Introduction: In this Point & Counterpoint, Thomas A. Barthold, Chief of Staff, Joint Committee on Taxation, Washington, DC; G. William Hoagland, Senior Vice President, Bipartisan Policy Center, Washington, DC; and Jane G. Gravelle, Senior Specialist in Economic Policy, Congressional Research Service, Washington, DC, reprise a panel discussion presented by the Teaching Taxation committee at the 2015 May Meeting. In the first point, Mr. Barthold discusses how the use of dynamic analysis permits the JCT to recognize that a tax policy change that is estimated to significantly change GNP will also affect the taxable base and thus tax revenues. In the second point, Mr. Hoagland explains how dynamic scoring adds additional information to the budget process and why it is here to stay. Dr. Gravelle provides the counterpoint by showing how the adoption of a single-point estimate for a dynamic scoring model that necessarily incorporates a variety of predictive assumptions provides a misleading picture of the likely responses to tax changes, especially when modeling assumptions are not transparent and inappropriate short-run and intergenerational effects are employed.

POINT: "The use of dynamic analysis permits the JCT to recognize that a tax policy change that significantly changes GNP will also affect the taxable base and thus tax revenues."

Macroeconomics and Revenue Estimating at the Joint Committee on Taxation

By Thomas A. Barthold, Chief of Staff, Joint Committee on Taxation, United States Congress, Washington, DC

The Joint Committee on Taxation (JCT) staff is nonpartisan and serves the entire Congress. One of the staff's key responsibilities is to provide revenue estimates. These are estimates of the change in federal receipts that would result from proposed tax legislation. The objective is to produce accurate, consistent, and impartial revenue estimates that can be relied upon by members of Congress in making legislative decisions.

I. What Is a Revenue Estimate

A revenue estimate is an estimate of the change in projected federal baseline receipts that would result from a change in law. The reference point for a JCT revenue estimate is the Congressional Budget Office (CBO) 10-year projection of federal receipts, referred to as the "receipts baseline." The receipts baseline serves as the benchmark for measuring the effects of proposed tax law changes. The baseline assumes that present law remains unchanged during the 10-year budget period. Thus, the receipts baseline is an estimate of the federal receipts that would be collected over the next 10 years in the absence of statutory changes. The JCT is required to estimate the revenue effects of proposals relative to the projected CBO receipts baseline.

A common misunderstanding that arises in reporting revenue estimates to policy makers is the distinction between a revenue estimate and receipts forecast. Generally, when the economy is growing, the CBO forecast of baseline receipts is growing. A negative revenue estimate of a tax proposal does not mean that the JCT is predicting receipts will fall. The negative revenue estimate means that receipts are predicted to grow more slowly if the proposal is enacted than they are projected to grow under present law in the baseline receipts forecast. Receipts would only decline if the revenue estimate predicted a loss in revenues that was greater than the underlying growth in baseline receipts.

II. Conventional Analysis

The JCT staff develops economic models to simulate future taxpayer behavior under the present law baseline and under the proposal to provide the Congress with estimated revenue effects. Each economic model is a theoretical construct representing economic processes by observable variables (using underlying economic data from a variety of sources) and a set of logical and quantitative relationships between the variables. There can be many different ways to model an economic problem. Models will differ in what features the modeler considers important and what features the modeler considers to be able to be ignored to make solution of the model practicable.

The JCT uses many different models.

• An individual tax model to forecast the effects on revenues from proposed changes in the individual income tax and from employment taxes. This model is based on data drawn from over 350,000 individual income tax returns supplemented with associated information returns and other data.
• A corporate model to forecast revenues from the corporate income tax. This model is based on ten years of data from every large domestic corporation and a random sample of smaller corporations.
• An estate and gift tax model to forecast the effects of proposed changes to estate and gift taxes. This model is based on estate and gift tax returns filed for a recent year and matched to income tax returns of the decedent.
• Separate models for each excise tax (as well as many other models for specific smaller taxes, credits, or exclusions).

 These models are all microeconomic models. That is, they analyze behavior at the level of the individual taxpayer or individual business unit. They take as given the baseline projections of such macroeconomic variables as aggregate labor supply, aggregate investment, inflation, and the like. Press reports of JCT estimates of tax proposals are most commonly based on these microeconomic models. The JCT refers to this as "conventional analysis."

A frequently expressed misconception about the JCT's conventional revenue estimates is the notion that they assume taxpayers will not change their behavior in any way in response to tax policy changes. One of the conventions that both the JCT and CBO use for revenue and expenditure estimates is that they are done against a fixed forecast of aggregate economic activity. The JCT generally assumes that a proposal will not change total aggregate production and therefore holds forecasted Gross National Product (GNP) fixed. However, the JCT's conventional revenue estimates are not "static" revenue estimates:  they always take into account many likely behavioral responses by taxpayers to proposed changes in tax law. Such behavioral effects include shifts in the timing of transactions and income recognition, shifts between business sectors and entity form, shifts in portfolio holdings, shifts in consumption, and tax planning and avoidance strategies, as follows:

• Changes in the time of transactions and income recognition:
  • Realization of capital gains in response to changes in the tax rates applicable to capital gains.
  • Issuance of corporate dividends in response to changes in dividend tax rates.
  • Changes in the composition of a taxpayer's total compensation between taxable income (e.g., wages) and non-taxed compensation (e.g., health benefits).
  • Acceleration of bonuses in anticipation of an individual income tax rate increase.
• Changes between business sectors or the legal form of doing business:
  • Organizing as a partnership in response to rising corporate rates or falling individual tax rates.
  • Shift in investment from more heavily taxed sectors to more lightly taxed sectors.
• Changes in types of portfolio investments:
  • Shifts from bonds to stocks or vice versa in response to tax rate changes on interest, dividends, or capital gains.
  • Shifts from taxable to tax-favored savings investments.
• Changes in the amount, types, and timing of consumption:
  • Reduced consumption of items that experience an excise tax increase.
  • Increased consumption of goods that are tax-favored, such as employer-sponsored health insurance and mortgage indebtedness.
• Tax planning and tax avoidance strategies:
  • Use of foreign tax credits and income allocation rules.
  • Reliance on performance-based compensation in response to a corporate deduction limitation (sec. 162(m)).
  • Structuring of compensation to obtain capital gains rather than income taxed at ordinary rates.

To summarize the JCT's conventional revenue estimating methodology: the JCT provides estimates relative to baseline receipts projected for future years under present law, not relative to receipts in years prior to the enactment of the proposal; the JCT generally assumes a fixed GNP forecast; and the JCT incorporates many types of microeconomic behavioral responses in its revenue estimates. ¹

III. Macroeconomic Analysis

Beyond raising funds for the federal government, members of Congress often intend for proposed tax policy changes to alter microeconomic behavior or to alter the future growth prospects of the economy. Conventional analysis generally addresses only microeconomic behavior: it does not account for possible changes in the underlying CBO macroeconomic projections.

Beginning in 2003, House Rule XIII(3)(h)(2) has required the JCT to provide a macroeconomic impact analysis of all tax legislation reported by the Ways and Means Committee. In early 2015 the House of Representatives modified its rules to require a single point estimate for the 10-year budget period of the deficit effect due to the macroeconomic response to certain "major" legislation. The new rule also requires qualitative analysis for 20 years after the budget period. With the adoption of the Concurrent Resolution on the Budget for Fiscal Year 2016, the Senate has adopted a comparable rule.

Over the past decade, the JCT has used several different models to simulate the macroeconomic effects of changes to tax policy to account for the sensitivity of the analysis to different modeling assumptions and frameworks. The JCT has used three different general equilibrium models:  the macroeconomic equilibrium growth model (MEG), an overlapping generations lifecycle model (OLG), ² and a dynamic, stochastic general equilibrium growth model with infinitely lived agents (DSGE). ³

These three models share similar neoclassical economic foundations. In MEG and DSGE, consumption is modeled according to stylized life-cycle consumption patterns. That is, it is assumed that people adjust savings and labor such that their consumption fluctuates less over time than income might. In OLG, the life-cycle consumption pattern is more explicitly modeled with separate decisions by 55 different cohorts. Labor-supply decisions change with changes in the marginal and average changes in after-tax wages. Saving and consumption respond to after-tax returns to saving and after-tax income. Business investment responds to expected return to investment and to after-tax cost of capital, which in part depends on availability of savings. While sharing these common features, the three models differ in the types of simplifying assumptions they make in some dimensions in order to more carefully examine economic outcomes in other dimensions.

A. Macroeconomic Equilibrium Growth Model (MEG)

In the MEG model the availability of labor and capital determines total national output in the long run. Prices adjust so that demand equals supply in the long run. In the short run, resources may be temporarily under-employed or over-employed as people and businesses adjust to outside changes in the economy.

In the MEG model, labor-supply responses to changes in after-tax wages (elasticities) are separately modeled for four different groups:

• High-income primary earners;
• High-income secondary earners;
• Low-income primary earners; and
• Low-income secondary earners.

Household saving and consumption are assumed to respond to the after-tax return to saving and after-tax income. Business production and housing production are modeled separately, and investment responds to changes in the user cost of capital.

MEG is an open economy model. Accordingly, cross-border capital flows and changes in net exports affect domestic economy outcomes.

Individuals in MEG are myopic. They do not anticipate changes in the economy or government policy.

B. Overlapping Generations Lifecycle Model (OLG)

Unlike the MEG model, the OLG model assumes that prices adjust to any changes in economic conditions (such as a change in fiscal policy) so that supply equals demand in every period and resources are always fully utilized after accounting for the cost of adjusting the capital stock. The model does not allow for unemployment. There is explicit modeling of international trade in goods and services, allowing for investments in tangible and intangible capital both in the United States and abroad.

In the JCT's OLG model, economic decisions are modeled separately for each of 55 adult-age cohorts. It is assumed that each cohort makes decisions about consumption and savings, including the amount of leisure to consume, based on changes in after-tax wages as well as the after-tax return to savings.

The OLG model has separate production sectors for business and housing. Investment in the business sector responds to the expected future value of the firm as determined by the after-tax return to capital. Investment in the housing sector responds to the expected after-tax return to housing investment as well as to the consumption value of housing.

International capital flows are modeled through a separate "multinational" sector that models flows of investment and output between domestic firms and their foreign subsidiaries, as well as flows between foreign firms and their U.S. subsidiaries.

OLG is a perfect foresight model, in that individuals in each period are presumed to have information about future fiscal conditions and their decision-making is affected by those conditions.

C. Dynamic, Stochastic General Equilibrium Model (DSGE)

Unlike the MEG and OLG models, a significant feature of the DSGE model is that it attempts to account for uncertainty in economic decision making. Taxpayers examine all possible states of the future economy. For example, an increase in volatility of future asset returns is assumed to change the investment decisions of taxpayers in the DSGE model. When policy variables are given stochastic or random components, the DSGE model provides the JCT staff with implications that the MEG and OLG models will not.

In the DSGE model, supply equals demand in the short run and in the long run. Consequently, there is always full employment. However, the model includes "sticky" prices and adjustment costs that cause output to be more sensitive to demand.

DSGE does not model international capital flows. It is assumed that all economic activity is contained within the U.S. economy.

Economic decisions are modeled separately for savers and non-savers. Non-savers do not own capital assets and have no access to the credit markets. Non-savers are generally lower income taxpayers. Non-savers may respond to tax policy changes differently than savers.

Decision-makers in DSGE make decisions based on rational expectations of future fiscal conditions. They can observe an array of possible future conditions; but because the model includes uncertainty, they do not have perfect information about the future path of the economy.

IV. Conclusion

The JCT uses its detailed microsimulation models, starting from the conventional estimates of a proposed tax change, to calculate changes in after-tax wages, after-tax rates of return to saving, and the user cost of capital. These calculated changes become inputs to the macroeconomic models to determine the possible effects that a proposed change in tax law may have on macroeconomic outcomes. Generally, if a tax policy significantly changes GNP, that change would affect the taxable base and thus tax revenues. Taking these effects into account is what many commentators refer to as "dynamic analysis."

Under the Concurrent Fiscal Year 2016 Budget Resolution, the JCT reports the estimated effects of the proposed legislation and then reports an additional year-by-year projection of the effects on Federal receipts from the projected increase or decrease in U.S. GNP that is projected by the JCT's macroeconomic analysis. ⁴

The JCT continuously works to make sure that our models reflect the most recent data and economic research. For example, over the past few years the JCT has added a more detailed international trade sector to the OLG model, distinguishing the location of tangible capital investments from the location of investments in intangible assets. The JCT is exploring adding another business investment sector to the MEG model and further refining labor-supply responses for low-income, middle-income, and high-income primary and secondary earners.

POINT: "Dynamic Scoring adds additional information to the budget process and is here to stay."

Dynamic Scoring: Is It Much Ado About Something?

By G. William Hoagland, Senior Vice President, Bipartisan Policy Center, and Former Staff Director, U.S. Senate Budget Committee, Washington, DC

I. Introduction

Reforms to the federal budget process were advocated almost immediately following the enactment of the historic Congressional Budget and Impoundment Control Act (CBA) of 1974. Both significant reforms and lesser ones have been adopted over the 41-year history of the Act. Proposed reforms historically can be classified into two categories: (1) those designed to force a legislated fiscal outcome, and (2) those designed to improve the accuracy and credibility of figures used in the process.

The first category is reflected in legislation such as the Balanced Budget and Emergency Deficit Control Act of 1985 (also known as the Gramm-Rudman-Hollings Act) and its subsequent legislation, the Budget Enforcement Act of 1990 and the Budget Control Act of 2011. These laws set in place limits or targets on annual federal spending and deficits, subject to automatic procedures (e.g. pay-as-you-go or pay-go, and sequesters) to achieve targets should enacted policies fall short.

The second category is best exemplified by the Federal Credit Reform Act of 1990. It was designed to better reflect the impact of long-term credit programs on the federal budget versus—up to that time—cash-based accounting of credit programs.

Advocated as a method to improve the accuracy in scoring federal legislation, "dynamic scoring" was advanced by Republican-controlled congresses beginning in 1995. The drive to modify the long-standing conventional scoring that assumed that overall output remained unaffected by the underlying legislative changes grew over the next two decades. The persistent push to "reform" culminated this year in two ways. First, a new House procedural rule (H. Res. 5) was adopted. Second, that House rule was subsequently carried into the Senate by the adoption of the FY 2016 Concurrent Budget Resolution (S. Con. Res. 11) that effectively applies the House Rule to Senate procedures. Both the House Rule and the Budget Resolution require the CBO and the JCT to incorporate the budgetary effects of changes in macroeconomic variables in major legislation considered in the 114th Congress. (See Box 1, following.)

Title III – Budget Enforcement
Section 3112. Honest Accounting: Cost Estimates for Major Legislation to Incorporate Macroeconomic Effects

(a) CBO and JCT Estimates – During the 114th Congress… estimates by the CBP under section 402…..or by the JCT to the CBO under section 201(f) for any major legislation…shall, to the greatest extent practicable, incorporate the budgetary effects of changes in economic output, employment, capital stock, and other macroeconomic variables…

(b)(1)… a qualitative assessment…In the 20-year fiscal year…
(b)(2)…identification of critical assumptions and source of data underlying estimate.
(c)(1)(A) in the Senate -- "major legislation" means … (before incorporating macroeconomic effects and not including timing shifts) in a fiscal year… (i)(I) 0.25 percent of VDP, (II) greater than $15,000,000,000 for a treaty, (ii) designated by (I) the Chairman of the Senate or (II) Chairman or Vice Chairman JCT;
(c)(1)(B) in the House (i)…..0.25 of GDP or designated…

Report Language: "In the Senate, these estimates would be provided for informational purposes only. In the House, the Chair …shall exercise the authority under (c)(1)(B)(ii)."

The nearly two-decade trek for advocates of dynamic modeling reflects the classic work of John D. Kingdon, a political scientist who demonstrated how advocates of reform adopt tactics to succeed over the long haul. ¹ Kingdon argued:

[Advocates for change prepare for better times by] softening up – repeatedly calling for adoption of the preferred policy. They attempt to connect their preferred policy to the series of policy problems that seriously concern decision-makers. They explain the preferred policy's mechanism for reducing those problems, supporting their claims of expert analyses. They modify their preferred policy after listening to the concerns and objections voiced by those with the power to reject their proposals. They do all this not in expectation of immediate success, but in hopes of being ready when political conditions are more favorable. At that time, their proposal might attain the heady status of 'an idea whose time has come.' ²

II. Does Dynamic Scoring Matter?

For macroeconomic scoring, aka dynamic scoring, its time has come. What impact will it have on policy makers' decision processes?  Will it result in better information and therefore better legislation?  How will it impact the consideration of legislation in the Congress?  What will dynamic scoring really mean in practice?  As the non-partisan referees of budget analysis, CBO and JCT's scores carry substantial influence. Their cost findings can, and often do, make or break a piece of legislation. History is littered with legislation that might have passed if it had not scored poorly as increasing federal debt and deficits.

Some analysts contend that dynamic scoring could significantly alter the projected cost of legislation, thereby giving a potential boost to some proposals that look more attractive when scored dynamically, while hurting the chances of others that may show larger deficit increases (or less deficit reduction). In reality, however, this scoring change will only be applied to a limited number of "major" pieces of legislation. Major is defined as any legislation having a conventional score resulting in a change greater than 0.25% of GDP in any fiscal year (approximately $45 billion) or any direct spending (non-appropriations) legislation designated by the Chairmen of the two Budget Committees as major. ³ That said, those would also be some of the most consequential—and potentially controversial—bills that are considered.

Because discretionary appropriations are controlled by aggregate spending limits currently set in law, the dynamic scoring analysis would not apply unless those limits are increased and the appropriations are clearly allocated for a specific purpose (e.g., infrastructure, education, national security) that triggers the rule's 0.25%-of-GDP metric (e.g. $45 billion).

III. Experience to Date in the 114th Congress

Since the adoption of H. Res. 5 and S. Con. Res. 11, only one legislative item has been subject to the new scoring procedure—the Tax Relief Extension Act of 2015, reported ordered from the Senate Finance Committee on July 21, 2015. ⁴   This legislation would extend tax provisions that expired in 2014 through December 31, 2016. Since extending the expired provisions would lower current law revenues by over $154 billion in 2016, this extender counts as "major" legislation. Three specific extenders—bonus depreciation, expensing of real property, and modifications of the research credit—account for the majority of the assumed revenue that would be lost under the bill. Over the ten-year scoring period, the extension of the expiring tax provisions was estimated to have a direct impact on increasing the federal deficit by $97 billion, but including the macroeconomic feedback effects (primarily, an assumption of an increase in capital stock) resulted in a $10 billion offset to the direct deficit estimated. Therefore, JCT estimated that the enactment of the extenders bill would increase the deficit by $87 billion over the next decade, a 10% reduction in the impact calculated under the conventional estimating procedure.

A report requested by the Chairman of the Senate Budget Committee to estimate the budgetary effects of repealing the Affordable Care Act (ACA) provides a second example of the new dynamic scoring procedures following their adoption this year. The requested report, prepared jointly by the CBO and JCT, incorporated macroeconomic feedback. ⁵   To date, however, no such legislation has been considered in the Congress. Over the ten-year scoring period, the direct impact of a total repeal of the ACA was estimated to increase the federal deficit by $353 billion, but including the macroeconomic feedback effects (primarily, an assumption that the repeal would boost the supply of labor), the increase in the deficit over the time period was estimated to be only $137 billion. This represents a 60% reduction in the expected deficit compared to the result under conventional methodology.

IV. Conclusions from Experience to Date

Both analyses provide the first indication of how the CBO and JCT will apply the new estimating methodology.

Conclusion #1. Not surprisingly, the first simple conclusion from these two examples is that the CBO and JCT staff are equipped and prepared to estimate major legislation for the Congress with macroeconomic feedback effects. The staffs are prepared to provide useful information on the economic impact of major legislation which has not generally been available to legislators when considering major bills. Further, this is early in the process, and the presentation of the actual cost estimates using the feedback methodology will evolve over time.

In the case of the tax extenders bill, there was explicit inclusion of estimates of increases in capital stock, with resulting increased revenues. As estimated by the JCT staff, the primary effect of the bill on the economy would be an initial increase in the stock of business capital by 0.3% in the first half of the 10-year estimating period, resulting in an increase in production, output, and therefore receipts of 0.1% over the decade.
For the repeal of the ACA, the impact the legislation would have on boosting the supply of labor was also made explicit and suggested an increase in GDP that would not have been identified under the conventional scoring procedures. CBO estimated that the law's repeal would raise economic output mainly by boosting the supply of labor, resulting in output increasing 0.7% over the next decade.

Conclusion #2. A more subtle but important conclusion from the two examples to date is that the estimates will be point or central estimates (not ranges) within a ten-year scoring window. As a result, the cost estimates with the macroeconomic impact will be applied to floor consideration of all major legislation both in the House and Senate, notwithstanding S. Con. Res. 11 report language explicitly stating that in the Senate the macroeconomic feedback effects would be for "informational purposes only."

While indications of the magnitude of uncertainty surrounding the central estimate have been and will be highlighted in the cost estimates, for purposes of considering legislation in the two chambers the central estimate will apply for purposes of determining whether certain Budget Act points-of-order will apply. The central estimate is the last estimate to appear in a CBO cost-estimate and presents the legislation's impact including the macroeconomic feedback.

This is particularly critical in the Senate where a 60-vote point-of-order (e.g. pay-go) would lie against any legislation increasing the federal deficit. In the House of Representatives such points-of-order can be waived with a simple majority vote. The Senate Parliamentarian does not pick and choose between two estimates, but will look to one score and absent any precedent to date, it is highly likely that the last estimate in a CBO score, the "total estimated changes, including macroeconomic feedback" will be the one estimate used to determine procedural points-of-order. The first test of the new procedure will be joined this fall with the tax extender bill once it is brought to the Senate chamber.

Conclusion #3. A final conclusion, at least when applied to the two policies analyzed (e.g. extending expiring tax provisions, repealing the ACA), simply is that the positive macroeconomic feedback effects did not offset the larger negative effects of increased federal deficits and debt.

Advocates of dynamic scoring who might have expected that growth changes resulting from the new methodology would offset completely the conventional estimated deficit of the proposed legislation might be disappointed. At least as it relates to the two substantive items estimated to date, extending expiring tax cuts would still increase deficit projections, albeit less so than under conventional methodology. Further, repealing the ACA did not translate into a reduction in the deficit, though it lessened the conventional estimate.

Enactment of either the tax extenders or repeal of the ACA would today require legislators either to waive the point-of-order that presents an increase in the deficit or to find ways to offset the resulting deficit from the economic feedback scoring. Interestingly, those offsets, if included in the legislation, could "de-trigger" the rule or, if considered sequentially to the legislation, might be found to have negative economic growth effects under the new methodology.

The two proposals analyzed to date under the new rule need not be expository as it relates to the federal deficit. The comprehensive immigration bill passed by the Senate in 2013 would have significantly increased the U.S. labor force. When the CBO applied dynamic scoring (prior to the adoption of this new procedure), it estimated a major reduction in the deficit of nearly $200 billion for the first decade after enactment.

V. Much Ado About Nothing?

No. Dynamic scoring is an important and valuable tool now added to the policy makers' tool box. Dynamic scoring is here to stay. As demonstrated by its application to date, it will provide new insights into the broader economic impact of critical legislation. Legislators will be provided additional data to evaluate the merits or demerits of legislation. How they choose to use that additional information will vary depending on the underlying policy proposal. But it will not necessarily make their decision-making process easier or less political.

Achieving the shared goal of increased national growth will continue to require trade-offs and compromises within the political process, regardless of what estimating methodology is used.

COUNTERPOINT: "Adoption of a single-point estimate for a dynamic scoring model that necessarily incorporates a variety of predictive assumptions provides a misleading picture of the likely responses to tax changes, especially when modeling assumptions are not transparent and inappropriate short-run and intergenerational effects are employed."

Issues in Dynamic Scoring

By Jane G. Gravelle, Senior Specialist in Economic Policy, Congressional Research Service, Washington, DC. The views in this article do not reflect the views of the Congressional Research Service.

I. Introduction

After many years of debate and a dozen years of providing advisory estimates on tax bills, dynamic scoring has been officially adopted in the U.S. Congress.

Conventional revenue and spending estimates are not static: they reflect numerous behavioral responses to changes. Conventional estimates, however, keep overall output (GDP) fixed, so they cannot consider changes due to increased labor and capital, and the taxes on those induced earnings. Dynamic scoring allows the incorporation of these effects into cost estimates.

The first drive towards dynamic scoring began in 1995 when Republicans took control of the Congress. One of the first actions of the new Congress was to hold joint hearings on dynamic scoring. ¹ Subsequently, the Joint Committee on Taxation (JCT) convened a group of modelers to study dynamic scoring. ² In 2003, House rules required macroeconomic analysis of tax proposals and both JCT and the Congressional Budget Office (CBO) produced the first of a series of dynamic analyses that year. ³  

House Resolution 5, adopted in January 2015, provided rules for the House of Representatives for the 114th Congress. It requires incorporation of macroeconomic effects in official scoring of spending and tax legislation (excluding appropriations bills) reported by a committee. The scoring rule applies to legislation with an annual budgetary effect of at least 0.25% of projected GDP or as requested by the Chair of the Budget Committee for spending measures or by the Chair/Vice Chair of the Joint Committee on Taxation (the Chair of the Ways and Means Committee) for revenue measures. This rule supplants the previous requirement for an advisory macroeconomic analysis of tax changes, which usually presented a range of effects. Under the new House rule, macroeconomic effects are to be incorporated into official scores that drive budget rules associated with the budget resolution.

These rules were also included in the Budget Resolution for FY2016, S. Con. Res. 11. In the Senate, however, the dynamic estimates remain advisory and do not constitute official estimates.

The CBO prepares cost estimates for spending, while the JCT prepares estimates for tax measures. The CBO and JCT will continue to be responsible for each type of estimate and its feedback effect (the change in revenue or spending due to changes in the size of the economy).

The CBO, working with the JCT, presented the first dynamic score under the new regime for the repeal of the Affordable Care Act. That estimate was in response to a request from the Chairman of the Senate Budget Committee and not associated with a formal score for a piece of legislation. ⁴ The JCT also presented a score for extending a number of temporary provisions that had expired. ⁵

II. Lack of Consensus

From 2003 to the present, advisory estimates reflected a wide range of effects, even varying in sign. Current rules require a point estimate, incorporated into official estimates in the House. If the first estimate produced (for the Affordable Care Act) is a guide, these new estimates will not quantify the assumptions made in arriving at estimates or providing sensitivity analysis, no longer providing information on the importance of alternative assumptions.

Four issues arise in estimating dynamic scores. The first issue is the types of effects to include and the time horizon to use. The second issue is the types of models to use. The third issue is the magnitude of behavioral responses built into the models. The fourth issue is the appropriate measure of the marginal tax rate.

A. Types of Effects

Three types of effects may influence the feedback effect: (1) the short-run demand-side stimulus, (2) the crowding-out effect where the increase or decrease in the deficit reduces or increases funds available for investment, and (3) supply-side effects, where labor supply and savings respond to changes in tax rates. Demand stimulus effects arise from employing unemployed resources. They are transitory and positive for a tax cut. Crowding out is the effect of increased deficits in reducing capital available for private investments. (Crowding in arises from surpluses). This effect—negative for a tax cut—is small initially but grows continually over time. Supply-side effects are the effects on the supply of work, savings, and investment from tax changes. They are generally positive for a tax cut and uncertain for tax reform. They are typically primarily due to labor supply in the budget horizon as capital takes some time to accumulate (unless investment flows from abroad).

A simulation of a $500 billion tax cut prepared by the JCT in 2005 ⁶ illustrates some of these short-term and long-term effects. In the case of an individual income tax cut incorporating all three effects, the revenue feedback effect (percentage reduction in cost due to macroeconomic effects) was 23%. If the short-run stimulus effect were eliminated, the feedback effect fell to 9.7%. If both stimulus and crowding-out effects were eliminated, the feedback effect was 13.5% (which is the supply-side effect). The stimulus and supply-side effects increased output, while the crowding-out effect reduced it. In the short run, with all effects, a 0.2% to 0.4% positive effect on output occurred in the budget horizon, but a 0.2% to 0.3% negative effect occurred in 30 years due to crowding out.

Even in the case of the estimates for former Ways and Means Chairman Dave Camp's Tax Reform Act of 2014 (H.R.1), which was roughly revenue neutral, demand-side effects were larger than supply-side effects. ⁷ In the in-house macroeconomic equilibrium growth (MEG) model that accommodated all of these effects, the supply-side effects in the budget horizon were 0.1% to 0.2% of GDP (for the lower and higher labor-supply elasticities ⁸ ) but the demand-side effect was 0.3% of GDP. ¹⁸  

Each effect has uncertainties. There is a case for eliminating short-run stimulus because these effects can be offset by actions of the Federal Reserve. Most of the economists testifying at the 1995 hearings either explicitly or implicitly indicated that demand-side effects should be excluded. The effects, at least in theory, also depend on how close the economy is to full employment. Moreover, estimates of the size of the stimulus (termed a "multiplier," for how much each dollar of spending or tax cut is translated into additional dollars of output) vary substantially: for an individual tax rate cut, the high estimate is six times the value of the low one. ¹⁰ Moreover, the effects also occur (and can be larger) for appropriated spending, which is exempted from dynamic scoring. Including these effects in the modeling of tax cuts while not including them for appropriations would therefore favor tax cuts over increases in appropriations. Finally, and perhaps most importantly, these effects are transitory and are inappropriate for assessing the effects of a permanent tax change.

The extent of crowding out of investment depends on the amount of borrowing from abroad. Crowding out also occurs with spending. Crowding out is a longer run phenomenon and may not have much effect in the budget horizon, but it may be crucial to evaluating the long-run effects of a permanent tax change. Appropriations changes also produce crowding-out effects.

Supply-side effects vary depending on the model type and the magnitude of responses built into the model. Labor-supply effects occur quickly, while capital-stock effects occur more slowly, especially if arising from savings. Increases in the capital stock due to attracting capital from abroad may occur more quickly than those arising from savings. Positive supply-side effects may also occur with some types of spending, including appropriated spending (e.g., spending on infrastructure may increase productivity).

B. Types of Models

JCT and CBO use two basic types of models:  a growth model (often called a Solow model) with labor supply and savings responses, and intertemporal models in which a representative agent optimizes choices for leisure and consumption over a lifetime (overlapping generations or OLG models) or over an infinite horizon. ¹¹  A Solow model is sometimes combined with a model that permits short-run stimulus effects, or a separate model may be used to measure those effects. Ideally, a model of international capital flows that takes into account feedback effects from the U.S. economy and the rest of the world would be incorporated to capture international capital flows. It appears, however, that the international model has not yet been fully developed.

For the estimates of the three effects, JCT uses the MEG model that combines demand-side effects with a Solow growth model. In the 2005 study cited earlier, JCT reported an 11.3% feedback effect with an alternative (lower) measure of labor-supply response. JCT also uses an OLG model that can only account for supply-side effects. The feedback effect for the OLG model was 18.6%—about a third higher than the supply-side effect in the MEG model.

The structure of the OLG model is more theoretical, and this sometimes produces results that are difficult to verify empirically. In the OLG model, agents choose leisure and consumption over time by maximizing a utility function; supply effects flow from this process.

The intertemporal model is favored by some economists because of its theoretical purity. It cannot, however, capture demand-side or crowding-out effects. Since the model is forward looking, it must initially be solved for an infinitely long steady state. A change in the deficit in either direction is not possible (because it would lead to an infinitely large positive or negative asset position for the government). As a result, a tax change that affects the deficit cannot be modeled in isolation. Some other policy must always be taken into consideration—a spending change, a transfer, or a future tax change accompanying the estimate that offsets any deficit or surplus effects. The choice matters: each of these will produce a different result.

The pristine structure of the intertemporal model, admired by some, is also a straitjacket. For example, intertemporal substitution elasticities (percentage change in the ratio of relative consumptions divided by the percentage change in price ¹² ) are the same whether periods are far apart or close together. There is no empirical evidence for the substitutability consumption between periods that are far apart, but that response greatly influences effects if the proposal being evaluated changes tax burdens on capital income.

Intertemporal models in use assume a representative agent (in the case of the OLG model, a single agent per generation). Thus, unlike the MEG model, they cannot distinguish between the suppliers of labor and capital (i.e., workers and investors). This limitation can be important when estimating the effect of tax cuts: along with an incentive to work more because the marginal wage is higher, higher incomes cause both leisure and consumption to increase (and labor to decrease). The effect on labor supply depends on those offsetting effects, and the income effect may not be appropriately captured in an intertemporal model.

Because of the structure of the utility function in intertemporal models, direct labor-supply effects are not observed; instead, they are a function of a combination of parameters selected by modelers. Modelers do not always choose parameters that generate results consistent with empirical evidence. The model structure also makes labor supply responsive to the interest rate, something one modeler called "shooting in the dark." ¹³

Open economy corporate models should respond to feedback effects from the rest of the world as well as from the U.S. economy. For example, if the corporate tax were lowered and international capital were mobile, some capital would flow into the United States from abroad and some would flow into the U.S. corporate sector from U.S. unincorporated sectors (including housing). These flows would likely reduce pretax returns in the U.S. corporate sector (where capital is now more abundant) but also have an effect in the U.S. noncorporate sector and in other countries. The flow of capital would eventually reach an equilibrium that reflects the mobility of international capital, the substitution between different products and the ease with which firms can substitute capital for labor. JCT and CBO do not have such a model. The Gravelle and Smetters model, ¹⁴ which does contain a rest-of-the-world sector along with multiple sectors abroad and in the United States, suggests that even a large change in the corporate tax rate has negligible effects on output because of these effects.

Even the Gravelle-Smetters model, which has an extensive structure, does not fully account for the effect of lowering the corporate rate. Lower rates cause debt finance to be less attractive and reduce the inflow of debt finance. Thus, a corporate rate reduction could possibly lead to an overall contraction in the U.S. capital stock. ¹⁵

Open economy issues were quite important in determining the effects of the Camp tax reform proposal, the Tax Reform Act of 2014. ¹⁶ The estimates showed a greater discrepancy between the supply side effects in the MEG and OLG models. Revenue feedback from supply side effects was $50 billion to $100 billion (increased GDP of 0.1% to 0.2%) over ten years in the MEG model, and almost $700 billion (increased GDP of 1.6%) in the OLG model. These larger differences may reflect, in part, an increase in the labor supply response in the OLG model (which may in turn be due to lack of an income effect and in part a new aspect of the OLG model that treats the shift in the ownership of intangible assets (e.g., patents and copyrights) due to changes in corporate taxes as having real effects on output. Since intangible assets can already be used costlessly in every location, this shift should not have an output effect. Similar issues were raised by William McBride of the Tax Foundation. ¹⁷

C. Behavioral Responses

Within any given model, the magnitude of the behavioral response drives the result. The range of demand-side multipliers was discussed previously. However, there is also uncertainty not only about the magnitude but also about the direction of labor-supply effects because of income and substitution effects. Most estimates are small, but significant differences are reported. For example, in the JCT simulations of the Camp proposal, labor supply increased by 0.4% to 0.5% in the MEG model compared to 1.3% in the intertemporal OLG model. This difference is in small part due to a higher labor substitution elasticity, but it is probably largely due to the lack of income effects in the OLG model due to closing the model with transfers and not separating suppliers of labor from suppliers of capital.

Most of these models use positive elasticities net of income and substitution effects so that labor supply grows with wages. However, it is hard to accept any positive labor-supply response since increases in real wages over a long period of time did not produce an increase in the work week.

In addition, empirical estimates of responses of many types, including those in intertemporal models that also capture substitution of consumption and leisure over time, may be overstated due to publication bias. When a statistical estimate finds a result contrary to theory, it is difficult to get that study published. Since there is a distribution across samples of effects, cutting them off at zero biases the body of estimates. For example, a study found that one type of elasticity—the short-term substitution between consumption in different periods—fell from 0.5 to between zero and 0.2 when a correction was made based on data that was sufficient to estimate the distribution including missing studies. ¹⁸

As noted above with the introduction of open economy elements in the OLG model, new and untested innovations in behavioral effects can be significant. The "firm specific capital" substitution in the JCT's OLG model for the Camp proposal was a large part of the difference between an output effect of 0.2% and one of 1.6%. This substitution effect was assigned a very high elasticity. If behavioral effects are incorporated in official scores, it is especially important to vet these sorts of innovations in modeling. This innovation is questionable not only with respect to its magnitude, but also regarding its existence. A change in the ownership of intellectual property should have no effect on real output in different locations. Lipitor™ is made from the same formula whether the patent is in Ireland or the United States.

D. Marginal Tax Rates

There is a need for more work on modeling marginal effective tax rates with tax reform. Base-broadening provisions that are part of tax reform can have marginal effects. Many of these associated with investment, such as slowing depreciation, have long been recognized, as have phase-outs in tax rates. Individual base broadening that increases the share of income being taxed (e.g. by eliminating itemized deductions) has not been incorporated, and these effects need to be captured. In general, because of these effects of base broadening, it is difficult to design a revenue-neutral, distributionally neutral tax reform that increases output. ¹⁹

III. Issues Going Forward

The Senate Budget Resolution states in section 3112(b)(2) that the macro estimates include "an identification of the critical assumptions and the source of data underlying that estimate." While the recent CBO/JCT analysis of the Affordable Care Act was not a formal score of a bill, it may be a guide as to what to expect from a dynamic score. Regrettably, it did not include critical assumptions. JCT estimated the effects of the non-insurance-related provisions, but it did include the effects of the Cadillac tax on high-cost insurance plans. CBO estimated the effects of insurance-related provisions. The study indicated that short-run demand effects were included and that Solow models would be used for supply-side effects. Although the study referred to previous papers, it did not report the precise multipliers or the precise labor-supply responses actually used. In particular, JCT has never reported its demand analysis in the form of multipliers. JCT uses two labor-supply elasticities (a high and low elasticity), but provides no indication which one is used in the study. The CBO labor-supply elasticity is twice the size of the high JCT labor-supply elasticity. Were different elasticities used for different parts of the analysis? It is not possible to tell from the information in the study. These examples demonstrate the importance of greater transparency as an issue going forward.

Another issue that arose in the analysis of the Affordable Care Act that needs to be considered in the future is its use of short-term stimulus effects, which are questionable. It did not, however, employ intertemporal models (although that may not be the case in the future).

Similar points can be made about the recent JCT score for extending expiring tax provisions. This score also included short-run effects. While in this case information was provided on labor-supply responses and other parameters, no sensitivity analysis or estimates of inputs were available to understand what was driving these responses. The estimate followed the Affordable Care Act precedent in not employing intertemporal models.

Aside from these issues of lack of transparency in methodology, improper use of short-run stimulus effects, and potential improper use of intertemporal models, there remain a number of issues of concern. Dynamic scoring cannot provide a reasonable basis for analysis without correct measurement of the marginal effective tax rates. To be useful in judging the impact of tax changes, dynamic scorers need to prepare longer run estimates. Single-point scoring is more misleading than informative, given the many assumptions that can have dramatic effects on results:  therefore, it is important that dynamic scoring provide ranges of effects with transparency about the various assumptions used. Finally, dynamic scoring of the economy cannot be adequately presented without construction of a full-scale corporate model with a rest-of-the-world sector. ■