Does the U.S. Federal Deficit Respond More to Revenues,Expenses,or GDP Growth? A Time-Varying Coefficient Analysis

Abstract

The feasibility of fiscal stimulus depends on levels of sovereign debt and future impacts on budget deficits, regardless of whether it is effective at increasing economic growth. This paper tests whether changes in revenues, expenses, GDP growth rates, private investment, and monetary stimulus affect the United States federal deficit one year after the change, taking account of fiscal multiplier effects. A state space model with time-varying coefficients from 1963 through 2019 shows that the deficit responds somewhat more to expenses, especially social insurance outlays, implying that entitlement reform may be the single best policy for fiscal solvency. For much of the time period, revenues and expenses are insignificant except in times of major upheaval and rapid change, showing the political nature of the problem and demonstrating that the solution must involve both revenues and expenses simultaneously, with careful attention to multiplier effects.

Keywords

government budget deficit revenues expenses

Introduction

The effect of fiscal stimulus on GDP, whether from increased spending or tax cuts makes for a good political talking point, but effects on budget deficits are less popular to discuss. When debt levels become unsustainable, discussing fiscal contraction becomes inevitable, which means either tax increases or spending cuts designed to reduce the deficit. This paper addresses this issue, namely whether changes in revenues or expenses exert a greater effect on the U.S. federal deficit over subsequent years. Either course of action – tax cuts or spending increases – could have fiscal multiplier effects which could adversely affect future revenues, possibly leading to a smaller long term impact on the deficit than is desired.

Arguments about revenues, expenses, and GDP center on the fiscal multiplier and the Laffer Curve. The fiscal multiplier primarily concerns future GDP growth, and the Laffer Curve, which is subject to much controversy, pertains to the potentially nonmonotonic relationship between tax rates and revenues. This paper, in contrast, simultaneously looks at revenues and expenses and their effects on future budget deficits, which follow any effects on GDP, and which may exist without substantial changes in GDP. The econometric model in this paper does not a priori assume specific multipliers or the presence of a Laffer Curve effect, and it finds some basis in fiscal multiplier effects, less support for the Laffer Curve, and points strongly toward time-inconsistent political behavior as a cause of persistent deficits.

Strong opinions abound about the fiscal multiplier and Laffer Curve. Although the Laffer Curve has some good intuition, its simplicity severely limits its policy relevance. By implicitly assuming homogeneous labor supply and a single marginal tax rate, it overlooks the complexity of the economy and the structure of American tax codes that includes myriad deductions and exemptions, as well as different rates for different types of income. As a result, reductions in top marginal tax rates motivated by the Laffer Curve may simply reduce revenues, and they could have other adverse effects such as increased inequality. At the same time, direct revenue effects from tax rate changes may exist independent of any indirect effects through GDP, and these likely run in opposite directions. Moreover, even if a tax cut did increase revenues, this could stimulate legislators’ desire to spend more, and spending increases can affect the perceived necessity for additional revenues. The recent history of American budget deficits does not show substantial reductions following tax cuts. This could be a result of starting to the left of the peak, provided that one single identifiable peak exists, faulty arguments about which tax rates to change, or increased spending. Regardless of the cause, increased deficits could portend future fiscal crises.

By evaluating the deficit apart from economic growth, this paper shows the feasibility of fiscal stimulus, which is independent of its effect on economic growth. To determine whether revenues or expenses have the greater effect on the deficit, this paper uses a state space model with time-varying coefficients. This approach controls for varying political conditions over time, as well as changes in the composition of revenues and expenses. By quantifying effects on the deficit from changes in revenues and expenses while controlling for the other, specific changes in tax policy or spending levels can be evaluated. This sheds light on both the fiscal multiplier and Laffer Curve arguments for their effects on subsequent budget deficits.

This paper separately evaluates the total deficit and the on-budget deficit, which excludes social insurance contributions and outlays. Relationships between variables change over time, showing the value of the time-varying coefficient approach. In some cases, both revenues and expenses are insignificant, presumably because decisions about revenues and expenses are made simultaneously. Though effects on GDP undoubtedly exist, future deficit reduction becomes a tenuous argument due to the time inconsistency of politicians. Thus, Laffer Curve arguments may be dubious from a revenues standpoint, but from a budget deficit standpoint, they are empirically even more tenuous, per this paper’s analysis.

When they are significant, expense changes exert a somewhat larger impact on future deficits than revenues, though the problem is political and multifaceted, and there exist asymmetric feedback effects. The state-space time-varying coefficient in this paper’s analysis controls for GDP growth, private nonresidential investment, and monetary stimulus. These controls matter, and GDP growth often lacks positive effects on structural deficits. This is a useful insight because, although it is debated whether economies grow more slowly amid high sovereign debt levels, even steady GDP growth may not solve a structural budget deficit. Investment matters somewhat, but only when including social insurance receipts and outlays in the calculation of the deficit. This indicates that investment as a source of persistent growth does reduce deficits when the spending is more or less automatic as opposed to politically motivated. Monetary stimulus ends up influencing deficits through interest expenses, which become more important as total debt levels grow with persistent deficits.

By estimating time-varying coefficients, this paper leaves the fiscal multiplier in the unseen middle of the model. Should this effect be present, or if there are any Laffer Curve effects, there may not be a sufficient basis for policies that are intended to reduce deficits. In this paper’s time-varying coefficient approach, the absence of deficit reduction effects could be due to the absence or weakness of multiplier or Laffer Curve effects, the presence of contemporaneous political pressures for spending, or time inconsistency in the legislative appropriations process. One contribution of this paper is to evaluate the wisdom of revenue and expenditure policies for their effect on deficits, as opposed to GDP and employment effects.

This paper proceeds with a review of relevant literature to include fiscal multipliers, deficit-oriented issues and policies, and political incentives regarding fiscal matters. An overview of data sources follows, from which the paper proceeds with an explanation of the state space model and its results.

Literature Review

Sovereign debt and budget deficits are important issues when evaluating economic growth, the role of government in the economy, and the ability to respond to future crises. Reinhart and Rogoff (2010) attracted much attention for their work indicating that economic growth rates fall when sovereign debt reaches a certain level, showing the risks of accumulated budget deficits. Subsequently, Herndon et al. (2014) conducted a replication analysis of Reinhart and Rogoff (2010) which corrected a calculation error and showed that growth rates did not substantially change when public debt exceeded 90% of GDP. Reinhart and Rogoff (2010) responded to their critics and conducted follow-up work on the topic (Reinhart et al., 2012). Their conclusions, however, remain controversial and the issue is far from settled. Later, Heimberger (2022) conducted a meta-regression analysis of similar studies and was unable to rule out a null effect, casting further doubt on Reinhart and Rogoff, despite their answer to their earlier critics.

Deficits and debt levels are related but different issues. Deficits, the focus of this paper, do cause higher levels of sovereign debt as a tautology, but the absolute increase may not increase the fiscal burden if the debt-to-GDP ratio is fairly constant or if interest remain low. Moreover, there is a chance that deficits could induce faster GDP growth and thereby reduce the debt burden, a result found by Taylor et al. (2012), though that was specific to deficits incurred specifically for investment. Relatedly, Peppel-Srebrny (2021) showed that sovereign debt for investment should lower bond yields, unlike other deficits. Despite potential positive effects, a growing federal debt burden in the United States shows that deficits have been accumulating faster than economic growth can offset the burden. Thornton (2012) traces this problem back to the 1970s, when tax cuts coincided with increased spending, leading to a structural problem. Although recent U.S. federal deficits have not been primarily due to investment, investment is an explanatory variable in this paper’s because it could affect future deficits through GDP growth and resulting tax revenue.

The Congressional Budget Office evaluates fiscal impacts of legislation over a ten-year horizon, but Peterson (2024) believes that material reductions in this debt require a much longer horizon, perhaps twenty or thirty years, with emphasis on entitlement reform. Reducing the debt burden requires either massive economic growth or bringing the budget into surplus either through revenues or expenses, or both. Chen and Imrohoroglu (2015) believe that this will require labor income tax rates exceeding 40%, which will have a significant adverse effect on economic growth. Focusing on expenses, Makin and Layton (2021) point out that the fiscal response to the COVID-19 pandemic was so devastating in terms of federal debt that fiscal consolidation, or austerity, will be necessary at some future point. Taking an interesting angle, Bohn (1998) showed that the federal government followed an intertemporal constraint when excluding interest payments, although the fiscal landscape of the United States has changed dramatically since that time. Moreover, interest payments are an inevitable component of current deficits when total levels of sovereign debt are high.

The extent to which government spending or tax cuts exert a positive effect on GDP growth is much contested. For this paper, the question is more nuanced because the issue is whether such spending materially affects future deficits, not GDP. Should the multiplier be less than one, there is an obvious problem. A positive effect on the deficit would require a cumulative multiplier exceeding one by at least the amount of the effective tax rate and no increase in expenses. Auclert et al. (2024) showed that cumulative multipliers can exceed one, consistent with Bouakez et al. (2023), who showed that multipliers are larger in multisector economies, such as the United States, due to input-output links. Similarly, Leeper et al. (2017) showed that multipliers can be larger under a regime of passive money and active fiscal policy. Linking taxes and spending, Ferriere and Navarro (2025) believe that multipliers are larger if spending is financed by taxes on high-income earners, thus linking these effects to the progressivity of the tax code. These results about large multipliers are far from universal, though. Brinca et al. (2016) believe that fiscal multipliers depend on the quantity of people facing binding credit constraints. Focusing on indebted governments as opposed to individuals, Huidrom et al. (2020) showed that fiscal multipliers are smaller when the government is in a weak fiscal position. Slack is often present in the economy, especially when stimulus is perceived as a necessity, and Ramey and Zubairy (1984) believe that this slackness causes the multiplier to be less than one. Effects may not be symmetric, as Barnichon et al. (2022) note, who showed that the contractionary multiplier is substantially more than one, but the expansionary multiplier is less than one. This is worrisome because austerity measures may not quickly solve a debt crisis if this is the case, and expansionary fiscal stimulus would also be counterproductive.

Austerity has its own proponents and detractors. Alesina et al. (2018) show that spending cuts have small output costs and may even be expansionary, but adjustments on the tax side are more costly for output and could even cause recessions. Calcagno (2012) sharply countered this view, believing that austerity policies are very misguided, a conclusion that this paper supports. This paper looks at the effects on deficits one year after policies are implemented, leaving time for effects to become clear, and for politicians to take additional action. Austerity measures can be deeply unpopular and may be hard to implement or maintain, and time-inconsistent politicians may simultaneously pursue policies with contrary effects. Gabriel et al. (2023) even show that austerity leads to increased vote share for extreme parties, fragmentation, and decreased voter turnout. These policies may have even caused Brexit (Fetzer, 2019).

Political factors are inseparable from any discussion of revenues and expenses and their effects on debt and deficits. A flypaper effect has been observed for intergovernmental grants when, after receiving a grant, spending permanently increases by more than the grant. This effect was first put forward by Hamilton (1983) and consistently noted since then (Hines and Thaler, 1995). Conversely, Becker (1996) showed that these results were very sensitive to specification and there may be no such effect. This depends somewhat on circumstances of tax collection efficiency and local autonomy (Rios et al., 2022). This paper is about the United States federal government, which gives intergovernmental grants as opposed to receiving them. The underlying idea of a flypaper effect may still be relevant, though. If money sticks where it first lands, like a fly on flypaper, then an increase in revenues may not bring the budget balance any closer to surplus because politicians may be inclined to spend the money and keep the deficit at the pre-existing status quo. Thus, even though higher revenues could reduce the deficit, this only happens when spending stays flat or decreases, which is a political decision, subject to the same sort of incentives that Niskanen (1968) put forward when characterizing bureaucrats as budget-maximizers with a use-it-or-lose-it constraint. The political nature of this process was discussed by Meyers (2014), who showed that the federal budget process in the United States now favors political brinkmanship and attempts to outmaneuver the other side.

Data

This paper focuses on the United States federal budget deficit. The dependent variable of interest is the federal deficit, for which this paper uses two calculations and estimates the same models separately for each. Table 1 lists all variables entering the model and their sources and units. The first calculation of the deficit is total revenues less total expenses, which does include changes in the capital account, and which this paper terms the total deficit. It is essential to include the capital account because many of its transactions reflect government expenses, such as federal grants-in-aid to states for block grant programs or transfers from the federal Highway Trust Fund to states for road construction. The second calculation, termed the on-budget deficit, omits social insurance contributions from total revenues and social benefits from total expenses.

Table 1.

Sources of Variables.

Variable	Source	Identifier
Total Revenues	BEA $^{1}$ Table 3.2	Line 40, pct. of nom. GDP
Total Expenses	BEA Table 3.2	Line 43, pct. of nom. GDP
Total Deficit	BEA Table 3.2	Diff. b/t preceding two rows
Revenues exc. Social Insurance	BEA Table 3.2	Diff. b/t lines 40 and 10, pct. of nom. GDP
Expenses exc. Social Insurance	BEA Table 3.2	Diff. b/t lines 43 and 27, pct. of nom. GDP
On-budget Deficit	BEA Table 3.2	Diff. b/t preceding two rows
Real Per Capita GDP Growth	FRED $^{2}$ GDPC1	Annualized pct. change
Private Nonresidential Investment	FRED Series PNFI	Pct. of nom. GDP
Fed. Purchases of Treasury securities	FRBOG $^{3}$ FA713061125.Q	Pct. of M2 Money Supply
Nominal $^{4}$ GDP	FRED Series GDP	$/billions
Nominal M2 Money Supply	FRED Series M2SL	$/billions

Bureau of Economic Analysis.

Federal Reserve Economic Data.

Federal Reserve Board of Governors.

Percentages of nominal GDP or M2 normalize for inflation.

The difference between these two calculations is not equal to the change in the balance of social insurance trust funds, resulting in an on-budget deficit or surplus that differs from what is in many other publications. When social insurance contributions exceeded revenues, the balance was invested in U.S. Treasury securities that paid a market return. The social insurance trust funds were thus intragovernmental debt, or an earmarked obligation to pay future expenses, which obligation earned interest. Thus, many federal interest payments over the years served to increase these trust funds, or intragovernmental debt. By completely omitting social insurance contributions and payments, the on-budget deficit shows federal expenses and revenues as though these programs never existed. Interest payments to these funds are thus eliminated from this calculation of the on-budget deficit. As a result, the on-budget surplus was considerably larger in the 1990s using this calculation that omits these interest payments compared to what is typically reported.

When estimating the deficit, lagged revenues and expenses are the variables of greatest interest. If changes in either can predict future deficits, then there are Keynesian multiplier effects, Laffer curve effects, or political decisions that follow these changes. Revenues and expenses are expressed in first differences as a percent of GDP.

Each model includes three other controls – private nonresidential investment, real per capita GDP growth, and Federal Reserve purchases of Treasury securities (monetary easing). Investment is first differenced as a percent of GDP, and Fed purchases are expressed as a percent of M2 money supply (purchases are changes in asset levels are thus analogous to a difference). Including these purchases is an important control because asset purchases are designed to affect interest rates, which affect government expenses, but they have a smaller effect when the money supply is large. These monetary stimulus purchases have increased in absolute numbers and as a percent of GDP, but there has been no such trend as a percent of M2.

For stationarity, all variables always enter their respective models in first differences. Table 2 shows results of ADF tests on all variables to demonstrate stationarity in first differences, but not in levels, which is the basis for the use of the first difference transformation in this paper’s models. Table 2 does not report a value in levels for Federal Reserve purchases of Treasurys because such purchases are a change and are analogous to a first difference; the corresponding level would be total holdings of Treasury securities.

Table 2.

Augmented Dickey-fuller Stationarity Tests on Variables.

	Level	First Difference
Variable	ADF Stat.	ADF Stat.
Total Deficit	–1.007	–5.003 $^{* * *}$
On-budget Deficit	–1.690 $^{*}$	–5.924 $^{* * *}$
Total Revenues	–0.196	–5.943 $^{* * *}$
Revenues exc. Social Ins.	–0.809	–4.789 $^{* * *}$
Total Expenses	–0.288	–4.334 $^{* * *}$
Expenses exc. Social Ins.	–1.158	–5.175 $^{* * *}$
Pvt. Nonresidential Investment	0.343	–5.857 $^{* * *}$
Fed Purch. Treasury Securities	–	–6.548 $^{* * *}$

$^{*} p < 0.1$ , $^{* *} p < 0.05$ , $^{* * *} p < 0.01$

In all ADF tests, Schwarz-Bayesian Information Criterion is minimized with one lag.

Note that all variables in this table are expressed as a percent of GDP.

Purchases of securities are only expressed in differences as total assets are the level.

All explanatory variables enter each model as four-quarter lags except GDP growth, which is the percentage change from four quarters prior to the current period. This reflects lags when tax rates change, which is typically at the start of a calendar year. It also provides time for GDP growth to be reflected in tax payments, for investment to begin yielding returns, and for revenue and expense decisions to have detectable results. Whenever revenues, expenses, investment, or Fed monetary action change, the models in the next section capture the effect on the deficit one year later, after feedback effects can take place, conditional on GDP growth during that time.

Deeper lag orders run the risk of obscuring subsequent political action that could affect the deficit, and fewer lags may not let enough time elapse for effects to become apparent. The choice of four may seem somewhat arbitrary, but it performs well on metrics such as avoiding residual autocorrelation. Nevertheless, because arguments could be made for more or fewer than four lags, other specifications are estimated as robustness checks in Section “Robustness Tests”, although diagnostics showed they were inferior to using four lags based on residual autocorrelation.

Econometric Analysis

The objective of the econometric approach in this section is to produce coefficients that yield fitted values of the deficit that track the observed values closely enough to eliminate any predictable variation in the residual error term. Complications center on three main factors, which are the political nature of underlying processes, macroeconomic feedback effects, changes in the composition of revenues and expenses over time, and typical econometric issues such as omitted variable bias and causality.

The components of the federal deficit, revenues and expenses, are affected by the current state of the economy and tax and spending regimes, political action by time-inconsistent actors subject to election cycles, and exogenous factors. As a result, neither the deficit nor its two components, revenues and expenses, follows a process that lends itself to easy modeling. Many political events involving both revenues and expenses throughout the period of analysis from 1963 to 2019 have been very substantive, and they include changes in the tax code, the composition of federal expenditures, demographic shifts that affect social insurance revenues and expenses, recessions, and fluctuations in private investment. As a result, innovations in the time series of revenues and expenses are not independent and identically distributed, let alone normally distributed. Changes in revenues, expenses, and Fed purchases of assets are all subject to this problem. GDP growth and investment are less susceptible, but both are influenced by a large set of explanatory variables.

Feedback effects are another complication, and these can occur on the demand side through the Keynesian multiplier or the supply side through the Laffer Curve. It is important to note that, because there are multiple competing explanations, significant results do not confirm the presence of Keynesian multiplier or Laffer Curve effects. Significance tests can reject the null of no effect, but the precise nature of the effect is not part of the hypothesis test.

When either of these demand-side or supply-side possibilities is realized, or if there is another unseen cause, changes in revenues or expenses have future effects on the deficit through future effects on either or both of revenues and expenses. The magnitude and direction of these changes may depend on the size of the policy change that precipitated them, prior levels of one or more variables, and contemporaneous changes in one or more variables.

A simple ARIMA model of changes in the deficit as a percent of GDP overlooks potential changes in the coefficients over time, which is a real possibility given the substantive political changes mentioned above. Changes in the composition of revenues and expenses would present a similar problem. A state space model with time-varying coefficients can address these concerns, and Kalman filtering of coefficients can provide estimates and fitted values that do eliminate residual autocorrelation.

Causal inference often presents a problem, and there is a question of whether deficits could cause changes in the level or composition of expenditures. When deficits or debt levels are large enough to be a noteworthy political issue, they may influence decisions about revenues or expenses. To reduce the chances of this problem, explanatory variables are lagged as is typical in most time series estimation. Current deficits should not cause prior expenses and revenue changes. This problem is not completely mitigated, however, as current deficits may affect future revenues and expenses which then affect future deficits. Thus, explanatory variables in some time periods may not be purely independent. The state space approach with time-varying coefficients helps, but when interpreting the model, it is important to understand that revenues and expenses are not purely a matter of choice as recent deficits may be affecting them. A complete analysis of this reverse causality issue is somewhat beyond the scope of this paper, but it is a direction worthy of future research.

Omitted variable bias can be an issue, and it can result from either the omission of meaningful explanatory variables or a suboptimal lag structure. Moreover, there may be common shocks to revenues and expenditures, such as the response to the COVID pandemic or the financial crisis that precipitated the Great Recession. The explanatory variables in the paper’s model include revenues and expenses, as well as things that could reasonably affect GDP such as investment, and Federal Reserve intervention to affect interest rates, which affects federal interest expenditures. When choosing a specification in Section “Model Diagnostics and Specification Selection”, the main criteria are stationarity and the absence of autocorrelation in residuals. Though not completely eliminating the risk of omitted variable bias, absence of a predictable pattern in the error term shows that there is not an obvious problem in with the set of explanatory variables. Contemporaneous shocks to both revenues and expenditures from the same source are less of an issue for econometric specification than for interpretation of results. When there are significant effects of revenues and expenses at the same time, care must be taken to evaluate the origin of the changes in revenues and expenses, and this is true more generally, as other variables that could reasonably affect the deficit likely do so through their effects on revenues or expenses.

State Space Model

The state space framework estimates an output equation as a function of underlying states that change over time. The dependent variable of the mean equation, which is the equation of primary interest, ( $y$ in (1) below) is the federal deficit. The explanatory variables, described in Section “Data”, are in $x$ . The underlying states of the model are the coefficients in $β$ which is given as a function of time $t$ . These coefficients are updated each period, reflecting the changes highlighted above.

y_{t} = x_{t - 4} β (t) + ε_{t} .

(1)

Independent variables are lagged by four quarters, as explained in Section “Data” so that coefficients show their effects on the deficit one year later. For robustness, I also estimated two additional specifications, one with two lags and one with eight lags, which are discussed in Section “Robustness Tests”. The primary specification, with four lags, has less autocorrelation than either two or eight in the state and output equations, providing an econometric basis for choice.

There is no intercept in $x$ because first differencing the other variables removes any deterministic intercept, so the significance of such a variable implies a linear trend in the growth of the deficit. A time-varying intercept would have no meaning that can be interpreted apart from unpredictable changes in the deficit, which should be confined to the error term, which can be tested for autocorrelation and stationarity.

The state equation, which shows the evolution of $β$ over time, is (2).

β_{t} = B β_{t - 1} + W, W \sim MVN (0, Q) .

(2)

The matrix

B

is estimated to produce estimates of

β

and

y

using the recursive maximum likelihood Kalman filter approach of Shumway and Stoffer (2025). This recursive approach uses the current measurement of the deficit,

y_{t}

, and the estimated prior value of the state vector,

β_{t - 1}

. Because

β_{t}

is the effect of

x_{t - 4}

y_{t}

, the model estimates the effect of lagged

x

on the current deficit

y_{t}

, utilizing the prior period’s

β

as a starting point. The effect of

x_{t - 5}

y_{t - 1}

is the point for estimating the effect of

x_{t - 4}

y_{t}

The matrix $B$ may or may not be constrained. The process of model selection is aided by estimating two specifications, one with $B$ unconstrained and another restricting it to be diagonal and unequal. When $B$ is unconstrained, one coefficient can affect other coefficients’ future values in addition to its own. Should this be the case, a change in, say, the coefficient on expenses in $x$ could affect the coefficient on revenues. Thus, a change in expenses could affect the way in which the deficit responds to revenues. This is more than just Keynesian-style feedback effects or Laffer Curve effects of tax cuts, which could exist even with diagonal $B$ . For example, supposing that feedback effects exist, consider an increase in federal expenses. This could lead to an increase in revenues, but no change in either coefficient apart from the state process, which diagonal $B$ imposes. With $B$ unconstrained, an increase in expenses, perhaps from more transfer payments as the population ages, could affect the coefficient on revenues, implying that revenues may have a greater or lesser effect on the deficit than before, depending on whether they have a greater or lesser multiplier effect.

The most pressing problem is that the variables in $x$ have many changes that are large in magnitude, but they affect the deficit in different directions and may affect one another. As a result, innovations in the state equations are not always normally distributed. The estimates of $β$ directly affect the estimated values of the deficit, $\hat{y}$ . The smoothed output of the Kalman filter can be conditioned on data up to time $t$ or the entire time series $T$ . Both approaches are useful for this paper’s analysis. When estimating the deficit conditional on data up to time $t$ , the Kalman filter starts at $t$ and goes backward, so the $β$ values reflect time-varying foreseeable effects on the deficit. When conditioning on the entire time series $T$ , the $β$ show time-varying ex post effects because the filter starts at the end of the time series. Concerning other matrices, the error term $W$ is time-invariant, and it must be unconstrained for $β$ to vary over time. Additionally, $Q$ is unconstrained, allowing for covariance of coefficients.

Model Diagnostics and Specification Selection

For each calculation of the deficit, one omitting transfer payments and the other including them, four specifications are possible. These include conditioning Kalman filter estimates of $β$ on $t$ or $T$ , and imposing diagonality on $B$ or leaving it unconstrained. All specifications include four lags, as explained above. Table 3 contains diagnostics for these specifications, including Augmented Dickey-Fuller (ADF) tests for unit roots and Ljung-Box tests for autocorrelation in the state and output equations. The task is to choose the most reliable of these based on residual stationarity and lack of autocorrelation. ADF tests for residual stationarity show that no specification has a unit root in the residuals of either the innovations or state equation (2) or the residuals of the output equation (1). Constraints on $B$ always lead to autocorrelation. When leaving it unconstrained and conditioning Kalman filter estimates of $β$ on $t$ , there is no autocorrelation in the state equation residual for either the total or on-budget deficit. For the output equation, the on-budget deficit equation lacks autocorrelation, but the total deficit still has some, although less than with other specifications.

Table 3.

Model Diagnostics and Selection.

	On-Budget Deficit				Total Deficit
Model	1	2	3	4	5	6	7	8
Restrictions on $B$	–	–	Diag.	Diag.	–	–	Diag.	Diag.
Time conditioning	$t$	$T$	$t$	$T$	$t$	$T$	$t$	$T$
	State Eq. Innovations				State Eq. Innovations
ADF Stat.	$- {8.4202}^{* *}$		$- {7.9091}^{* * *}$		$- {9.8638}^{* * *}$		$- {7.6833}^{* * *}$
Ljung-Box $Q$	2.1356		${3.1962}^{*}$		1.1023		${8.3489}^{* * *}$
	$\hat{y} - y$				$\hat{y} - y$
ADF Stat.	$- {8.5817}^{* * *}$	$- {13.1545}^{* * *}$	$- {7.9034}^{* * *}$	$- {12.2454}^{* * *}$	$- {8.9075}^{* * *}$	$- {13.9106}^{* * *}$	$- {7.7266}^{* * *}$	$- {12.548}^{* * *}$
Ljung-Box $Q$	2.1763	${22.722}^{* * *}$	${3.1251}^{*}$	${26.110}^{* * *}$	${5.1196}^{* *}$	${13.917}^{* * *}$	$- {18.407}^{* * *}$	${10.045}^{* * *}$
Preferred Spec.	✓				✓
$B_{2, 1}$ (rev $\to$ exp)	0.2974		–		–0.5320		–
$B_{1, 2}$ (exp $\to$ rev)	–0.2572		–		–0.1964		–

$^{*} p < 0.1$ , $^{* *} p < 0.05$ , $^{* * *} p < 0.01$

The preferred specification is clearly to condition the Kalman filter on $t$ and leave $B$ unconstrained, and only the results from this specification are reported. Because future values are not included, forward facing effects of policy action are, to some extent, foreseeable at a given moment, save for the fact that coefficients sometimes change rather abruptly. Because some autocorrelation remains in the equation for the total deficit, policy recommendations are more tenuous, as there could be some omitted variable bias, but this should not be surprising because entitlement spending is more or less “automatic” and tied more to demographics than political choices. This speaks to the need to test for effects on both the on-budget and total deficits.

Results

Coefficients do change dramatically over time, supporting the time-varying approach in this paper. Magnitudes are small because most variables are expressed as a percent of GDP in first differences. Results are shown in time series plots of $β$ coefficients from (1), with estimates conditioned on $t$ , and $B$ unrestricted. Plots display point estimates for the coefficients in each time period as well as 95% confidence intervals, computed as 1.96 standard errors in either direction from the point estimate.

When leaving $B$ unrestricted, there exists asymmetric, nonzero covariance between the coefficients on revenues and expenses shown in Table 3, and the direction is not the same for the on-budget and total deficits. For either calculation of the deficit, an increase in the expense coefficient is followed by a decrease in the revenue coefficient. This implies that, if the deficit responds negatively (positively) to an increase (decrease) in expenses, that revenues should have more of an offsetting effect. This implies Keynesian multiplier effects, and these tend to be larger for the on-budget deficit, as shown in Table 3. The reverse is not true, however, for the on-budget deficit, for which an increase in the revenues coefficient is followed by an increase in the expenses coefficient. Thus, if revenues impact the deficit, expenses in turn affect the deficit more as well, and there is a flypaper-type effect for the central government. This is not the case for the total deficit, for which expenses matter less after a revenue increase. To some degree, this is expected because increases in social insurance contributions are not immediately followed by more transfer payments. These asymmetric effects in $B$ are among the most important results in this paper.

For many variables, coefficients are often insignificant except when major changes occur. This implies that deficits are attributable to a combination of all variables, not just one. Structural budget deficits will only be brought closer to balance or surplus with a deliberate focus on all of these variables.

Revenues

Revenues tend to be fairly insignificant for future deficits most of the time with a few notable aberrations, shown in Figures 1 and 2. Some of this can be explained by the asymmetric feedback effects of revenues and expenses in the discussion of matrix $B$ above. A positive coefficient on revenues means that higher revenues are linked to a higher budget balance, meaning a smaller deficit or larger surplus.

The 1970s are rather peculiar, with an inverse relationship between revenues and deficits in the 1973-75 recession. Revenues increased as the budget balance fell. There are a few possible explanations. The first is that expenses mattered more, which is reflected in the next section. Another is that inflation and bracket creep increased revenues as a percent of GDP and simultaneously depressed consumer spending power. This would lead to a feedback effect in which GDP falls because of a decline in consumer spending, which would cause deficits to expand as expenses continue. During the early 1980s, as income taxes fell sharply under President Ronald Reagan, there was some link between revenues and larger deficits in an intuitive direction.

The early 1990s, before the tech bubble, also showed a significant inverse relationship, somewhat like the 1970s, but not of exceptionally large magnitude. This was during the early 1990s recession when taxes were increased. Shortly after this period, budget cuts also took place. The explanations of the 1970s and 1990s are probably similar. The key takeaway is that the budget deficit only significantly responds to revenues when there is a substantial decrease. Revenue increases have no clear effect, probably owing to the political nature of revenue and expense decisions.

For both calculations of the deficit, the relationship was positive in the early 2000s for the total deficit, after income tax rates were cut. The on-budget deficit showed no similar response, and payroll tax rates were not cut. Current revenues presumably were enough to fund governmental operations, but they were needed to repay intragovernmental debt, leading to an expanding total deficit.

This positive relationship was also significant during the Great Recession and recovery, but this time it was significant for both measures of the deficit. As revenues fell, the deficit grew (budget balance fell). Thus, deficits at this time were, to a large extent, attributable to a fall in revenues. The Budget Control Act of 2011, which placed caps on discretionary spending, was a likely contributor to this relationship by mitigating a concurrent cause of the deficit, namely rising expenses, during this critical time. This may be one reason why both measures of the deficit responded after the Great Recession, but only the total deficit responded in the early 2000s.

Expenses

Budget deficits are reflected in negative numbers, and so are expenses. Thus, a positive coefficient on expenses is expected, because an increase in expenses leads to a more negative number, which should make the budget balance more negative too.

The late 1970s, especially for the on-budget deficit, are somewhat puzzling, as the relationship was inverse, shown in Figures 3 and 4. The deficit rapidly expanded at this time, and expenses rose as a percent of GDP while revenues fell, but GDP also contracted. Combined with bracket creep, this result implies that the deficit did respond to expenses, but the result was confounded by changes in GDP. The clear result for expenses is that significant responses of the deficit are limited to times when expenses change substantially. This is often during recessions, but it need not be, as shown by the mid-1980s.

During the 1980s, expenses mattered for the total deficit, but less so for the on-budget deficit. Thus, the military buildup under President Ronald Reagan was not by itself a cause for the persistent trend toward budget deficits, which were somewhat more attributable to revenue shortfalls, implying a fiscal multiplier from expenses. It is imperative to note that, had expenses declined commensurately, the deficit would not have expanded as a tautology. Results in the figures for revenues and expenses simply show that changes in revenues and expenses led to future deficits in the 1980s, and that revenues mattered more at that time.

Expenses were substantial contributors to the deficit during the Great Recession alongside revenues. Expenses continued to be significant for some time after revenues began to recover, and they were a problem for both the total and on-budget deficits. The positive response in the early 2000s, which is also visible for revenues, likely reflects changes in tax rates and expenses under President George W. Bush.

Looking at revenues and expenses together, both matter at similar times, but coefficients on revenues do tend to be larger. This implies that revenue shortfalls influence future deficits more than expense increases and lends support for a larger expansionary multiplier than contractionary. For both revenues and expenses, coefficients are larger for the on-budget deficit, which is not surprising as social insurance trust funds have more steady revenues and expenses.

Real GDP Growth

The coefficient on GDP growth captures the effect of GDP growth from the time when expenses and revenues are measured, $t - 4$ , and the current period, $t$ , or the effect of economic growth after a change in revenues or expenses, shown in Figures 5 and 6. If significant, then economic growth or contraction materially affects the deficit after accounting for expenses and revenues.

During the 1970s recession, when there were counterintuitive effects of revenues on the deficit, effects of GDP growth and expenses are quite logical. The positive significant relationship between GDP and the deficit came at a time of declining real GDP, so as GDP fell, the deficit moved in the same direction (further into the red). Thus, even if revenues increased due to bracket creep, depressed spending power led to declining revenues from Keynesian feedback effects, and increased government spending compounded the problem. The effect was short-lived, and GDP growth eventually did reduce the deficit as the recession drew to a close in 1975.

During the early 1980s recession, GDP growth and the deficit were inversely linked, though coefficients were not very large. The reduction in the top marginal income tax rate was from 70% down to 28%, with later reductions bringing it down to 26% over several years. The 1980s tax cuts may have had some small Laffer Curve effects as the rate cuts were larger and had a higher starting point than later cuts in the early 2000s, though the cuts may have moved from the right of the peak to the left, with an ambiguous negative effect. This is clearly not the case with the later tax cuts in the early 2000s. Laffer Curve effects are often a tenuous explanation, but given current American tax rates, Laffer Curve stimulus effects to increase revenues are not a suitable justification for future tax cuts, valid as they may have been in the 1980s. It is also important to note that large tax cuts can, and sometimes will, stimulate the economy and lead to higher employment, although this may not offset revenue losses.

GDP growth is not a cure-all for budget deficits, especially when they are structural. The 1990s productivity surge had some positive effects on the deficit, which went into surplus. Indeed, this is better explained by GDP growth than revenues or expenses, both of which were insignificant. Unfortunately, politics matter more than revenue generating potential. GDP growth is only significant during the 1990s for the on-budget deficit. This GDP growth came before working-age Baby Boomers pushed the trust funds into a large surplus, so current revenues were not needed to repay the principal of the trust funds. At this time, however, spending began to increase as this surplus was required to become intragovernmental debt, or an earmarked commitment to fund the obligations out of future current revenues.

The seeds of a major structural deficit were sown at this time of rapid GDP growth. Prior to this time, there was little intragovernmental debt to repay out of current expenses. Since then, intragovernmental debt repayments have consumed increasing amounts of current revenues, to the point that the federal budget would be in surplus were it not for the 1990s borrowing. Had the social insurance trust funds been invested in other assets and received a market return, their solvency would not rely on current tax revenues. This is even after accounting for a major fall in asset prices during the Great Recession, from which financial markets more than recovered not long afterward.

During the recoveries from the Great Recession and early 2000s recession, GDP growth and the deficit were inversely linked, and coefficients were larger. Compared to the 1980s, the reduction in top marginal tax rates in the early 2000s had a lower starting point, 39.6%, which was reduced to 35%. As with the 1980s, spending increases followed the tax cut, though this was more visible in the total deficit in the early 2000s because of a demographic tide, and in the on-budget deficit in the 1980s due to rising military spending.

As growth rebounded during the recovery, expenses increased and the deficit expanded. Revenues explain some of this, but the significance of GDP growth is troubling. This shows that, during economic recovery, fiscal multipliers were not large enough or that Laffer Curve effects, which justified tax cuts, did not materialize. After the economy had largely recovered from the Great Recession, economic growth positively affected the federal budget balance. After rising intragovernmental debt squandered earlier rapid economic growth in the 1990s, tax cuts and spending increases were yet another regrettable step toward persistent budget deficits in the future, though demographic trends were far more relevant than a decade prior.

Private Nonresidential Investment

Long-run growth requires investment in new productive capacity. For the on-budget deficit, the relationship is not always significant, and it has fluctuated at times, shown in Figures 7 and 8. The total deficit’s response to investment is harder to interpret. Negative coefficients in the 1970s, imply that investment did not generate enough tax revenue to reduce the deficit, though this can be confounded by higher government spending. It is more of a problem for the total deficit, showing that social insurance outlays are expenses that do not significantly change with investment, even if current expenses do. The positive coefficient from 2012 through 2015 implies that investment contributed to the post-recession growth, but that this likely affected social insurance contributions more than on-budget revenues, perhaps because the early 2000s tax cuts with their lack of Laffer Curve effects were not yet undone. It is difficult to point to investment as a strategy to reduce the deficit, especially because GDP growth does not consistently have a clear positive effect.

The only time when the relationship between investment and the deficit was persistently negative was the Great Recession. Investment cannot lead to a reduced deficit if GDP growth does not materialize. During a recession, GDP may start to recover before investment due to slack in the economy, which is a likely explanation.

Federal Reserve Purchases of Treasurys

When the Federal Reserve purchases Treasurys on the open market, it does so with the objective of influencing interest rates. By keeping interest rates low, interest expenses are reduced, which reduces payments to the public and to the social insurance trust funds. With a higher money supply, the federal government can borrow without crowding out private investment. Thus, increased purchases of Treasurys by the Fed should reduce deficits, even after controlling for revenues, expenses, investment, and GDP growth. These purchases are a better metric of central bank action to affect interest costs than interest rates alone because they involve direct changes in the money supply and can prevent crowding out from increased borrowing.

For most of the time series, these purchases are either insignificant or positive, shown in Figures 9 and 10, implying that more monetary easing brings the budget balance toward surplus. The only strange result is that Fed purchases of Treasurys have a negative coefficient for the on-budget deficit in the 1970s, but it was positive for the total deficit. If monetary stimulus pushed the on-budget balance more toward deficit, but the total toward surplus, this implies fiscal accommodation at a time when social insurance receipts and outlays were fairly well matched. The opposite was the case in 2013-2015, when fiscal accommodation only helped with the use of current revenues to make whole the trust funds by repaying intragovernmental debt.

The monetary easing of the mid-1980s, led toward a positive relationship. This may not have been intentional, as the easing came after abnormally tight monetary policy to rein in inflation, and the spending increases had political justification apart from newly lowered interest rates. This positive relationship was even more poignant during the early 2000s and during the Great Recession, when rates were cut from a much lower starting point. Despite their unprecedented size, deficits during the Great Recession were accompanied by abnormally low interest rates on government debt; interest expenses were lower in dollar terms than in the 1990s when debt levels were lower. Much of this was due to the Fed’s action, but even without this, interest rates may have been fairly low. Financial markets were falling precipitously, and Treasury securities were in high demand because they were still perceived as safe.

Robustness Tests

Re-estimating this paper’s model with different lag orders, namely two and eight, provides a useful and informative robustness test. Diagnostics for these specifications, which continue with leaving $B$ unconstrained and time conditioning on $t$ , are in Table 4. Neither of these specifications removes as much serial correlation as when using four lags, shown by significant Ljung-Box $Q$ -statistics. Residual serial correlation when using eight lags implies that things are happening between the change in revenues and expenses and the deficit, indicating that omitting the fourth lag is causing some omitted variable bias. Conclusions from these robustness tests are nonetheless useful, though subordinate to the primary four-lag specification.

Table 4.

Model Diagnostics for Robustness Tests.

	On-budget Deficit		Total Deficit
Lag Order	2	8	2	8
Restrictions on $B$	–	–	–	–
Time conditioning	$t$	$t$	$t$	$t$
	State Eq. Innovations		State Eq. Innovations
ADF Stat.	$- {7.990}^{* * *}$	$- {5.057}^{* * *}$	$- {9.233}^{* * *}$	$- {7.735}^{* * *}$
Ljung-Box $Q$	${54.697}^{* * *}$	${121.820}^{* * *}$	${18.390}^{* * *}$	${15.766}^{* * *}$
	$\hat{y} - y$		$\hat{y} - y$
ADF Stat.	$- {7.993}^{* * *}$	$- {6.431}^{* * *}$	$- {7.414}^{* * *}$	$- {7.249}^{* * *}$
Ljung-Box $Q$	${22.624}^{* * *}$	${42.720}^{* * *}$	${49.651}^{* * *}$	${36.926}^{* * *}$
$B_{2, 1}$ (rev $⟶$ exp)	0.001	$- 0.001$	0.004	0.030
$B_{1, 2}$ (exp $⟶$ rev)	$- 0.004$	$- 0.007$	0.002	$- 0.035$

$^{*} p < 0.1$ , $^{* *} p < 0.05$ , $^{* * *} p < 0.01$ .

When using two lags, all variables are nearly always insignificant, showing that two quarters are not long enough for budget deficits to become pronounced. In practical terms, a six-month continuing resolution should not be expected to have long-lived effects on the deficit. Graphs for revenues and expenses and their respective effects on the on-budget and total deficit are in the appendix.

When using eight lags, effects on the on-budget deficit are nearly always insignificant, but effects on the total deficit are not. For the total deficit, all variables have significance for Federal Reserve purchases of Treasury securities. Significance patterns are similar, yet different in an intuitive way, shown in Figures 11–14. Rapid spikes, either positive or negative, are no longer present with this longer lag order, but more gradual changes are. Graphs for revenues and expenses with eight lags for the on budget deficit are in the appendix, but graphs for the four significant variables and the total deficit are displayed below. Concerning the lack of significance for the Fed’s purchases, monetary stimulus may reduce interest expenses, but it does not reduce the deficit, perhaps because it makes borrowing seem less costly.

The changes in significance patterns show that shocks, which are temporary, often have temporary effects on the deficit, even if new policies and programs persist. This is reinforced by the lack of significance altogether for the on-budget deficit, and the absence of rapid spikes for the total deficit. Should there be a positive shock to federal spending, perhaps because of a recession, this is a good result. The negative flip side, however, is that any material reduction in spending likely will not reduce the deficit over the long term. Recent political brinkmanship over the federal debt ceiling has at times led to agreements on spending cuts, yet the deficit has continued to increase.

There are two potential and complementary reasons why discretionary spending cuts have only short-lived effects on the deficit. The first and most important is that discretionary spending is not the primary driver of the federal deficit, which in recent years has been a result of rising entitlement spending. Thus, more gradual changes, like demographic pressures, do lead to more persistent effects on the deficit. The second reason is that political decisions can be overridden later, and politicians’ preferences and promises are time-inconsistent. Political wrangling and legislative compromises historically have not shown much promise of reducing the deficit, although there is no a priori reason why this should be impossible. Even the Budget Control Act of 2011, which looks fairly effective when including four lags in the model, may not matter as much over longer horizons.

Policy Implications

These results lead to several important policy conclusions. The first is that revenue increases are not likely to balance the federal budget because expenses are involved. There is no clear link between major increases in revenues and a reduced deficit, and the only significant relationship between revenues and deficits is when revenues decrease. The lack of significance for revenue increases could be because of contractionary fiscal multipliers, but given that increases in expenses are linked to increased deficits, it is more likely that time-inconsistent politicians spend the increased revenue. Deficits do respond to decreases in revenues, implying that spending does not pull back when revenues fall, which is also shown by the asymmetric covariance of coefficients described above. The existence of multiple competing explanations, ranging from the absence of causal impacts, time-inconsistent political behavior, or even endogenous countervailing effects, offers a direction for further research.

The second policy conclusion is that expenses matter for deficits, both on-budget and total, and they matter most during downturns. Spending is politically palatable during political downturns, but the model does not explicitly test whether political decisions that are specific to downturns are the cause of deficits. In recent years, coefficients on expenses for the on-budget and total deficits have been fairly similar, implying that fiscal solvency will require entitlement reform. The 1980s military buildup was not a major contributor to the deficit when controlling for revenues, GDP growth, investment, and monetary stimulus. When combined with a brief negative relationship between expenses and the deficit in the 1990s, fiscal multipliers have an effect that must be considered when evaluating the effect of government expenses on future deficits because the resulting decline in economic activity will also hit revenues.

Entitlement reform, painful as it may be, does not seem to present this same problem. This is reflected in the asymmetric covariance of coefficients described above. Increases in the coefficient for on-budget expenses are followed by an increase in the coefficient on revenues. This is far less pronounced for the total deficit, which could be a result of the automatic nature of recurring entitlement spending or the fact that it comprises such a large proportion of total federal spending. The implication of the model, though is that structural entitlement reform is likely the best route to fiscal sustainability in the United States based on the recent trajectory of revenues and expenses.

Austerity in areas other than entitlement reform does not appear to be a valid solution to deficit reduction. This could be a result of fiscal multipliers that cause spending cuts to lead to revenue decreases. In the United States, the size of government expenditures as a proportion of GDP, although less than in some other developed countries, is substantial enough that cuts could ripple far into the economy, causing noticeable revenue problems. If large deficits lead to a fiscal crisis that causes markets to panic and interest rates to rise, austerity may become inevitable. Given the results in this paper, austerity should not be viewed as a policy of choice for deficit reduction largely because of these multiplier effects.

Entitlement reform appears more promising as a means of reducing the deficit than austerity in other spending categories, but it too would not be without pain and the potential for feedback effects. There has not been substantial bidirectional variation in entitlement spending, which has been on an upward trajectory. When the Social Security and Medicare trust funds are depleted, decisions will have to be made about future funding of these programs, though cuts will be deeply unpopular and politically unpalatable. At this point, if there are cuts, it will become clearer whether entitlement reform is indeed the road to deficit reduction, though this model’s preliminary results indicate that it is.

The third policy conclusion is that pro-growth policies to bring in more revenue through economic growth may not materially reduce deficits. This is not, however, because increases in economic activity to not lead to more government revenue, which should be the case almost as a tautology. This result is likely because this paper’s model separates any causal impact or lack thereof from endogenous policy responses. Economic growth is necessary for rising living standards, but government expenses can rise with economic activity just like revenues, and the time inconsistency of politicians may mean that increased revenues will be spent, not saved. Rather than implying causality, this empirical observation is most useful as a word of caution when evaluating whether any pro-growth policy will be fiscally beneficial, especially when many relevant political decisions are yet to be made. On a related note, the negative coefficients at some points imply that post-recession growth may be accompanied by spending increases, which further points to political decisions about expenses as the source of large budget deficits. Further causal analysis about whether pro-growth policies reduce the deficit is a point for future research.

Although fiscal multipliers and Laffer Curve effects may exist, deficits are ultimately a multifaceted political problem. GDP growth and changes in revenues and expenses may exert a statistically significant impact, but ensuring fiscal solvency and future stability will require evaluating revenues, expenses, and GDP growth together.

Conclusion

Fiscal multipliers, which have been studied at great length, can justify government spending or tax cuts as a means to boost economic growth. Should growth be high enough to increase revenues, the increased deficit may be offset. This paper goes a step further to evaluate the effect of revenues, expenses, GDP growth, private investment, and monetary stimulus on the federal deficit. Using a state-space time-varying coefficient model to evaluate changes in these five variables on the deficit one year later, this paper concludes that the deficit is a multifaceted problem. It cannot be addressed through revenues or expenses alone due to its political nature.

Revenues and expenses are linked, and although expenses have a more significant impact on deficits, fiscal multipliers do exist, and there are asymmetric feedback effects between revenues and expenses. These results cast doubt on the wisdom of austerity as a solution to a fiscal crisis. Moreover, effects similar to the flypaper effect for intragovernmental transfers imply that revenue increases tend to be followed by spending increases, so tax hikes alone will not solve the problem. GDP growth to eliminate a deficit is not a viable solution because government expenses tend to grow with GDP. Monetary stimulus can reduce interest expenses, but this becomes less feasible as debt levels grow. Structural entitlement reform is the most promising, though quite painful, strategy for fiscal sustainability because it will have the greatest impact on the deficit with the fewest feedback effects.

Footnotes

Acknowledgement

I am grateful to an anonymous referee for suggestions regarding econometric specifications that materially improved the paper and strengthened its analysis.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors received no financial support for the research, authorship and/or publication of this article.

ORCID iD

Robert Gmeiner

Author Biography

Robert Gmeiner, PhD, is an assistant professor of economics at the University of South Dakota Beacom School of Business.

Appendix

References

Alesina

Alberto

Favero

Carlo A.

Giavazzi

Francesco

. 2018. “What Do We Know About the Effects of Austerity?” AEA Papers and Proceedings 108: 524–30.

Auclert

Adrien

Rognlie

Matthew

Straub

Ludwig

. 2024. “The Intertemporal Keynesian Cross.” Journal of Political Economy 132: 4068–121.

Barnichon

Regis

Debortoli

Davide

Matthes

Christian

. 2022. “Understanding the Size of the Government Spending Multiplier: It’s in the Sign.” Review of Economic Studies 89: 87–117.

Becker

Elizabeth

. 1996. “The Illusion of Fiscal Illusion: Unsticking the Flypaper Effect.” Public Choice 86: 85–102.

Bohn

Henning

. 1998. “The Behavior of U.S. Public Debt and Deficits.” Quarterly Journal of Economics 113: 949–63.

Bouakez

Hafedh

Rachedi

Omar

Santoro

Emiliano

. 2023. “The Government Spending Multiplier in a Multisector Economy.” American Economic Journal: Macroeconomics 15: 209–39.

Brinca

Pedro

Holter

Hans A.

Krusell

Per

Malafry

Laurence

. 2016. “Fiscal Multipliers in the 21st Century.” Journal of Monetary Economics 77: 53–69.

Calcagno

Alfred

. 2012. “Can Austerity Work?” Review of Keynesian Economics 0: 24–36.

Chen

Kaiji

Imrohoroglu

Ayse

. 2015. “Debt in the U.S. Economy.” Economic Theory 64: 675–706.

10.

Ferriere

Axelle

Navarro

Gaston

. 2025. “The Heterogeneous Effects of Government Spending: It’s All About Taxes.” Review of Economic Studies 92: 1061–125.

11.

Fetzer

Thiemo

. 2019. “Did Austerity Cause Brexit?” American Economic Review 109: 3849–86.

12.

Gabriel

Ricardo Duque

Klein

Mathias

Pessoa

Ana Sofia

. 2023. “The Political Costs of Austerity.” Review of Economics and Statistics. https://doi.org/10.1162/rest_a_01373.

13.

Hamilton

Bruce W.

1983. “The Flypaper Effect and Other Anomalies.” Journal of Public Economics 22: 347–661.

14.

Heimberger

Philipp

. 2022. “Do Higher Public Debt Levels Reduce Economic Growth?” Journal of Economic Surveys 37: 1061–89.

15.

Herndon

Thomas

Ash

Michael

Pollin

Robert

. 2014. “Does High Public Debt Consistently Stifle Economic Growth? A Critique of Reinhardt and Rogoff.” Cambridge Journal of Economics 38: 257–79.

16.

Hines

James R.

Thaler

Richard H.

. 1995. “Anomalies: The Flypaper Effect.” Journal of Economic Perspectives 9: 217–26.

17.

Huidrom

Raju

Kose

M. Ayhan

Lim

Jamus J.

Ohnsorge

Franziska L.

. 2020. “Why Do Fiscal Multiplilers Depend on Fiscal Positions?” Journal of Monetary Econmics 114: 109–25.

18.

Leeper

Eric M.

Traum

Nora

Walker

Todd B.

. 2017. “Clearing Up the Fiscal Multiplier Morass.” American Economic Review 107: 2409–54.

19.

Makin

Anthony J.

Layton

Allan

. 2021. “The Global Fiscal Response to COVID-19: Risks and Repercussions.” Economic Analysis and Policy 69: 340–9.

20.

Meyers

Roy T.

2014. “The Implosion of the Federal Budget Process: Triggers, Commissions, Cliffs, Sequesters, Debt Ceilings, and Shutdown.” Public Budgeting and Finance 34: 1–23.

21.

Niskanen

William A.

1968. “Nonmarket Decision Making: The Peculiar Economics of Bureaucracy.” American Economic Review 58: 293–305.

22.

Peppel-Srebrny

Jemima

. 2021. “Not All Government Budget Deficits are Created Equal: Evidence From Advanced Economies’ Sovereign Bond Markets.” Journal of International Money and Finance 118: 102460.

23.

Peterson

Dana M.

2024. “U.S. Sovereign Debt Crisis: Mounting Challenges, Practical Solutions.” Business Economics 59: 143–8.

24.

Ramey

Valerie A.

Zubairy

Sarah

. 1984. “Government Spending Multipliers in Good Times and in Bad: Evidence From US Historical Data.” Journal of Political Economy 126: 850–901.

25.

Reinhart

Carmen M.

Reinhart

Vincent M.

Rogoff

Kenneth S.

. 2012. “Public Debt Overhangs: Advanced Economy Episodes Since 1800.” Journal of Economic Perspectives 26: 69–86.

26.

Reinhart

Carmen M.

Rogoff

Kenneth S.

. 2010. “Growth in a Time of Debt.” American Economic Review 100: 573–678.

27.

Rios

Vicente

Hortas-Rico

Miriam

Pascual

Pedro

. 2022. “What Shapes the Flypaper Effect? The Role of the Political Environment in the Budget Process.” Local Government Studies 48: 793–820.

28.

Shumway

Robert H.

Stoffer

David S.

. 2025. Time Series Analysis and Its Applications With R Examples. 5 edition. London: Springer,.

29.

Taylor

Lance

Proaño

Christian R.

de Carvahlo

Laura

Barbosa

Nelson

. 2012. “Fiscal Deficits, Economic Growth, and Government Debt in the USA.” Cambridge Journal of Economics 36: 189–204.

30.

Thornton

Daniel L.

2012. “The U.S. Deficit/Debt Problem: A Longer Run Perspective.” University of Louisville Law Review 50: 625.