| Abstract: |
February 2, 2015{{p}}How Does Student Loan Debt Affect Light Vehicle
Purchases?{{p}}{{p}}Christopher Kurz and Geng Li{{p}}{{p}}Light vehicle sales
plummeted during the recent financial crisis and the ensuing recession.1
However, since the first quarter of 2009, sales of automobiles and light
trucks have rebounded, spurring production and investment, as a result, gains
in employment. Looking ahead, a sustained robust level of vehicle sales will
likely provide an important signal of, and be a direct contributor to, a
broader economic recovery. Among many factors that affect vehicle sales,
recent research has shown that household indebtedness helps explain the drop
and subsequent recovery in vehicle sales.2 Moreover, against the backdrop of
rapidly growing student loan balances, the potential effect on household
spending of a heightened burden of student debt has received increased
attention from analysts and in the press. For example, analyzing the Consumer
Credit Panel FRBNY/Equifax (CCP) data, Brown and Caldwell (2013) find evidence
that suggests that the marked rise in student loan burdens in recent years may
be weighing on debt-funded vehicle purchases.{{p}}{{p}}The CCP data, while
providing timely information on borrowing for a large sample of Americans,
have certain limitations that stand in the way of making a direct and
comprehensive inference about student debt and auto purchases. First, only
debt-funded auto purchases, which account for about half of all auto purchases
in U.S., can be inferred from the CCP data. Second, purchases of new and used
vehicles cannot be separated in the CCP data, a distinction that matters when
measuring economic activity. Third, except a borrower's age, the CCP data do
not contain any demographic and socioeconomic characteristics of the consumers
and their households, which are known to be important determinants of auto
purchases. Our study relies on another data source, the Consumer Expenditure
Survey (CE), to explore the relationship of household debt (student loans in
particular) and consumer expenditures. Specifically, we will introduce the
unique features of the CE data, discuss the quality of household balance-sheet
related data therein, and explore the impact of student loan indebtedness on a
household's decision to finance the purchase of a vehicle and whether the
vehicle is new or used.{{p}}{{p}}I. The Consumer Expenditure
Survey{{p}}{{p}}The Consumer Expenditure Survey is the only regular U.S.
household survey that collects detailed information on household demographic
and socioeconomic characteristics, consumption expenditures, and balance
sheets. The survey covers a large nationwide representative sample: about
7,000 households have been surveyed each quarter in recent years. Each sampled
household is surveyed four times a year before being replaced. Johnson and Li
(2009) provide a detailed description of the household debt information that
the CE collects and find that these data compare favorably with their
counterparts in the Survey of Consumer Finances (SCF), which is widely viewed
as the best available data source of U.S. household balance sheets (but
contains very limited expenditure data). Johnson and Li (2010, forthcoming)
further demonstrate that both the debt service-to-income ratios calculated
using the CE data and the mortgage-type information CE collects have
significant informational merit for understanding consumers' liquidity
conditions. Our analysis below uses data for households in the CE sample who
participated in all four quarterly interviews, who were headed by a person
between 20 and 65 years old, and whose annual household before tax-income was
between $3,000 and $300,000.{{p}}{{p}}a. Student loan information in the
CE{{p}}{{p}}Prior to the release of 2013 CE (the latest release), student loan
balance information was lumped together with several other types of household
credit, such as personal loans and pension plan loans. However, the SCF data,
which have detailed information on each of such loans, reveal that the vast
majority of the combined balances of these household debts likely represent
student loans.3 Therefore, we use the combined loan balances as an
approximation of student loan balances. Since the second quarter of 2013,
student loan debt balances and services have been reported separately in the
CE. Our analysis focuses on how a household's student loan indebtedness at the
time of their first survey period is related to that household's vehicle
purchases during the subsequent four-quarter interval.{{p}}{{p}}Figure 1
contrasts the share of households with positive student loan balances
estimated using the CE data with that in the SCF data from 1996 to 2013.4 We
find that the CE-series underestimate the prevalence of student loan borrowing
relative to the SCF-series by a significant margin. However, the two series
share a very similar trend, both showing a sizable increase (of a similar
magnitude) in the share of households with some outstanding student debt.
Furthermore, as shown in Figure 2, the average balances of student loan debt
among the borrowers are remarkably close in the two surveys, in terms of both
the level and the trend over time.{{p}}Figure 1: Share of Student Loan
Borrowing Households in the SCF and CE Data{{p}}Figure 1: Share of Student
Loan Borrowing Households in the SCF and CE Data. Figure 1 shows the fraction
of student loan holders within the Survey of Consumer Finances and the
Consumer Expenditure Survey from 1996 to 2013. While both lines trend upward,
the Survey of Consumer Finances data is above the Consumer Expenditure Survey
fraction throughout the figure by roughly 6-7 percent. The Survey of Consumer
Finances fraction ranges from roughly 12 percent at the beginning of the
sample to roughly 20 percent at the end of the sample.{{p}}Figure 2: Average
Student Loan Balances among Borrowers in the SCF and CE Data{{p}}Figure 2:
Average Student Loan Balances among Borrowers in the SCF and CE Data. Figure 2
plots the average student loan balances in the Survey of Consumer Finances and
the Consumer Expenditure Survey from 1996 to 2013. Both balances are trend up
together and the levels are similar. The series trend up from about 10-13
thousand dollars to about 27 thousand dollars.{{p}}{{p}}In addition, both the
CE and SCF data show that many households across the age spectrum owe student
debt. Accordingly, our analysis will examine the effect of such debt on auto
purchases among all consumers aged between 20 and 65, instead of focusing on
only the young borrowers.{{p}}{{p}}b. Light vehicle purchases data in the
CE{{p}}{{p}}The CE records detailed information on the household vehicle
inventory, and, important for our analysis, allows us to identify whether a
vehicle purchased while the household was in the survey was used or new, and
whether there was a loan involved in the purchase. Table 1 summarizes the
share of households buying a vehicle in a given year, the split of new versus
used, and whether the purchase was financed. As the table indicates, over the
period of 1996–2013, about 28 percent of households in our CE sample purchased
an auto or a light truck in a given year, split about one-third, two-thirds
between new and used vehicles. In addition, more than three-quarters of the
new purchases were financed by a loan, whereas only about 45 percent of used
purchases were debt-financed.5{{p}}Table 1: Share of Households Buying a Light
Vehicle{{p}}Table 1: Share of Households Buying a Light Vehicle. Fraction of
Households: 28%; New: 32%; New - Financed: 76%; New - Cash: 24%; Used: 73%;
Used - Financed: 44%; Used - Cash: 56%. Source: Authors' estimates using the
Consumer Expenditure Survey data. Note: The new and used shares sum to more
than 100 percent, as some Households bought both a new and a used vehicle in a
year.{{p}}{{p}}II. The Effect of Student Loan Debt on Vehicle
Purchases{{p}}{{p}}We begin with a simple logistic regression where an
indicator variable for whether the household purchased a vehicle is regressed
on measures of household indebtedness--the amount of debt owed relative to the
household's income--and a vector of demographic and socioeconomic
characteristics. Our variables of indebtedness are measured as of the
beginning of the panel, representing pre-determined balance-sheet conditions
with respect to subsequent auto purchases. Specifically, our measures include
the total "other" debt balance-to-income ratio and the student loan debt
balance-to-income ratio, which are calculated, respectively, as the ratio of
the level of debt and to household pre-tax income. The "other" debt measure
excludes student loans and is the sum of mortgage, home equity, auto, and
credit card debt outstanding. Importantly, separating student loan borrowing
from total indebtedness will address whether or not student loans impart a
differential effect upon consumer expenditures as opposed to overall
indebtedness. To allow for nonlinear effects of indebtedness on auto
purchases, we also include, for each of the two ratios, a high-debt-to-income
indicator that is set to equal to 1 if the ratio is greater than the 90th
percentile of its distribution across households and zero otherwise. The
control variables include age, race, education, employment, marital status,
the number of children, the log of income, homeownership, and a dummy for
being a student. To proxy for expected earnings we include an interacted term
of the age and education controls. The logistic model specification also
controls for the student and work status for a spouse (if present) as well.
Calendar-year fixed effects are also included in the regression.{{p}}{{p}}The
debt-related variables of interest from the simple logistic regression are
reported in column 1 of Table 2. For each variable, we report the estimated
coefficient, standard error, and its associated odds ratio, which is the
estimated coefficient-implied probability of purchasing a vehicle divided by
the probability of not purchasing.6 The results indicate that, on average, the
total other debt-to-income ratio does not have a significant effect on auto
purchases except when it becomes particularly high: The households with
particularly high levels of total other debt relative to their income are on
average 15 percent less likely to purchase a vehicle in a given year. However,
the likelihood of purchasing a vehicle increases with the amount of student
loan debt owed by a household except for very high levels of student debt.
Indeed, a 10 percent increase in student loan-to-income ratio increases the
likelihood of auto purchases by 4 percent for the average indebted consumer
even after controlling for an individual's current income, educational
attainment, and our age-education proxy for expected earnings. That said, the
regression shows that having very high levels of student loan debt, as defined
by being above the 90th percentile of student-loan to income indebtedness, is
associated with a significantly lower probability of purchasing a vehicle (45
percent lower). Taken altogether, while the average consumer's student loan
debt-to-income ratio is positively related to auto purchases, those highly
indebted of student loans are significantly less likely to purchase a new or
used vehicle.7{{p}}{{p}}As shown earlier, a consumer may buy a used or a new
vehicle, and may make the purchase with or without financing. The effects of
household indebtedness on auto and light truck purchases may differ by the
underlying vehicle and financing choices. To address this possibility, we
employ a multinomial logistic (MNL) framework, an extension of the binary
logistic regression that allows for more than two categories of the dependent
or outcome variable. The five outcomes are modeled as {not purchasing an auto,
purchasing a used auto with cash, purchasing a used auto with financing,
purchasing a new auto with cash, purchasing a new auto with financing} and are
denoted respectively alternatives j∈{0, 1, 2, 3, 4}. The probability of the
mode, j, of purchasing a vehicle for household i is modeled
as:{{p}}\displaystyle p_{i j} = \frac{\text{exp}(X_i\beta_j)}{1 + \sum^4_{j=1}
\text{exp}(X_i \beta_j)} ,{{p}}Where pj is the probability of choosing option
j and βj is the vector of coefficients pertaining to choice j.{{p}}{{p}}The
estimated coefficients, standard errors, and the associated odds ratios are
reported in columns 2 through 5 in Table 2. The odds ratios are calculated
relative to the outcome of not purchasing a vehicle. Several of the estimated
coefficients of the MNL model have a different sign than those of the binary
logistic regression, underscoring that indebtedness indeed affects vehicle
purchases differently depending on vehicle and financing choices.{{p}}Table 2:
Logit Coefficient Estimates of Vehicle Purchase Choice{{p}}Logit Multinomial
Logit{{p}}(1) (2) (3) (4) (5){{p}}Used - cash Used - financed New - cash New -
financed{{p}}Total Debt to Income 0.00 -0.01 0.01 -0.02 0.01{{p}}Standard
Error (0.00) (0.00) (0.00) (0.00) (0.00){{p}}Odds Ratio 1.00 0.99 1.01 0.98
1.01{{p}}High Total Debt -0.17 -0.01 -0.51 0.58 -0.18{{p}}Standard Error
(0.06) (0.09) (0.10) (0.19) (0.10){{p}}Odds Ratio 0.85 0.99 0.60 1.78
0.83{{p}}Student Loan Debt to Income 0.04 0.04 0.05 -0.22 0.04{{p}}Standard
Error (0.01) (0.01) (0.02) (0.09) (0.02){{p}}Odds Ratio 1.04 1.04 1.05 0.80
1.04{{p}}High Student Loan Debt -0.62 -0.34 -1.04 -11.78 -0.94{{p}}Standard
Error (0.25) (0.35) (0.41) (997.05) (0.49){{p}}Odds Ratio 0.54 0.71 0.35 0.00
0.39{{p}}N 50,111{{p}}50,111{{p}}{{p}} Note: The specifications are run using
data from 1996 to 2013. In addition to the above estimates, controls are
included for age, race, education, employment, marital status, student status
, children, income, and homeownership. The specification also controls for the
student and work status for a spouse (if present). *,**,** represent
significance at the 1, 5, and 10 percent, respectively.{{p}}{{p}}Turning to
the specific results, a higher total other debt-to-income ratio is associated
with a lower probability of purchasing a vehicle--regardless whether it was a
new or used one--with cash, but increases the probability of purchasing a
vehicle when the purchase is financed. These results are broadly consistent
with the notion that consumers with a higher total other debt-to-income ratio
may have less cash available to finance a vehicle purchase or simply prefer
debt-financing for other reasons. That said, households with a high
debt-to-income ratio are significantly less likely to purchase a vehicle (new
or used) with financing.8{{p}}{{p}}We now turn our focus to the impact of
student loans on vehicle purchases. First, aside from new cash-purchases, the
student loan debt-to-income ratio has a positive and significant effect on
purchases of motor vehicles. For example, for new purchases that are financed,
a 10 percent increase in student loan indebtedness would increase the
likelihood of a household purchasing an automobile by 4 percent in a given
year (column 5). In the case of new cash-purchases, an admittedly small
segment of overall sales, the negative coefficient on student loan debt
mirrors that of total debt-to-income, indicating that consumers with higher
debt outstanding tend to be more cash-constrained. Second, those with a
student loan-to-income ratio at or above the 90th percentile have, on balance,
a lower likelihood of buying a vehicle, and such effects are economically and
statistically significant when the purchase was financed with a loan.9 ,
10{{p}}{{p}}To summarize, we use household-level data from the CE to shed new
light on how student loan debt affects auto purchases. We find that the level
of student loan a household owes indeed affects their choice of whether to buy
a new or used vehicle and how to finance that purchase. On balance, apart from
those with a very high level of student loan debt, consumers with a higher
student loan-to-income ratio tend to have a higher likelihood of buying a
vehicle, except when buying a new vehicle with cash. To be sure, while our
results control for a rich set of household characteristics, they do not
explicitly identify whether or not the higher likelihood of purchasing a
vehicle results from an unobserved variable, or features specific to student
loan debt. That said, the results are in contrast to the popular view that
student loan debt may hinder vehicle purchases and indicates that more
research is warranted in this area.{{p}}{{p}}References{{p}}{{p}}Brown and
Caldwell (2013) "Young Student Loan Borrowers Retreat from Housing and Auto
Markets," Liberty Street Economics, Federal Reserve Bank of New York, May,
2013.{{p}}{{p}}Johnson, Kathleen W. and Geng Li. (2009) "Household Liability
Data in the Consumer Expenditure Survey." Monthly Labor Review, vol. 132,
no.12, 18-27.{{p}}{{p}}Johnson, Kathleen W. and Geng Li (2010),"The Debt
Payment to Income Ratio as an Indicator of Borrowing Constraints: Evidence
from Two Household Surveys," Journal of Money, Credit, and Banking, vol. 42,
no. 7, pp. 1373-90.{{p}}{{p}}Johnson, Kathleen W. and Geng Li (forthcoming)
"Are Adjustable-Rate Mortgage Borrowers Borrowing Constrained?" Real Estate
Economics.{{p}}{{p}}Johnson, Kathleen W., Karen Pence, and Daniel Vine (2014)
"Auto Sales and Credit Supply," Finance and Economics Discussion Series
2014-82, Federal Reserve Board.{{p}}{{p}}Mian, Atif, and Amir Sufi (2010)
"Household Leverage and the Recession of 2007 to 2009," IMF Economic Review,
May, 2010.{{p}}{{p}}Mian, Atif, and Amir Sufi (2011) "Consumers and the
Economy, Part II: Household Debt and the Weak U.S. Recovery," FRBSF Economic
Letter, January, 2011{{p}}{{p}}{{p}}1. "Light vehicles" refer to automobiles
and light trucks, hereafter referred to as "vehicles" or "autos" Return to
text{{p}}{{p}}2. For example, see Mian and Sufi (2010, 2011). Return to
text{{p}}{{p}}3. For example, the 2007 SCF data show that among the households
that have a positive balance on either student loans, personal loans, or
pension plan loans, 80 percent of them have a positive balance of student
loans and the student loan balances account for 87 percent of total balances
of such loans. Return to text{{p}}{{p}}4. The SCF values for the years in
which no survey was conducted are linearly interpolated. Return to
text{{p}}{{p}}5. The numbers for used and new vehicle financing are consistent
with the findings of Johnson, Pence, and Vine (2014). Return to
text{{p}}{{p}}6. For the student loan and 'other' debt-to-income variables,
the odds ratio is calculated for a 10 percent increase in the particular debt
ratio. Return to text{{p}}{{p}}7. Importantly, the results are robust to
employing alternative measures of student loan indebtedness, such as the level
of student loan debt or a categorical indicator of a positive student loan
balance. Return to text{{p}}{{p}}8. Interestingly, when total debt-to-income
ratio is controlled for, the high total debt households appear to be more
likely to buy a new vehicle with cash. Return to text{{p}}{{p}}9. The effect
of high student loan debt on new cash purchases is imprecisely estimated
because too few high-student loan debtors bought a new vehicle with cash in
our sample. Return to text{{p}}{{p}}10. There is little difference in median
pre-tax income and overall total indebtedness between new cash and new
financing purchasers. That said, if new cash purchasers maintain a student
loan balance, their median student loan to income ratio, at about 8 percent,
is roughly 1/2 of the student loan to income ratio for households financing
new auto purchases. Return to text{{p}}{{p}} Disclaimer: FEDS Notes are
articles in which Board economists offer their own views and present analysis
on a range of topics in economics and finance. These articles are shorter and
less technically oriented than FEDS Working Papers.{{p}}Search Working
Papers{{p}}{{p}}{{p}}{{p}}Skip Meet Economists Section{{p}}{{p}}Meet the
Economists{{p}}All Economists{{p}}By Field of Interest{{p}}Financial
Economics{{p}}International Economics{{p}}Macroeconomics{{p}}Mathematical and
Quantitative Methods{{p}}Microeconomics{{p}}{{p}}Skip stay connected
section{{p}}{{p}}Stay Connected{{p}}Twitter{{p}}YouTube{{p}}RSS
Feeds{{p}}Subscribe{{p}}{{p}}{{p}}Last update: February 2, 2015 |