Is College Pricing Power Pro-cyclical?

Is College Pricing Power Pro-Cyclical? Levi Altringer Jeffrey Summers Department of Economics Linfield College (Accepted for publication by Research in Higher Education, May 2015) We define pricing power as a college’s ability to increase its net tuition revenue by raising its stickerprice for tuition. The greater is the positive effect of sticker-price increases on net tuition revenue, the greater is the pricing power. We gauge variation in the pricing power of private, non-profit baccalaureate colleges by estimating this effect over the academic years 2000-01 through 2012-13 – a span which includes two recessions. Drawing on IPEDS data for 118 colleges, our empirical results reveal that the additional real net tuition revenue earned per dollar increase in the real sticker-price fell by 5% following the 2001-02 recession, but had fully recovered to its pre-recession level by 2004-05. In contrast, since the end of the Great Recession of 2007-09 the real return to sticker-price increases has fallen by 12% with no recovery in sight by the end of our sample period. So while private colleges’ pricing power was pro-cyclical during the sample period’s first recession, the later, more severe Great Recession has produced a larger, longer lasting negative effect on pricing power. We consider the implications for private colleges of this lost pricing power.

1

I. Introduction For a number of years, private colleges have combined increases in their tuition sticker-prices and their institutional financial aid grants as complementary components of their tuition pricing strategy. The goal of this so-called high price/high aid strategy is to generate needed net tuition revenues while shaping enrollment. A succinct description of the strategy was recently provided by Rivard (2014a) who noted that…”The high tuition, high discount business model is often confusing to students and parents, but it is how things are done at most private colleges: the colleges charge high prices and then offer students they want huge discounts.” The strategy’s effects on net tuition revenues have been a research focus. Redd (2000) analyzed the correlation between tuition rates, institutional aid, enrollment, and net tuition revenues at private colleges and universities. He found that schools with more moderate aid increases experienced larger gains in enrollment and net tuition revenues. Breneman, Doti, and Lapovsky (2001) found that a 5% increase in the tuition rate, holding enrollment constant, led to an 11.7% increase in institutional aid and resulted in a 0.5% increase in net tuition revenue at private schools. The implied elasticity of net tuition revenue with respect to the tuition rate was 0.10. Summers (2004) studied private liberal arts colleges and concluded that tuition sticker-price and aid increases were combined at these schools so as to increase their net tuition revenues. But Doti (2004) argued that the strategy would exhibit diminishing returns because there was a limit to how much additional revenue schools could raise through continued increases in aid.1 The strategy’s effectiveness in raising net tuition revenues clearly requires that private colleges have pricing power - which we define as the ability to increase net tuition revenue by raising the sticker-

As state funding for public insitutions has waned, those schools have increasingly relied on tuition and financial aid increases to raise revenues while maintaining access. Hillman (2012) studies the effects of tuition discounting on net tuition revenue at public four-year colleges and universities. 1

2

price for tuition. The greater is the positive effect of sticker-price increases on net tuition revenue, the greater is the pricing power. But in the last few years observers have asserted that changes wrought by the Great Recession of 2007-2009 have reduced this pricing power – perhaps permanently. This calls into question the future usefulness of a high price/high aid strategy. As Lapovsky (2013) notes … “Some schools with high discount rates have decided that their high price/high aid strategy needs to be changed: they have lowered their published price and decreased their discount rate, thus keeping net tuition revenue constant or even increasing it.” Summarizing a recent Moody’s Investors Service survey, Corkery (2013) observed that….“The demand for four-year college degrees is softening, the result of a perfect storm of economic and demographic forces that is sapping the pricing power at a growing number of U.S. colleges and universities.” Moody’s conclusion was based on its observation that ...”Families remain willing to pay for college, but their capacity to pay higher prices has been largely tapped and has dramatically dampened the sector’s capacity to grow tuition revenue.” (Moody’s Investors Service 2013). This sentiment is echoed by Alexander (2014) who asks if the demand for U.S. higher education has peaked. And many of higher education’s CFO’s have questioned the long run sustainability of a their institutions’ business models (Lederman 2013). Recent changes in selected private colleges’ tuition pricing strategies – replacing annual tuition increases with freezes, modifications, or outright reductions – provide evidence supporting an assertion of lost pricing power. The examples of Wittenberg, Middlebury, and Sewanee can be cited in this regard. Rather than increase tuition and financial aid year after year as in the past, Wittenberg University is in its second consecutive year with the sticker-price frozen at the 2012-13 level. In 2010, Middlebury College adopted a tuition-pricing policy it called CPI+1 which placed a cap on tuition and fee increases at 1% above the rate of inflation (Rivard 2014b).

3

One of the earlier and most notable post-Great Recession changes in tuition pricing strategies occurred at Sewanee - The University of the South. In February 2011 Sewanee announced that for the 2011-12 academic year it was reducing its sticker-price for tuition by 10%. Although other private colleges further down the institutional hierarchy had experimented with sticker-price reductions in the past, Sewanee was apparently the first of the more selective baccalaureate colleges to do so. Explaining why, Sewanee’s Vice Chancellor John McCardell noted that sticker-price increases had to be paired with unsustainable increases in financial aid if the school was to remain price competitive. “The current dynamic is for Sewanee to spend heavily on merit aid for top students who are admitted both here and to far less expensive public institutions. And that has forced up the discount rate -- the percentage off sticker price that reflects the various forms of aid provided to students -- to a trajectory that can't continue. For this fall's class, the discount rate was 45 percent, up from 37 percent two years earlier. While 30 percent of students pay the full sticker price, that trend line is not positive for private liberal arts colleges. Others are wrestling with similar issues. As we raise our fees, we see declines in the number of full-paid students. We see a slow death scenario. None of us are going to die next year, but in the long run it's unsustainable." (Jaschik 2011).

While some loss of pricing power at private colleges following the Great Recession may have been expected, we know of no empirical study checking for its existence and estimating its size. In this paper we do so in a way that enables us to investigate whether that loss was more or less severe than the loss which occurred in earlier downturns of the business cycle. We thus provide an historical perspective against which one may judge current circumstances. Drawing on IPEDS data for 118 private colleges, our empirical results reveal that the additional real net tuition revenue earned per dollar increase in the real sticker-price fell by 5% following the 2001-02 recession, but had fully recovered to its pre-recession level by 2004-05. In contrast, since the end of the Great Recession of 2007-09 the return to sticker-price increases has fallen by 12% with no recovery in sight at the end of our sample period. So, while private colleges’ pricing power was pro-cyclical during the sample period’s first recession, the more severe Great 4

Recession has produced a relatively longer lasting, negative effect on pricing power. We will consider the implications for private colleges of this lost pricing power. The layout of our paper is as follows. The next section presents a brief review of the literature focused on business cycle effects in higher education. The third section provides a theoretical derivation of the factors that determine how sticker-price tuition increases affect net tuition revenues. We argue that these factors are likely to vary over the course of the business cycle in a way that makes colleges’ pricing power pro-cyclical. The paper’s fourth section discusses our data and empirical method. This is followed by the presentation of our results and consideration of their implications for private colleges. The paper’s final section offers a summary conclusion and identifies questions for future research.

II. The Business Cycle and Higher Education Ours is apparently the first paper to focus on the business cycle’s effects on pricing power at private colleges. However, a number of studies focus on various other business cycle effects at both public and private colleges and universities. We briefly consider selected examples of these studies. Breneman (2002) provides an historical overview considering the five recessions that occurred over the period from the early 1970s through 2001. He discusses the effect of cyclical variation in public financial support for higher education and notes that public institutions largely responded to cyclical declines in this support by boosting tuition. He observes that the 1980s and 1990s were periods of rapid tuition inflation at private colleges and universities and noted that strategic tuition discounting (the high aid part of the high price/high aid pricing strategy) as a way of managing enrollment was developed during those decades. The causes and effects of variation in state financial support for public higher education over the course of the business cycle have been a focus for many papers. Humphrey (2000) found that 5

state appropriations for higher education are sensitive to the business cycle. For the period 1969-1994 his estimated elasticity of state appropriations with respect to real state per capita income was 1.39. Kane, Orzag Apostolov, Inman, and Reschovsky (2005) studied the interaction between state higher education appropriations and other state budget items from 1977 to 2003. Their empirical results indicate that state spending on higher education, and on other state programs, decline as a state’s unemployment rate rises. By contrast, they show that Medicaid spending increases with unemployment, and provide evidence that Medicaid spending crowds out higher education spending. Their estimates suggest that each one dollar increase in per capita Medicaid expenditures reduces state spending on higher education by 39 to 58 cents . Delaney and Doyle (2011) empirically test Hovey’s (1999) balance wheel theory. This theory hypothesizes that state higher education appropriations rise more than do other appropriations as state budgets increase – as happens during expansions - and that they fall more than do other appropriations when state budgets decrease – as happens during recessions. The authors test this by specifying a cubic relationship between higher education expenditures and all other state expenditures while controlling for various combinations of economic, political, and higher education-specific factors. Their very interesting, robust results support Hovey’s theory. Bhatt, Rork, and Walker (2011) study the substitutability of general fund appropriations and earmarked revenues as sources of state higher education funding over the period 1996-2008. They specify a panel model in which the dependent variable is the share of higher education spending that flows from the states’ general funds. The explanatory variables include real earmarked revenues per student, the unemployment rate as a control for the business cycle, and political and tax base controls. They find that appropriations and earmarked revenues are substitutable, but that the degree of substitutability does not vary over the business cycle.

6

In contrast to the amount of research focused on public schools, the business cycle’s effects on private schools has received much less attention. Doti (2002) studied these effects using data for private schools from NACUBO’s Tuition Discounting Surveys over the period 1990-2001. Controlling for tuition and fees and grants, he found that increases (decreases) in real GDP lead to increases (decreases) in freshman enrollment at these schools. This indicates that changes in real economic activity over the course of the business cycle directly impact enrollment, and by extension, net tuition revenues, at private institutions. Doti (2004) uses NABUCO data for 107 private schools in the years 1992 and 2002 to study how the relationship between tuition discounting and net tuition revenues changed over the decade. While the paper’s focus is not on business cycle effects per se, the second year of its two-year comparison lies at the end of the 2000-02 recession. This suggests that the paper’s results may reflect business cycle effects. Those results indicate that the sample schools’ ability to deploy financial aid as a tool for raising net tuition revenue declined over the decade, with the greatest decline occurring at less selective schools. Our paper provides an estimate of the Great Recession’s effect on pricing power at private colleges, which is just one of the myriad effects caused by that recession. Brown and Hoxby (forthcoming) have prepared a study of numerous other effects. The book’s website indicates that:2

“The recent financial crisis had a profound effect on both public and private universities, which faced shrinking endowments, declining charitable contributions, and reductions in government support. Universities responded to these stresses in different ways. This volume presents new evidence on the nature of these responses, and on how the incentives and constraints facing different institutions affected their behavior. The studies in this volume explore how various practices at institutions of higher education, such as the drawdown of endowment resources,

2

http://www.press.uchicago.edu/ucp/books/book/chicago/H/bo19198130.html

7

the awarding of financial aid, and spending on research, responded to the financial crisis. The studies examine universities as economic organizations that operate in a complex institutional and financial environment. The authors examine the role of endowments in university finances and the interaction of spending policies, asset allocation strategies, and investment opportunities. They demonstrate that universities’ behavior can be modeled using economic principles.”

III. Theory Private colleges raise their tuition sticker-prices in order to increase their net tuition revenues. An explanation for this behavior can be motivated by appeal to the existence of both Bowen and Baumol effects at these schools (Bowen 1980, Baumol and Bowen 1966). Bowen effects cause colleges to raise tuition in search of the additional revenues needed to fund an unlimited number of activities they judge to be quality enhancing. Baumol effects are produced when labor productivity growth occurring outside of higher education exceeds labor productivity growth inside of higher education. This productivity growth differential raises the income available to employment outside of higher education for the very educated workers who are so intensively employed by colleges. This forces schools to increase their employees’ compensation packages in order to attact and retain the educated workers they require. The need to fund these increasing compensation costs serves as impetus for the schools’ desire to raise revenues by increasing their sticker-prices.3

In addition to the rising compensation costs, exogenous increases in technology costs and regulatory compliance costs can also put pressure on schools to raise their sticker-prices. Empirical evidence for the relative importance of Bowen and Baumol effects is mixed. Archibald and Feldman (2011) provide evidence for the primacy of Baumol effects in higher education. Martin and Hill (2014) find evidence for the relative importance of Bowen effects at research universities. 3

8

Our theoretical exposition centers on a derivation of how a one-dollar increase in the stickerprice for tuition affects the dollar value of a colleges’ net tuition revenues.4 We derive the effect on net tuition revenue of a change in the sticker-price and show that the size of this effect importantly depends upon two factors. These factors are; 1) the response of enrollment to changes in the average net tuition price measured as the sticker-price minus the average per-student institutional financial aid grant; and, 2) the response of a college’s average per-student financial aid grant to changes in the sticker-price. We will show that the more responsive enrollment is to an increase in the average net tuition price, and the more colleges increase their per-student grant aid as the sticker-price increases, the less will total net tuition revenue rise with each dollar increase in the sticker-price. We then argue that these factors are likely to vary over the business cycle in a way that makes the return to stickerprice increases pro-cyclical. To begin, consider equations (1) and (2) which specify the average net tuition price paid by students at a private college, PN, and the total net tuition revenue the college earns, NTR. 𝑃𝑁 = 𝑃𝑆 − 𝐴𝐼𝐷(𝑃𝑆 )

(1)

𝑁𝑇𝑅 = 𝑃𝑁 × 𝐸𝑁𝑅(𝑃𝑁 )

(2)

In equation (1) the average net tuition price paid by students is the difference between the stickerprice, 𝑃𝑆 , and the college’s average per-student financial aid grant, 𝐴𝐼𝐷(𝑃𝑆 ). Consistent with the use of a high price/high aid pricing strategy in which the sticker-price and aid rise together, AID is specified as a function of PS where dAID/dPS > 0. In equation (2) we see that net tuition revenue is the product of a college’s average net tuition price, PN, and its enrollment, ENR(PN). Since students’ enrollment decisions are responsive to changes

Please note that our analysis is carried out in terms of unit changes rather than percentage changes. We focus on deriving the marginal effect of a one-dollar increase of the sticker-price on the dollar value of net tuition revenues, not on deriving the elasticity of net tuition revenues with respect to a change in the sticker-price. 4

9

in the average net tuition price, enrollment is expressed as a function of the average net price where we make the standard assumption that enrollment and the net price are inversely related, implying that dENR/dPN < 0. Taking the derivative of equation (2) with respect to 𝑃𝑆 , as in equation (3), we derive the marginal effect of a change in the sticker-price on net tuition revenue.

𝑑𝑁𝑇𝑅 𝑑𝑃𝑆

=(

𝑑𝑃𝑁 𝑑𝑃𝑆

𝑑𝐸𝑁𝑅

) × [𝐸𝑁𝑅 + ((𝑃𝑁 ) × (

𝑑𝑃𝑁

))]

(3)

where: dPN/dPS = (1 – dAID/dPS)

(4)

Since a college interested in raising its net tuition revenue will increase its per student aid grant by no more than one dollar for every dollar increase in the sticker-price, in equation (4) we have 1 ≥ dAID/dPS > 0. This means that in equation (3), 1 > dPN/dPS ≥ 0. We have defined pricing power as a college’s ability to increase its net tuition revenue by raising its sticker-price for tuition. If colleges have pricing power, the change in NTR caused by the change of PS in equation (3) will be positive. Note however that the size of this positive effect will vary depending upon how responsive enrollment is to a change in the average net tuition price, dENR/dPN, and how responsive is a college’s average per-student grant aid to a change in its sticker-price, dAID/dPS. Specifically, dNTR/dPS falls as dENR/dPN becomes more negative and as dAID/dPS approaches one. This is what we may expect to observe in recessions when students and parents become more sensitive to net price changes and colleges raise their financial aid offers in an effort to cushion the potential negative enrollment impact of their sticker-price increases. The opposite would occur as a result of expansions. This suggests that dNTR/dPS will vary over the business cycle – falling as a result of recessions and rising as a result of expansions. We therefore hypothesize that colleges are likely to lose pricing power due to recessions and gain it as a result of expansions. 10

A relevant theoretical issue to consider at this juncture is whether business cycle turning points and changes in pricing power occur at the same time, or whether changes in pricing power phase in with a lag after turning points have already occurred. Lagged changes in pricing power are theoretically possible. Our derivation indicates they would occur if the expected changes in dENR/dPN and dAID/dPS occur with lags. This would happen if it takes some time for the effects of recessions and expansions to spread and set in sufficiently to elicit behavioral responses from students, parents, and colleges. If lags between business cycle turning points and changes in pricing power are present, it is important that we choose an empirical specification that allows us to control for them. 5 A final theoretical issue of note is that the sizes of pro-cyclical swings in pricing power are likely to vary across colleges as a function of their relative positions in the institutional hierarchy. Wealthier, more selective schools with longer queues of applicants attract students who are less pricesensitive (dENR/dPN is less negative) and these colleges have less need to cushion their sticker-price increases with offers of increased financial aid (dAID/dPS is smaller). This means that cyclical variation in colleges’ pricing power is a function of colleges’ selectivity. In the next section we specify an empirical model that enables us to estimate the effects of sticker-price changes on net tuition revenues while allowing for lagged effects and controlling for selectivity differences across schools.

IV. Data and Empirical Method Data Our tuition, net tuition revenue, and selectivity data is drawn from the U.S. Department of Education’s Integrated Postsecondary Education Data System (IPEDS) for the academic years 2000-

A reviewer of this paper correctly observed that as the job market sours in a recession, the opportunity cost of attending college falls helping to make students less responsive to net price changes. If this happens at private colleges, it would be a source of lags between the recessions’ start and the decline in pricing power. 5

11

01 through 2012-13.6 The sample is composed of the IPEDS-reporting, private, non-profit colleges designated by the 2010 Carnegie classification system as Baccalaureate—Arts and Sciences colleges that offer only a bachelor’s degree. Focusing on bachelor’s degree-only schools makes the empirical analysis tractable. These schools have only one sticker-price and their net tuition revenues flow from only one source – their bachelor’s degree-seeking undergraduates.

Schools offering both the

bachelor’s and advanced degrees have multiple sticker-prices (at least one for each degree-type) and multiple sources of net tuition revenue. We avoid the need to disentangle the effects of multiple sticker-price changes on multiple net revenue flows by limiting our sample to bachelor’s degreegranting schools. The IPEDS data for these schools was downloaded and inspected for inconsistencies that clearly indicated reporting errors. After omitting one school due to such errors and another because it does not charge tuition, we ended up with a potential sample of 118 schools. The sticker-price and net tuition revenue figures were converted to year 2001 prices using the Higher Education Price Index (HEPI). HEPI was chosen as the deflator because it measures the average relative level in the prices of a fixed market basket of goods and services purchased by colleges and universities excluding expenditures for research. As such, it measures changes in the costs colleges face for the inputs used to produce their educational outcomes.7 Our estimated effects of sticker-price increases on net tuition revenue are thus scaled relative to these input costs. Table 1 presents descriptive statistics for our sample.

The sample period is determined by data availability as 2000-01 and 2012-13 are the earliest and most recent years for which the needed data is available on IPEDS. Sticker-price is the IPEDS published tuition and required fees variable. Net tuition revenue is the IPEDS tuition and fees total variable which measures the amount of tuition and educational fees revenue, net of any allowances applied in the general purpose financial statements (i.e., institutional financial aid grants). Our selectivity variable is the percent of applicants who are accepted as reported by IPEDS. 6

Information about the HEPI can be found at The Common Fund Institute website https://www.commonfund.org/commonfundinstitute/hepi/pages/default.aspx 7

12

Empirical Method We seek to estimate the effects of sticker-price changes on net tuition revenue across our sample schools, in each year of our sample, in a way that allows us to observe the hypothesized procyclical behavior of these effects. To explain our method for doing so, we begin by considering the cross-section, regression function for net tuition revenue expressed in equation (5).

NTRi = β0 + β1 PS,i + β2 (PS,i x ADMITi) + ei

(5)

In (5) the subscript i designates schools and ei is a stochastic error term. The variables NTR and PS are as previously defined the total net tuition revenue and the sticker-price, while ADMIT is the school’s admissions rate – measured as the percentage of applicants to a school who are accepted. The lower is ADMIT, the more selective is the college. The marginal effect of a change in the stickerprice from equation (5) is given in equation (6).

dNTRi/dPS,i = β1+ β2 ADMITi

(6)

In (5) and (6) we expect β1 > 0 and β2 < 0, the latter coefficient indicating that becoming more selective (lowering ADMIT) raises the return to sticker-price tuition increases. Our empirical method is to specify thirteen versions of cross-section equation (5), one for each academic year of our sample from 2000-01 through 2012-13, and estimate these thirteen crosssection equations as a system using Zellner’s (1962) Seemingly Unrelated Regression (SUR) procedure. SUR allows us to estimate the thirteen equations as a system taking into consideration the fact that the error terms in each equation may be be correlated across the equations. Taking this cross-equation

13

correlation into consideration will improve the efficiency of our estimates relative to those yielded by least-squares estimation of each cross-section equation separately. The procedure also allows us to test the hypotheses that the coefficients β1 and β2 have constant values across the thirteen cross-section equations for t = 2000-01 through 2012-13. We will be able to reject one or both of these hypotheses if the marginal return to sticker-price increases given in equation (6) is pro-cyclical as we have hypothesized. An accessible discussion of the SUR technique can be found in Griffiths, Hill, and Judge (1993).8 An alternative estimation strategy would be to merge the cross-section and time-series dimensions of our dataset and estimate a panel model with school fixed effects. Doing so would raise a question about the proper way to test for pro-cyclical variation in pricing power in this model. A straightforward approach could be to include a slope dummy for PS and (PS x ADMIT) keyed to the recession years of 2000-01, 2001-02, 2007-08, and 2008-09.9 But controlling for cyclical effects by using slope dummies keyed to recession years is an ad hoc approach. Our theoretical discussion in the previous section leads us to expect that pricing power is pro-cyclical. But we have also argued that swings in pricing power may phase in and out over the course of business cycles with lags. If there are lags, the use of slope dummies keyed specifically to the recession years as controls would result in model misspecification.10

Some readers may be more familiar with SUR estimation of time-series equations stacked by cross-section rather than of cross-section equations stacked by time as in our model. However, we are not the first to estimate an SUR of this form. For another example, see Brehm and Saving (1964). 8

This panel model would take the form: NTRit = α0 + α1PSit + α2ADMITit + α3(DUM x PSit) + α4(DUM x ADMITit) + µi + ϵit where DUM is a dummy variable that takes on the values of 1 in the recession years of 2000-01, 2001-02, 200708, and 2008-09 and 0 otherwise, µi are school fixed effects, and ϵit is the stochastic error term. If one thought the sample’s two recessions had different effects, two slope dummies could be included – one for the period 2000-02 and the other for 2007-09. 9

We did merge our data and estimated a panel model with fixed effects and two slope dummies keyed to the sample’s two recessions. The marginal effects of changes in PS and (PS x ADMIT) for non-recession years had the expected signs 10

14

A major benefit of the SUR specification is that it makes no a-priori assumption about whether cyclical swings in pricing power occur either with or without lags. But because the procedure allows the coefficients β1 and β2 in equation (5) and (6) to vary in each cross-section equation over time, we can estimate whether or not lags are present. And as we shall see, the SUR results indicate that lags are indeed present. We therefore prefer the SUR specification to an alternative panel model with ad hoc controls for cyclical effects.11

V. Results and Implications Results Breusch and Pagan (1980) provide a Lagrange multiplier test of the hypothesis that the variance-covariance matrix of the error terms for our thirteen, cross-section equation system is diagonal. The test-statistic, which is distributed chi-square, indicates we can reject that hypothesis at better than the 1%-level.12 This implies that the error terms are correlated across the equations and there is an efficiency gain from estimating our thirteen cross-section versions of equation (5) as a system using SUR. Table 2 presents the empirical results from doing so. The column labeled “Year” refers to the academic year t of each cross-section equation where t = 2000-01 through 2012-13. The column designated “N” shows the number of schools included in the estimation of each year’s equation. The maximum number of possible schools is 118, but some of these schools did not report all of their data for every year of our sample period. To preserve degrees of freedom, we estimated each year’s and were statistically significant. But the slope dummy terms were collinear, which proved to be another problem with the panel specification. 11

We thank a reviewer of this paper who encouraged us to carefully articulate our rationale for the SUR specification.

The calculated chi-square test statistic is 6,097. The 1%-critical value for 78 degrees of freedom is 106.96. A useful primer on the Breush-Pagan test is provided by Kennedy (2008), p. 186-87. 12

15

equation using all of the schools with fully-reported data for the year. As shown in the N column, that number varied between 101 and 112 of the possible 118 schools. Scanning down the coefficient estimate columns for β1 and β2 in Table 2, we see that in each year the estimated coefficients have the expected signs, and with the exception of the β2 values for 2004-05 and 2005-06, are all statistically significant at better than the 5%-level. Wald tests indicate that while we cannot reject the hypothesis that the β1 coefficients are the same across time, we can reject at the 5%-level the hypothesis that the β2 coefficients are the same over time.13 This opens the possibility that pricing power may be pro-cyclical. In order to visualize any pro-cyclical behavior that may be present, Figure 1 displays the returns to sticker-price increases and a graph of those returns calculated by fitting equation (6) with the estimated values of β1 and β2 for each year at three different selectivity levels. The series “Median” maps the fitted values for the school with the full sample’s median value of ADMIT in each year. This splits the sample in two where one half of the sample includes lower selectivity schools having values of ADMIT above the full sample’s median and the other half includes higher selectivity schools having ADMIT values below the full sample’s median. In Figure 1, the series “High” refers to the median of the higher selectivity schools. The series “Low” tracks the median for the schools of lower selectivity. Note that the return to sticker-price increases across the three selectivity levels mirror each other closely, with the return always being highest at High selectivity and lowest at Low selectivity schools as we expected. We have asked if colleges’ pricing power is pro-cyclical. As shown in Figure 1, the return to sticker price changes fell following our sample’s first recession and subsequently recovered. The

The calculated chi-square test statistic for the hypothesis that β1 is the same over time is 6.30. For the hypothesis that β2 is the same over time, the calculated statistic is 21.90. The 5% chi-square critical value for 12 degrees of freedom is 13

21.03.

16

results therefore support an affirmative response to our question with respect to this first recession. But it is not clear that pricing-power during the Great Recession will turn out to be pro-cyclical. Figure 1 shows that the return to sticker-price increases did fall after the Great Recession, but it had not yet begun to recover by the end of our sample period in 2012-13. Three additional, interesting observations can be gleaned from our results. First, note that pricing power fell with a lag following the start of both the sample’s first and second recessions. Our theoretical derivation showed that these lagged changes would be caused by lagged behavioral responses on the part of students, parents, and colleges. The estimates indicate that pricing power declined in the academic year after the first recession had ended with the marginal return to a onedollar increase of the sticker-price falling from $1,209 in 2001-02 to $1,146 in 2002-03. But the behavior of pricing power during the Great Recession was different. The estimates show that it fell in the recession’s second year of 2008-09, stabilized the following year, and then began an uninterrupted decline all the way to the end or our sample period in 2012-13. A second interesting observation has to do with the relative size and length of the decrease in pricing power associated with each of the sample’s recessions. Evaluated at Median selectivity, the estimated return to a one dollar sticker-price increase fell by 12% from $1,203 to $1,070 over the postGreat Recession years of 2009-10 through 2012-13. This decline was more than twice as large and has so far lasted about one-third longer than the 5% decline from $1,209 to $1,146 associated with the earlier, less severe recession. This suggests that more severe recessions produce greater losses of pricing power that last for longer periods of time. Third, sticker-price increases raised less additional net tuition revenue at the end of the sample period than they did at its beginning. At Median selectivity, the return to a one-dollar increase in the sticker-price of $1,070 in 2012-13 was almost 10% less than the return of $1,176 in 2000-01. And it

17

was 13% less than the sample maximum of $1,211 that was returned in 2007-08. Let’s consider some possible implications for private colleges of this lost pricing power.

Implications For private colleges that have relied on a high price/high aid pricing strategy as the mechanism for raising desired net tuition revenues, the apparent collapse in pricing power since the Great Recession signals that the strategy’s usefulness has been diminished – perhaps permanently. If so, it is not clear what a new pricing strategy would look like. But a muddling through that includes the use of tuition-setting experiments designed to elicit information about students’ tuition price sensitivities may be an interim response adopted by some colleges. The Wittenberg, Middlebury, and Sewanee experiments we discussed in the introduction may serve as examples. They could also provide some lessons. While Wittenberg has been able to live with a frozen sticker-price over the last two years, Middlebury is modifying its CPI+1 policy by excluding room and board fees from the CPI+1 calculation. Middlebury spokesman Bill Burger has indicated that…”the college was fine-tuning its policy,” and that…”it was never a formal policy meant to last forever.” (Rivard 2014b). And while Sewanee did reduce its sticker-price by 10% for the 201112 academic year, it has increased since then and is now almost back to its 2009-10 level (Rivard 2013). If experiments like these prove inadequate to the task of at least stabilizing pricing power, colleges will have to find ways of adjusting to an environment of constrained net tuition revenue growth. The most preferred way of doing so may be to ease the constraint’s effects by drawing on revenues from improved investment returns on their endowments. This response may be sufficiently effective only at the wealthiest, most selective schools. A second preferred option might be to replace lost tuition revenue with more external financial support acquired through enhanced fundraising efforts. Fundraising has long been part of “business 18

as usual” at private colleges. But in the current macroeconomic environment, it is quite possible that intensification of those efforts would prove to be sufficiently successful only at the most selective colleges with the wealthiest alumni bases. Other schools will need other options. Carlson (2014) presents survey results for how 291 private colleges indicated they would respond to tuition revenue shortfalls. In that sample, 70% of the schools planned to improve their enrollment management systems; 60% indicated they would start new programs to attract students; and 53% said they would choose to direct more resources toward marketing. All three responses will raise total costs if they are not financed by internal resource reallocation. And they might deliver something close to zero-sum revenue effects. With a stagnant to declining pool of qualified, traditional 18-21 year old students in many schools’ recruitment areas, the success of one school’s enhanced recruitment and marketing efforts may occur as efforts at another school produce disappointing results. One school’s robustly enrolling, new program may appear even as a different new program somewhere else continues to scramble for students. A way around this outcome for some schools could be found by tailoring their efforts to attract more nontraditional students. But success in attracting new student-types is not guaranteed, for it is a safe bet that other schools will be trying to do this as well. This leaves us with the least frequently indicated ways in which the surveyed schools said they would adjust. All require that they rationalize their costs. Only 23% intended to eliminate low enrollment programs; 20% would use layoffs, early retirements, and furloughs; and only 17% intended to reduce campus services and operations (which presumably may include the deferral of facility maintenance costs).14

Data from The Delta Cost Project indicate that spending on facilities operation and maintenance across higher education fell by between 4% and 8% in 2010. See the Project’s perspectives paper “Climbing Walls and Climbing Tuition” available 14

19

Although it is difficult for them to do it, there is evidence that private colleges have been rationalizing their costs. Figure 2 shows the average growth rates of real per-student expenditures from 2004-05 through 2012-13 at our sample colleges. In the years of economic expansion prior to the Great Recession, these real growth rates were positive. But the effect of the Great Recession is clear. The average level of real expenditure growth turned negative in 2009-10. There was a rebound to zero growth the next year, but on average there has been no real expenditure growth since the recession. If the loss of pricing power caused by the Great Recession continued beyond the end of our sample period, it is likely that real expenditure growth remained depressed as well.

V. Conclusion We defined pricing power as a college’s ability to increase its net tuition revenue by raising its sticker-price for tuition. The greater is the positive effect of sticker-price increases on net tuition revenues, the greater is the pricing power. We then presented a theoretical derivation of the factors that determine how colleges’ sticker-price tuition increases affect their net tuition revenues. Next came our argument that these factors are likely to vary over the course of the business cycle in a way that makes colleges’ pricing power pro-cyclical. Our empirical results showed that the net revenue returned by sticker-price tuition increases at private colleges was pro-cyclical during the sample period’s first, less severe recession. And the estimated return to sticker-price increases fell after the more severe Great Recession. But it had not yet stopped falling by the end of our sample period in 2012-13. In that year each one-dollar increase in the sticker-price yielded 9% less in additional net tuition revenue than it did in 2000-01, and 13% less than it did right before the Great Recession began. Simply put, our results indicate that sticker-

at http://www.deltacostproject.org/sites/default/files/products/Delta-Cost-Climbing-Walls-Climbing-Tuitions.pdf for additional detail.

20

price increases at private colleges raise less additional net tuition revenue today than they used to. Private colleges are experiencing a pricing power drought. We conclude by identifying two research questions to follow up on as the future unfolds. First, will pricing power at private colleges eventually recover from the effects of the Great Recession, and how will the schools respond if it does not? Second, if there is a recovery in pricing power, will the schools return to business as usual, or will they change in ways that reduce their exposure to the effects of future business cycles?

21

References Alexander, B. (2014) Has higher ed peaked?, Inside Higher Ed, April 7. Archibald, R. and Feldman, D. (2011) Why does college cost so much? Oxford. New York Baumol, W. and Bowen, W. (1966) Performing arts: the economic dilemma. Twentieth Century Fund, New York. Bowen, H. (1980) The costs of higher education: How much do colleges and universities spend per-student and how much should they spend? Jossey-Bass. Washington D.C. Bhatt, R., Rork, J. and Walker, M.B. (2011) Earmarking and the business cycle: The case of state spending on higher education, Regional Science and Urban Economics, 41, 352-359. Brehm, C. and Saving, T. (1964) The demand for general assistance payments, American Economic Review, 59, 1002-1018. Breneman, D., Doti, J., and Lapovsky, L. (2001) Financing private colleges and universities: the role of tuition discounting. in: Paulsen, M. and Smart, J. (eds.) The financing of higher eduction: theory, research, policy, and practice. New York, Algora. Breneman, D. (2002). For colleges, this is not just another recession. Chronicle of Higher Education, June 14. Breusch, T. and Pagan, A. (1980) The Lagrange multiplier test and its application to model specification in econometrics, Review of Economic Studies, 47, 239-253. Brown, J. and Hoxby, C. (2014) How the financial crisis and great recession affected higher education, University of Chicago Press, Chicago IL. Carlson, S. (2014) Goals for enrollment and tuition revenue elude many colleges, Chronicle of Higher Education, October 13. Corkery, M. (2013). Colleges lose pricing power, Wall Street Journal, January 9. Delaney, J. and Doyle, W. (2011) State spending on higher education: Testing the balance wheel over time, Journal of Education Finance, 36, 343-368. Doti, J. (2002). No ivory tower. NACUBO Business Officer. October, 35-38. Doti, J. (2004). Is higher education becoming a commodity? Journal of Higher Education Policy and Management, 26, 363-369. Hillman, N.(2013) Tuition discounting for revenue management, Research in Higher Education, 53, 263-281

22

Humphrey, B. (2000) Do business cycles affect state appropriations to higher education? Southern Economic Journal, 67, 398-413. Hovey, H. (1999) State spending for higher education in the next decade: The battle to sustain current support, National Center for Public Policy and Higher Education. Griffiths, W., Hill, R., and Judge G, (1993) Learning and practicing econometrics, John Wiley and Sons. Jaschik, S. (2011) Price check at Sewanee, Inside Higher Ed, February 18. Kane, T., Orzag, P., Apostolov, E., Inman, R., and Reschovsky, A. (2005) Higher education and public universities: Role of Medicaid and the business cycle, Brookings-Wharton Papers on Public Affairs, 99-146. Kennedy, P. (2008) A Guide to Econometrics 6E, Wiley-Blackwell. Lapovsky, L. (2013) The higher education business model: Innovation and financial sustainability, TIAA-CREF Institute. Lederman, D. (2013) CFO survey reveals doubts about financial sustainability, Inside Higher Ed, July 12. Martin, R. and Hill, R. (2014) Baumol and Bowen effects at research universities, Social Science Research Network. Redd, K. (2000) Discounting to disaster: tuition discounting, college finances, and enrollments of low income undergraduates, USA Group Foundations New Agenda Series. Rivard, R. (2013) Two years after dramatic price drop, Sewanee’s tuition has climbed back up, Inside Higher Ed, October 9. Revard, R. (2014a) As prices rise, colleges are offering students steeper discounts again, Inside Higher Ed, July 2. Rivard, R. (2014b) Middlebury backs away from attempt to control rising prices, Inside Higher Ed, April 3. Summers, J. (2004) Net tuition revenue generation at private liberal arts colleges, Education Economics, 12, 219-230. Moody’s Investors Service (2013) US higher education outlook negative in 2013, January 16. Zellner, A. (1962) An efficient method of estimating seemingly unrelated regression equations and tests for aggregation bias, Journal of the American Statistical Association, 57, 348-368.

23

Table 1 – Descriptive Statistics Year 2000-01 2001-02 2002-03 2003-04 2004-05 2005-06 2006-07 2007-08 2008-09 2009-10 2010-11 2011-12 2012-13

PS

NTR

ADMIT

$17,970 (6,038) 18,559 (6,121) 18,725 (6,080) 18,998 (6,201) 19,400 (6,320) 19,540 (6,358) 20,252 (6,502) 20,398 (6,953) 20,850 (6,605) 21,856 (6,815) 22,317 (6,888) 22,892 (7,056) 22,947 (7,349)

$16,776,934 (11,780,241) 17,531,247 (12,335,602) 17,170,723 (12,297,569) 17,605,810 (12,438,065) 18,184,787 (12,764,992) 18,350,509 (12,913,971) 18,856,751 (13,344,105) 18,740,630 (13,209,392) 18,882,847 (13,455,704) 19,239,689 (13,559,112) 19,239,315 (13,613,248) 19,941,075 (13,746,939) 20,031,228 (14,007,147)

65.37 (20.47) 65.53 (20.02) 63.42 (19.76) 61.46 (19.40) 60.69 (19.39) 59.09 (19.86) 57.49 (20.50) 57.05 (20.00) 58.62 (19.71) 56.55 (19.37) 57.42 (19.65) 57.51 (20.99) 57.67 (21.79)

Each cell presents the sample mean and (standard deviation) for the column variable in the row year. PS and NTR are measured in real 2001 dollars deflated using the Higher Education Price Index.

24

Table 2 – Empirical Results Coefficient Estimates Year

β0

β1

β2

Adjusted R2

N

2000-01

-6,564,404 (-4.81) -7,106,420 (-6.09) -6,155,071 (-5.66) -6,021,629 (-6.17) -6,962,346 (-7.17) -6,945,437 (-7.33) -7,301,711 (-6.69) -7,847,832 (-5.79) -7,801,610 (-5.25) -8,757,818 (-5.21) -8.793.613 (-5.15) -7,775,953 (-4.52) -6,782,918 (-3.94)

1,310.21 (16.38) 1,302.31 (18.41) 1,295.02 (19.79) 1,255.70 (20.82) 1,251.18 (20.90) 1,246.10 (21.66) 1,279.85 (21.46) 1,305.63 (20.25) 1,277.24 (18.21) 1,290.98 (17.46) 1,267.23 (17.25) 1,243.76 (17.01) 1,240.21 (16.90)

-1.86 (-2.33) -1.33 (-1.96) -2.15 (-3.45) -1.60 (-2.94) -0.67 (-1.24) -0.59 (-1.17) -1.17 (-2.18) -1.52 (-2.21) -1.37 (-2.18) -1.42 (-2.55) -1.49 (-3.10) -2.03 (-4.68) -2.65 (-4.99)

0.49

101

0.50

106

0.49

107

0.50

106

0.53

106

0.52

106

0.52

105

0.55

106

0.49

108

0.46

110

0.46

112

0.45

112

0.46

111

2001-02 2002-03 2003-04 2004-05 2005-06 2006-07 2007-08 2008-09 2009-10 2010-11 2011-12 2012-13

Results are from SUR estimation of the thirteen cross-section regressions speficied in equation (5). N is the number of schools included in the estimation for a given year. t-statistics are in parentheses. 5% two-sided, critical t-value for df = 100 is 1.99.

25

Figure 1: Fitted Values of Equation (6)

Year 2000-01* 2001-02* 2002-03 2003-04 2004-05 2005-06 2006-07 2007-08* 2008-09* 2009-10 2010-11 2011-12 2012-13

Selectivity Median $1,175 1,209 1,146 1,149 1,208 1,209 1,209 1,211 1,191 1,203 1,174 1,117 1,070

Low $1,161 1,196 1,126 1,132 1,200 1,201 1,191 1,193 1,176 1,188 1,160 1,098 1,046

High $1,215 1,239 1,194 1,184 1,221 1,220 1,232 1,244 1,217 1,217 1,203 1,165 1,138

* indicates a recession year according to the National Bureau of Economic Research

1300 1250 1200 1150 Median

1100

High

1050

Low

1000 950

2013

2012

2011

2010

2009

2008

2007

2006

2005

2004

2003

2002

2001

900

26

Figure 2 – Real Average Annual Expenditure Growth at Sample Schools

Real Growth (%) 4.00% 2.00% 0.00% 2004-05 2005-06 2006-07 2007-08 2008-09 2009-10 2010-11 2011-12 2012-13 -2.00% -4.00% -6.00% -8.00%

Expenditures are measured as the sum of instruction, student services, academic support, and institutional support expenditures per FTE student as reported by IPEDS. Real values are calculated using the Higher Education Price Index.

27