Comments and Discussion.
KAREL MERTENS (1) The 2017 tax law is the most significant reform of individual and business income taxation in decades, and will be a topic of study for years to come. In their paper, Robert Barro and Jason Furman lay the groundwork with a quantitative analysis of the long-run impact on economic activity through the lens of neoclassical growth theory and a careful calibration of the user cost effects of the law. Their empirical exercise appears straightforward, but it quickly raises difficult questions with important quantitative implications: What is the appropriate discount rate for firms? Will the temporary provisions in the law be extended, and what are agents' expectations? How will the government balance the intertemporal budget, and will there be crowding out? To what extent will production shift from pass-through entities to corporations, and how does this affect productivity? After tackling all these issues, Barro and Furman calculate that GDP will rise by 0.4 percent after 10 years under the law as written, and by 1.2 percent if all the law's provisions are eventually made permanent.
The neoclassical modeling approach is tractable and transparent, and it may well capture most of the relevant tax incentive effects on capital formation. It remains at the core of many of the structural macroeconomic models used for evaluating tax policy reforms. Barro and Furman discuss the key sensitivities to modeling assumptions, and point to several features that are missing from the basic framework, such as the extensive margin effects of the reduction in the corporate tax rate due to the lumpy nature of many investment decisions, the possible effects on total factor productivity as a result of the changed incentives for investment in research and development, or the impact on foreign direct investment flows and international tax competition. Given these limitations, there clearly remains considerable uncertainty about the estimates of the long-term effects of the 2017 tax reform.
Although the focus is primarily on the long run, Barro and Furman also calculate the short-term effects of the business tax reform using estimated convergence rates toward the postreform steady state. A full analysis of the short-run dynamics would require additional detailed assumptions about expectations of future tax rates, the monetary policy response, and the like. Guided by the existing reduced-form evidence on the effects of changes in individual marginal rates, Barro and Furman conclude that the short-term effects of the corporate tax cut will be dwarfed by an increase of 0.9 percentage point in annual GDP growth for 2018-19. The personal tax cuts, which are currently set to expire in 2025, are more important than the corporate cuts for the overall debt impact of the law, and their impact may play a role in determining future political outcomes. In this sense, they may also matter in the longer run.
Barro and Furman rely on empirical studies by Barro and Charles Redlick (2011) and by Mertens and Jose Montiel Olea (2018) to assess the effects of the personal tax cuts. These studies are part of a broader body of empirical literature estimating the aggregate causal effects of tax changes. Much of this literature has deployed empirical tools originally developed to study monetary policy, but has adapted them in recent years to address the identification challenges associated with tax policy. Some of the findings have been used by policymakers to evaluate recent tax proposals (Council of Economic Advisers 2017). The remainder of this comment provides additional perspective on the possible GDP effects of the 2017 tax law based on the results from this literature. I describe the main reduced-form models, show how they may be used to make projections about the impact of the 2017 tax law, and discuss their interpretation and limitations. For brevity, details are omitted but can be found online (Mertens 2018), together with replication materials. (2)
The object of interest is the causal effect [[gamma].sub.h] on GDP observed h [greater than or equal to] 0 years after a reform of size [DELTA][[tau].sub.t] in period t, given by
(1) [[DELTA]y.sub.t+h] = [[gamma].sub.h][DELTA][[tau].sub.t] + [DELTA][[bar.y].sub.t+h],
where [DELTA][y.sub.t+h] is future output growth and [DELTA][[bar.y].sub.t+h], is counterfactual future output growth in the absence of the tax reform. In order to use equation 1 for out-of-sample projections, the tax policy interventions [DELTA][[tau].sub.t] in this equation are inputs that are measurable in advance. Also required are estimates of the dynamic causal effects [[gamma].sub.h] associated with these inputs. Such estimates are commonly obtained from structural macroeconomic models, but can also be obtained from regression-based models that avoid the need to make detailed assumptions.
Starting with Christina Romer and David Romer (2010), a growing number of studies have exploited historical tax reforms as quasi-experiments to identify the effects of tax changes on aggregate economic activity. (3) Romer and Romer construct a time series, [DELTA][[??].sub.t] containing revenue impact measures for all the major postwar tax reforms in the United States. Estimating [[gamma].sub.t] in equation 1 by regressing [DELTA][y.sub.t+h] on [DELTA][[??].sub.t], is, however, problematic because tax reforms are not random events. Important determinants of GDP growth, such as government spending and recessionary shocks, also influence tax policy. To avoid problems of simultaneity, Romer and Romer classify each of the tax reforms according to the primary motivation of policymakers, on the basis of a variety of historical sources. Next, they construct a series of exogenous tax reforms [DELTA][[??].sup.exo.sub.t], containing only those reforms classified as either ideological or as arising from long-term deficit concerns, while omitting interventions responding to the business cycle or short-term (and typically defense-related) government spending changes. The causal effects of tax reforms can be obtained as the least-squares estimates of [[delta].sub.h] in
(2) [DELTA][y.sub.t+h] = [[delta].sub.h][DELTA][[??].sup.exo.sub.t] + [u.sub.t+h],
where [u.sub.t+h] is a residual term. This approach is valid as long as the tax reforms retained in [DELTA][[??].sup.exo.sub.t] are indeed uncorrelated with the residual [u.sub.t+h] capturing all other determinants of output growth in t + h. In practice, the [[delta].sub.h]s can be estimated by separate regressions for every h, by a single regression for h = 0 and additional lags of [DELTA][[??].sup.exo.sub.t] as regressors, or by inverting the coefficients of a single regression for h = 0 and lags of [DELTA][y.sub.t] as additional regressors. The regressions may include other control variables, or they may be part of a system of dynamic equations with many other endogenous variables.
The first three rows of panel A in my table 1 present estimates of the GDP impact of the 2017 tax law for different variants of the direct regression approach given in equation 2. The estimates are not based on the original series of Romer and Romer (2010) for [DELTA][[??].sup.exo.sub.t], but on an alternative version proposed by Mertens and Morten Ravn (2012) that omits tax changes with long implementation lags to avoid an additional source of bias. (4) The models use postwar quarterly data, and the series for [DELTA][[??].sup.exo.sub.t] are static estimates of total revenue effects as a ratio of GDP in the previous quarters. The scaling by GDP means that the estimates of [[gamma].sub.h], have the familiar interpretation as tax multipliers. Because almost all provisions in the 2017 tax law become effective in the 2018 tax year, the act clearly fits into the category of reforms with short implementation lags included in Mertens and Ravn's (2012) version of [DELTA][[??].sup.exo.sub.t]. The motivation for the 2017 law also seems predominantly ideological, such that it appears reasonable to make use of the estimated effects derived from the exogenous tax reforms of Romer and Romer (2010).
The projections in rows 1 through 3 of panel A in my table 1 are calculated by applying the estimated tax multipliers for each of the three models to a reduction in total tax revenues in the first quarter of 2018 equal to -1.1 percent of GDP, which is based on scoring by the Joint Committee on Taxation. The results in the first row are from the distributed lag specification given by Romer and Romer (2010). The results in the second row are from a multivariate vector autoregressive model that includes [DELTA][[??].sup.exo.sub.t] as an exogenous regressor to each equation, following Carlo Favero and Francesco Giavazzi (2012). Finally, the third row shows results for the same vector autoregressive model but adds a distributed lag of [DELTA][[??].sup.exo.sub.t], following Mertens and Ravn (2012).
Each of the three direct regression approaches yields very similar projections for output growth in 2018, which is predicted to rise by about 1.3 percentage points. The models differ significantly, however, in dynamics after the first year. The Romer and Romer (2010) model shows a continued positive impact on GDP growth rates through 2019 and 2020, and a total cumulative increase of 2.74 percentage points by 2020. The Favero and Giavazzi (2012) model, by contrast, shows a moderate reversal of GDP levels beyond 2018, and a much more modest cumulative three-year growth impact of 0.82 percentage point. Finally, the Mertens and Ravn (2012) model yields a positive effect on GDP growth that persists in 2019, a sharp reversal in 2020, and a cumulative three-year increase in GDP of 1.39 percentage points.
The other common approaches in the literature can be explained by considering a simple joint system for taxes and output:
(3) [DELTA][y.sub.t+h] = [[zeta].sub.h][DELTA][T.sub.t] + [e.sub.t+h],
(4) [DELTA][T.sub.t] = [theta][DELTA][y.sub.t] + [DELTA][[tau].sub.t].
Equation 3 relates future output growth to changes in observable measures of the burden of taxation [DELTA][T.sub.t], such as tax revenues (Blanchard and Perotti 2002), average tax rates (Mertens and Ravn 2013), or average marginal tax rates (Barro and Redlick 2011). Equation 4 makes explicit that these tax measures are endogenous and vary not only because of occasional tax reforms but also because of changes in economic activity [DELTA][y.sub.t]. This is obvious when T is tax revenues, but it is also true when using average or average marginal tax rates, for example, because of tax progressivity. Another reason that T is endogenous is that tax policy interventions are also systematically related to [DELTA][Y.sub.t]--for instance, by some policy rule. In this case, [DELTA][y.sub.t] in equation 4 can be viewed as the unobserved residual in such a rule that is uncorrected with [e.sub.t+h] for h = [+ or -] 0, 1,..., and with the policy reaction coefficient absorbed in [theta]. When 0 = 0, least-squares estimates of [[zeta].sub.h] in equation 3 are not interpretable as the causal effect of tax changes because of simultaneity.
The literature has addressed this identification problem in three alternative ways. Olivier Blanchard and Roberto Perotti (2002) use an outside estimate of [theta] to back out [DELTA][[tau].sub.t] as the residual in equation 3, and then use this residual as an instrument to estimate [[zeta].sub.h] Barro and Redlick (2011) and Mertens and Ravn (2014) instead use the Romer and Romer (2010) series [DELTA][[??].sup.exo.sub.t] as an instrument to estimate [[zeta].sub.h]. Mertens and Ravn (2014) also use the resulting estimate of [[zeta].sub.h] to obtain [e.sub.t] and use it as an instrument to estimate [theta]. Finally, Dario Caldara and Christophe Kamps (2017) use nontax instruments (for example, oil or monetary shocks) to estimate [theta], and then use the implied residuals given in equation 4 as an instrument to estimate [[zeta].sub.h]. In practice, most studies embed the relationship between equations 3 and 4 in richer systems with dynamic terms and more endogenous variables than just output and taxes. The main reason is that the required identification assumptions become less stringent, and arguably only plausible, in these richer settings.
Two complications arise when using parameter estimates from equations 3 and 4 for out-of-sample projections of the impact of tax reforms. Each of these complications becomes clearer after combining both equations to obtain
The first difficulty stems from a key difference with the direct regression approach used in equation 2, which is that the indirect approaches treat [DELTA][[tau].sub.t] as unobserved. Moreover, richer dynamic models identify [DELTA][[tau].sub.t] as a macro-economic shock--that is, as a surprise deviation in taxes from the prior period forecasts that are unrelated to any other structural source of forecast error, such as monetary policy, productivity, or financial market shocks. The tax shocks are only loosely related to historical tax reforms, which are all at least to some degree anticipated preceding enactment. In addition, sizable tax shocks may also occur when forecasted tax changes are not realized, and the series for [DELTA][[??].sup.exo.sub.t] may also more generally contain measurement error. Any of these concerns causes bias in a direct regression of equation 5 after replacing [DELTA][[tau].sub.t] with [DELTA][[??].sup.exo.sub.t]. Although the indirect approaches, at least in principle, avoid these problems, one disadvantage is that the tax shocks measured by the residual in equation 4 are only identified in-sample. This means that, for the purpose of out-of-sample projections, an additional assumption is required regarding the size of the surprise tax shock induced by the 2017 tax reform.
Another, more minor, complication arises because of macroeconomic feedback on taxes. Typically, the focus in empirical work is on estimates of [[zeta].sub.h], which have the interpretation of the causal effects of a tax shock leading to a unit change in the measure of taxes of interest, [DELTA][T.sub.t]. With macro-economic feedback [theta] [not equal to] 0, this differs from the effect of a unit change in [DELTA][[tau].sub.t]. (5) However, after establishing the size of the shock induced by the reform, it is the latter that is needed for projecting the GDP impact. Equation 5 shows that this projection requires knowledge of [(1 - [[zeta].sub.h][theta]).sup.-1][[zeta].sub.h], and therefore not only of [[zeta].sub.h] but also of [theta]. The distinction is almost automatically relevant, because if [theta] = 0, there would not be an identification problem to begin with. Empirically, it is the case that [theta] > 0 (higher output leads to higher taxes) and [[zeta].sub.h] < 0 (higher taxes lead to lower output), such that [(1 - [[zeta].sub.h][theta]).sup.-1] ([) < 1. (6) In practice, the difference therefore leads to a reduction in the effects relative to those measured by [[zeta].sub.h].
Rows 4 through 6 in panel A of my table 1 report results for each of the three indirect approaches outlined above, as implemented within a structural vector autoregressive model. The estimates shown are for a tax shock that occurs in 2018:Q I and equals the entire revenue impact of-1.1 percent of GDR The implicit assumption is therefore that the prospects for the eventual tax reform had no influence on economic activity before 2017:Q4. This assumption is perhaps questionable, because proposals with the basic contours of the reform were made well in advance. Conversely, the odds of passage in Congress as well as the extent of the cuts in individual taxes remained highly uncertain until very late in the legislative process. (7) In any case, the projections in rows 4 through 6 can easily be adjusted to reflect alternative assumptions regarding the size of the tax surprise induced by the 2017 tax law.
Panel A in my table 1 shows that the Blanchard and Perotti (2002) and Caldara and Kamps (2017) identification approaches yield relatively similar projections for GDP growth, which is predicted to increase by about 0.9 percentage point in 2018. The projections of the Mertens and Ravn (2014) model, which uses the Romer and Romer (2010) exogenous reforms for identification, indicate GDP growth that is higher by 1.57 percentage points for 2018. All three models show relatively small effects beyond 2018, with a slightly more pronounced reversal in the Mertens and Ravn (2014) model. The projected cumulative effect on 2020 GDP levels ranges from 0.77 to 1.13 percentage points higher. The final row of panel A in my table 1 provides the simple average of the projections of all six tax multiplier models, which shows a growth impact of 1.21 and 0.36 percentage points in 2018 and 2019, respectively. The average of the projections suggests that in 2020 (and beyond), the 2017 act becomes a modest drag on economic growth.
The tax multiplier models discussed so far only consider the effects of changes in total tax revenues. The usefulness of the resulting projections depends on how similar the 2017 tax reform is in terms of the persistence and composition of the tax changes identified in the sample by the various models. In the postwar period, federal tax changes have typically either included sunsets or offsetting provisions, or else have been reversed by bracket creep or subsequent legislation. The revenue-to-GDP ratio, as a result, has remained fairly stable, and the projections in my table 1 implicitly assume a trajectory of future taxes that is correspondingly reverting to average levels. The many sunsets included in the 2017 act suggest that its enactment has generated historical typical expectations for future taxes, at least on the individual side. Important differences in the composition of tax changes relative to other reforms, however, may also matter for determining growth effects. The remainder of the estimates in my table 1 provide projections based on a number of additional models that account for different aspects of reforms.
One dimension in which the 2017 act differs substantially from most other postwar reforms is the magnitude of the business tax cuts. Panel B in my table 1 shows projections that are based on the baseline model of Mertens and Ravn (2013), which separately identifies the effects of changes in the personal and corporate provisions of the tax reforms. The identification approach is similar to that of Mertens and Ravn (2014), but makes use of separate instruments for the individual and corporate provisions of the exogenous tax reforms presented by Romer and Romer (2010). The identification allows for correlation between personal and corporate tax changes, and exploits heterogeneity in composition across U.S. postwar tax reforms to isolate the dynamic causal effects of each type of tax change. The model only indirectly identifies shocks to both average tax rates, and the projections in my table 1 use the numbers from the Joint Committee on Taxation's report to determine the size of the shocks. Specifically, the first (second) row shows the effects of an unexpected shock of -0.8 percent (-7.4 percent) in the average personal (corporate) tax rate in 2018:Q1. Assessing the effects of the international reform is particularly difficult, because it has no historical counterpart in the estimation sample. The projections in panel B for the international provisions, which are based on assuming an additional positive corporate tax shock of 3.6 percent in 2018:Q1, should therefore be interpreted with great caution. (8) The last row in panel B of my table 1 shows the combined effect of all provisions implied by the estimates based on Mertens and Ravn (2014).
The main implication of accounting for the composition of the 2017 law in terms of the individual and corporate provisions is that the projection for 2018 GDP growth is noticeably larger, at an increase of 1.79 percentage points. The predicted three-year cumulative impact of 1.39 percentage points, conversely, is roughly similar to the average of projections in the tax multiplier models, which indicates a somewhat more pronounced reversal of GDP growth in 2019 and 2020. The projections suggest a relatively large, but short-lived, effect on growth as a result of the business tax reform. This reflects the fact that the corporate tax shocks discussed by Mertens and Ravn (2013) are identified largely by transitory changes in the after-tax cost of new investment that can create strong incentives for intertemporal substitution." Whether such incentives are currently at play depends on firms' expectations regarding future corporate tax rates and depreciation allowances. In any case, the results are not directly informative about the long-run impact of the reduction in the corporate tax rate, which is an important limitation relative to Barro and Furman's approach.
Another distinguishing feature of the 2017 law is the substantial cuts in marginal tax rates for individuals, at least in the short run. Panel C in my table 1 shows projections based on two studies that estimate the growth effects of changes in marginal tax rates, rather than revenues or average tax rates. These numbers are for the individual tax reform only, and do not incorporate the growth effects of the business and international tax reforms. Identification in both cases relies on using the exogenous tax reforms discussed by Romer and Romer (2010) to construct instruments for income-weighted average marginal tax rates (AMTRs). Barro and Redlick (2011) use Romer and Romer's original average tax rate series, whereas Mertens and Montiel Olea (2018) construct new instruments for changes in AMTRs. The estimate for Barro and Redlick's (2011) model in my table 1 is obtained by multiplying an AMTR cut of 2.75 percentage points with an estimated 0.5 percent two-year increase in GDP for every percentage point decrease in the AMTR. (10) The remaining projections in panel C are obtained by assuming a shock to the overall AMTR of -2.75 percentage points. This number is calculated using the same methods as Mertens and Montiel Olea, and is higher than the cut of 2.30 percentage points assumed by Barro and Furman.
The first set of projections from Mertens and Montiel Olea uses a model that only identifies the effects of AMTR shocks. The second set of projections is based on a model that separately identifies the effects of changes in marginal and average tax rates. The methodology in this case is analogous to that of Mertens and Ravn (2014), with distinct instruments for marginal and average tax rates, while accounting for the fact that both are correlated. Both sets of projections are roughly similar, and suggest a growth impact of about 1.30 percentage points in 2018, and an additional 1 percentage point of GDP growth in 2019. The cumulative three-year increase in real GDP is predicted to be about 2.4 percentage points. An estimate of the total growth impact of the 2017 tax law can in principle be obtained by adding the projections in panel C to those in the second and third rows of panel B. The main conclusion is that taking into account the substantial cuts in marginal tax rates in the 2017 law suggests a larger impact on GDP growth than suggested by the tax multiplier models in panel A.
Most postwar changes to individual tax rates differ substantially by income level, with typically much larger changes in top statutory rates. The 2017 law differs in that it cuts tax rates more uniformly than is typically the case. Panel D in my table 1 shows results based on two studies that allow the aggregate effects of tax changes to depend on the distribution of tax changes by income level. Mertens and Montiel Olea (2018) separately identify the effects of shocks to average marginal tax rates for the top 1 percent and bottom 99 percent income groups, again by adopting the methodology used by Mertens and Ravn (2014), with distinct instruments for changes in AMTRs for both groups. The projections in the first row are based on a shock to the top 1 percent and bottom 99 percent AMTRs of -2.66 and -2.78 percentage points, respectively. Owen Zidar (forthcoming) instead follows a direct regression approach based on an extension of Romer and Romer's (2010) series that separates the revenue impact of tax changes affecting the top 10 percent and bottom 90 percent of the income distribution. The results in my table 1 use the estimates reported by Zidar and assume a 50/50 split between the top 10 percent and bottom 90 percent, as suggested in a distributional analysis by the Tax Policy Center (2017).
According to the results in panel D of my table 1, the main implication of accounting for the distributional aspects of the 2017 law is that the growth impact is more delayed, and occurs largely in 2019 and 2020 rather than 2018. Zidar's estimates even suggest a negative effect in 2018, although it is based on an estimate that is not statistically significant. By 2020, Mertens and Montiel Olea's model predicts a level of GDP that is higher by 3 percentage points, while Zidar's regressions suggest an increase in GDP by about half that amount. As for panel D, an estimate of the total growth impact of the 2017 tax law can be obtained by adding the projections in panel B for the corporate provisions.
The main advantage of the reduced-form approaches is that all the relevant channels through which taxes affect economic activity are in principle reflected in the model parameters, without the need for large numbers of detailed theoretical assumptions. Expectations of future tax rates can be particularly hard to verify, but are likely important in shaping the short-run effects. Reduced-form models do not require explicit assumptions regarding the dynamics of expected future tax rates, which are instead part of the estimation. Recent methodological advances also make it possible to account for changes in multiple policy instruments. Conversely, the approach is valid only to the extent the reform is unexpected, exogenous, and reasonably similar to the historical variation and dynamics of tax policy that underlie the model estimates. For these reasons, it is better suited to assess the impact of the individual tax component of the 2017 tax law than the impact of the corporate and international tax components.
The main conclusion from the projections from the reduced-form models is that most specifications yield a sizable growth effect for 2018, and more modest effects afterward. The projections complement those available from structural models, and it should be noted that uncertainty in the underlying parameters means that the range of plausible outcomes typically remains wide. That being said, much progress has been made in monetary economics by combining reduced-form evidence and quantitative structural models, and there is no reason to believe that the same strategy cannot similarly advance the study of tax policy. In any case, in the absence of any major macroeconomic shocks, the trajectory of GDP over the next few years will be informative.
REFERENCES FOR THE MERTENS COMMENT
Auerbach, Alan J. 1989. "Tax Reform and Adjustment Costs: The Impact on Investment and Market Value." International Economic Review 30, no. 4: 939-62.
Barro, Robert J., and Charles J. Redlick. 2011. "Macroeconomic Effects from Government Purchases and Taxes." Quarterly Journal of Economics 126, no. 1: 51-102.
Blanchard, Olivier, and Roberto Perotti. 2002. "An Empirical Characterization of the Dynamic Effects of Changes in Government Spending and Taxes on Output." Quarterly Journal of Economics 117, no. 4: 1329-68.
Caldara, Dario, and Christophe Kamps. 2017. "The Analytics of SVARs: A Unified Framework to Measure Fiscal Multipliers." Review of Economic Studies 84, no. 3: 1015-40.
Cloyne, James. 2013. "Discretionary Tax Changes and the Macroeconomy: New Narrative Evidence from the United Kingdom." American Economic Review 103, no. 4: 1507-28.
Cloyne, James S., and Paolo Surico. 2017. "Household Debt and the Dynamic Effects of Income Tax Changes." Review of Economic Studies 84, no. 1: 45-81.
Council of Economic Advisers. 2017. "The Growth Effects of Corporate Tax Reform and Implications for Wages." Washington: White House.
Favero, Carlo, and Francesco Giavazzi. 2012. "Measuring Tax Multipliers. The Narrative Method in Fiscal VARs." American Economic Journal: Economic Policy 4, no. 2: 69-94.
Gil, Paula, Francisco Marti, Richard Morris, Javier J. Perez, and Roberto Ramos. Forthcoming. "The Output Effects of Tax Changes: Narrative Evidence from Spain." SERIEs.
Guajardo, Jaime, Daniel Leigh, and Andrea Pescatori. 2014. "Expansionary Austerity? International Evidence." Journal of the European Economic Association 12. no. 4: 949-68.
Hayo, Bernd, and Matthias Uhl. 2014. "The Macroeconomic Effects of Legislated Tax Changes in Germany." Oxford Economic Papers 66, no. 2: 397-418.
House, Christopher L., and Matthew D. Shapiro. 2008. "Temporary Investment Tax Incentives: Theory with Evidence from Bonus Depreciation." American Economic Review 98, no. 3: 737-68.
Htissain, Syed M., and Lin Liu. 2017. "Macroeconomic Effects of Government Spending Shocks: New Narrative Evidence from Canada." Working paper. https://sites.google.com/site/syedmhussainl984/research
-- 2018. "Comparing the Effects of Discretionary Tax Changes between the US and the UK." B. E. Journal of Macroeconomics 18, no. 1: article 20160041.
Lopes, Jose. 2016. "The Federal Tax Multiplier in Canada: A Narrative Approach." Job market paper, Cornell University, https://sites.google.eom/a/cornell.edu/josemariolopes/research
Mertens, Karel. 2018. "The Near Term Growth Impact of the Tax Cuts and Jobs Act." Working Paper no. 1803. Federal Reserve Bank of Dallas.
Mertens, Karel, and Jose Luis Montiel Olea. 2018. "Marginal Tax Rates and Income: New Time Series Evidence." Quarterly Journal of Economics 133, no. 4.
Mertens, Karel, and Morten Ravn. 2012. "Empirical Evidence on the Aggregate Effects of Anticipated and Unanticipated US Tax Policy Shocks." American Economic Journal: Economic Policy 4, no. 2: 145-81.
-- 2013. "The Dynamic Effects of Personal and Corporate Income Tax Changes in the United States." American Economic Review 103, no. 4: 1212-47.
-- 2014. "A Reconciliation of SVAR and Narrative Estimates of Tax Multipliers." Journal of Monetary Economics 68, suppl.: S1-S19.
Nguyen, Anh D. M., Luisanna Onnis, and Raffaele Rossi. 2017. "The Macro-economic Effects of Income and Consumption Tax Changes." Sheffield Economic Research Paper Series no. 2017008. Sheffield: University of Sheffield, Department of Economics.
Riera-Crichton, Daniel, Carlos A. Vegh, and Guillermo Vuletin. 2016. "Tax Multipliers: Pitfalls in Measurement and Identification." Journal of Monetary Economics 79: 30-48.
Romer, Christina D., and David H. Romer. 2010. "The Macroeconomic Effects of Tax Changes: Estimates Based on a New Measure of Fiscal Shocks." American Economic Review 100, no. 3: 763-801.
Tax Policy Center. 2017. "Distributional Analysis of the Conference Agreement for the Tax Cuts and Jobs Act." Washington.
Zidar, Owen. Forthcoming. "Tax Cuts for Whom? Heterogeneous Effects of Income Tax Changes on Growth and Employment." Journal of Political Economy.
(1.) The views expressed in this comment are those of the author and do not necessarily reflect the views of the Federal Reserve Bank of Dallas or the Federal Reserve System.
(2.) The supplemental materials for this and all other papers in this volume may be found at the Brookings Papers web page, www.brookings.edu/bpea, under "Past BPEA Editions." Also see https://karelmertens.com/research.
(3.) The same approach has been used to study the effects of tax policy changes in the United Kingdom (Cloyne 2013; Cloyne and Surico 2017; Nguyen, Onnis, and Rossi 2017; Hussain and Liu 2018), Germany (Hayo and Uhl 2014), Canada (Lopes 2016; Hussain and Liu 2017), Spain (Gil and others, forthcoming), or multiple countries (Guajardo, Leigh, and Pescotori 2014; Riera-Crichton, Vegh, and Vuletin 2016).
(4.) Tax reforms often legislate tax changes that only become effective with a delay. Mertens and Ravn (2012) provide evidence for anticipation effects on economic activity of preannounced tax changes, which generally lead to a violation of the exogeneity requirements.
(5.) The same distinction is also highlighted by Caldara and Kamps (2017).
(6.) In models with more endogenous variables, the effect usually goes in the same direction.
(7.) Based on the Romer and Romer (2010) exogenous reforms with short implementation lags, Mertens and Ravn (2012) cannot reject the null hypothesis of no effects on GDP in quarters before enactment, which suggests that anticipation effects during the legislative process are limited.
(8.) Since the expected revenue increases stem from the repatriation of income at lower rates than the prior statutory rate, it may even be considered to be a cut in taxes.
(9.) See Auerbach (1989) and House and Shapiro (2008) for theory and evidence. 10. Note that Barro and Redlick (2011) do not estimate the equivalent of [theta] from equations 3 and 4. Therefore, the estimate reported in my table 1 is in this case not adjusted for scale as discussed, and is instead based directly on the instrumental variables estimate of the slope coefficient [[zeta].sub.h]. The results in my table 1 based on Mertens and Montiel Olea (2018) are adjusted to account for macroeconomic feedback effects.
Table 1. Estimates of the Effects of the 2017Tax Law on Real GDP Growth, 2018-20 Effect on GDP growth 2018 2019 2020 Cumulative, (percentage points) 2018-20 A. Based on tax multiplier estimates Direct regressions Romer and Romer (2010) 1.34 0.83 0.57 2.74 Favero and Giavazzi (2012) 1.23 -0.11 -0.30 0.82 Mertens and Ravn (2012) 1.31 1.17 -1.08 1.39 Structural vector autoregressive models Blanchard and Perotti (2002) 0.93 0.30 -0.18 1.05 Mertens and Ravn (2014) 1.57 -0.09 -0.36 1.13 Caldara and Kamps (2017) 0.86 0.06 -0.15 0.77 Average 1.21 0.36 -0.25 1.32 B. By individual and corporate tax provisions Mertens and Ravn (2013) Individual tax reform 0.87 -0.25 -0.12 0.51 Business tax reform 2.04 0.03 -0.16 1.92 International tax reform -1.00 -0.02 0.08 -0.94 Total 1.79 -0.21 -0.19 1.39 C. Based on responses to individual marginal tax rates Barro and Redlick (2011) Average marginal tax rate 1.38 n.a. n.a. Mertens and Montiel Olea (2018) Average marginal tax rate 1.29 1.01 0.13 2.44 Average marginal tax rate and 1.38 1.07 -0.05 2.39 average tax rate D. Based on estimates allowing for income dependence Mertens and Montiel Olea (2018) Top 1 percent and bottom 99 0.54 1.06 1.45 3.04 percent average marginal tax rates Zidar (forthcoming) Top 10 percent and bottom -0.77 1.70 0.62 1.55 90 percent incomes Source: Author's calculations.
KENT SMETTERS Congratulations to the editors for encouraging two distinguished economists on opposing sides of the recent tax debate to coauthor a paper on the topic. And, of course, especially hearty congratulations to the authors for actually doing it! This paper serves as a role model for future discussions of "hot" topics. Although the authors offer different conclusions, they also agree on a lot, which is also instructive for policymakers.
The Barro--Furman modeling is very elegant. The simplicity of the household side makes it easy for the authors to treat extensions to the firm side with rigor, including the debt/equity choice, multiple types of capital, and pass-through entities. They are careful in their calibration. (1) The model is transparent, a rarity in Washington policymaking, especially during the recent tax debate.
WHY ARE THE GROWTH EFFECTS SO SMALL? The Barro--Furman model, along with its calibration strategy, strike me as fairly "progrowth" in nature: (2) The base model is Barro--Ricardian (no debt effects, although later adjusted); the government's budget constraint is balanced using lump sum taxes; though prices are deterministic, the initial interest rate includes a risk premium that is being taxed before the tax cut (and with no tax loss offsets); labor supply is fixed (income effects tend to dominate substitution effects in the long run); the household sector is represented by a single, infinitely lived agent facing infinite long-run savings elasticity; and there is no international tax competition (allowing the value of [r.sup.k] in the model to remain fixed after the tax cut).
The most obvious question, therefore, is Why are the growth effects so apparently small on the corporate side? My table 1 summarizes a large part of the answer by reporting the output-weighted average effective tax rate (ETR) across U.S. corporate industries, between 2018 and 2040, both before and after the passage of the 2017 tax law.
Notice that the ETR falls sharply in 2018 after passage of the law but returns most of the way to its projected value from before the law by 2023, and especially by 2027. The reason can be found in my table 2, which includes a partial list of the numerous tax provisions--including some of their phaseouts--contained in the 2017 tax law used to generate the ETR values in my table 1. Several provisions in 2018 substantially reduce the ETR, including bonus depreciation. However, within just five years, bonus depreciation starts phasing out. Moreover, "pay-for" provisions, such as the amortization of research and experimental expenditures, also raise the ETR.
Before the 2017 tax law, many capital investments would have been depreciated over the next decade. But since the law, depreciation is being accelerated through expensing. As a result, much of the short-run reduction in ETR values simply reflects a shift in the timing of depreciation allowances rather than a permanent reduction in the ETR, thereby creating only a small win for business in present value.
This timing shift also largely explains why the Joint Committee on Taxation (JCT), the government's leading tax experts and official tax scoring agency, priced the provisions in the "changes to the treatment of investment" category at just $86 billion in lost revenue between 2018 and 2027. The Penn Wharton Budget Model (PWBM 2017a) priced it higher, at $180 billion, the difference largely reflecting the PWBM's different modeling of income-shifting and reclassification. Still, by either measure, these provisions are cheap in comparison with the cost of, for example, permanent expensing. But permanence would have required a standard bill and 60 votes in the Senate.
The temporary reduction in the ETR also largely explains Barro and Furman's table 14, which appears to indicate a general agreement among different models pertaining to the 2017 tax law's growth effect. Some of the models listed in their table 14 are not publicly documented, so it is challenging to make a detailed comparison. Nonetheless, if the tax cuts were permanent, the Ramsey framework used by Barro and Furman would generally produce much larger long-run increases to GDP relative to reduced-form models or models with a fixed savings rate.
Of course, one reason for the bigger potential gain in the Ramsey model is its infinite long-run savings elasticity. But another reason is simply its ability, as a structural model, to consider novel tax changes. In contrast, nonstructural reduced-form or fixed-savings-rate models are generally calibrated to unrelated historical data and unrelated policy changes, which have been more muted. Put differently, the reason for the apparent model similarity in their table 14 is not because the different models are similar but because the 2017 tax law is not a particularly large reform.
THE ROLE OE DEBT The authors discuss the potential role of debt, with Furman believing that it plays a larger role in limiting growth through mitigating capital formation than does Barro. The authors nicely provide a robustness check, where more debt modestly increases the interest rate. Ultimately, the role of debt in the modeling exercise is an empirical question.
Empirically, estimates by Laurence Kotlikoff and Lawrence Summers (1981, 1988) indicate that almost 80 percent of wealth is transferred intergenerationally, with the other 20 percent being motivated by standard life cycle considerations, indicating a potentially large role for Ricardian equivalence. Franco Modigliani (1988) flips these percentages, arguing that the life cycle explains most wealth accumulation. The excellent review by William Gale and John Karl Scholz (1994) essentially splits the difference, noting challenges in data availability and definitions. All these estimates are based on aggregate wealth data.
Emanuela Cardia (1997) shows that many tests for Ricardian equivalence using aggregate consumption data have low power. Using micro data, however, Joseph Altonji, Fumio Hayashi, and Kotlikoff (1997) find that only $0.13 of each $1 redistributed from children to parents is rebated back to children.
John Laitner's (1992) excellent but often-overlooked paper--a model with the potential for earnings differences between generations along with a nonnegativity constraint on bequests--effectively reconciles the micro and macro evidence. In his model, a considerable amount of intergenerational transfers can be made, even though the marginal impact of an additional $1 of debt behaves more like a traditional life cycle model, thereby crowding out capital formation.
Robust international capital flows, of course, can also mitigate the negative effects of debt on capital formation. Since 1990, the marginal foreign take-up of debt has averaged about 40 percent (PWBM 2016), which motivates the 40 percent open assumption in the PWBM's overlapping-generations (OLG) model. However, U.S. debt is the ubiquitous safe asset throughout the world, serving, for example, as a reserve asset for foreign insurers. Foreign take-up of U.S. equity is much lower, consistent with the home bias puzzle (Feldstein and Horioka 1980; Obstfeld and Rogoff 2001).
However, what is probably the "scariest" problem associated with rising debt--foreign investors losing confidence (a bank run)--is not captured in modern, smooth tax models. (3) Although these types of concerns are generally associated with emerging economies, Carmen Reinhart and Kenneth Rogoff (2015) show that even large, advanced economies are not robust to these problems.
RAMSEY VERSUS SAMUELSON AND DIAMOND The presence of infinite horizons in the Ramsey model with capital accumulation allows for the presence of only a single household (think Nietzsche's Ubermensch (4)) to rule the economy. (5) This modeling approach, however, produces several challenges.
First, modeling individual-side (nonbusiness) tax provisions is hard. With uninsurable income heterogeneity, identical average marginal tax rates at the household level can be achieved using a wide array of individual-side tax systems, ranging from completely flat to very progressive. However, the risk-sharing properties (relevant for welfare calculations) and precautionary savings (relevant for macro considerations in a second-best, Mirrleesian economy) differ significantly between these different tax designs (Nishiyama and Smetters 2005).
Second, the Ramsey model is often viewed as a long-run equilibrium model, as noted by the authors. As a result, it is often paired with shorter-term disequilibrium models, with the final model output reported as a linear combination across model results. The experts at the JCT, for example, report blended output from three models, with more weight shifting to the neoclassical model over time. The JCT uses very rounded weights to blend its models, presumably because the blend weights are very hard to estimate. Given the infrequent nature of major policy changes--about once every 30 years for major tax reforms (Auerbach and Smetters 2017)--it would likely require centuries of stationary data to estimate the right blend weights, especially because the blend weights should also be a function of the economy (for example, an output gap) and the actual policy experiment (a la the Lucas critique).
We can see the challenges of model blending in the Barro-Furman analysis when trying to bridge their Ramsey and reduced-form model output. If they stacked (summed) their GDP level effects between their Ramsey model and their reduced-form model based on that of Barro and Charles Redlick (2011), then the headline, l0th-year GDP that they report in their table 1 would be nearly four times larger--1.55 percent instead of 0.4 percent. The implied annual growth rates would also be larger, and their corresponding dynamic revenue losses would be smaller. If they regard their two model results as unstacked, then they predict a larger GDP level effect by 2019 than by 2027, requiring a negative growth rate in the intervening years to connect the dots. These differences become even more dramatic if we consider their reduced-form model based on that of Karel Mertens and Jose Montiel Olea (2018), which produces even larger short-run gains than the Barro-Redlick model.
Ultimately, Barro and Furman choose to report their GDP levels and dynamic revenue losses on an unstacked basis. To be sure, this choice could be reasonable for the 2017 tax law, given its sunsets--although their reduced-form models do not distinguish between a temporary and permanent tax cut. Still, it is a highly subjective judgment that can, as just noted, produce very different final results. For example, even if one believes in large short-run responses (either due to Keynesian or supply-side effects), it is hard to provide an economic interpretation for why some of the initial level gains would have to be given back in subsequent years after a tax cut, especially for the provisions-permanent scenario with no sunsets. (6)
Third, intra- and intergenerational distributional analysis in the Ramsey model is difficult. For intragenerational analysis, simply grossing up all pretax wages from a static distribution by an identical factor misses important nonhomothetic factors, including capital income being much more concentrated than labor income. Intergenerational distributional analysis--generational accounting (Auerbach, Gokhale, and Kotlikoff 1994), or the closely related fiscal imbalances measure (Gokhale and Smetters 2003)--is made irrelevant by the Ricardian property.
For better or worse, distributional analysis is important in any tax reform debate. As a result, the Ramsey macroeconomic results would again need to be blended with distributional outcomes from a different model. Unfortunately, that "different model" is often very simple. In particular, as in the recent tax debate, the media usually reports distributional measures such as "the top 1 percent of income earners get X percent of the tax cut," despite the fact that this type of measure does not correspond to a meaningful change in the Mirrleesian after-tax, after-transfer income distribution.
Fourth, the Ramsey model requires infinite long-run savings elasticity. Besides infinity being a "big number," it is also unclear how to divide steady-state savings between foreign capital flows and the domestic Ramsey agent to, for example, distinguish between gross domestic product and gross national product.
All these problems vanish in the life cycle OLG frameworks of Paul Samuelson (1958) and Peter Diamond (1965). (7) Although the OLG framework is also an equilibrium model, this point is misunderstood. In my view, all well-specified models are always in (transitional or steady-state) equilibrium. Rather, what people really mean by disequilibrium is whether the model can produce Keynesian effects--for example, unemployment or output gaps.
However, substantial income heterogeneity can be captured in the OLG model, which can also produce Keynesian effects. Poorer households operate closer to their natural borrowing limit (the present value of their safe income that can be legally borrowed against--for example, not Social Security). Under the standard Inada utility condition, it takes the presence of just one reasonably large enough idiosyncratic shock (for example, unemployment risk) in the ergodic set to ensure little to no borrowing. Poorer households, therefore, endogenously have larger marginal propensities to consume, which can be matched against the empirical data. The OLG model can also be modified with search frictions and sticky nominal wages to capture unemployment and an output gap. The economy's openness can also be dialed as desired (for example, to capture the home bias puzzle), because the OLG model's saving elasticity is not infinite asymptotically.
My figure 1 shows the simulations from the PWBM's dynamic OLG model for the 2017 tax law under different initial interest rate assumptions (discussed below). For these purposes, convex adjustment costs (both domestic and international) are turned off, and other fairly "progrowth" assumptions a la Barro and Furman are made, including a closure rule that stabilizes the debt-to-GDP ratio starting in 2040 by reducing "wasteful" government spending.
The combination of capital flows and household labor supply response produces short-run supply-side GDP results similar to Barro and Furman's reduced-form Barro-Redlick model. (8) The Barro--Redlick model predicts that the individual-side tax provisions will raise GDP by 1.15 percent by 2019, with no effect thereafter. If we then stack the individual-side effect onto the business-side effects that they report in their table 1, total GDP increases by 1.55 percent by 2027 without debt effects, and by 1.35 percent with debt effects. Using a similar initial interest rate assumption (the high-r case), the PWBM's model value (with debt effects) is 1.06 percent by 2019 and 1.2 percent by 2027. Over the longer run, these values begin to diverge (1.6 percent for the PWBM by 2040, and 2.5 percent for Barro and Furman), mostly due to nonlinear debt effects in the OLG framework.
The OLG model also allows one to consider the short- and long-run effects of both individual-side and business-side tax reform within a single integrated model, thereby avoiding ad hoc model blending. The tighter integration of the individual and business sides naturally also supports income reclassification. And the OLG life cycle model can also capture intra- and intergenerational dynamic distributional analysis within the same framework, with more accuracy than simple distributional models. For example, Auerbach, Kotlikoff, and Darryl Koehler (2018) have recently shown that the 2017 tax law produces very little distributional impact once life cycle considerations are incorporated, in sharp contrast to the popular press's coverage of the law.
MODELING EXPECTATIONS As Barro and Furman note, dealing with expectations with the 2017 tax law is a bit tricky due to sunsets. Of course, an equilibrium with fully rational expectations would solve the political economy problem within the macroeconomic problem, thereby producing a single simulation. But taking a position on the political economy model (for example, a representative democracy or median voter) is challenging, so I agree with the authors' approach to consider two different options. Still, I think they cheat themselves a little bit in the law-as-written scenario by assuming consistent expectations.
To be sure, if the purpose of this exercise is to help government agencies dynamically score the 2017 tax law, then the simulations should correspond to the actual law, as written--otherwise, the budgeters are playing policymakers. However, it is still legitimate to ask whether economic actors, when forming their expectations, will assume that the 2017 tax law will be extended, given the policy rhetoric. Toward this end, in my figure 1, the PWBM calculated two simulations where, when combined, sunsets arrived as a surprise on their sunset dates. This approach typically sets an upper bound of the possible macroeconomic gains relative to fully anticipated sunsets. Yet this approach incorporates arguably plausible expectations about tax extenders while still being consistent with the actual law." Consistently, the PWBM's analysis of the 2017 tax law generally falls between Barro and Furman's law-as-written and provisions-permanent scenarios.
STATIC IS STILL KING One of the main goals of dynamic scoring is to estimate the net amount of revenue losses from a tax cut that must be covered, potentially, by future generations. Economists enjoy dynamic models. However, for this purpose, static analysis is the key input into dynamic analysis, and is still king. During the process leading to the passage of the 2017 tax reform law, the Tax Foundation and the PWBM provided independent static analyses of the JCT's official calculations. The Tax Foundation and the JCT both priced the law similarly, at slightly less than $1.5 trillion over the first decade. In contrast, the PWBM (2017a) priced it almost $500 billion more. More recently, the Congressional Budget Office (2018) adjusted the JCT's static estimate upward by $400 billion. In their analysis, Barro and Furman, however, start with the JCT's original static estimate and adjust it downward for dynamic effects, producing an ultimate dynamic score of just $1.2 trillion.
The PWBM's (2017a) more pessimistic analysis stems, in part, from its different modeling of income-shifting and reclassification. Still, knowing what I know now (for example, about the growing ability of states to bypass state and local tax deduction limitations, and some new international tax reduction techniques), I believe that the PWBM's estimates are probably not pessimistic enough. Any dynamic score below $1.8 trillion is, in my opinion, very unlikely to materialize over the next decade.
FUTURE WORK The choice of the initial interest rate in the standard neoclassical model without aggregate uncertainty plays a big role in predicting GDP gains, yet its correct calibration is theoretically ambiguous. I agree with Barro that discounting future risky corporate cash flows at the risk-free rate makes little sense. At the same time, investors in the neoclassical model face no price risk (that is, there are no risky cash flows), so the usual justification for the presence of an equity premium in the model's initial interest rate is not present. Still, other parts of tax calibration are based on average values from a risky world. The core problem is that the real-world model generating the data includes aggregate uncertainty, which is necessary to generate risky cash flows, whereas the stylized neoclassical model does not. In other words, the real-world model has a higher dimension than the neoclassical model, making the calibration mapping challenging. Toward this end, the PWBM reported its results using a higher initial interest rate assumption (high r), similar to that of Barro and Furman, as well as using a lower interest rate assumption (low r). However, future research must tackle this problem directly by incorporating aggregate uncertainty into the model, thereby dealing with the well-known "curse of dimensionality" challenge for general equilibrium models.
The role of entrepreneurship, however, probably remains the biggest black hole in tax modeling. Entrepreneurs often risk substantially more personal wealth than could ever be recovered with future tax loss offsets, and entrepreneurs typically lose. Yet entrepreneurship is a primary driver of growth (and apparently of income inequality) in the United States. A successful entrepreneur generates economic rents, so the corporate tax rate now matters, even with full expensing. To the extent that these rents mostly reflect inefficient market power (for example, network effects), taxing these rents is likely efficient. But to the extent that these rents are required to induce risky investment (for example, patent protection for successful new drug development), taxing them is likely inefficient. To date, tax economists do not have many insights into this important topic.
REFERENCES FOR THE SMETTERS COMMENT
Altonji, Joseph G., Fumio Hayashi, and Laurence J. Kotlikoff. 1997. "Parental Altruism and Inter Vivos Transfers: Theory and Evidence." Journal of Political Economy 105, no. 6: 1121-66.
Auerbach, Alan J., Jagadeesh Gokhale, and Laurence J. Kotlikoff. 1994. "Generational Accounting: A Meaningful Way to Evaluate Fiscal Policy." Journal of Economic Perspectives 8, no. 1: 73-94.
Auerbach, Alan J., Laurence J. Kotlikoff, and Darryl Koehler. 2018. "The New Tax Bill: Winners and Losers." Working paper. https://www.kotlikoff.net/node/633
Auerbach, Alan J., and Kent Smetters, editors. 2017. The Economics of Tax Policy. Oxford University Press.
Barro, Robert J., and Charles J. Redlick. 2011. "Macroeconomic Effects from Government Purchases and Taxes." Quarterly Journal of Economics 126, no. 1: 51-102.
Cardia, Emanuela. 1997. "Replicating Ricardian Equivalence Tests with Simulated Series." American Economic Review 87, no. 1: 65-79.
Congressional Budget Office. 2018. "The Budget and Economic Outlook: 2018 to 2028." Washington. https://www.cbo.gov/publication/53651
Diamond, Peter A. 1965. "National Debt in a Neoclassical Growth Model." American Economic Review 55, no. 5: 1126-50.
Feldstein, Martin, and Charles Horioka. 1980. "Domestic Saving and International Capital Flows." Economic Journal 90, no. 358: 314-29.
Gale, William G., and John Karl Scholz. 1994. "Intergenerational Transfers and the Accumulation of Wealth." Journal of Economic Perspectives 8, no. 4: 145-60.
Gokhale, Jagadeesh, and Kent Smetters. 2003. "Fiscal and Generational Imbalances: New Budget Measures for New Budget Priorities." Policy Discussion Paper no. 5. Federal Reserve Bank of Cleveland.
Kotlikoff, Laurence J., and Lawrence H. Summers. 1981. "The Role of Intergenerational Transfers in Aggregate Capital Accumulation." Journal of Political Economy 89, no. 4: 706-32.
--. 1988. "The Contribution of Intergenerational Transfers to Total Wealth: A Reply." In Modelling the Accumulation and Distribution of Wealth, edited by Denis Kessler and Andre Masson. Clarendon Press.
Laitner, John. 1992. "Random Earnings Differences, Lifetime Liquidity Constraints, and Altruistic Intergenerational Transfers." Journal of Economic Theory 58, no. 2: 135-70.
Mertens, Karel, and Jose Luis Montiel Olea. 2018. "Marginal Tax Rates and Income: New Time Series Evidence." Quarterly Journal of Economics 133, no. 4.
Modigliani, Franco. 1988. "The Role of Intergenerational Transfers and Life Cycle Saving in the Accumulation of Wealth." Journal of Economic Perspectives 2, no. 2: 52-10.
Nishiyama, Shinichi, and Kent Smetters. 2005. "Consumption Taxes and Economic Efficiency with Idiosyncratic Wage Shocks." Journal of Political Economy 113, no. 5: 1088-115.
Obstfeld, Maurice, and Kenneth Rogoff. 2001. "The Six Major Puzzles in International Macroeconomics: Is There a Common Cause?" NBER Macroeconomics Annual 15: 339-112.
PWBM (Penn Wharton Budget Model). 2016. "Setting Behavioral Responses in PWBM's Dynamic Simulations." Philadelphia: University of Pennsylvania, Wharton School.
--. 2017a. "The Tax Cuts and Jobs Act, as Reported by Conference Committee (12/15/17): Static and Dynamic Effects on the Budget and the Economy." Philadelphia: University of Pennsylvania, Wharton School.
--. 2017b. "The Tax Cuts and Jobs Act, as Reported by Conference Committee (12/15/17): Tax Effects by Industry." Philadelphia: University of Pennsylvania, Wharton School.
Reinhart, Carmen M., and Kenneth S. Rogoff. 2015. "Financial and Sovereign Debt Crises: Some Lessons Learned and Those Forgotten." Journal of Banking and Financial Economics 2, no. 4: 5-17.
Samuelson, Paul A. 1958. "An Exact Consumption-Loan Model of Interest with or without the Social Contrivance of Money." Journal of Political Economy 66, no. 6: 467-82.
Uzawa, Hirofumi. 1968. "Time Preference, the Consumption Function, and Optimum Asset Holdings." In Value, Capital and Growth: Papers in Honour of Sir John Hicks. Edinburgh University Press.
(1.) The authors also avoid a common mistake where the estate tax is mixed into the effective tax rate.
(2.) The results reported by the Penn Wharton Budget Model for the 2017 tax law also assume a government closure rule that stabilizes the debt-to-GDP ratio in 2040 by reducing "wasteful" government spending.
(3.) Both the Barro-Furman model and the PWBM use the "magic of the closure rule" to avoid this problem, by assuming that the government eventually does something--either using lump sum taxes (Barro-Furman) or reducing "wasteful" government spending (PWBM)--to generate intertemporal balance. Barro and Furman write that lump sum taxes could translate into policy terms through, for example, "reductions to Social Security or Medicare benefits or other government transfer programs." This statement reflects a Ricardian view where changes in pay-as-you-go spending do not have an impact on household saving. A more general justification, although maybe more conservative in ideology, is to associate their mechanism with reducing "wasteful" government spending that is economically neutral and does not enter the representative household's utility function. Nonstructural models do not require closure because they are not complete, instead relying on reduced-form rules. A very common mistake made by nonstructural models, however, is to confuse deficits with debt; it is the stock of debt that matters for the crowding out of capital formation. Moreover, because debt was on an increasing path even before the 2017 tax law, the marginal impact of new debt can have a nonlinear impact on capital formation in the OLG model, which will not be picked up with a simple empirical rule.
(4.) This Ramsey household should nol be confused with the social planner common in, for example, Mirrleesian-style screening models. The Ramsey household acts in its own interest, whereas the social planner aggregates heterogeneous utilities according to some social rule (thereby, acting more akin to Kierkegaard's "knight of faith").
(5.) With multiple infinite-horizon households, the one with the smallest rate of time preference would accumulate all assets asymptotically. If all households have the same time preference, then the steady-state wealth distribution is no longer asymptotic. Uzawa (1968) attempts to solve this problem by allowing the rate of time preference to increase in wealth (for example, rich dynasties have spoiled children). In contrast, Barro and Furman. if anything, suggest the opposite direction, thereby implicitly rejecting this fix.
(6.) The Barro-Furman and Mertens-Montiel Olea reduced-form modeling only has an impact of two years, and therefore does not distinguish between their law-as-written and provisions-permanent scenarios. For their provisions-permanent scenario, the 10th-year GDP level gain is similar to the Barro-Redlick model value in 2019, although much less than their Mertens-Montiel Olea model. Even for the Barro-Redlick model, negative growth rates would be required before the 10th year in the unstacked scenario.
(7.) The Congressional Budget Office, the JCT, and the PWBM use the OLG model, either in part or in full, as their main model.
(8.) The 2017 tax law produces small Keynesian effects in the PWBM. Most of the short-run effects were supply-driven.
(9.) At very high levels of depreciation with no convex adjustment costs, you could get a reversal in gains between the two cases due to intertemporal substitution. But that scenario is unlikely.
Table 1. Effective Corporate Tax Rates for All Industries, 2018-40 (a) Percentages 2018 2023 2027 2040 Before the 2017 tax law 21.18 23.53 22.95 21.93 After the 2017 tax law 9.16 17.33 18.88 16.06 Source: PWBM (2017b). (a.) The model incorporates incentives to reclassify and intertemporally shift book income. Table 2. Several Major Provisions of the 2017 Tax Law That Have an Impact on Effective Tax Rates Start year Provision 2018 Corporate tax rate drops to 21 percent Equipment and software expensing increases Bonus depreciation is extended and expanded Net interest deductions are limited Net operating loss deductions are limited Domestic production activities deduction is repealed 2022 Amortization is allowed for research and experimentation expenditures Rules change for net interest deduction limitations 2023 Bonus depreciation extension and expansion phaseout begins 2026 Bonus depreciation extension and expansion phaseout completes Source: PWBM (2017b).
GENERAL DISCUSSION Alan Blinder had three comments related to the authors' assumptions. The first was the assumption that the net deficit effect will be financed with lump sum taxes. How the net deficit effect will ultimately be financed is uncertain; but according to Blinder, we know with absolute certainty that it will not be financed with lump sum taxes. Instead, it will be financed with something distortionary. He suggested a cleaner assumption would be to finance the net deficit effect with something akin to the average amount of distortion in the tax system. But the assumption of no distortions, he argued, is "absolutely wrong."
Second, Blinder was concerned about the Ramsey framework's assumption that the long-run real interest rate is constant, equal to the time-preference rate. Under this assumption, the long run is in essence the "infinity run," which in turn is subject to a whole host of its own assumptions--for instance, that everyone's behavior is essentially the same. He argued that under any policy-relevant time horizon--such as a decade or two--there is an upward-sloping supply curve of capital. Thus, the real interest rate likely moves above the time-preference rate, all things remaining equal, implying a smaller effect on capital formation. Echoing discussant Kent Smetters, Blinder argued that the basic real rate for federal borrowing should be thought of as the risk-free rate; and because the risk-free rate has been falling for decades, he concluded that the authors' interest rate assumptions are biased.
Third, Blinder echoed the point made by Furman that when there is full expensing, the corporate tax rate is irrelevant; lowering it only creates windfall benefits. Although coupling full expensing with cutting the corporate tax rate does not create a bias regarding the magnitudes of the effect on the growth rate, in a way it creates a bias toward the attractiveness of the policies.
William Gale applauded the authors for a heroic effort on a difficult issue and thanked them for producing a helpful paper. He noted that the paper reinforces the notion that the tax cut does not pay for itself. "This paper is just another nail in that coffin, but a very sophisticated, well-researched nail," he stated. However, Gale expressed concern that the authors may have overestimated the effects of the new tax law in the provisions-permanent scenario because of their choice for the prereform baseline, which assumes the temporary provisions in place before the 2017 tax law would have expired. To more accurately isolate the effects of the 2017 tax law in the provisions-permanent scenario, Gale argued that the authors should instead assume that expiring provisions are also made permanent in the baseline. This comparison would standardize assumptions about policymakers' behavior--namely, that they tend to extend temporary tax cuts--and would allow for a cleaner estimate of the impact of the 2017 tax law per se. In table 11, the authors estimate that extending prereform bonus depreciation rules would raise GDP by 0.3 percent after 10 years. Using this as the baseline, a more accurate estimate for the effect of the 2017 tax law on GDP in the provisions-permanent scenario would be 0.7 percent--from table 13, the authors' estimate of a 1.0 percent increase in GDP after 10 years with crowding out, less the 0.3 percent that would have occurred if the prereform temporary provisions had been extended.
Olivier Blanchard expressed two points. First, he wondered if one should be worried about a fall in research and development, which is presumably linked to total factor productivity. But perhaps the magnitude of the effect is small enough that it can be ignored. Second, he questioned the validity of the Cobb--Douglas functional form assumption, which implies unit elasticity of the capital--output ratio with respect to the user cost. (1) Blanchard noted that there is very little direct evidence on the long-run elasticity of capital with respect to the user cost. "We're taking on faith a fairly big part of the mechanism, and we might want to see whether there is any evidence that would support the assumption," he concluded.
John Haltiwanger turned the discussion toward the role of endogenous innovation and productivity, and in turn the role of entrepreneurs in that process. He noted that there is increasing empirical evidence among innovation-intensive sectors of the economy that start-ups and high-growth young businesses play a disproportionate role in accounting for productivity growth, innovation, and job creation, and that entrepreneurs at that stage are overwhelmingly pass-through entities. Therefore, it is not clear whether the tax cuts will have a large effect on innovation and productivity in that respect. A paper by Daron Acemoglu and others shows that subsidies to large, mature incumbents can have significant negative effects on innovation and productivity growth. (2) He suggested that the authors evaluate the new tax law in light of this finding.
Donald Marron endorsed the style of the paper and encouraged more authors to write in a way that brings people with different perspectives together. He criticized the type of policy analysis that happens "at the level of op-eds," which, he noted, tend not to be as precise or informed, and are not often collaborative. He asked the authors what indicators we should track in, say, five years' time to know which author's predictions are more correct. Most economists are comfortable with expectations of about 2 percent annual GDP growth, but the uncertainties are large compared to the differences in the present paper; so looking at top-level GDP growth is not likely to be what determines the most accurate estimates.
Robert Hall noted that the conversation thus far seemed to neglect a fact "embodied in the Bible of corporate finance": that the cost of capital is not the cost of funding. (3) Rather, the cost of capital incorporates the risk of the investment, and thus should not be used as a funding rate. This may explain why so-called hurdle rates are much higher than interest rates. Therefore, it would seem that the authors implicitly do not consider the personal taxation of C corporations in their model. Furman responded that the model's assumption of an infinite supply of capital implies that the personal taxation of C corporations does not change and that marginal finance is tax-free. Hall dissented, but agreed to talk more with the authors at a later time. "You're taking a strong stand on a controversial issue in public finance," he stated.
Next, Hall believes that a point missing from the tax reform discussion, in general, is that the introduction of expensing of business taxation of investment creates a consumption tax. Additionally, there has been a movement toward a consumption tax at the individual level, in the form of tax deferral. To avoid providing an inefficient subsidy of capital formation, there should be either a consumption tax administered at the personal level--that is, no tax on saving--or a first-year tax right off, but not both. There seems to be equal enthusiasm for both movements, he noted.
Finally, on the assumption of an infinite elasticity of the supply of capital (advocated by Barro), Hall pointed to research by Greg Kaplan and Giovanni Violante that thoroughly examines the issues of heterogeneity. (4) According to Hall, this work implies that Barro is "about two-thirds correct"; that is, about two-thirds of wealth is held under the conditions approximated by Barro's model. Therefore, he concluded, "We ought to take two-thirds of Barro's number, and one-third of Furman's number to get the truth."
Robert Gordon wondered if the authors' estimated growth rate for 2018 and 2019 of 0.9 percent was too high, a priori, given that the tax cuts amount to 0.75 percent of GDR The implied tax multiplier of 0.9 / 0.75 = 1.2 seemed counterintuitive to him, given that research by Valerie Ramey, among others, implies that government spending multipliers are barely above 1. (5) Alan Blinder and Mark Zandi estimated the corporate tax cut multiplier resulting from the American Recovery and Reinvestment Act to be roughly 0.3. (6) Nonetheless, supposing that the stimulus from the tax reform will be 0.9 percent growth, as Barro and Furman suggest, the Bipartisan Budget Act of 2018 adds another $300 billion of government spending, which is expected to add another 0.75 percent to growth, and also add to that an "unknown spillover" due to the one-third increase in the value of the stock market over the last two years, which is expected to raise consumption. Adding these stimuli to the steady observed GDP growth rate of 2 percent a year yields an implied demand-side GDP growth rate of about 4 percent a year for 2018 and 2019.
Gordon wondered, with all this demand-side growth, from where will the counterbalancing supply-side growth come? He noted that hours of work have been growing at around 1.6 percent a year for the past seven years, but this growth of hours required a reduction in the unemployment rate of 0.8 per year. A continuing reduction in the unemployment rate of 0.8 per year implies an unemployment rate in early 2020 of 2.5 percent, which would be unprecedented. By definition, the remainder of supply-side growth must come from productivity. Although modest productivity growth is expected, due to the extra investment generated by the tax cuts, he wondered whether it was even remotely plausible to imagine a jump in productivity growth in the total economy from the 0.6 percent achieved in the last 7 years to above 2 percent a year in 2018-19, which would be required to balance the demand-side stimulus. If supply-side growth does not occur, inflation may rise and cut off some of the demand-side stimulus on real activity, he concluded.
Mark Mazur liked the paper, and thinks it reflects a growing consensus on the way to look at the effects of the 2017 tax reform. He expressed two minor quibbles. The first was related to the supposed efficiency gains due to businesses reorganizing as C corporations. The majority of the shift will involve S corporations and limited liability companies converting to C corporations, which effectively comes down to corporations "checking a box" to indicate they want to be taxed at the entity level, which Mazur argued should have no effect on efficiency. Second, he explained that, though international provisions are often portrayed as raising revenues over the 10-year budget period, most of the revenue is raised all at once, after a lump sum tax is levied on unrepatriated foreign earnings. Therefore, he thinks it is incorrect to treat the international provisions as having a revenue effect over an extended period.
Richard Cooper was reminded of the unfavorable comparison of U.S. corporate taxes with those of foreign countries during the debate leading up to the tax reform. He argued that the authors' model was sufficiently general, in that it should apply to the European countries as well as the United States. Thus, the effects of similar reforms in Europe should apply in the same way as in the authors' model. He hypothesized that this was not actually true, and suggested more testing was needed. Although European investment has been increasing, he suggested that this is because world demand is increasing, not because of European tax changes.
George Perry appreciated the paper's demonstration that the tax cuts will not pay for themselves. But while he understood the paper's attempt to avoid political judgments, he believes the analysis would be more useful if it addressed the fiscal gap that will undoubtedly emerge as a result of the new tax law. Without making predictions or judgments about the future political climate, such an analysis would add realism to the paper to point out that maintaining the present tax package would require large cuts in spending, and to note that the cuts would have to fall heavily on the middle and lower income groups that now benefit from federal transfer, health, and retirement programs.
Robert Barro was absent from the discussion due to illness, so N. Gregory Mankiw, "having been Robert's colleague for 30 years," took the opportunity to respond to some of the points raised on his behalf. On Blinder's point about future increases in distortionary taxes, Barro might have responded that government spending could be cut in the future to reduce the distortions. That is, there are two sides to the issue--spending and taxes--and Barro likely views the lump sum assumption as a moderate compromise. On Gordon's point about the corporate tax multiplier, Barro might have argued that Gordon was relying on the textbook Keynesian model, in which tax multipliers have to be smaller than spending multipliers. Although Mankiw, the author of several of the most widely used economics textbooks, "may love the Keynesian textbook model," there is evidence that tax multipliers are actually bigger than spending multipliers. He pointed to research by Christina Romer and David Romer that suggests tax multipliers are much larger than spending multipliers, which does not fit the textbook Keynesian model. (7) If anything, this speaks poorly of the Keynesian model, and suggests that incentives may play a more significant role.
Furman briefly responded to several of the questions raised. He agreed with Mankiw's rebuttal of Blinder's point about the efficiency cost of financing and distortionary taxes. Furman believes it does not make sense to assume that the efficiency cost of the financing will be the same as in the previous system; discretionary spending could be cut to offset the tax distortions. Further, a value-added tax would be a more efficient way to collect revenue.
The most significant misimpression Furman heard during the discussion involved the relative size of tax changes for businesses and individuals. He argued that it is appropriate to include pass-throughs with corporations, as the authors modeled; it is also important to disregard issues of timing. In the case of the law as written, there is in essence no change to the individual side in the long run because the provisions expire after 2025, so the individual side is not that important in the grand scheme of things. Further, as a result of these expiring provisions and other complications arising from the use of the chained consumer price index, it is very difficult to precisely analyze the individual income tax changes.
Furman was enthused by Haltiwanger's points about the role of endogenous innovation and productivity. In looking at the advantages of corporate form, Furman admitted that he and Barro had considered only one side of the issue, and thought it would be interesting to look at the other side. On Marron's point about the medium-run indicators of the model's predictive success, Furman believes nothing in the macroeconomic data is going to confirm one way or the other which of the authors is correct in his predictions, because the predicted effects are very small compared with the amount of variation. Micro data may be useful to show which corporations were affected. He also agreed with Cooper that it would be interesting to apply the model to Europe.
On Blinder's point about the infinity run, Furman joked that "infinity is a long time from now, and we are quite explicit that we only get 40 percent of the way there in 10 years." He believes this assumption should allay some of the concerns about the issues inherent in assuming a long run of infinity.
Finally, Furman was intrigued by Smetters's comments about the discount rate, acknowledging that he and Barro did not thoroughly explore this issue in the paper. The authors treat depreciation allowances like a corporate bond, in which a certain payment is promised every year, implying a risk-free rate. However, it is conceivable that the discount rate could be a function of the tax system. For example, the final tax bill removed net operating loss carrybacks. Most economists would have preferred an improvement of net operating loss carrybacks--as was proposed in the House of Representatives' version. Perhaps, then, the removal of net operating loss carrybacks should be modeled as increasing the discount rate that one should use for depreciation allowances. It may be important to think about not just when write-offs will occur but also how to change the discount rate that is used for the write-offs.
(1.) The Cobb-Douglas functional form assumption dates back to a seminal paper by Dale Jorgenson. See Dale W. Jorgenson, "Capital Theory and Investment Behavior," American Economic Review 53, no. 2 (1963): 247-59.
(2.) Daron Acemoglu, Ufuk Akcigit, Harun Alp, Nicholas Bloom, and William R. Kerr. "Innovation, Reallocation and Growth," Working Paper no. 18993 (Cambridge, Mass.: National Bureau of Economic Research, 2017).
(3.) Richard A. Brealey, Stewart C. Myers, and Franklin Allen, Principles of Corporate Finance, 12th ed. (New York: McGraw-Hill Education, 2017).
(4.) Greg Kaplan and Giovanni L. Violante, "A Model of the Consumption Response to Fiscal Stimulus Payments," Econometrica 82, no. 4 (2014): 1199-239; Greg Kaplan. Giovanni L. Violante, and Justin Weidner, "The Wealthy Hand-to-Mouth." Brookings Papers on Economic Activity, Spring 2014: 77-138.
(5.) For a review of the literature, see Valerie A. Ramey, "Can Government Purchases Stimulate the Economy?" Journal of Economic Literature 49, no. 3 (2011): 673-85; see also Valerie A. Ramey and Sarah Zubairy, "Government Spending Multipliers in Good Times and in Bad: Evidence from US Historical Data," Journal of Political Economy 126. no. 2 (2018): 850-901.
(6.) Alan S. Blinder and Mark Zandi, "How the Great Recession Was Brought to an End," July 27, 2010, https://www.princeton.edu/~blinder/End-of-Great-Recession.pdf.
(7.) Christina D. Romer and David H. Romer, "The Macroeconomic Effects of Tax Changes: Estimates Based on a New Measure of Fiscal Shocks." American Economic Review 100, no. 3 (2010): 763-801.
|Printer friendly Cite/link Email Feedback|
|Publication:||Brookings Papers on Economic Activity|
|Date:||Mar 22, 2018|
|Previous Article:||Macroeconomic Effects of the 2017 Tax Reform.|
|Next Article:||Liquidity Crises in the Mortgage Market.|