Accounting for the Rise in College Tuition Grey Gordon and Aaron Hedlund Indiana University and University of Missouri
NBER/CRIW October 17, 2015
Motivation
6000
(mean) nettuit_per_fte 7000 8000 9000
10000
Real net tuition per FTE at 4-year, non-profit colleges:
1985
1990
1995 2000 Academic Year
2005
What is causing net tuition to rise?
2010
Motivation
Many theories exist. Supply side: I
Baumol’s cost disease — costs increase, productivity does not.
I
Cuts in government aid — reductions passed on to student.
I
Bowen rule — “arms race of spending” (Ehrenberg 2002).
Demand side: I
Bennett hypothesis — colleges capture student aid rents.
I
College premium increases — rents captured.
Our goal: test these theories quantitatively.
Method We combine I
a mostly standard lifecycle model with
I
Epple, Romano, Sarpca, and Sieg (2013)’s model of colleges.
In this paper, only one college, a monopolist. Rent extraction is exaggerated. We feed in estimates or statutory law for exogenous processes: I
college costs,
I
college non-tuition revenue (including government aid),
I
borrowing limits, interest rates, and grants,
I
and the college earnings premium.
Results
Between 1987 and 2010, net tuition increased 78%. All theories together account for a tuition increase of 106%. Separately, holding else equal at 1987 values, I
The supply-side theories decrease tuition by 6%.
I
Changes in student aid cause tuition to increase by 102%.
I
The college earnings premium causes tuition increases of 24%.
“Optimistic” predictions if aid and college premium stop changing.
Model Youths are born with sY , a vector of parental income and ability. Youth problem (in college): Yj (l, sY ) =
max
c+φ≥0,l 0 ≥l
u(c + φ) + β
πYj+1 (l 0 , sY )+ (1 − π)Es 0 |j,sY V (0, l 0 , tmax , s 0 , 0)
s.t. c + T (sY ) + φ ≤ |{z} e Y + ξEFC (sY ) + ζ(sY ) + bs + bu | {z } |{z} | {z } | {z } R&B
earn.
transfers
gov grant
ann. borrow
statutory limits on borrowing (ls0 , lu0 , ls , lu , bs , bu ) = f (l 0 , l)
Worker problem is a mostly standard lifecycle problem
Worker Problem
Model
The decision to enroll is made at time zero: max{Y1 (0, sy ) + q + α, Es|sY V (0, 0, 0, s, 0)} | {z } | {z } college
q ≡ college quality α ≡ preference shock
work
Model The college problem: max q(θ, I )
I ≥0,T (·)
s.t. E N + T N = F + C(N1 ) + I N Endogenous θ ≡ average ability I ≡ investment N1 ≡ freshmen N ≡ NPV of freshmen T ≡ average net tuition Exogenous E ≡ endowment (non-tuition revenue) F ≡ fixed cost C(·) ≡ college “custodial costs” We parametrize q(θ, I ) as χq θχθ I 1−χθ . Full college problem
Data and Estimation We use NLSY97, IPEDS/Delta Cost Project, and take estimates from the literature. Change in exogenous variables that form basis for our experiments: Exogenous variable log college premium student loan interest room and board average gov grant subsidized limit unsubsidized limit non-tuition revenue per student fixed cost of college (billions) marginal cost, relative change
Label λ i φ ζ¯ l¯s l¯u E F 2 2 C /C1987
Note that unsubsidized loans began in 1993.
1987 .46 4.7 3072 488 23994 0 17843 12 1
2010 .66 3.0 9129 1779 23000 40805 18418 30 4.7
Data and Estimation
We estimate a number of parameters inside the model: Param ξ χθ χq α
Description transfer size ability input quality level pref. shock
Value .208 .252 2.68 .003
Target avg tuition ρ(p.inc,enroll) enroll rate % with loans
Data 5788 .295 .379 35.7
Model 6100 .316 .325 42.7
and some others I didn’t show you earlier. To keep tuition down with the monopolist, need low transfers and more marginal students, which introduces bias. Nontargeted, 1987/2010
Results Net tuition, investment, and HS grad enrollment 0.5 35000
Net tuition (model) Net tuition (data) Investment (model)
30000
Investment (data)
0.45
Enrollment (model) Enrollment (data)
2010 dollars
0.4
20000 0.35 15000
0.3 10000
5000 1990
1995
2000
2005
Year
Unsubsidized loans began in 1993.
0.25 2010
Enrollment rate
25000
Results Net tuition, investment, and HS grad enrollment 0.5 35000
Net tuition (model) Net tuition (data) Investment (model)
30000
Investment (data)
0.45
Enrollment (model) Enrollment (data)
2010 dollars
0.4
20000 0.35 15000
0.3 10000
5000 1990
1995
2000
2005
Year
Possible explanations for exaggerated tuition:
0.25 2010
Enrollment rate
25000
Results Net tuition, investment, and HS grad enrollment 0.5 35000
Net tuition (model) Net tuition (data) Investment (model)
30000
Investment (data)
0.45
Enrollment (model) Enrollment (data)
2010 dollars
0.4
20000 0.35 15000
0.3 10000
5000 1990
1995
2000
Year
1) Tuition smoothing?
2005
0.25 2010
Enrollment rate
25000
Results Net tuition, investment, and HS grad enrollment 0.5 35000
Net tuition (model) Net tuition (data) Investment (model)
30000
Investment (data)
0.45
Enrollment (model) Enrollment (data)
2010 dollars
0.4
20000 0.35 15000
0.3 10000
5000 1990
1995
2000
2005
Year
2) Colleges learning about willingness to pay?
0.25 2010
Enrollment rate
25000
Results Net tuition, investment, and HS grad enrollment 0.5 35000
Net tuition (model) Net tuition (data) Investment (model)
30000
Investment (data)
0.45
Enrollment (model) Enrollment (data)
2010 dollars
0.4
20000 0.35 15000
0.3 10000
5000 1990
1995
2000
2005
Year
3) Earnings premium forecast errors?
0.25 2010
Enrollment rate
25000
Results Statistic College costs College endowment Borrowing limits Interest rates Non-tuition cost Grants College premium Mean net tuition Enrollment rate % taking out loans Ability of graduates Investment Ex-ante utility
1987
Experiment * * * * * *
$6100 0.33 42.7 0.76 $21550 -40.98
* $7583 0.29 50.5 0.78 $22793 -40.99
$12345 0.27 100.00 0.80 $27338 -40.97
$5762 0.48 51.1 0.66 $20034 -40.78
2010 * * * * * * * $12559 0.48 100.00 0.74 $26837 -40.36
Monopolist extracts almost all rent: look at utility and borrowing.
Results Statistic College costs College endowment Borrowing limits Interest rates Non-tuition cost Grants College premium Mean net tuition Enrollment rate % taking out loans Ability of graduates Investment Ex-ante utility
1987
Experiment * * * * * *
$6100 0.33 42.7 0.76 $21550 -40.98
* $7583 0.29 50.5 0.78 $22793 -40.99
$12345 0.27 100.00 0.80 $27338 -40.97
$5762 0.48 51.1 0.66 $20034 -40.78
So, Bennett hypothesis moves tuition drastically.
2010 * * * * * * * $12559 0.48 100.00 0.74 $26837 -40.36
Results Statistic College costs College endowment Borrowing limits Interest rates Non-tuition cost Grants College premium Mean net tuition Enrollment rate % taking out loans Ability of graduates Investment Ex-ante utility
1987
Experiment * * * * * *
$6100 0.33 42.7 0.76 $21550 -40.98
* $7583 0.29 50.5 0.78 $22793 -40.99
$12345 0.27 100.00 0.80 $27338 -40.97
$5762 0.48 51.1 0.66 $20034 -40.78
2010 * * * * * * * $12559 0.48 100.00 0.74 $26837 -40.36
Increased costs cause more enrollment, lower ability, lower tuition.
Results
Micro evidence on pass-through rate from FSLP: I
Turner (2014): 12% (Pell).
I
Long (2004): up to 30% (Hope Scholarship GA).
I
Lucca, Nadauld, and Shen (2015): up to 65% (broad msr.).
I
Cellini and Goldin (2014): for-profit 78% higher tuition at FSLP-eligible schools.
For us, rough aggregate pass-through rates: I
Grants: 85% = (12559-11454)/(1779-488).
I
Borrowing Limits: 13% = (12559-9949)/(.5*40805-900)
Results
How can increased college costs result in lower tuition?
College costs
Intuition: I
Cost increase driven by F .
I
Tuition for current students is maxed out.
I
Reduce average cost by increasing enrollment, which lowers tuition by a composition effect.
The form of the cost increase matters for Baumol cost disease. The estimation does allow for implicit substitution of students. Baumol might explain tuition increases for individual colleges.
Conclusion
In an Epple et al. type model with a college monopolist, I
existing theories can explain the full tuition increase,
I
demand-side theories can explain the increase on their own,
I
and supply-side theories work in the wrong direction.
In future research, need multiple colleges to I
discipline market power,
I
reduce bias in parameter estimates,
I
allow for welfare implications, and
I
determine drivers of tuition by school type.
Results Second order effects can be large. Tuition cdf 1
1987 College costs fixed College endowment fixed 0.8
Borrowing limits fixed
Cumulative frequency
Interest rates fixed Non-tuition cost fixed 0.6
Grants fixed College premium fixed 2010
0.4
0.2
0 2000
4000
6000
8000
10000
12000
14000
16000
2010 dollars
Grants, borrowing limits drastically increase tuition at bottom tail.
Results Tuition function in 1987
12000
10000
Tuition
8000
6000
4000
2000 0 0.2 0.4 Ability
250000 0.6
200000 150000
0.8
100000 1
50000 0
Back
Parental income
Results Tuition function in 2010
22000
20000
Tuition
18000
16000
14000
12000
10000 0 0.2 0.4 Ability
250000 0.6
200000 150000
0.8
100000 1
50000 0
Back
Parental income
Results Enrollment comparison between 1987 and 2010 Pr(attend)>=.5 in 1987 and 2010 Pr(attend)>=.5 only in 2010 200000
Pr(attend)>=.5 only in 1987 Pr(attend)<.5 in 1987 and 2010
Parental income
150000
100000
50000
0 0
0.2
0.4
0.6
0.8
1
Ability
High ability, middle income priced out in 1987, not 2010.
Back
Model Unsimplified college problem: max q(θ, I )
I ≥0,T (·)
s.t. E + T = F + C + I α(sY ) = Prob(enroll|sY , T (sY ), q(θ, I )) θ = E(xα)/E(α) C = Nc (N1 J) T = NI (E(T α)) E = NI (E E(α)) I = NI (I E(α)) where E(x) the expectation over newborns, Pis JY −1 Nf (x) := j=0 (1 + r )−j f (π j x) computes a net present value, and I is the identity function. Back
Data and Estimation College cost function estimates: Estimated aggregate cost function 2010
Total cost (billions of 2010 dollars)
30
2005
25
20
15
2000 1995 1990 1987
1.8
1.85
1.9
FTE students / age 18 population
Back
Results Loans
Consumption
1987,attain=2
35000
70000
1987,attain=5
30000
60000
2010 dollars
2010 dollars
2010,attain=2
25000
2010,attain=5
50000
20000 15000
40000
10000
30000
0 20
30
40
50
60
70
80
90
Population with loans Fraction of workers
1 0.8 0.6 0.4 0.2 0 20
30
40
50
60
70
80
90
20000 20
Fraction of workers with loans
5000
30
40
50
60
70
80
90
Default or bad standing population 0.5 0.4 0.3 0.2 0.1 0 20
30
40
50
60
70
80
90
≈ 50% never default. Debt overhang for those who dropped out.
Model Workers, conditional on not defaulting: V R (a, l, t, s) =
max u(c) + βEs 0 |s V (a0 , l 0 , t 0 , s 0 , f 0 = 0)
c≥0,a0 ≥a
s.t. c + a0 /(1 + r (a0 )) + p(l, t) ≤ e(s)(1 − τ ) + a l 0 = (l − p(l, t))(1 + i), t 0 = max{t − 1, 0} s ≡ characteristics (age, years of completed college). a0 , a ≡ private credit r (a0 ) ≡ interest on credit (borrowing ⇒ 12.7%, saving ⇒ 2%). l 0 , l, t ≡ student loans and years remaining before loan paid off. p(l, t) ≡ prescribed student loan payment (“on-time payment”). τ ≡ tax (experiments are revenue neutral). V ≡ value from best of repaying and defaulting next period. Default problem is similar, but p(l, t) replaced by γe(s)(1 − τ ), principal l increases upon default, and duration t gets reset to tmax . Back
Data and Estimation
Avg. net tuition Enrollment rate Graduation rate % taking out loans Corr(p.income,enroll) Investment per student Avg. annual loan size College grad ability Corr(ability,enroll) Back
Model 1987 $6100 0.325 0.554 42.7 0.316 $21550 $4663 0.764 0.632
Tuition function (1987)
Data 1987 $5788* 0.379* 0.554* 35.7* $20251 $7144 -
Tuition function (2010)
Model Final SS $12559 0.483 0.554 100.0 0.276 $26837 $6873 0.735 0.782
Data 2010 $10293 0.414 0.594 52.9 0.295* $23750 $8414 0.716 0.522
Enrollment patterns
Data and Estimation The FTE-weighted averages of these measures over time: Trends of key aggregates 25000
0.55
20000
2010 dollars
15000
Net tuition Investment Endowment
0.45
Custodial cost Enrollment (FTE)
10000
Enrollment (HS grad)
0.4 5000
0 1990
1995
2000
2005
0.35 2010
Enrollment rate
0.5
Data and Estimation We estimate the custodial cost function following a similar procedure to Epple, Romano, and Sieg (2006): Estimated aggregate cost function 2010
Total cost (billions of 2010 dollars)
30
2005
25
20
15
2000 1995 1990 1987
1.8
1.85
FTE students / age 18 population
1.9
Results
Consider the FOC of the college problem absent preference shocks: T (sY ) = C 0 (N) + I − E +
qθ (θ, I ) (θ − x(sY )) qI (θ, I )
Direct effect: C 0 ↑⇒ T ↑, so tuition increases. Indirect effect: F + C (·) increases, placing pressure on budget constraint, causing I to fail. So, tuition falls. Back
Literature Nonexhaustive literature roughly divided into strands: Cost disease: Baumol (1967), Archibald and Feldman (2008) Government approp.: Heller (1999), Chakrabarty et al. (2012), Koshal and Koshal (2000), Titus et al. (2010), Cunningham et al. (2001) Bennett: McPherson and Shapiro (1991), Singell and Stone (2007), Rizzo and Ehrenberg (2004), Turner (2012,2014), Long (2004,2006), Cellini and Goldin (2014), Lucca et al. (2015), Frederick, Schmidt, and Davis (2012) College premium: Autor, Katz, and Kearney (2008), Katz and Murphy (1992), Goldin and Katz (2007), Card and Lemieux (2001), Andrews et al. (2012), Hoekstra (2009) Structural higher ed.: Abbott et al. (2013), Athreya and Eberly (2013), Ionescu and Simpson (2015), Ionescu (2011), Garriga and Keightley (2010), Keane and Wolpin (2001), Fillmore (2014), Fu (2014), Jones and Yang (2015), Epple, Romano, and Sieg (2006), Epple, Romano, Sarpca, and Sieg (2013).