Together, the sweeping tax cuts enacted in the United States at the end of 2017 and the spending package enacted in February of this year are expected to add $276 billion in fiscal stimulus to the US economy this year, or 1.4 percent of GDP. But the actual impact on US economic growth may turn out to be lower than meets the eye. As we explain in this blog, the impact of fiscal stimulus on economic activity and output varies with the business cycle. Existing projections of the impact of the Trump stimulus measures—the "multiplier" effects—miss (or ignore) the fact that the US economy is operating at nearly full potential. Based on the current state of the US economy, we predict that the Trump fiscal stimulus will yield an extra boost to GDP of 0.5 percent by 2020, instead of 2.1 percent if this dimension is not taken into account.
Budget Impact of Tax and Spending Measures
President Trump signed the Tax Cuts and Jobs Act of 2017 (TCJA) and the Bipartisan Budget Act (BBA) into law on December 22, 2017 and February 9, 2018, respectively. The Joint Committee of Taxation (2017), a nonpartisan committee of the US Congress, estimates that through substantial cuts to statutory tax rates for individuals, pass-through businesses, and corporations, TCJA will increase the budget deficit by $136 billion in fiscal year 2018 and $280 billion in fiscal year 2019. Translated into calendar year, this corresponds to $206 billion in 2018, almost 1.1 percent of GDP, and $275 billion in 2019, 1.4 percent of GDP (table 1). Macroeconomic Advisers (2018) estimates that by raising "caps" on discretionary budget authority for fiscal years 2018 and 2019, providing disaster relief, and authorizing four more years of funding for the Children's Health Insurance Program, BBA will further increase the budget deficit by $70 billion, or 0.36 percent of GDP, in 2018. Together, TCJA and BBA are thus estimated to induce a fiscal stimulus of $276 billion or about 1.4 percent of GDP in calendar year 2018 (table 1).
|Table 1 Impact on budget balance|
|TCJA (billions of dollars, FY)||–135.7||–280||–258.8|
|TCJA (billions of dollars, CY)||–205.7||–274.7||–249.3|
|TCJA (as percent of 2017 GDP, CY)||–1.06||–1.42||–1.29|
|BBA (billions of dollars, CY)||–70||–85||–85|
|BBA (as percent of 2017 GDP, CY)||–0.36||–0.44||–0.44|
|Total (billions of dollars, CY)||–275.7||–359.7||–334.3|
|Total (as percent of 2017 GDP, CY)||–1.42||–1.86||–1.72|
|CY = calendar year; FY = fiscal year
Notes: 2017 GDP = $19,386.8 billion (Bureau of Economic Analysis, January 26, 2018). Calendar year estimates for TCJA calculated using following formula:
|CY = FY +.25*FY. Otherwise, CY[t]= 0.75*FY[t] +.25*FY[t+1].|
|Sources: Joint Committee on Taxation (2017) for TCJA in FY; Macroeconomic Advisers (2018) for BBA in CY; authors’ calculations for TCJA in CY using formula above.|
Large Effects with Average Multipliers
We start from a simple framework similar to that of Mertens (2018), where the causal effect of a fiscal policy change of size Δfpt in period t on GDP growth is given by γh in the following equation:
1. Δyt+h = γhΔfpt + Δȳt+h
where Δyt+h is future growth observed after h ≥ 0 years, and Δȳt+h is counterfactual future growth in the absence of the change in fiscal policy.
Estimates of γh can be obtained from regression-based models that exploit historical fiscal policy changes. The main advantage of this approach compared with more structural approaches is that there is no need to make explicit behavioral and equilibrium assumptions. In particular, there is no need to assume a particular dynamic for expected future tax rates, which is instead part of the estimation.
The key difficulty in the regression-based approach is simultaneity. Given that many if not most changes in fiscal policy are made in reaction to current or expected cyclical conditions, a naïve regression of Δyt+h on all changes in fiscal policy would lead to inconsistent estimates of the parameter γh. The idea is thus to exploit changes in fiscal policy that are quasi-random or "exogenous," Δfptexo. In this case, the causal effects of fiscal policy can be obtained as the least-squares estimates of δh in regressions of the form:
2. Δyt+h = δhΔfptexo + Δut+h
Panel A of table 2 reports different estimates of linear fiscal multipliers obtained from direct regressions similar to equation (2). The first three rows reproduce the linear multipliers used in Mertens (2018). Applying an average of these tax multipliers (δh = –1.56) to the projected revenue impact of the TCJA (Δfptexo = Δτt = –1.06% of 2017 GDP in Year 1, where Δτt is the change in tax liabilities induced by the TCJA), Mertens (2018) finds that real GDP is expected to be higher by 1.65 percent by the end of 2020, with most of the growth frontloaded in 2018. 
The fourth row of panel A presents fiscal multipliers from our own recent empirical work on large fiscal expansions in OECD economies over the postwar period (Cohen-Setton, Gornostay, and Ladreit 2018). In this work, we find large fiscal expansions in a panel of 16 OECD countries between 1960 and 2007 by looking at large declines in cyclically adjusted fiscal balances. We then identify fiscal expansions that were not motivated by countercyclical reasons and use them in estimations similar to equation (2). As in Mertens (2018), our linear multipliers predict a sizeable increase in GDP of 1.36 percent by the end of 2020.
Small Effects with Multipliers that Account for the State of the Economy
These projections, however, fail to account for the possibility that fiscal multipliers may vary over the business cycle. A recent literature has investigated this question in the context of government spending multipliers. But there is no reason to believe that the mechanisms that make multipliers dependent on the state of the economy apply only to changes in government expenditures. Panel B of table 2 reports what this literature calls "state-dependent" multipliers from two recent studies (Auerbach and Gorodnichenko 2012, Ramey and Zubairy 2018) and from our work. One advantage of our multipliers over these two studies is that the fiscal policies underlying our model estimates are closer to TCJA and BBA.
|Table 2 Estimates of fiscal multipliers
(percentage point change in real GDP growth per 1 percent of GDP change in primary balance in Year 1)
|Study||Year 1||Year 2||Year 3||Cumulative||Type of fiscal change/definition of state|
|A. Linear tax multipliers|
|Romer and Romer (2010)||–1.27||–0.78||–0.54||–2.59||tax / n.a.|
|Favero and Giavazzi (2012)||–1.16||0.10||0.29||–0.77||tax / n.a.|
|Mertens and Ravn (2012)||–1.24||–1.10||1.02||–1.31||tax / n.a.|
|Cohen-Setton, Gornostay, and Ladreit (2018)||–0.39||–0.57||–0.32||–1.28||stimulus / n.a.|
|Average without Cohen-Setton, Gornostay, and Ladreit (2018)||–1.22||–0.59||0.26||–1.56|
|Average with Cohen-Setton, Gornostay, and Ladreit (2018)||–1.01||–0.59||0.11||–1.49|
|B. State-dependent multipliers|
|Auerbach and Gorodnichenko (2012)||–0.91||–0.50||–0.30||–1.70||spending /recession|
|Ramey and Zubairy (2018)||–0.22||–0.38||–0.04||–0.64||spending / slack|
|Cohen-Setton, Gornostay, and Ladreit (2018)||–0.34||–0.51||–0.73||–1.57||stimulus / recession|
|Cohen-Setton, Gornostay, and Ladreit (2018)||–0.67||–0.48||–0.32||–1.47||stimulus / slack|
|Auerbach and Gorodnichenko (2012)||–0.15||0.03||0.08||–0.05||spending / expansion|
|Ramey and Zubairy (2018)||–0.71||0.12||–0.08||–0.68||spending / no slack|
|Cohen-Setton, Gornostay, and Ladreit (2018)||0.31||–0.38||0.20||0.14||stimulus / expansion|
|Cohen-Setton, Gornostay, and Ladreit (2018)||–0.32||–0.46||–0.11||–0.89||stimulus / no slack|
n.a. = not applicable
Like Auerbach and Gorodnichenko (2012), we find (cumulative) multipliers that are large and above 1 in bad times but much lower in good times. As illustrated in table 3, accounting for the present US economic conditions generates a first-order difference in the estimated impact of TCJA and BBA. State-dependent multipliers yield a boost in output that is only one-third or one-quarter of the boost predicted by linear multipliers, which do not account for the stage of the business cycle. One can disagree on which state-dependent multiplier is better or on whether there is still slack in the US labor market, but it is misleading to assume away the state of the economy when evaluating the likely impact of the Trump fiscal package.
|Table 3 Summary of results of the impact of TCJA and BBA (percentage points relative to counterfactual real US GDP growth)|
|Impact of TCJA|
|Using linear multipliers||1.08||0.62||–0.12||1.58|
|Using state-dependent multipliers||0.23||0.18||–0.02||0.39|
|Impact of BBA|
|Using linear multipliers||0.37||0.21||–0.04||0.54|
|Using state-dependent multipliers||0.08||0.06||–0.01||0.13|
|Impact of TCJA and BBA|
|Using linear multipliers||1.44||0.83||–0.16||2.11|
|Using state-dependent multipliers||0.31||0.24||–0.03||0.52|
|Source: Authors’ calculations.|
Auerbach, Alan J., and Yuriy Gorodnichenko. 2012. Measuring the Output Responses to Fiscal Policy. American Economic Journal: Economic Policy 4, no. 2: 1–27.
Cohen-Setton, Jeremie, Egor Gornostay, and Colombe Ladreit. 2018 (forthcoming). Large Fiscal Expansions in OECD Countries: Identification and Effects. PIIE Working Paper. Washington: Peterson Institute for International Economics.
Joint Committee on Taxation. 2017. Estimated Budget Effects Of The Conference Agreement For H.R.1, The "Tax Cuts And Jobs Act." JCX-67-17 (December 18). Washington.
Macroeconomic Advisers. 2018. The Bipartisan Budget Act of 2018: Implications for the Forecast. Macro Focus, March 21.
Mertens, Karel. 2018. The Near-Term Growth Impact of the Tax Cuts and Jobs Act. Mimeo. Available at https://karelmertenscom.files.wordpress.com/2018/03/tcja_impact_march23.pdf (link is external).
Ramey, Valerie A., and Sarah Zubairy. 2018. Government Spending Multipliers in Good Times and in Bad: Evidence from US Historical Data. Journal of Political Economy 126, no. 2: 850–901.
1. Mertens (2018)'s central estimate is that TCJA will yield a level of GDP 1.3 percent higher by the end of 2020. This number is obtained by applying the same fiscal impulse (–1.1 percent of GDP) to tax multipliers obtained by averaging over six studies: three studies that use a direct regression models and are reported here in table 1 and three other studies that use structural vector autoregressive (SVAR) models. In Cohen-Setton, Gornostay, and Ladreit (2018) we follow a direct regression approach and so restrict our comparison to the estimates obtained from direct regression models in Mertens (2018).
2. A draft of the paper is available from the authors upon request.