The Effect of Student Debt on Consumption: A State-Level Analysis

The Eﬀect of Student Debt on Consumption: A

State-Level Analysis

Berrak Bahadir

∗

and Dora Gicheva

∗

October 2019

Abstract

Using state-level data for the 2003–2018 period, we show that changes in the state-

level student debt-to-income ratio are associated with lower consumption growth over

the next four years. We use exogenous increases in the annual limits for subsidized

federal student loans and grants to identify a causal relationship between student loans

and consumption. Our results are robust to controlling for mean reversion and using

alternative lag structures.

JEL Codes: D14, E21, I22

Keywords: Student loans, household credit, consumption

∗

Department of Economics, University of North Carolina at Greensboro

1 Introduction

According to data from the New York Fed Consumer Credit Panel, outstanding student

debt is currently at $1.49 trillion, representing the second largest type of household credit

after mortgage debt. As such, student loans constitute a critical component of households’

budgets, especially since repayment starts soon after the completion of schooling, when

earnings tend to be relatively low. Student debt assumes even more importance for young

borrowers who are credit constrained and are forced to lower consumption when paying oﬀ

student loans.

Using state-level data for the 2003–2018 period, we investigate the macroeconomic im-

plications of the rise in outstanding student debt by studying the eﬀects of student loans on

aggregate consumption. Because the relationship between student debt and consumption is

confounded by educational attainment, our empirical strategy uses the exogenous increase

in annual limits for subsidized Staﬀord loans following the Higher Education Reconciliation

Act of 2005 and several increases in the maximum Pell Grant to identify a causal impact.

The microeconomic literature has been uncovering statistically and economically signiﬁ-

cant relationships between student debt and household decisions such as marriage (Gicheva,

2016; Sieg and Wang, 2018) or homeownership (Cooper and Wang, 2014; Mezza, Ringo,

Sherlund and Sommer, 2019). While this literature ﬁnds that student loans lead borrowers

to postpone important life decisions that are closely tied to consumption choices, the link

between student debt and household consumption has not been examined directly.

Macroeconomic studies have not examined the consequences of student loans but rather

focus on total household credit or only on mortgage debt (Beck, B¨uy¨ukkarabacak, Rioja and

Valev, 2012; Schularick and Taylor, 2012; Mian, Suﬁ and Verner, 2017). Recent trends in

the data underline the importance of studying student loans separately from mortgages and

other types of consumer credit. As Figure 1 shows, student loans have grown substantially

since 2003, while the growth in other types of debt has been mostly negative since the Great

Recession. As a result, student loans have become an increasingly important component of

household credit with potentially unique consequences for consumption dynamics.

Figure 1: State-Level Student Debt and Other Household Debt As Fractions of Income

1.2

Student debt/income

Student debt

Other debt

2004 2006 2008 2010 2012 2014 2016 2018

1.1

0.1

0.9

0.05

0.8

Year

Other debt/income

Weighted average student debt-to-income and other household debt-to-income ratios. Sources: Federal

Reserve Bank of New York (debt) and Bureau of Economic Analysis (per-capita personal income).

2 Empirical Methodology and Data

The model we are interested in estimating for state i in year t is:

it+3

− c

] = α

+ β[StudentLoans

it−1

− StudentLoans

it−4

] + γ · [X

it−1

− X

it−4

(1)

+ (ν

it−τ

− c

it−τ −1

]) + ε

τ =1

where c denotes the natural log of consumption; StudentLoans measures the ratio of per-

capita outstanding federal and private student loans to personal income; X

is a vector of

time-varying controls at the state level including changes in other household credit and the

share of the state’s population with some college and with a four-year degree; α

is a state

ﬁxed eﬀect; and ε

is a random error term. Some of the models we estimate include lagged

consumption growth to account for possible reversion toward the mean; previous studies

such as Mian et al. (2017) use a similar approach. We allow for clustering of the errors at

the state level.

An exogenous increase in borrowing is expected to boost aggregate demand and may

hence have an immediate positive eﬀect on consumption. Possible negative eﬀects are ex-

pected to occur when borrowers start to service their debt. To capture this medium-run

eﬀect, we use the change in outstanding student loans from year t − 4 to t − 1 to predict

changes in consumption between t and t + 3.

The length of this lag structure also corre-

sponds to the time it takes for a typical student to ﬁnish college and start paying oﬀ the

debt.

Student debt may have a positive association with consumption through increased educa-

tional attainment and earnings potential. Unobserved heterogeneity also plays an important

role, for example through the eﬀects of family resources on both student borrowing and sub-

sequent labor market outcomes. We rely on an instrumental variable (IV) approach, which

uses the fact that the limit for subsidized Staﬀord loans increased from $2,625 to $3,500 for

freshmen and from $3,500 to $4,500 for sophomores in the 2007–08 academic year,

and the

maximum Pell grant, which we expect to have negative relationship with student debt, has

been increasing by a diﬀerent amount most years.

To isolate cross-state variation in out-

standing student debt due to the above federal policy changes, we interact the increases in

the federal loan and grant limits with, respectively, the state-speciﬁc share of undergraduate

The literature has used a three- to four-year horizon of private credit changes to examine the eﬀect of

credit expansions on macroeconomic outcomes; see Mian et al. (2017). Our results are robust to using two

and four-year lags.

We use an increase of $875 in our empirical analysis.

The limits for unsubsidized Staﬀord loans increased by $2,000 for graduate students starting in the fall

of 2007 and for undergraduate students in the following year. We verify that our results are robust to using

this increase as an additional instrumental variable in the model, but its relationship with student debt is

weaker so we omit it from our preferred speciﬁcations. These results are available on request.

students borrowing at the subsidized Staﬀord limit and the state-speciﬁc share of undergrad-

uate students receiving any Pell grant in the 1999-2000 academic year, or prior to the start of

the study period. The latter two variables serve the role of exposure measures, which predict

how strongly changes in federal policies should aﬀect state-level changes in the per-capita

ratio of student loans to personal income. While an increase in the maximum Staﬀord loan

is unlikely to aﬀect students borrowing lower amounts, Pell limit increases tend to result in

higher grant aid for all Pell recipients.

To account for the fact that policy changes likely

take time to aﬀect borrowing behavior and the amount of accumulated student debt, we use

changes in loan and grant limits between years t − 5 and t − 2 to instrument for student loan

changes between t − 4 and t − 1.

Annual state-level data on outstanding student debt and other household debt come from

the New York Federal Reserve Bank’s Consumer Credit Panel, which is constructed from

a random sample of Equifax credit reports. We use per-capita consumption data from the

Bureau of Economic Analysis. We construct average educational attainment at the state

level using data from the American Community Survey. Finally, the shares of students in

each state who borrowed the maximum subsidized Staﬀord loan or received any Pell Grant

are based on data from the National Postsecondary Student Aid Study.

3 Results

We document the relationship between student loans and consumption in Table 1. Columns 2

through 4 present baseline results from OLS speciﬁcations with an increasing set of controls.

In the most parsimonious speciﬁcation, 1 percentage point (slightly more than a standard

deviation) increase in the change in student debt to income ratio is associated with 0.9

Lucca, Nadauld and Shen (2018) use similar policy variations to study how postsecondary institutions

change their prices in response to changes in student loan availability.

percentage point (less than a third of a standard deviation) decrease in the growth rate of

consumption over the next three years. When we control for educational attainment, which

is reasonable to assume to be positively correlated with both student debt and consumption

growth, the coeﬃcient estimate decreases to -1.3. Including lagged consumption growth

allows us to control for state-level variations in macroeconomic conditions simultaneous with

the changes in student debt; inclusion of these variables strengthens the negative relationship

between student debt and consumption.

The ﬁnal two columns of Table 1 report results from the IV speciﬁcations, with ﬁrst-stage

results presented in the bottom panel. We ﬁnd that both instruments are signiﬁcant deter-

minants of student loans with the expected signs. The results show that 1 percentage point

increase in the change in student debt to income ratio is associated with 2.5 percentage point

lower consumption growth rate during the subsequent three years in the speciﬁcation with-

out controls for lagged consumption changes, and 3.7 percentage point lower consumption

growth with these controls. These eﬀects are larger in magnitude than the OLS estimates,

which is consistent with the expected bias driven by changes in educational attainment.

Further, the magnitude of these results cannot be accounted for by a direct increase in ed-

ucation spending since data from the Consumer Expenditure Survey suggest that spending

on education accounted for only 2.3% of aggregate expenditures in 2018.

4 Conclusion

Using state-level data for the 2003–2018 period, we show that an increase in the student

debt-to-income ratio contributes to lower consumption growth in the medium run. A possible

mechanism for the results is that credit constrained young borrowers, who start paying oﬀ

student loans soon after they graduate when earnings are relatively low, are forced to lower

Table 1: Relationship Between Student Debt and Other Household Debt and Consumption

Per Capita

(1) (2) (3) (4) (5) (6)

Mean OLS OLS OLS IV IV

[sd]

Dependent variable: Δ

ln C

t+3

0.033

[0.031]

(StudentDebt/Income)

t−1

0.018 -0.914*** -1.288*** -1.770*** -2.480*** -3.684***

[0.008] (0.231) (0.260) (0.248) (0.327) (0.411)

(OtherDebt/Income)

t−1

-0.033 -0.123*** -0.120*** -0.116*** -0.114*** -0.093***

[0.149] (0.015) (0.014) (0.017) (0.013) (0.016)

%CollegeGrad

t−1

0.008 0.467** 0.115 0.503** 0.051

[0.010] (0.186) (0.150) (0.196) (0.144)

%SomeCollege

t−1

0.007 0.471*** -0.058 0.748*** 0.145

[0.014] (0.110) (0.107) (0.120) (0.126)

Δ ln C

t−1

,Δ ln C

t−2

,Δ ln C

t−3

No No Yes No Yes

First-Stage Results

SubsidizedLimit

t−2

× Exposure 0.043*** 0.035***

(0.006) (0.006)

P ellLimit

t−2

× Exposure -0.08*** -0.008***

(0.004) (0.003)

OtherDebt/Income

t−1

0.004 0.008***

(0.003) (0.003)

%CollegeGrad

t−1

-0.030 -0.065**

(0.036) (0.029)

%SomeCollege

t−1

0.128*** 0.050

(0.027) (0.032)

F stat of excluded instruments

Hansen J statistic

P-value of J statistic

27.4

1.359

0.244

20.8

1.768

0.184

Notes: * p<0.10, ** p<0.05, *** p<0.01. Time subscripts denote the end period and number of lags;

for example, Δ

StudentDebt/GDP

t−1

stands for (StudentDebt/GDP )

t−1

− (StudentDebt/GDP )

t−4

. The

models include state ﬁxed eﬀects. The reported errors are clustered at the state level. N = 450.

their consumption, generating signiﬁcant eﬀects at the aggregate level. This mechanism

is consistent with the ﬁndings of prior studies suggestive of student borrowers being credit

constrained after graduation (e.g. Rothstein and Rouse, 2011) and underlines the importance

of binding credit constraints for aggregate macroeconomic outcomes.

Our study is the ﬁrst to combine the literature on the unintended consequences of student

debt and the existing research on the macroeconomic eﬀects of household debt. To our

knowledge, we are also the ﬁrst to directly examine the link between student debt and

consumption using an exogenous variation in student borrowing. Last but not least, our

results are informative of the degree to which student loan debt can aﬀect non-borrowers

through its impact on macroeconomic conditions.

References

Beck, Thorsten, Berrak B¨uy¨ukkarabacak, Felix K Rioja, and Neven T Valev,

“Who gets the credit? And does it matter? Household vs. ﬁrm lending across countries,”

The BE Journal of Macroeconomics, 2012, 12 (1).

Cooper, Daniel and J Christina Wang, “Student Loan Debt and Economic Outcomes.

Current Policy Perspective No. 14-7.,” Federal Reserve Bank of Boston, 2014.

Gicheva, Dora, “Student loans or marriage? A look at the highly educated,” Economics

of Education Review, 2016, 53, 207–216.

Lucca, David O, Taylor Nadauld, and Karen Shen, “Credit supply and the rise in

college tuition: Evidence from the expansion in federal student aid programs,” The Review

of Financial Studies, 2018, 32 (2), 423–466.

Mezza, Alvaro, Daniel Ringo, Shane Sherlund, and Kamila Sommer, “Student

Loans and Homeownership,” Journal of Labor Economics, 2019, forthcoming.

Mian, Atif, Amir Suﬁ, and Emil Verner, “Household debt and business cycles world-

wide,” The Quarterly Journal of Economics, 2017, 132 (4), 1755–1817.

Rothstein, Jesse and Cecilia Elena Rouse, “Constrained after college: Student loans

and early-career occupational choices,” Journal of Public Economics, 2011, 95 (1-2), 149–

163.

Schularick, Moritz and Alan M Taylor, “Credit booms gone bust: Monetary policy,

leverage cycles, and ﬁnancial crises, 1870-2008,” American Economic Review, 2012, 102

(2), 1029–61.

Sieg, Holger and Yu Wang, “The impact of student debt on education, career, and

marriage choices of female lawyers,” European Economic Review, 2018, 109, 124–147.