FISCAL STIMULUS AND Yuliya Demyanyk, Elena Loutskina, and Daniel Murphy*

Abstract—In the aftermath of the –induced , pol- In this paper, we use detailed new data on Department of icymakers have questioned whether fiscal stimulus is effective during pe- Defense (DOD) spending to evaluate whether government riods of high consumer indebtedness. This study empirically investigates this question. Using detailed data on Department of Defense spending for spending during the stimulates local eco- the 2007–2009 period, we document that the open-economy relative fiscal nomic growth differently across geographies with varying multiplier is higher in geographies with higher consumer debt. The results levels of prerecession consumer indebtedness. We find that suggest that in the term (2007–2009), fiscal policy can mitigate the adverse effect of consumer (over)leverage on real economic dur- consumer debt is an important determinant of the fiscal mul- ing a recession. We then exploit detailed microdata to show that both het- tiplier during the Great Recession. During the 2007–2009 erogeneous marginal propensities to consume and slack-driven economic period, the DOD spending multiplier is higher in geogra- mechanisms contribute to the debt-dependent multiplier. phies with higher prerecession consumer debt-to-income ra- tios than in geographies with lower prerecession consumer debt-to-income ratios.2 We then exploit detailed microdata I. Introduction to evaluate whether aggregate and aggregate sup- HE ability of to mitigate ply economic mechanisms contribute to the debt-dependent Thas always been a hotly debated topic among academics, multiplier. practitioners, and policymakers. The 2007 crisis brought new Our analysis is based on DOD spending data that cover pur- arguments to the table as it acutely highlighted the role that chases and obligated funds from $25 to multimillion-dollar consumer indebtedness plays in a recession (Mian & Sufi, contracts since 2000. We observe the start and end dates of 2011). The dramatic rise in U.S. household leverage from the contracts, the primary contractor locations, and the postal about a 1.2 debt-to-income ratio in late 1990 to about 1.65 code in which the majority of the work was performed. Armed in 2006 not only set the stage for the Great Recession but with the granular DOD spending data, we combine the em- also contributed to a decline in aggregate and pirical approaches of Mian and Sufi (2015) and Nakamura ultimately slowed the (Mian, Rao, & Sufi, and Steinsson (2014) to implement an instrumental variable 2013; Mian & Sufi, 2015). It is unclear whether fiscal stimulus analysis that evaluates how prerecession consumer debt-to- is effective in this environment. income ratios and the change in DOD spending from 2007 to ’ debt and subsequent deleveraging are fre- 2009 affect economic output over this period. The detailed na- quently invoked to argue that expansionary fiscal pol- ture of this new DOD spending data allows us to conduct the icy might be ineffective during consumer-debt-overhang- analysis at the core-based statistical area (CBSA) level and induced slumps.1 These arguments are bolstered by evidence hence better capture the heterogeneity in consumer leverage. that household deleveraging may be associated with low Moreover, it permits us to focus on the recessionary period spending propensities (Sahm, Shapiro, & Slemrod, 2015; with high total consumer debt and rapid deleveraging by con- Jappelli & Pistaferri, 2014) that should lead to low short-run ducting a purely cross-sectional analysis. fiscal multipliers. At the same time, the proponents of de- Our results suggest that DOD spending multipliers ex- mand stimulus argue that expansionary fiscal policy is more hibit significant heterogeneity across CBSAs with different effective during periods of consumer deleveraging due to high prerecession levels of consumer leverage. The difference in spending propensities (Eggertsson & Krugman, 2012). While the multiplier between the 75th and 25th percentiles of the some theoretical literature sheds light on this debate, few pa- consumer-leverage distribution is about the same as the aver- pers empirically examine this question. age CBSA open-economy fiscal multiplier.3 The results sug- gest that at least in a short to medium run, expansionary fiscal stimulus during a deleveraging recession can mitigate the ad- verse effects of consumer debt overhang on : Received for publication September 13, 2017. Revision accepted for pub- a 1 percentage point increase in government spending rela- lication June 25, 2018. Editor: Yuriy Gorodnichenko. ∗Demyanyk: University of Illinois at Chicago; Loutskina: University of tive to local income offsets the adverse effects of consumer Virginia, Darden School of Business; Murphy: University of Virginia. indebtedness by about 16%. The views expressed are those of the authors and do not necessarily reflect the official positions of the of Cleveland or the Federal Reserve System. A supplemental appendix is available online at http://www.mitpress 2Throughout our analysis, we exploit total consumer indebtedness that journals.org/doi/suppl/10.1162/rest_a_00796. accounts for all types of debt balances: mortgages, auto , card 1In his 2010 speech promoting austerity, U.K. Chancellor of the Exche- debt, and other forms of consumer credit. For simplicity, we refer to it as quer George Osborne asserted, “We have to move away from an economic consumer debt. modelthatwasbasedonunsustainableprivateandpublicdebt....There 3We adopt Nakamura and Steinsson’s (2014) terminology in referring is no choice between going for growth today and dealing with our to multipliers estimated from cross-regional variation as “relative open- tomorrow. Indeed we will not have meaningful growth unless we show we economy multipliers.” See Chodorow-Reich (2019) for a discussion of the can deal with our debts” (https://conservative-speeches.sayit.mysociety.org relationship between estimates of open-economy multipliers and fiscal mul- /speech/601526). tipliers derived from national aggregate data.

The Review of and Statistics, October 2019, 101(4): 728–741 © 2019 by the President and Fellows of Harvard College and the Massachusetts Institute of Technology https://doi.org/10.1162/rest_a_00796

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/rest_a_00796 by guest on 01 October 2021 FISCAL STIMULUS AND CONSUMER DEBT 729

In the second half of the paper (section IV), we examine with the local consumer debt-to-income ratio, despite con- the validity of two economic mechanisms that could con- sumer leverage having no direct effect on in tribute to the debt-dependent multiplier. First, fiscal multipli- this sector. The dependence of this multiplier on prerecession ers might depend on debt-driven heterogeneity in marginal consumer indebtedness suggests that slack dependence may propensities to consume (MPCs). Galí, López-Salido, and also contribute to debt-dependent multipliers during the Great Vallés (2007) and Eggertsson and Krugman (2012) suggest Recession. that consumption of high-debt, credit-constrained house- Overall, our results indicate that the benefits of fiscal holds responds strongly to fiscal stimulus, while consump- stimulus—higher income and employment—are higher in tion of nondebt-constrained agents is relatively unaffected by geographies suffering from consumer debt overhang. While additional income. The resulting higher MPCs among highly we explore only the relative multiplier and do not evaluate levered leveraged consumers should lead to higher fiscal mul- the -term costs of fiscal stimulus (e.g., public debt and tipliers. In contrast, recent empirical studies document that future burdens), our results offer an important implica- high-debt households use additional income to pay down debt tion: the ills of private debt overhang can be mitigated, at rather than to spend (Sahm et al., 2015; Jappelli & Pistaferri, least in the short run, by government spending. 2014). These studies suggest that deleveraging can be associ- is relatively more effective at stimulating income and em- ated with less effective fiscal policy if debt-ridden households ployment in areas with high consumer debt-to-income ra- are characterized by lower MPCs. tios compared to areas with low consumer debt-to-income We use detailed microdata to evaluate whether govern- ratios. ment spending leads to higher consumption responses for This paper contributes to a number of strands of literature high-debt households. Specifically, we exploit individual- on fiscal policy and consumer behavior. First, we contribute to level measures of consumer debt-to-income ratios along with the debate about the efficacy of fiscal policy during consumer- two measures of household consumption: (a) individual-level debt-overhang-induced slumps. Inspired by the 2007 crisis, consumer credit card balances and (b) postal code–level new an emerging theoretical literature explores optimal policy car registrations.4 We find that consumption of high debt-to- during recessions that feature financial frictions and hetero- income households responds more positively to an increase geneous consumers (Guerrieri & Lorenzoni, 2017; Eggerts- in DOD spending during the crisis period than consump- son & Krugman, 2012). On the empirical side, Bernardini tion of low debt-to-income households. We also show that and Peersman (forthcoming) and Klein (2017) explore time- this consumption is unlikely to be funded via an increase in series variation in aggregate macroeconomic characteristics borrowing. to document that the fiscal multiplier is higher during peri- Second, we evaluate whether debt-dependent multipliers ods of higher aggregate consumer indebtedness. Bernardini, arise due to local labor market slack. Geographies with higher De Schryder, and Peersman (2017) is the study closest to debt-to-income households experienced deeper cuts in real ours. They use panel state-level data to explore how local economic activity and higher levels of unemployment during business cycles and affect fiscal multipliers. the Great Recession (Mian & Sufi, 2011, 2015). In the pres- We augment this literature and offer new insights into the ence of the resulting slack, fiscal stimulus should be more economic mechanisms contributing to debt-dependent fiscal effective in stimulating the local economy because it is less multipliers. likely to crowd out private sector employment (Michaillat, Second, the expanding literature on state-dependent mul- 2012; Murphy, 2017). tipliers employs structural vector autoregressions and na- Directly testing the slack channel is not possible in our tional aggregate statistics to evaluate whether fiscal policy setting because slack during the recession is endogenous to is more effective in recessions than in expansions (Auerbach government spending and consumer debt. Instead, we adopt & Gorodnichenko, 2012; Ramey & Zubairy, 2018). In con- an indirect approach and document debt-dependent fiscal trast to these studies, we offer an alternative cross-sectional multipliers in the tradable sector of the economy. This sec- empirical design that allows us to evaluate heterogeneity in is subject to local labor market conditions (slack) but fiscal multipliers in a given period. The granularity of our is unlikely to directly benefit from local household spend- novel data also offers additional insights into the mechanisms ing. We further isolate specific industries that do not di- responsible for state-dependent multipliers. rectly benefit from local individual consumption such as the Finally, a growing literature empirically evaluates con- National and International Affairs sector (NAICS sumer responses to various forms of stimulus (Kaplan & 9811). We find that the positive effect of DOD spending on Violante, 2014; Misra & Surico, 2014; Shapiro & Slemrod, local employment in the national security sector increases 2003; Aaronson et al., 2012; Parker et al., 2013; Agarwal et al., 2007; Cloyne & Surico, 2017). We augment this litera- ture by documenting debt-dependent MPCs during a delever- 4A wide set of literature exploits credit card balances to proxy for individ- aging recession: highly leveraged households tend to con- ual consumption levels (Mian et al., 2013; Aaronson, Agarwal, & French, 2012; Agarwal, Liu, & Souleles, 2007). Mian and Sufi (2011) use car reg- sume more in response to increases in DOD spending than istrations to capture household consumption. less levered households.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/rest_a_00796 by guest on 01 October 2021 730 THE REVIEW OF ECONOMICS AND STATISTICS

II. Data and Sample Selection ipation effects and therefore is more relevant for estimating the effects of fiscal stimulus. A. Government-Spending Data We also build a measure of actual fiscal outlays: DOD spending. In the presence of hand-to-mouth consumers (Galí In this paper, we use the new database of DOD contracts et al., 2007; Eggertsson & Krugman, 2012), actual govern- available at USAspending.gov. This official government web- ment disbursements are potentially more relevant for the site contains detailed information on DOD contracts signed propagation of fiscal policy. To build this measure, we al- since 2000. The data are based on DD-350 and DD-1057 5 locate the obligated (and partially deobligated) amount of military procurement forms. The database covers purchases the contract equally across all months of the contract du- and obligated funds from $25 disbursements to large pro- ration and then aggregate the monthly data into geographic curements up to $32 billion. Each observation in the data spending estimates over considered time periods. In the rest set corresponds to a unique individual contract between the of the paper, we report estimates based on both DOD spend- DOD and a prime contractor. We observe the total contract ing and DOD obligations and for simplicity refer to both as amount (obligated funds) and the duration of the contract: government spending.7 from a minimum of one day in cases of outright purchase of ready-made or services to more than a decade in cases of large military contracts (the latter of which account for less B. Real Economic Data than 0.2% of contracts). Furthermore, we observe the loca- tion, industry, and tax ID of the prime contractor and, in most To build various measures of real economic growth, we cases, information on the locations (postal codes) wherein exploit two data sets. First, we obtain annual GDP data for the majority of the work was actually performed. We also 372 CBSAs from the Bureau of Economic Analysis (BEA). observe deobligated amounts (terminated contracts) that we Second, we expand our analysis to income and employment remove by matching them to original obligations based on data from the Quarterly Census of Employment and Wages contractor ID, primary contractor postal code, and a dollar (QCEW) data set provided by the Bureau of Labor Statistics. amount of the original contract falling within 0.5% of the de- The traditional fiscal-multiplier literature focuses on GDP, obligated amount. In the case of a match, we consider both yet the limited number of CBSAs covered in the data, as contracts null and void. This restriction removes 4.7% of con- well as the inability to capture fine geography and industry tracts from the sample.6 variation in GDP, significantly limit our ability to conduct the These DOD data are uniquely suited to evaluate our core analysis using GDP alone. In contrast, BLS data allow us to question. DOD spending constitutes more than half of dis- build two core dependent variables—growth in income and cretionary government spending. It is the third largest source growth in employment—across counties, states, and a much of government spending (18% of the U.S. ) after So- larger set of CBSAs, as well as across different industries. cial Security (25%) and Medicare/Medicaid (24%) and thus We exploit this feature of the data in our tests. constitutes a significant force of fiscal stimulus during a re- Alongside the aggregate economic indicators within a cession. Not surprisingly, a number of studies in the literature given geography, we conduct the analysis by sector of the have exploited aggregate DOD spending in evaluating the ef- economy. Specifically, we evaluate how tradable and non- fect of fiscal policy on economic growth (Barro & Redlick, tradable sectors react to consumer indebtedness and fiscal 2011; Ramey, 2011; Auerbach & Gorodnichenko, 2012). stimulus. To do so, we classify industries into tradable, non- The information on contract timing in the data permits tradable, strict nontradable, construction, and other, follow- us to build two measures of DOD spending. First, we con- ing the classification scheme of Mian and Sufi (2012). struct DOD obligations that equal the total amount of new contracts signed in a given period less the amount associated C. Measure of Consumer Indebtedness with terminated contracts (deobligations). In this measure, we disregard the maturity of the contracts and the timing of ac- In this paper, we rely on consumer leverage as a measure of tual DOD disbursements. Prior literature (e.g., Ramey, 2011) credit-constrained households. Specifically, we use the 2006 suggests that information in DOD obligations captures antic- (and, for robustness, 2007) county-level consumer debt-to- income ratios published by Mian et al. (2013). This measure captures total consumer indebtedness by for all 5Nakamura and Steinsson (2014) show that DD-350 and DD-1057 spend- types of debt balances: mortgages, auto loans, credit card ing covers in excess of 96% of total DOD spending and accounts for almost debt, and other forms of consumer credit. When appropriate, all of the time-series variation in DOD spending at the state-year level. 6We do not eliminate partial contract deobligations for two reasons. First, we aggregate this measure to larger economic geographies the data start in 2000, which prevents us from effectively filtering out deobli- (CBSA or state) using population-weighted averages. gations that are close to the sample start date. Second, despite the presence of unique contractor IDs, it is impossible to identify prior contracts that were deobligated if the deobligation amount is below the original contract 7The changes in DOD spending and DOD obligations are highly cor- amount. We implemented a wide set of plausible alternative approaches related with a correlation coefficient equal to 0.87 during the 2007–2009 and find the results of our IV analysis to be quantitatively and qualitatively period. This can be attributed to over half of DOD contracts having short similar to those reported in this paper. (less than one year) maturity.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/rest_a_00796 by guest on 01 October 2021 FISCAL STIMULUS AND CONSUMER DEBT 731

Our choice of proxy for debt-constrained households is as reported by the BLS. Following Mian and Sufi (2015), driven by prior literature that explicitly links consumer lever- we also control for local demographic characteristics: the age and deleveraging pressure during the Great Recession percentage of white people in the local population, median (e.g., Mian et al., 2013). We validate this measure by docu- household income, median home values, the percentage of menting that high debt-to-income households do, on average, owner-occupied housing units, the percentage with less than delever. Table 4 of this paper shows that the direct effect of a high school diploma, percentage with only a high school debt-to-income on credit balances is negative and significant diploma , the unemployment rate, a dummy for urban areas, across all considered forms of consumer debt.8 Furthermore, and the poverty rate at the respective geographic level. considering consumer leverage prerecession mitigates tradi- Government contracts are notoriously political and hence tional reverse-causality concerns. It is highly unlikely that potentially endogenous to local economic conditions. Politi- the depth of the economic downturn in the 2008–2009 period cians from more recession-prone or deeper-recession geogra- can affect predetermined consumer leverage in 2006 (2007) phies might lobby for larger DOD allocations for their con- Finally precrisis leverage is measurable at the start of the re- stituencies. To address this endogeneity problem, we adopt cession, making it an actionable measure for policymakers. the Bartik-style instrument approach proposed in Nakamura and Steinsson (2014):

D. Validating Government-Spending Data Average(G ) GPost − GPre GInstrument = it × . (2) i v Pre Before we proceed with our analysis of the core question A erage(Gt ) Y of the paper, we validate the new DOD spending data and The instrument is the predicted change in government spend- our cross-sectional empirical design. Specifically, we report a ing based on a location’s average share of national defense baseline analysis of the open-economy fiscal multiplier using spending and the total aggregate change in national defense our new data and then compare the results to previous findings spending.9 The instrument relies on the aggregate variation in documented in the literature. In this analysis, we extend the defense spending while eliminating the ability of the appro- instrumental-variable empirical approach of Nakamura and priation process to reallocate DOD spending in response to Steinsson (2014) to our cross-sectional setting, local economic conditions. The identifying assumption is that Y Post − Y Pre GPost − GPre the buildup and drawdown of national defense spending as- i i = α + β i i + Controls + , (1) sociated with, for example, wars in and are Y Pre Y Y Pre i i i i not responses to economic conditions in any particular city. Note that the instrument changes with each specification de- where Y Post is 2009 income (employment or GDP) in geog- i pending on the normalization variable (income or GDP) and raphy i and Y Pre is 2007 income (employment or GDP) in i whether the specification utilizes a DOD-obligations-based geography i. Since DOD spending in both 2008 and 2009 measure of government spending or a DOD-spending-based affected the local real economy in 2009, we evaluate the measure. increase in government spending from the 2006–2007 pe- Table 1 presents summary statistics of our core variables of riod (GPre) to the 2008–2009 period (GPost ). We normalize i i . Panel A reports the growth in various real economic both the dependent variable (difference in economic output) characteristics between the 2006–2007 and 2008–2009 pe- and the core variable of interest (difference in government riods using CBSA-level aggregates. We find that over this spending) by the same beginning-of-the-period measure of period, consumer income declined on average by 0.91%. We economic output. Specifically, we normalize the change in observe significant heterogeneity, with some CBSAs expe- government spending by total income (in cases of income riencing declines in aggregate wage income as high as 26% or employment specifications) or total GDP (in GDP speci- and some growing at a 31% rate. The average change in de- fications). The coefficients β , β , and β capture Income Empl GDP fense spending as a fraction of prerecession income is 1.1%, the government-spending multiplier for different economic with a standard deviation of 5%. On average, the level of variables of interest. We conduct this analysis at differ- defense spending is 2.7% of CBSA income, with a standard ent levels of geographic aggregation: county, CBSA, and deviation of 6.5%. The heterogeneity indicates that while for state. some CBSAs, DOD spending negligibly contributes to the To account for differences in industry structure across ge- local economy, the other CBSAs rather heavily depend on ographies, we control for the beginning-of-the-period share DOD spending. of nineteen two-digit NAICS industries in local employment Table 2 validates our data and empirical approach by re- porting the results of the instrumental variable analysis fol- 8The household net worth introduced by Mian et al. (2013) is likely lowing equation (2). Panel A reports the cross-sectional IV to better capture individual credit constraints, yet it is endogenous to local economic growth and recent literature questions whether the Saiz elasticity is a valid instrument for housing price changes (Davidoff, 2015). Mian et al. (2013) report that their results are robust to using simple prerecession con- 9We obtain qualitatively and quantitatively similar results if we exploit sumer leverage as a measure of household net worth shocks and associated the average geography share of DOD spending using only prerecession credit constraints. spending allocation shares as of 2006–2007.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/rest_a_00796 by guest on 01 October 2021 732 THE REVIEW OF ECONOMICS AND STATISTICS

TABLE 1.—SUMMARY STATISTICS N Mean SD p25 p50 p75 A: CBSA-Level Growth Rates

Personal Income Growth07−09 (%) 828 −0.91 6.12 −4.52 −0.65 2.63 Personal Income Growth07−09, Nontradables (%) 828 0.19 0.71 −0.17 0.14 0.56 Personal Income Growth07−09, Tradables (%) 828 −0.32 2.41 −0.62 −0.06 0.20 Personal Income Growth07−09, Construction (%) 828 −0.46 1.70 −1.09 −0.38 0.35 Personal Income Growth07−09, Other Sectors (%) 828 2.10 3.02 0.44 1.82 3.54 Employment Growth07−09 (%) 828 −4.96 4.30 −7.24 −4.70 −2.50 GDP Growth07−09 (%) 372 0.67 6.67 −3.03 1.23 4.65 B: CBSA-Level Changes Relative to Prerecession Income

CBSA DOD spending07−09/Income2007 (%) 828 2.65 6.48 0.15 0.59 2.23 DOD Spending07−09/Income2007 (%) 828 1.12 4.95 −0.07 0.07 0.60 IV for DOD Spending07−09/Income2007 (%) 828 0.61 1.43 0.05 0.14 0.48 DOD Obligations07−09/Income2007 (%) 828 1.01 6.13 −0.09 0.04 0.54 IV for DOD Obligations07−09/Income2007 (%) 828 0.44 1.05 0.03 0.10 0.35 Debt-to-Income2006 824 1.60 0.60 1.19 1.44 1.83 C: TransUnion Individual-Level Characteristics

Debt-to-Income2006 6,689,130 1.24 1.44 0.06 0.46 2.36 Total Debt07−09/Income2007 (%) 6,689,130 −5.5 136.0 −36.1 −4.2 23.7 Mortgage Debt07−09/Income2007 (%) 6,689,130 −0.5 100.2 −7.8 0 0 Credit Card Debt07−09/Income2007(%) 6,689,130 −1.2 8.1 −1.9 0 0.6 Auto Loans07−09/Income2007 (%) 6,689,130 −1.1 14.7 −5.8 0 0 D: R.L. Polk Auto Registrations. ZIP Code-Level

Growth in Auto Registrations07−09 (%) 22,509 −41.3 29.3 −56.2 −39.7 −25.3 (A) Summary statistics for variables capturing real economic growth from 2007 to 2009 (panel A). 2006/07-2008/09 DOD spending. (C) 2006 consumer debt-to-income. Panel B: Growth in various individual-level debt accounts (panel C). Postal-code-level auto registrations (panel D).

TABLE 2.—GOVERNMENT SPENDING AND LOCAL ECONOMIC GROWTH A. Income-Based Multiplier across Different Geographic Units County CBSA State *** *** DOD Spending07−09 0.09 0.37 2.08 (12.43) (3.86) (1.54) *** *** * DOD Obligations07−09 0.08 0.35 1.65 (9.75) (2.91) (1.75) Geography-level controls Yes Yes Yes Yes Yes Yes Number of observations 1,743 1,743 828 828 51 51 First-stage regression coefficient 0.40 0.69 2.12*** 2.95** 3.73*** 5.75*** Kleibergen-Paap LM-test 1.47 1.34 13.25*** 9.1*** 5.86** 5.63** Kleibergen-Paap F-test 205.34*** 91.76*** 20.29 12.48 2.47 2.36 B. CBSA-Level Multiplier across Different Measures of Real Economic Growth Income Employment GDP *** *** DOD Spending07−09 0.37 –0.23 –0.78– (3.86) – (3.60) – (1.31) – *** *** DOD Obligations07−09 –0.35 0.23 –0.54 – (2.91) (2.91) – (0.95) CBSA-level controls Yes Yes Yes Yes Yes Yes Number of observations 828 828 828 828 372 372 First-stage regression coefficient 2.12*** 2.95*** 2.12*** 2.95** 1.52*** 2.70*** Kleibergen-Paap LM-test 13.25*** 9.10*** 13.25*** 9.10*** 5.60*** 2.32 Kleibergen-Paap F-test 20.29 12.48 20.29 12.48 10.62 3.22 This table presents the IV analysis following regression equation (1). A: The county-, CBSA-, and state-level results where the dependent variable is growth in income between 2007 and 2009. B: CBSA-level results for growth in income, employment, and GDP. Standard errors are clustered by state in the county-level and CBSA-level regressions. T -statistics are reported in parentheses. ∗∗∗ p < 0.01, ∗∗ p < 0.05, and ∗ p < 0.1.

analysis of the effect of DOD obligations and DOD spend- From the first-stage regression, we report only the core co- ing on 2007–2009 wage-based income growth at the county, efficient of interest, as well as the Kleibergen-Paap LM-test CBSA, and state levels. Panel B reports the results of CBSA- and F-test statistics for weak instruments.10 level analysis for different measures of real economic output. Both analyses incorporate a wide set of controls for prereces- 10Robust Kleibergen-Paap F-statistics are produced by weakivtest in Stata sion local industry structure and local economic conditions. and are available only for specifications with one endogenous regressor.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/rest_a_00796 by guest on 01 October 2021 FISCAL STIMULUS AND CONSUMER DEBT 733

Panel A of table 2 offers a number of important findings. data on government spending in the context of evaluating the First, we document positive multiplier estimates irrespective effect of fiscal stimulus on local economic output. of the level of analysis or the government-spending mea- sure used: DOD obligations versus DOD spending. Second, the multiplier coefficients are increasing with the size of the III. Consumer Indebtedness and the explored geographic unit. The county-level multipliers are Government-Spending Multiplier positive, very statistically significant, but economically small (0.04/0.09); CBSA-level estimates are considerably larger A. CBSA-level Analysis and Results (0.24/0.36), and state-level open-economy multipliers exceed Armed with validated data, we turn to the core question 1 (1.65/2.08). This can be attributed to the fact that our data of this study and investigate whether government spending provide information about prime contract vendors and do not becomes (in)effective when high-debt consumers are poten- capture the ability of vendors to subcontract or hire employ- tially credit constrained and have to deleverage. To evaluate ees across county or CBSA lines. With smaller, less-populous this question, we alter the baseline specification (1) by in- geographies, the government spending is more likely to spill corporating the effect of prerecession consumer debt and al- into or from other geographic areas. This potential measure- lowing for a consumer-debt-dependent government-spending ment error is likely to lead to attenuation of our multiplier multiplier, estimates.11 Consistently, the multipliers increase with the size of the geographical unit. This would have been problem- Y Post − Y Pre GPost − GPre atic if our goal was to obtain a precise estimate of the state- i i = α + β i i + γ 06 Pre 1 Pre DT Ii or city-level fiscal multiplier. Our objective instead is to de- Yi Yi termine whether the fiscal multiplier depends on consumer Post − Pre Gi Gi 06 debt. Since the measurement error stemming from spillovers + β2 × DTI + Controlsi Y Pre i would affect the interaction term between DOD spending and i consumer debt to the same extent, it should not affect our + i, (3) inferences. 06 We conduct our investigation of debt-dependent fiscal mul- where DTIi is the debt-to-income ratio in CBSA i in 06 tipliers at the CBSA level. The county-level analysis offers 2006. Notably, DT Ii is predetermined and exogenous to the the best way to capture heterogeneity in consumer debt but change in economic growth during the recessionary period. fails to capture an economically meaningful government- β2 is the core coefficient of interest; β2 > 0 would indicate spending multiplier. On the other hand, the state-level anal- that expansionary fiscal policy is more effective in geogra- ysis captures a meaningful government-spending multiplier phies suffering from consumer debt overhang, and β2 < 0 but is too coarse to capture heteorgeneity in consumer in- would suggest that fiscal policy is less effective in areas with debtedness. The CBSA-level analysis offers a balanced ap- high consumer debt. proach. In interpreting CBSA-level results, it is important to Given the potentially endogenous nature of government note that the local-income state-level estimates greater than spending, we instrument both the direct effect of government 1 correspond to CBSA-level multiplier estimates of 0.37 (see spending, as well as the interaction between the change in table 2A). government spending and the debt-to-income ratio. Specifi- With this observation in mind, we further validate our data cally, we employ two instruments: a Bartik-style instrument and empirical approach by conducting an additional CBSA- described in equation (2), as well as its interaction with the level analysis. Table 2B illustrates that our IV results are fairly 2006 debt-to-income ratio. Similar to table 2, we control for robust to alternative definitions of real economic activity: em- local industry structure prerecession and a wide set of prere- ployment growth and GDP growth. Overall, results presented cession CBSA-level economic conditions. in table 2 establish baseline estimates of open-economy mul- Table 3 reports the results of this IV analysis for income tipliers at the CBSA level and confirm the validity of the new growth. The results suggest that government spending cre- ates relatively more economic growth in areas with higher consumer leverage.We document a statistically and econom- 11 One can argue that with migration and trade in intermediate goods, ically significant positive coefficient β2 for each measure of county-level open economies should exhibit higher multipliers than state- economic activity (income, employment, or GDP). Figure level relatively more closed economies: it is easier for smaller regions to pull resources from surrounding areas, thus permitting larger output re- 1 summarizes the economic significance of the estimates sponses to fiscal stimulus. This argument, however, relies on government by presenting the implied magnitudes of the government- spending being confined within the borders of a given local economy. In our spending multiplier for different levels of consumer debt and setting, the measurable DOD spending can spill over to other geographic regions and the rate of such spillover is likely to increase as the region different measures of economic activity. In case of income size decreases. For example, 41% of the contracts are implemented in the growth, a 1-standard deviation increase in the debt-to-income postal code where the primary contractor is located. In contrast, 74% of the ratio (0.597) is associated with an increase in the DOD spend- contracts are implemented within the same state. As we move to smaller, × less populous geographic areas, this spillover would be more pronounced ing multiplier of 0.354 (0.608 0.597), or about the average leading to smaller coefficients of interest. CBSA fiscal income multiplier (0.37).

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/rest_a_00796 by guest on 01 October 2021 734 THE REVIEW OF ECONOMICS AND STATISTICS

TABLE 3.—DEBT-DEPENDENT FISCAL MULTIPLIERS Income Growth Employment Growth GDP Growth A. DOD Spending *** *** * *** *** ** DOD Spending03−05 0.37 0.37 −0.51 0.23 0.24 −0.25 0.78 0.74 −2.86 (3.86) (3.86) (1.66) (3.60) (3.64) (1.16) (1.31) (1.31) (2.48) *** *** *** *** ** *** Debt-to-income2002 −0.03 −0.04 −0.03 −0.03 −0.02 −0.04 (5.79) (6.50) (5.65) (5.65) (2.00) (3.41) ** ** *** DOD Spending03−05 0.61 0.34 2.54 × Debt-to-income2002 (2.36) (2.01) (2.89) CBSA-level controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Number of observations 828 824 824 828 824 824 372 372 372 Kleibergen-Paap LM test 13.25*** 13.21*** 8.73*** 13.25*** 13.21*** 8.73*** 5.60** 5.54** 8.02*** B. DOD Obligations *** *** * *** *** ** DOD Spending03−05 0.35 0.36 −1.04 0.23 0.23 −0.58 0.54 0.51 −5.01 (2.91) (2.89) (1.72) (2.91) (2.91) (1.55) (0.95) (0.94) (2.26) *** *** *** *** * *** Debt-to-income2002 −0.03 −0.04 −0.03 −0.03 −0.02 −0.04 (5.43) (5.83) (5.56) (5.33) (1.93) (2.75) ** * ** DOD Obligations03−05 1.03 0.60 4.35 × Debt-to-income2002 (1.98) (1.90) (2.12) CBSA-level controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Number of observations 828 824 824 828 824 824 372 372 372 Kleibergen-Paap LM-test 9.10*** 9.06*** 5.36** 9.10*** 9.06*** 5.36** 2.32 2.31 3.76* This table presents the results of the CBSA-level IV analysis of the effect of DOD spending on income, employment, and GDP growth between 2007 and 2009 following regression equation (3). Standard errors are clustered by state. T -statistics are reported in parentheses. ∗∗∗ p < 0.01, ∗∗ p < 0.05, and ∗ p < 0.1.

One can also look at the economic significance from the period, leading to a deeper local recession and biasing our perspective of DOD spending being able to mitigate the coefficient of interest downward. Indeed, we find that DOD adverse effects of consumer debt overhang. The direct ef- spending in general and our instrument in particular are neg- fect of consumer leverage is negative and economically and atively correlated with local income, housing prices, and the statistically significant.12 In the case of income multipli- percentage of the college-educated and white population. To ers, our results suggest that a 1-percentage point increase address this concern, we conducted robustness tests (not re- in DOD spending relative to income reduces the direct neg- ported for breivity) where we control for an exhaustive set of ative effect of consumer leverage by 0.006, or about 16% cross effects between the control variables and prerecession of (−0.033) the DTI coefficient. The results based on DOD DTI. If anything, the resulting coefficients of interest exhibit obligations exhibit stronger debt-dependent multipliers, with higher economic magnitudes, indicating the downward bias a 1-percentage point change in DOD obligations reducing the of the original estimates. negative effect of consumer leverage by 0.01, or about 30% Second, one can argue that our results might be attributed of the direct effect. One might expect the DOD spending mul- to DOD spending multipliers varying with other local eco- tiplier to exhibit stronger debt dependence if heterogeneous nomic and geographic characteristics distinct from leverage. MPC is the only relevant mechanism. If, however, as dis- Ideally, to address this concern, we would like to control for cussed below, debt-dependent multipliers are driven by local DOD spending growth interacted with all local economic co- slack, then the DOD obligations–based multipliers might ex- variates. Implementing such analysis is not feasible with our hibit stronger debt dependency due to firms’ hiring respond- sample size, as it requires us to instrument over thirty inde- ing more to anticipated outlays. pendent variables. To attenuate this concern, we conducted a series of tests where we controlled for the interactions be- tween DOD spending growth and control variables one at a B. Robustness Tests time. Overall, we find the results reported in table 3 to be One can argue that local economic and demographic condi- quantitatively and qualitatively robust to the inclusion of ad- 13 tions distinct from consumer leverage might drive our results. ditional interaction terms. First, our core results might arise due to local economic con- Combined, this evidence suggests that the debt-dependent ditions being correlated with DOD spending. For example, DOD multiplier we document is unlikely to be driven by the it is possible that the government systematically allocates a heterogeneity in local economic or demographic characteris- higher share of the DOD budget to poorer and more diverse tics distinct from consumer leverage. geographies and that residents of such geographies face par- ticularly strong borrowing constraints during the 2008–2009 13Controlling for median local household income (median local house prices) interacted with DOD spending does attenuate the interaction co- 12These results are qualitatively and quantitatively consistent with Mian efficient of interest in two out of six core specifications. Notably, median and Sufi (2015), who document that weakness in consumer balance sheets household prices are correlated with local debt-to-income ratio and can be contributed to local economic slumps. capturing part of the leverage effect.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/rest_a_00796 by guest on 01 October 2021 FISCAL STIMULUS AND CONSUMER DEBT 735

FIGURE 1.—EFFECT OF CONSUMER LEVERAGE ON THE GOVERNMENT A. Economic Mechanisms SPENDING MULTIPLIER Changes in aggregate demand in response to fiscal stimu- lus are associated with changes in private or con- sumption (or both). Private , for example, tend to respond to changes in interest rates associated with fis- cal stimulus, as well as to changes in the expected future marginal product of caused by productive public in- vestment (Baxter & King, 1993). Fiscal stimulus also can affect consumption. Some theories argue that it can decrease consumption through expectations of higher future and increases in real interest rates. Others point to increases in consumption through increases in expected income (Murphy, 2015; Rendahl, 2016), presence of credit-constrained hand- to-mouth consumers (Galí et al., 2007; Eggertsson & Krug- man, 2012), or declines in the real (Eggertsson, 2010). How do these theories inform the mechanisms that might be responsible for the debt-dependent multipliers we docu- ment? Theories that rely on interest-rate channels are unlikely to explain the evidence presented in table 3 since is constant across cities in the . Similarly, our focus on defense spending rules out local public invest- ment as a cause of heterogeneous multipliers. Our evidence is also inconsistent with arguments based on heterogeneous tax liabilities. It is unlikely that the current level of consumer leverage leads to a heterogeneous effect on individual future taxes within the United States. The remaining demand-side channels rely on heterogene- ity in the number of credit-constrained consumers. Eg- gertsson and Krugman (2012) present a Keynesian-style model that demonstrates the efficacy of expansionary fis- cal policy in a debt-overhang-driven recession. In their model, high-debt consumers are forced to deleverage due to a credit contraction. The resulting credit constraints that deleveraging households face imply that the consumption This figure reports the DOD spending multipliers implied by the estimates reported in table 3 at the 25th of high-debt (deleveraging) households responds more to percentile (1.19), 50th percentile (1.44), and 75th percentile (1.83) of the debt-to-income distribution. The dashed line represents the average (across the debt-to income distribution) estimated multiplier. government stimulus than the consumption of low-debt households. Similarly, Galí et al. (2007) present a model with hand-to-mouth consumers that dedicate all newly found in- IV. What Contributes to Heterogeneity in the come to consumption. Both studies imply that more debt- Government-Spending Multiplier? constrained consumers should have higher consumption re- sponses to fiscal stimulus and, by extension, lead to higher While it is important to know whether fiscal stimulus is ef- government-spending multipliers in areas with higher con- fective during consumer-debt-overhang-induced recessions, sumer leverage compared to those with relatively low con- it is no less important to understand what economic mech- sumer leverage. anisms contribute to the heterogeneity in the fiscal multi- To evaluate the validity of the MPC-driven economic ra- plier. Existing economic literature offers a number of chan- tionale for the debt-dependent multiplier, we turn to mi- nels through which government spending can affect real crodata exploited in the prior literature as proxies for con- output. In these theories, the efficacy of fiscal stimulus de- sumption: individual credit card balances (Aaronson et al., pends on its net effect on aggregate demand and whether 2012; Agarwal et al., 2007) and auto purchases (Mian et al., aggregate can accommodate the increase in aggregate 2013). Alongside consumption measures, these microdata al- demand. In this section, we present a discussion and empir- low us to capture debt-to-income ratios at a very granular ical evaluations of aggregate-supply and aggregate-demand individual level. Thus, we draw our inferences from het- economic mechanisms that can lead to debt-dependent DOD erogeneous consumer responses to DOD spending within spending multipliers. the same geographic unit (CBSA). Our identification stems

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/rest_a_00796 by guest on 01 October 2021 736 THE REVIEW OF ECONOMICS AND STATISTICS

from debt-driven heterogeneity in MPCs across individu- card balances between February 2008 and February 2010, als exposed to identical labor market and DOD spending normalized by individual-level prerecession income.15 We conditions. winsorize the credit card growth data at the 1% and 99% lev- els to eliminate the extreme observations, which are likely TU Credit card balances. First, we analyze the response to erroneous. To be consistent with the prior analysis, DT Ik DOD spending of individual credit card balances using the is measured as of February 2007 (the end of 2006). We ex- anonymized TransUnion (TU) panel data provided by the ploit previously defined measures of CBSA-level changes in Federal Reserve Bank of Cleveland. These data cover a ran- government spending. dom sample of about 10 million individuals residing in the CBSA fixed effects control for a wide set of local eco- United States and are reported as of February of each year. nomic conditions during the recession and absorb the direct The data offer a rich set of characteristics of consumers’ effect of government spending. More important, CBSA fixed financial behavior, including total consumer debt balances, effects control for local labor market conditions and, as such, credit card balances, auto balances, and mortgages. eliminate alternative explanations stemming from slack. Our The data also provide consumer characteristics, including identification is based on within-CBSA heterogeneous con- credit scores, income, and the postal code of an individual’s sumer responses to CBSA-level innovation in government residence. spending. Consistently, we cluster the standard errors at the The TU panel is uniquely suited to evaluate our core ques- CBSA level. tion of interest since it reports individual consumer income CBSA fixed effects and the fact that government spend- modeled by TU using a proprietary model. To our knowl- ing is exogenous to individual household financial decisions edge, no other data set offers actual or estimated consumer mitigate potential endogeneity concerns. However, it is still income for a representative and geographically diverse set of possible that, for example, a negative correlation between individual consumers. To check the accuracy of the income government spending and local income might bias our coef- βCC measure from TU, we first aggregated individual-level data ficients of interest 2 . To eliminate this potential bias, we to the county level and correlated the resulting measure with instrument the interaction term in this regression using an county-level income reported by the BEA. The correlation co- interaction between our Bartik-style instrument and individ- efficient is 68% and offers considerable confidence in the TU ual debt-to-income ratios. Furthermore, following Balli and 14 TU estimates. Second, we aggregate the data at the postal code Sørensen (2013), we demean the individual DT Ik within a level and observe a correlation coefficient of 86% between CBSA before interacting it with government spending. TU imputed income and IRS postal code–level income. Similar to our prior analysis, we control for prerecession Using individual total debt balances and individual in- local industry structure and a wide set of prerecession local come, we then build one of the core variables of interest in this economic conditions—though here, we measure these vari- TU ables at the county level. In addition, we control for available study, debt-to-income ratio (DT Ik ) at the consumer-level k as of February 2007. Finally, following the prior literature individual-level characteristics, including (log of) consumer (Aaronson et al., 2012; Agarwal et al., 2007), we capture income measured as of February 2007. Notably, we do not individual consumption by examining growth in credit card control for individual credit score, number of credit accounts, balances. The Survey of Consumer reports that for or credit utilization. These consumer characteristics either individuals with credit cards, about 67% of consumption is directly depend on the debt-to-income ratio or directly con- done via credit card accounts. About 69% of the individuals tribute to higher consumer leverage and thus are capturing in the TransUnion sample have credit cards, and 78% of those similar economic fundamentals, but less precisely. exhibit positive credit card balances. Panel A of table 4 reports the results of the analysis. Col- To evaluate whether individuals with higher prerecession umn 1 suggests that in response to DOD spending, consumers DTI exhibit higher MPCs, we implement the following re- with higher levels of prerecession leverage increase their con- gression analysis: sumption more than households with lower levels of lever- age. The magnitude of the coefficients of the interaction term TU Growth in Credit Card Balances = between DT Ik and the change in DOD spending is signif- k icantly larger than the direct negative effect of prerecession = βCC TU + βCC TU × 1 DT Ik 2 DT Ik DOD Spendingi consumer leverage. A 1-standard deviation increase in gov- + CBSA + County Controls + ε , (4) ernment spending (5%) almost fully mitigates the adverse i k effect of leverage on individual consumption (0.11 × 5% − 16 where growth in credit card balances for individual k resid- versus the direct DTI coefficient of 0.005). ing within CBSA i is the difference in individual-level credit 15While we would like to narrow the window and restrict our analysis to the period after the onset of the crisis in the third quarter of 2007 to the last 14Note that all credit bureaus report information only about individuals quarter of 2009, the nature of the data does not offer us this flexibility. who have a social security number and a , so the aggregate 16Table 4 also shows that high-debt consumers deleverage during 2007– incomes of TransUnion consumers and that reported by and to the BEA are 2009 to a higher extent than low-debt consumers: the direct effect of debt-to- expected to differ. income on credit balances is negative and significant across all considered

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/rest_a_00796 by guest on 01 October 2021 FISCAL STIMULUS AND CONSUMER DEBT 737

TABLE 4.—GOVERNMENT SPENDING AND CONSUMER CREDIT BALANCES A. Response of Consumer Credit Balances to DOD Spending Dependent Variable: Credit Total Mortgage Auto Cards Debt Loans Loans *** *** *** *** Debt-to-income2006 −0.004 −0.138 −0.116 −0.004 (11.79) (45.02) (41.61) (10.11) ** DOD Spending07−09 0.102 0.045 −0.106 0.028 × Debt-to-income2006 (2.21) (0.37) (0.80) (1.22) CBSA FEs and county controls Yes Yes Yes Yes Number of observations 6,412,628 6,412,628 6,412,628 2,820,262 Kleibergen-Paap LM test 11.06*** 18.50*** 18.60*** 18.93*** B. Subprime versus Superprime Borrowers Dependent Variable: Credit Cards Subprime Near Prime Prime Superprime *** *** *** *** Debt-to-income2006 −0.006 −0.004 −0.002 −0.001 (24.72) (14.52) (10.42) (3.81) *** ** DOD Spending07−09 0.120 0.070 0.033 −0.002 × Debt-to-income2006 (2.98) (1.92) (1.17) (0.07) Number of observations = 6,301,755; Kleibergen-Paap LM-test =11.11∗∗∗. This table reports the IV analysis individual-level debt growth (credit cards, mortgages, and auto loans) between 2007 and 2009 normalized by 2007 individual-level income (panel A). In panel B, the core variables of interest are interacted with dummies for subprime, near-prime, prime, and near-super-prime borrowers. All resulting interaction terms are instrumented. Standard errors are clustered by CBSA. T-statistics are reported in parentheses. ∗∗∗ p < 0.01, ∗∗ p < 0.05, and ∗ p < 0.1. Source: TransUnion.

To further examine whether credit constraints are respon- in all reporting periods considered in our analysis. All the sible for the documented debt-dependent MPCs, we conduct robustness tests offer results consistent with those reported the analysis for subgroups of individuals in different credit in table 4. score categories from the potentially most-credit-constrained One can still argue that credit card–based evidence may be subprime borrowers to the likely least-credit-constrained a manifestation of relaxation of borrowing constraints rather prime borrowers. than evidence of higher MPCs driven by hand-to-mouth con- Panel B of table 4 reports the results of the seemingly sumption. Notably, we are indifferent between the two ex- unrelated IV regression, exploring whether our core coeffi- planations, because any additional dollar of debt is financing cients of interest vary with individuals’ prerecession credit consumption and is therefore consistent with the notion that scores: subprime, near-prime, prime, and superprime cate- high-leverage consumers have higher MPCs. Yet to explore gories.17 Consistent with the idea that credit constraints lead the relaxation-of-borrowing-constraints hypothesis, we eval- to higher MPCs, we observe that DOD spending has the uate how DOD spending affects individual-level consumer largest effects on high-debt-to-income consumers if they are debt balances. also in the subprime credit score category—that is, the most Columns 2 through 4 of panel A in table 4 report the anal- credit-constrained group. We observe the effect dissipating ysis of different types of consumer debt and its response to as we move to the near-prime category, and it is virtually DOD spending following regression specification (4). Here, nonexistent economically or statistically in top credit score instead of growth in credit card balances, we evaluate the categories. The results add further validity to an argument growth in total outstanding debt balances, mortgage bal- that DOD spending facilitates consumption by more credit- ances, and auto loans. We find no evidence consistent with constrained individuals to a higher extent than that by less government spending relaxing the borrowing constraints of credit-constrained individuals. high-leverage households more than that of low-leverage Additional robustness tests indicate that these results are households. Once again, these results are robust to various robust to a wide set of regression specifications, data sources dependent-variable specifications and sources of credit bu- used, and approaches to measuring growth in credit card bal- reau data (additional robustness tests are not reported for ances. Furthermore, we find qualitatively similar results when brievity). we focus our analysis on subsets of consumers in an attempt to isolate individuals who are most likely to channel the ma- Auto registrations. One can argue that credit cards are not jority of their consumption through credit card accounts: only an ideal measure of individual consumption, as these ac- for individuals with at least one credit card account and only counts might still be capturing consumer borrowing behavior. individuals who have positive credit card account balances To mitigate this concern, we augment our analysis of credit card balances with the one that uses a direct measure of con- forms of debt. This evidence is consistent with tight credit constraints during sumption: postal code–level auto registration data from R. L. the 2007–2009 period (Eggertsson & Krugman, 2012) and thus validates Polk (see Mian et al., 2013). our choice of core variable of interest (debt-to-income). These data are collected from new automobile registrations 17Subprime, near prime, prime, and superprime are defined as individuals with VantageScore credit scores less than or equal to 700, between 700 and and provide information on the total number of new automo- 799, between 800 and 899, and above 900, respectively. biles purchased in a given geography every year. These data

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/rest_a_00796 by guest on 01 October 2021 738 THE REVIEW OF ECONOMICS AND STATISTICS

TABLE 5.—GOVERNMENT SPENDING AND AUTO REGISTRATIONS effective during periods of high unemployment (Michaillat, Growth in Auto Registrations 2012; Murphy, 2017). ** * In Michaillat (2012), an increase in government sector em- Debt-to-income2006 −0.020 −0.018 (2.08) (1.82) ployment increases labor market tightness and crowds out *** DOD Spending07−09 1.721 – private sector employment, thus diminishing the impact of × Debt-to-income2006 (3.84) – ** government spending on economic output. Yet during the pe- DOD Obligations07−09 – 1.817 × Debt-to-income2006 – (2.36) riods of high slack (high unemployment) in a local economy, CBSA FEs and county-level controls Yes Yes the new government-spending-driven jobs have little influ- Number of observations 22,509 22,509 ence on labor market tightness, leading to weak crowding Kleibergen-Paap LM test 21.04*** 9.00*** out of the private sector. Consistently, prior empirical litera- This table reports the results of the postal-code-level IV analysis of car registration growth between 2007 and 2009. Standard errors are clustered by CBSA. T -statistics are reported in parentheses. ∗∗∗ p < 0.01, ture documents that fiscal multipliers are higher in times of ∗∗ p < 0.05, and ∗ p < 0.1. Source: R. L. Polk. high unemployment (Auerbach & Gorodnichenko, 2012). The prior literature also leads us to expect higher eco- nomic slack in geographies with higher consumer indebted- capture actual consumption by households at a fairly gran- ness. After all, higher consumer debt in combination with ular postal code level. The addresses in the data are derived housing price declines contributed to household net worth from registrations, so the postal code measure represents the shocks, and the local employment slump via depressed house- postal code of the person who purchased the automobile, not hold consumption (Mian et al., 2013). Combined, these two that of the dealership. streams of the literature suggest that we might observe a debt- Table 5 reports the results of the following instrumental dependent DOD spending multiplier due to higher economic variable analysis: slack in 2008 and 2009 associated with higher prerecession consumer indebtedness. Log growth in auto registrationsz Implementing a direct empirical analysis of the slack chan- = βARDT ITU + βARDT ITU × DOD Spending nel is challenging since slack is endogenous to both local 1 z 2 z i economic conditions and government spending. We approach + CBSAi + County Controls + εz, (5) evaluating the validity of this economic mechanism indirectly by looking at the real output in sectors of the economy that do where growth in auto registrations at the postal code z within not directly benefit from household consumption but depend CBSA i is the (log) growth of the number of auto registra- on local labor market tightness. TU tions between 2009 and 2007. DT Iz is the postal-code- First, we examine multipliers for broad sectors that exhibit level consumer debt-to-income ratio that is based on ag- differential dependence on local : tradable gregated individual-level debt and income data reported by sectors, nontradable and strict nontradable sectors, as well TransUnion. We incorporate the same set of county-level as construction and other (unclassified) sectors of the econ- controls as in the previous analysis. The hypothesis implies omy.18 If only consumption-based mechanisms contribute that we should expect debt-constrained households to exhibit to the debt-dependent multiplier, then we should observe more auto registrations in response to government spending only debt-dependent multipliers in nontradable sectors of the βAR > ( 2 0). economy. Similar evidence in the tradable sectors would sug- The results presented in table 5 are consistent with debt- gest additional, potentially supply-side, mechanisms at work. dependent MPCs and indicate that postal codes characterized Table 6 presents the result of the analysis across different by highly leveraged consumers tend to experience a larger sectors following regression equation (4). To make the sector increase in auto registrations in response to DOD spending multipliers comparable, we normalize both DOD obligations during the 2007–2009 period compared to zip codes with less and spending and growth in sectors’ income by total CBSA- levered consumers. This increase in consumption is unlikely level income in 2007 (i.e., total CBSA income, not sector- to be funded with debt (panel A of table 4). Combined, tables specific income). 4 and 5 offer strong evidence of debt-dependent MPCs and Table 6 suggests that the multiplier increases with con- thus validate the MPC-driven economic mechanism underly- sumer leverage not only in industries affected by local con- ing the debt-dependent DOD spending multiplier. sumer spending, such as construction and (strict) nontrad- ables, but also in sectors that do not directly benefit from B. Aggregate Supply Economic Mechanisms local household spending. The debt-dependent multipliers in tradable sectors are significantly above 0 and of similar mag- Aggregate supply constraints, however, can counteract the nitude to those in nontradable sectors. The largest cross-effect effect of increases in aggregate spending. In the simplest one-period Ricardian endowment economy, consumption de- clines one-for-one with government purchases regardless of 18Mian and Sufi (2015) isolate the consumption-driven mechanisms be- hind the adverse effect of consumer debt on real economic growth by focus- the presence of hand-to-mouth consumers. In contrast, recent ing on the nontradable sector. Their test primarily relies on local consumer theoretical work argues that demand stimulus might be more expenditure almost by definition.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/rest_a_00796 by guest on 01 October 2021 FISCAL STIMULUS AND CONSUMER DEBT 739

TABLE 6.—DEBT-DEPENDENT MULTIPLIER:TRADABLE AND NONTRADABLE SECTORS Nontradable (strict) Nontradable Tradable Construction Other A. DOD Spending ** DOD Spending07−09 −0.042 −0.035 −0.075 −0.060 −0.177 (2.41) (1.39) (1.63) (1.26) (1.22) ** ** *** *** Debt-to-income2006 −0.001 −0.002 −0.006 −0.016 −0.010 (0.82) (2.44) (2.21) (6.77) (3.29) ** * ** * DOD Spending07−09 0.034 0.032 0.068 0.079 0.223 × Debt-to-income2006 (2.41) (1.64) (1.89) (2.10) (1.75) CBSA-level controls Yes Yes Yes Yes Yes Number of observations 824 824 824 824 824 Kleibergen-Paap LM-test 8.73*** 8.73*** 8.73*** 8.73*** 8.73*** B. DOD Obligations ** * DOD Obligations07−09 −0.062 −0.059 −0.118 −0.134 −0.388 (2.51) (1.66) (1.55) (1.60) (1.42) ** ** *** *** Debt-to-income2006 −0.001 −0.002 −0.006 −0.016 −0.011 (0.87) (2.46) (2.20) (6.59) (3.14) ** * * DOD Obligations07−09 0.051 0.052 0.103 0.137 0.387 × Debt-to-income2006 (2.38) (1.77) (1.60) (1.85) (1.61) CBSA-level controls Yes Yes Yes Yes Yes Number of observations 824 824 824 824 824 Kleibergen-Paap LM-test 5.36** 5.36** 5.36** 5.36** 5.36** This table reports the CBSA-level IV analysis of income growth in different sectors of the economy between 2007 and 2009 following the regression equation (3). Standard errors are clustered by state. T-statistics are reported in parentheses. ∗∗∗ p < 0.01, ∗∗ p < 0.05, and ∗ p < 0.1.

coefficient is documented for “other” industries that are diffi- TABLE 7.—STATE-DEPENDENT MULTIPLIERS:NATIONAL SECURITY SECTOR cult to classify into tradables, nontradables, or construction. A. DOD Spending The evidence suggests that economic mechanisms distinct *** *** * DOD Spending07−09 0.079 – 0.079 −0.141 from the consumer spending channel are likely contributing (2.90) – (2.92) (1.77) ** to the debt-dependent multiplier. Debt-to-income2006 – 0.001 0.001 −0.001 – (1.56) (2.28) (0.99) Second, to further validate the slack mechanism, we iso- ** DOD Spending07−09 –––0.152 late industries that benefit from supply-side channels but are × Debt-to-income2006 –––(2.28) unlikely to be affected by consumption-based mechanisms. CBSA control variables Yes Yes Yes Yes Isolating such industries is challenging since most indus- Number of observations 828 824 824 824 Kleibergen-Paap LM-test 13.25*** – 13.21*** 8.73*** tries ultimately produce consumer goods. We found only one industry that fits these requirements: the National Security B. DOD Obligations ** ** * and International Affairs sector (NAICS 9811). This sector DOD Obligations07−09 0.076 – 0.076 −0.266 (2.41) – (2.42) (1.85) is unique as it cannot directly benefit from increases in house- ** Debt-to-income2006 – 0.001 0.001 −0.001 hold consumption, but can benefit from local economic slack. – (1.56) (2.03) (1.08) ** Any debt-driven heterogeneity in the government-spending DOD Obligations07−09 –––0.252 × multiplier in this sector is unlikely to be explained by the Debt-to-income2006 –––(2.06) CBSA control variables Yes Yes Yes Yes individual consumption behavior. Number of observations 828 824 824 824 Table 7 reports the results. Consistent with the notion that Kleibergen-Paap LM-test 9.10*** –9.06*** 5.36** household consumption does not directly affect employment This table reports the CBSA-level IV analysis of 2007–2009 income growth in the National Security and International Affairs sector (NAICS 9811) between 2007 and 2009. Standard errors are clustered by and wages in this industry, we find that high prerecession state. T -statistics are reported in parentheses. ∗∗∗ p < 0.01, ∗ p < 0.05, and ∗ p < 0.1. consumer leverage does not have economically or statisti- cally significant effects on this sector of the economy dur- ing the recession. Yet we observe economically and statis- tically significant debt-dependent fiscal spending multipli- sults imply that both aggregate demand and aggregate supply ers in this sector. This evidence cannot be explained by any channels contributed to the debt-dependent multipliers dur- consumption-driven economic channel and is consistent with ing the recession. the slack channel. To summarize, while tables 6 and 7 do not offer direct C. Multiplier during Periods of Credit Expansion evidence in support of the slack channel, they do provide a strong indication that supply-side frictions such as local Notably, both economic mechanisms underlying debt- slack are likely to contribute to the debt-dependent multiplier. dependent multipliers are unique to the recessionary environ- Along with our evidence from consumer microdata, these re- ment and associated credit contraction. High debt-to-income

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/rest_a_00796 by guest on 01 October 2021 740 THE REVIEW OF ECONOMICS AND STATISTICS

TABLE 8.—DEBT-DEPENDENT FISCAL MULTIPLIERS DURING BOOMS fact that the abundance of consumer debt and the associ- A. DOD Spending ated increase in leverage reduces the efficacy of government ** ** spending in stimulating a local economy. It can also be in- DOD Spending03−05 0.42 0.42 2.09 (2.14) (2.11) (1.32) dicative that abundance of debt stimulates consumption and ** ** Debt-to-income2002 0.04 0.04 by extension reduces slack in the economy. (2.34) (2.38) DOD Spending03−05 −1.26 × Debt-to-income2002 (1.12) V. Conclusion CBSA-level controls Yes Yes Yes Number of observations 828 824 824 The Great Recession illustrates the importance of con- Kleibergen-Paap LM-test 6.47** 6.18** 9.70*** sumer balance sheets during an economic downturn. A num- B. DOD Obligations ber of academic studies document that accumulation of debt DOD Obligations03−05 0.51 0.49 8.27 by consumers set the stage for the 2007 crisis and then slowed (1.61) (1.61) (1.13) ** ** the economic recovery. In such an environment, both fiscal Debt-to-income2002 0.03 0.05 (2.12) (2.14) and monetary authorities face the challenge of designing a − DOD Obligations03−05 5.58 proper policy response, particularly because high consumer × Debt-to-income2002 (1.10) CBSA-level controls Yes Yes Yes debt balances are frequently invoked to question the efficacy Kleibergen-Paap LM-test 3.18* 3.63* 3.30* of expansionary fiscal policy. “You cannot solve a problem This table presents the CBSA-level IV analysis of the effect of DOD spending on income growth between created by debt by running up even more debt, say the critics” 2003 and 2005 similar to regression equation (3). Standard errors are clustered by state. T -statistics are reported in parentheses. ∗∗∗ p < 0.01, ∗∗ p < 0.05, and ∗ p < 0.1. (Eggertsson & Krugman, 2012). This paper empirically investigates how the geographic heterogeneity in prerecession consumer leverage affects the leads to higher MPCs when consumers are credit constrained open-economy relative fiscal multiplier during the 2007– (Eggertsson & Krugman, 2012). In contrast, during periods of 2009 recession. Using new transaction-level data on DOD credit expansion, individual indebtedness is not necessarily spending, we document that during the 2007–2009 reces- associated with credit constraints and high MPCs (Guerreri sionary period, the DOD spending multiplier is higher in & Lorenzoni, 2017). CBSAs with higher prerecession consumer debt-to-income Similarly, the relationship between leverage and subse- ratios than in CBSAs with lower prerecession consumer debt- quent slack was a particular feature of the Great Recession to-income ratios. The evidence suggests that at least in the and was attributed to households’ need to delever (Mian & short run (two years considered in this study), expansionary Sufi, 2015). During periods of credit expansion, leverage is fiscal stimulus has the capacity to mitigate the adverse effects likely to positively contribute to economic growth and lead to of consumer leverage on local employment and income. higher employment (lower slack) as agents borrow to finance We augment these results by presenting evidence consis- their current consumption. Therefore, finding a positive cor- tent with both aggregate demand and aggregate supply mech- relation between household leverage and the government- anisms. On the aggregate demand side, we find evidence spending multiplier during an expansionary period would supporting heterogeneous MPC-based explanations. Our re- suggest a debt-dependent multiplier that cannot be attributed sults show that in response to an increase in DOD spending, to binding consumer credit constraints. households with high debt-to-income ratios tend to increase To explore whether the debt-dependent multiplier is unique consumption more relative to households with low debt-to- to the recessionary environment, we implement a cross- income ratios. sectional analysis during the boom period of 2003 through We also find evidence consistent with local economic slack 2005. Specifically, we use 2002–2003 economic indicators contributing to the heterogeneity in government-spending as measures of preboom activity and 2004–2005 indicators multipliers. Higher consumer indebtedness depresses house- as measures of boom activity. For consistency with our prior hold consumption and contributes to local unemployment. analysis, we conduct the analysis at the CBSA level and use Yet it creates a more fruitful environment for fiscal stimu- the 2002 consumer debt-to-income ratio. lus because it leads to local excess capacity. In the presence The results presented in table 8 confirm that the debt- of local economic slack, government spending is unlikely to dependent multiplier we document is an attribute of a re- crowd out the private sector, which in turn leads to higher cessionary environment when credit constraints are likely government-spending multipliers. binding. During the 2003–2005 period, consumer leverage Overall, our results contribute to the debate about the ef- positively affected economic growth (consistent with Lout- ficacy of expansionary fiscal policy and add to our under- skina & Strahan, 2015), yet we do not observe a positive rela- standing of the economic mechanisms through which gov- tionship between government spending multipliers and local ernment spending operates. While we are unable to quantify consumer leverage. If anything, during this period, high debt the relative importance of these demand-side and supply-side is associated with lower fiscal multipliers (although the es- mechanisms, our evidence points to the importance of both timates lack statistical significance), which may reflect the channels and calls for future research in this area.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/rest_a_00796 by guest on 01 October 2021 FISCAL STIMULUS AND CONSUMER DEBT 741

REFERENCES Kaplan, Greg, and Giovanni L. Violante, “A Model of the Consumption Aaronson, Daniel, Sumit Agarwal, and Eric French, “The Spending and Response to Fiscal Stimulus Payments,” Econometrica 82 (2014), Debt Response to Minimum Wage Hikes,” American Economic Re- 1199–1239. view 102 (2012), 3111–3139. Klein, Mathias, “Austerity and Private Debt,” Journal of , Credit, Agarwal, Sumit, Chunlin Liu, and Nicholas S. Souleles, “The Reaction of and Banking 49 (2017), 1555–1585. Consumer Spending and Debt to Tax Rebates,” Journal of Political Loutskina, Elena, and Philip Strahan, “Financial Integration, Housing and Economy 115 (2007), 986–1019. Economic ,” Journal of Financial Economics 115:1 (2015), Auerbach, Alan, and Yuriy Gorodnichenko, “Measuring the Output Re- 25–41. sponses to Fiscal Policy,” American Economic Journal: Economic Mian, Atif, Kamalesh Rao, and Amir Sufi, “Household Balance Sheets, Policy 4 (2012), 1–27. Consumption, and the Economic Slump,” Quarterly Journal of Eco- Balli, Hatice Ozer, and Bent E. Sørensen, “Interaction Effects in Econo- nomics 128 (2013), 1687–1726. metrics,” Journal of Empirical Economics 45 (2013), 583–603. Mian, Atif, and Amir Sufi, “House Prices, Home Equity–Based Borrow- Barro, Robert J., and Charles J. Redlick, “Macroeconomic Effects from ing, and the U.S. Household Leverage Crisis,” American Economic Government Purchases and Taxes,” Quarterly Journal of Economics Review 101 (2011) 2132–2156. 126 (2011), 51–102. ——— “What Explains the 2007–2009 Drop in Employment?” Economet- Baxter, Marianne, and Robert G. King, “Fiscal Policy in General Equilib- rica 82 (2015), 2197–2223. rium,” American Economic Review 83 (1993), 315–334. Michaillat, Pascal, “Do Matching Frictions Explain Unemployment? Not Bernardini, Marco, Selien De Schryder, and Gert Peersman, “Heteroge- in Bad Times,” American Economic Review 102 (2012), 1721– neous Government Spending Multipliers in the Era Surrounding the 1750. Great Recession,” CESifo working paper 6479 (2017). Misra, Kanishka, and Paolo Surico, “Consumption, Income Changes, and Bernardini, Marco, and Gert Peersman, “Private Debt Overhang and the Heterogeneity: Evidence from Two Fiscal Stimulus Programs,” Government Spending Multiplier: Evidence for the United States,” American Economic Journal: 6 (2014), 84– Journal of Applied Econometrics (forthcoming). 106. Chodorow-Reich, Gabriel, “Geographic Cross-Sectional Fiscal Spending Murphy, Daniel P., “How Can Government Spending Stimulate Consump- Multipliers: What Have We Learned,” American Economic Journal: tion?” Review of Economic Dynamics 18 (2015), 551–574. 11:2 (2019), 1–34. ——— “Excess Capacity in a Fixed-Cost Economy,” European Economic Cloyne, James S., and Paolo Surico, “Household Debt and the Dynamic Review 91 (2017), 245–260. Effects of Income Tax Changes,” Review of Economic Studies 84:1 Nakamura, Emi, and Jón Steinsson, “Fiscal Stimulus in a Monetary Union: (2017), 45–81. Evidence from U.S. Regions,” American Economic Review 104 Davidoff, Thomas, “Supply Constraints Are Not Valid Instrumental Vari- (2014), 753–792. ables for Home Prices Because They Are Correlated with Many Parker, Jonathan A., Nicholas S. Souleles, David Johnson, and Robert Mc- Demand Factors,” Critical Finance Review 5:2 (2015), 177–206. Clelland, “Consumer Spending and the Economic Stimulus Pay- Eggertsson, Gauti, “What Fiscal Policy Is Effective at Zero Interest Rates?” ments of 2008,” American Economic Review 103 (2013), 2530– in Daron Acemoglu and Michael Woodford, eds., NBER Macroeco- 2553. nomics Annual 2010 (Chicago: University of Chicago Press, 2010). Ramey, Valerie, “Identifying Government Spending Shocks: It’s All in the Eggertsson, Gauti B., and Paul Krugman, “Debt, Deleveraging, and the Liq- Timing,” Quarterly Journal of Economics 126:1 (2011), 1–50. uidity Trap: A Fisher-Minsky-Koo Approach,” Quarterly Journal of Ramey, Valerie, and Sarah Zubairy, “Government Spending Multipliers Economics 127 (2012), 1469–1513. in Good Times and in Bad: Evidence from U.S. Historical Data,” Galí, Jordi, J., David López-Salido, and Javier Vallés, “Understanding the Journal of Political Economy 126 (2018), 850–901. Effects of Government Spending on Consumption,” Journal of the Rendahl, Pontus, “Fiscal Policy in an Unemployment Crisis,” Review of European Economic Association 5:1 (2007), 227–270. Economic Studies 83 (2016), 1189–1224. Guerrieri, Veronica, and Guido Lorenzoni, “Credit Crisis, Precautionary Sahm, Claudia R., Matthew D. Shapiro, and Joel Slemrod, “Balance-Sheet Savings and the Liquidity Trap,” Quarterly Journal of Economics, Households and Fiscal Stimulus: Lessons from the Payroll Tax Cut 132 (2017), 1427–1467. and Its ,” NBER working paper 21220 (2015). Jappelli, Tullio, and Luigi Pistaferri, “Fiscal Policy and MPC Heterogene- Shapiro, Matthew D., and Joel B. Slemrod, “Consumer Response to Tax ity,” American Economic Journal: Macroeconomics, 6 (2014), 107– Rebates,” American Economic Review 93 (2003), 381–396. 136.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/rest_a_00796 by guest on 01 October 2021