases
in the credit should include the revenue lost because of the reduction
in the parents' work and saving.

Dynamic estimation is not accurate unless it includes all of the
policy changes that are required to support the tax change. This leads
to our fifth guideline.

Guideline 5: Dynamic Estimation Should Account For
the Need to Finance Policy Changes

Because the government is a going concern, it need never actually
pay off its outstanding debt. Nevertheless, government debt cannot
indefinitely grow faster than national income. One implication of
this constraint is that, over the entire time frame of the economy's
existence, the present value of tax revenue must equal the present
value of noninterest government spending. (Present value takes into
account the time value of money--the fact that monetary sums can earn
interest over time.) In the long run, there can be no ``unfunded tax
cuts'' or ``unfunded spending increases.'' If current tax revenue is
reduced while current spending is left unchanged or if current
spending is increased while current revenue is left unchanged,
government debt increases. Servicing this debt requires that future
taxes be higher, future spending be lower, or both.

To be sure, it is infeasible for estimators to accurately predict
which adjustments future Congresses and Presidents may adopt.
Nevertheless, to avoid analyzing economically impossible policy
specifications, dynamic revenue estimates should recognize that some
such adjustments must occur. To ensure comparability across proposals,
it may be best to adopt a few stylized assumptions about the nature
of the financing. A few benchmark cases could then be considered;
perhaps one in which the debt service is financed with reductions in
government purchases, one in which it is financed with reductions
in transfer payments, and one in which it is financed with higher
income tax rates.

How a tax cut is financed can alter its effects in the short run
and the long run. If current revenues are reduced through a tax cut
while current spending is held fixed, the government's budget deficit
will get larger. Such a deficit-financed tax cut has a strong positive
effect on aggregate demand because consumers are likely to spend part
of the tax cut and there is no offsetting reduction in government
spending. It may do somewhat less to boost aggregate supply, however,
if the deficit raises interest rates and, as a result, lowers
investment. This effect is often called crowding-out, because a
government's deficit spending reduces, or crowds-out, the amount of
savings available for private firms to use for funding investment. On
the other hand, if current spending is reduced along with current
revenue, the aggregate demand effects of the tax cut are muted,
because the spending cuts lower aggregate demand. The boost to
aggregate supply is greater, however, because no crowding-out occurs.
To maximize the aggregate supply impact of the recent tax cuts, the
President has stressed the need to restrain government spending.

Guideline 6: Dynamic Revenue Estimation Should
Use a Variety of Models Until Greater Consensus Develops

One challenge facing estimators is that different models yield
different results. Comparing the results from different models is the
best way to resolve differences between possible approaches and to
test the sensitivity of results to changes in assumptions. To improve
our ability to distinguish among models, a set of models could be
applied to clearly defined and relatively simple hypothetical
policies. This would allow the different models' results to be
compared and would make it easier to attribute any variation among
their results to differences in their assumptions. As mentioned above,
the JCT did such an exercise in 1997 when exploring the possible
effects of fundamental tax reform. The CBO and the JCT also used a
variety of models in their dynamic analyses in 2003. Presenting
dynamic revenue estimates as supplementary information rather than
as part of the official revenue estimates facilitates the use of a
variety of models.

One reason that dynamic revenue estimation is subject to so much
uncertainty is that fiscal changes may have important effects that
are left out of standard models of economic growth. For example,
standard growth models take the rate of technological progress as
given. Some research, however, has suggested that technological
progress may be a by-product of capital accumulation; if so, changes
in capital income taxes can alter the rate of technological progress.
As another example, standard models take the economy's equilibrium
level of unemployment as given. Yet some research has indicated that
the equilibrium unemployment rate depends on productivity growth,
which can also be influenced by changes in capital taxation.
Incorporating such nonstandard effects into dynamic revenue
estimation is undoubtedly a formidable challenge, but if initial
results on these effects are confirmed by future research, this
challenge should not be avoided.

Conclusion

Fully dynamic revenue estimation that incorporates macroeconomic
behavioral changes is an important step forward in applying economic
insights to policy analysis. Significant progress has been made on
this front; continued progress is essential to sound policy making.

Appendix: The Model Used in the Capital-Tax Example

The model underlying the capital-tax example is the growth model
developed by Frank Ramsey in 1928. It is a leading textbook model, and
most of its assumptions are standard among models of economic growth.
For instance, output is produced by combining capital and labor,
and productivity growth increases how much output a given amount of
capital and labor can produce. Consumers maximize their welfare by
deciding how much of their income to save. Businesses maximize profit
and compete when hiring workers and selling products. Over the long
term, the saving rate determines the capital stock and, thus, the
level of output in the economy. The Ramsey model allows consumers
to choose their saving rate, while simpler models impose a constant
saving rate estimated from historical data.

Unlike some other models, the Ramsey model assumes that consumers
are members of families comprised of an infinite number of generations
and that they care about the well-being of their descendants. This
means that consumers consider the effects of their choices on their
children and subsequent generations. Some critics of the Ramsey model
view this assumption as unreasonable. However, the results presented
in the text do not change substantially if we assume that people care
less about each successive generation and, for generations far enough
into the future, hardly consider their welfare at all.

In the Ramsey model, the long-run equilibrium for the economy can
be described by two relationships. Firms invest in capital equipment
until the value of the output produced by the last unit of capital
equipment just equals the interest rate--the cost borne by the firm to
invest. The interest rate is, in turn, determined by consumers'
choices about their consumption and savings. These choices depend on
the growth rate of technology, the discount rate (a measure of how
much consumers prefer having a dollar today compared to a dollar in
one year), and consumers' flexibility with regard to spending in
different time periods. To solve the model, we must make an
assumption about how the government finances policy changes in the
long run. In the capital-tax example, we assume that the government
adjusts transfer payments accordingly.

Knowing this long-run equilibrium allows us to calculate the impact
of a cut in tax rates on tax revenue taking into account the
aggregate dynamic effects that this chapter has described. In
particular, we can summarize the difference between a dynamic
analysis and a static analysis with a few key parameters, or inputs,
to the model. We assume that the tax rates on labor and capital
income are each 25 percent, capital's share of total income is
one-third, and the elasticity of substitution between capital and
labor is one. Then, if dynamic effects are considered, a capital tax
cut reduces tax revenue in long-run equilibrium by half as much as
a static analysis would indicate.